Linguistik AktuelljLinguistics Today (LA) provides a platform for original monograph studies into synchronic and diachr...
53 downloads
1463 Views
5MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Linguistik AktuelljLinguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.
Series Editor VVernerAlbraham University of California at Berkeley University of Vienna
Rijksuniversiteit Groningen
Advisory Editorial Board Guglielmo Cinque (University of Venice) Gunther Grewendorf (T.W. Goethe-University, Frankfurt) Liliane Haegeman (University of Lille, France) Hubert Haider (University of Salzburg) Christer Platzack (University of Lund) Ian Roberts (University of Stuttgart) Ken Safir (Rutgers University, New Brunswick NJ) Hoskuldur Thrainsson (University of Iceland, Reykjavik) Lisa deMena Travis (McGill University) Sten Vikner (University of Stuttgart) C. Jan-Wouter Zwart (University of Groningen)
Volume 3I Peter Svenonius (ed.)
The Derivation of va and av
E
E
o
VOANDOV
Edited by PETER SVENONIUS
University ofTroms�
JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM
/
PHILADELPHIA
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences - Permanence of Paper for Printed Library Materials, ANSI
Z39.48-1984-
Library of Congress Cataloging-in-Publication Data The derivation of VO and OV / edited by Peter Svenonius. p.
cm. -- (Linguistik aktuell / Linguistics today, ISSN 0166-0829; v. 31)
Based on a workshop held May 1998, University ofTroms0. Includes bibliographical references and index. 1. Grammar, Comparative and general--Word order 2. Grammar, Comparative and general- Sentences. 3. Typology (Linguistics). 4. Generative grammar I. Svenonius, Peter. II. Linguistik aktuell ; Bd. 31. P295.D46
2000
415--dc21 ISBN
©
90 272 2752 7
(EUR)
/ 1 55619 915 5
(US) (alk. paper)
99-088392
2000 - John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfIlm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. . P.O.Box 75577 . 1070 AN AMSTERDAM · The Netherlands John Benjamins North America· P.O.Box 27519 . Philadelphia PA 19118-0519 . U SA
Table of Contents
Introduction
1
Peter Svenonius Word Order, Restructuring and Mirror Theory
27
Michael Brody OV is More Basic than VO
45
Hubert Haider Remnant Movement and OV Order
69
Liliane Haegeman V-movement and VP-movement in Derivations Leading to VO-order
97
Knut Tarald Taraldsen Deriving OV Order in Finnish
1 23
Anders Holmberg Parameter Change in Icelandic
153
Porbjorg Hr6arsd6ttir The Righ,t Periphery in SOV Languages : English and Dutch
181
Sjef Barbiers Nominal Structure: An Extension of the Symmetry Principle
219
Quantifier Movement in Icelandic
255
Naoki Fukui & Yuji Takano
Peter Svenonius Licensing Movement and Stranding in the West Germanic OV Languages
Roland Hinterholzl
293
TABLE OF CONTENTS
vi
Two Types of VO Languages
327
Matthew Pearson Index
365
Peter Svenonius University of Troms¢
This volume grew out of a workshop held at the University of Tromsjj in May of 1 998 (a detailed report on that workshop appears as Svenonius 1 998). The volume contains eleven papers dealing with various aspects of the differences between VO and OV languages, especially with the derivation of one or the other or both from the same or different base structures. Of the eleven, the papers by Fukui & Takano and Haegeman were not presented at the conference; papers by Alison Henry and Susan Pintzuk, which were presented there, could unfortunately not be included in the volume. I would like to take this opportunity to thank all of the participants both in the workshop and the creation of the volume, especially my colleagues Anders Holmberg and Tarald Taraldsen, and also the anonymous reviewers who assisted us in the process of critiquing the various papers for revision. 1 In this introduction I quickly and subjectively tour the history of the work on VO and OV word order which gave rise to the papers at hand. At appropriate points I discuss the current papers, placing them in their respective contexts. A majority of the papers, it will be noted, deal mainly or exclusively with Germanic languages. This bias was inherent in the composition of the workshop, and reflects in part the research programs of the faculty here in Tromsjj, in part a tendency in the recent wave of research that has occurred in the wake of Kayne's influential work, codified in his 1 994 book.
1.
Typology
The broad division of languages of the world into the subclasses VO and OV comes from typological studies. Greenberg ( 1 963), dividing a 30-language sample into classes in which the 'basic' or unmarked main clause order of
2
PETER SVENONIUS
S [ubject] , V [erb] , and O [bject] is VSO, SVO, or SOY, catalogued a litany of word order patterns that corresponded with the basic word orders: he noted, for example, that in languages in which SOY order is basic, an inflected auxiliary follows the main verb; complementarily, VSO correlates with Aux-V order (Universal number 1 6). Exceptions are now known to many of the universals Greenberg proposed, but most of them still survive as significant tendencies. Subsequent work, at least since Lehmann 1 973, has usually collapsed SVO and VSO (along with VOS) into one type, YO, contrasting with the OV type (though see e.g. Hawkins 1983 for arguments against a unified VO type). Many of Greenberg' s universals still hold up even for these more general categories VO and OV when examined against larger samples. For example, Dryer 1 992 shows that the correlation of VO with Aux-V, and of OV and V-Aux holds for about 90% of the languages for which Dryer had the relevant data (64 of 7 1 languages with inflected auxiliaries expressing tense or aspect). Vennemann 1 974 labeled the two types VX and XV, based on the fact that other elements tend to pattern with the object. For example, there is a systematic cross-linguistic tendency for PPs and DP complements to appear on the same side of the verb. The same is true for manner adverbs: they appear on the same side of a verb, in the usual case, as the direct object: the opposite side from the auxiliary verb (cf. Dryer 1 992). Furthermore, the tendencies extend into other categories. VO languages tend to have prepositions, not postpositions, and the opposite is true of OV languages (cf. Greenberg's Universals 3 and 4). Genitive possessors in the noun phrase have a clear predisposition to follow N in VO languages, and to precede N in OV languages (cf. Greenberg's Universal 2; cf. also Hawkins 1 983). The tendencies are unmistakeable, though in almost every case, exceptions can also be found. Subtler distinctions than the unmarked order of major elements may also distinguish VO from OV languages. For example, Greenberg noted some morphological tendencies : there is a correspondence between morphological case and OV order (Universal 4 1 ) . Others have proposed that syntactic mechanisms are sensitive to the VO/OV distinction. For example, Ross 1 967a argued that Gapping, a deletion transformation, operated to the right in VO languages, giving rise to the order SVO and SO, and to the left in OV languages, giving SO and SOV. He also argued that OV languages cannot have rightward movement rules. Similarly, Fukui 1 993 suggests that orderings which do not preserve head-finality in an OV language are more costly, in terms of derivational economy (Chomsky 1 992), than orderings which do. He also suggests that VO languages favor reorderings that preserve head-initiality (cf. also Steele 1 978, who notes that VSO and VOS are frequently alternatives for a single language, as are SOY and OSV).
INTRODUCTION
3
My contribution to this volume explores a particular instantiation of this general tendency; specifically, I take up the contrast in Icelandic between Object Shift, a leftward movement which preserves VO order, and Quantifier Move ment, which gives rise to OV structures. As Icelandic Quantifier Movement has been little studied, I document its characteristics. I argue that Norwegian has a form of Quantifier Movement which does preserve VO order, which bears on the question of whether languages like Dutch and Yiddish might have a version of Object Shift which perturbs VO structure.
2.
Analysis
Various accounts have been offered of the Greenbergian ordering correlations (see for example the various papers in Hawkins 1 988), ranging from strictly formal, grammar-internal accounts to historical accounts to accounts based on processing and other potentially extragrammatical factors (cf. e.g. Frazier 1 979 for an early processing-based account, Hawkins 1 994 for a recent one). Historical and functional accounts necessarily interact with formal syntactic accounts, though linguists differ according to which they take to be of primary concern. In early generative grammar, phrase structure rules were assumed to stipulate word order. Cross-categorial tendencies are potentially captured in an abstract X' theory like that of Chomsky 1 970, and with the advent of the notion of parameters it became common to assume a headedness parameter for deep structures: languages are parametrized according to whether dependents precede or follow the head (Chomsky 1 980: 179, 1 98 1 : 128ff., Stowell 1 9 8 1 : 74f.). On the other hand, it seems that in actual practice it has not been uncommon to assume that, for example, noun phrases in a given language are head-final while other categories are head-initial (e.g. Huang 1 982 for Chinese). On such a view, the headedness parameter describes why no single category varies (e.g. no 2 language has some head-initial verbs and some head-final ones), but does not account for cross-categorial tendencies (but see Jackendoff 1 977: 8 1-85, who suggests that a simplicity metric will favor grammars in which X' rules are cross-categorially general). Even given category-specific headedness parameters, there are many exceptions to the universals. One approach to these exceptions has been to take them to be historical residue, the result of a shift from consistent OV, say, to verb-before-object but retaining other OV characteristics. Another approach has been to eliminate the exceptions through fine-grained analysis. For example, if auxiliaries are taken as dependents of the verb, as in traditional grammar, then
4
PETER SVENONIUS
the fact that they appear on the opposite side of the verb from the object is confusing. Independent evidence, however, suggests that auxiliaries head their own projections (Pullum & Wilson 1 977; Gazdar, Pullum, & Sag 1 982), and that the main verb is in fact the dependent of its auxiliary. Given the analogy Aux:V::V:O, the word order facts fit neatly into place. As another, somewhat more speculative example, take the order of nouns and possessor noun phrases: N[oun]-G[enitive] order correlates with VO and GN with OV (cf. Greenberg's Universal 2). In Hawkins' ( 1 983) survey of 336 languages, about 96% of VI languages (those with basic order VSO or VOS) are NG, and about 88% of SOY languages are GN, showing a strong correlation. However, when SVO languages are examined, about 67% are NG and 33% are GN. This sort of result led Hawkins and others to doubt that VO order is a significant predictor of other features. But careful analysis gives some hope of bringing the SVO languages back into the fold. Generativists have analyzed parallels between sentences and noun phrases in a number of ways (Lees 1959; Chomsky 1970; Abney 1987). Looking at these analyses, it is clear that a possessor NP sometimes corresponds to the subject and sometimes to the object of the corresponding clause (cf. Koptjevskaja-Tamm 1993). In VI and SOY languages, whether a genitive element is objectlike or subjectlike would not make any difference with respect to NG/GN word order. But in an SVO language, it might be expected that subjectlike genitives precede the noun while objectlike ones follow. Thus, in English, the s-genitive is subjectlike (note that it is the natural choice for the subject of action nominal izations) and precedes N, while the of-genitive is objectlike, being a prepositional phrase, and follows N. 3 Such fine-grained analyses raise the prospect of stating exceptionless syntactic principles governing word order. An important subcase of the general strategy of eliminating exceptions through analysis is to argue that observed tendencies are regular at Deep Structure. German, for example, has objects both before and after verbs, depending on whether one looks at finite or non-finite verb forms and at main or embedded clauses. Bach 1 962 proposed that in German, surface VO order is derived from an OV Deep Structure. This kind of approach opened up the possibility of analyzing apparent exceptions to universals in terms of movement: the idea is that languages are consistent at Deep Structure in having head-initial or head-final characteristics, but transformations may give rise to surface inconsistencies. Ross 1 967 a noted that German was a counterexample to some generalizations about OV languages; he proposed that German is in fact a VO language, with OV order being derived. Japanese and Siouan, according to Ross, are 'Deep' OV languages, while German, Hindi, and Turkish are not.
INTRODUCTION
5
This type of analysis is very common in generative work. However, moving the 'true' nature of a language back to Deep Structure runs the risk of losing Greenbergian correlations, which are stated over surface word order. A headed ness parameter affecting Deep Structures can describe the Greenbergian correla tions only if things don't move around very much, or on the assumption that something like Structure Preservation (Emonds 1 976) prevents non-canonical orders from arising, in the usual case. Several specific proposals have been made that ensure that head-initial or head-final structures will in the usual case be preserved (e.g. those of Ross 1 967a; Steele 1 978; Fukui 1993 mentioned above). Another tack has been to parametrize Surface Structure conditions. Kayne 1983 proposed that government (relevant at various levels, including Surface Structure) is directional, and that the direction is parametrized by language. Koopman 1 984: 1 06ff. and Travis 1 984: 32ff. propose the same for Case assign ment (cf. also Li 1 990). They also argue that theta assignment is directional, and that the parameter for theta role directionality is independent of that of Case assignment. Since theta roles are assumed to be assigned at D-Structure, this subsumes some of the Deep Structure directionality parameter (cf. Koopman op. cit. p. 1 20). Though Travis explicitly argues that a language can have differ ent settings for headedness and theta-assignment, the natural conclusion of this line of thinking is nevertheless to do away with the headedness parameter for phrase structure altogether, letting other, independently motivated mechanisms take over. In this vein, Haider 1 992 proposes a general licensing principle, where licensing occurs to the right in VO languages and to the left in OV languages. If all sentence elements are subject to licensing, including for example manner adverbs, then the tendency for all subordinate elements (rather than just those which are Case-marked or theta-marked) to precede or follow their superordinate heads is expected. On the other hand, it is less clear exactly what is subject to the licensing principle. For example, the subject precedes the verb in many VO languages. Possibly, this could be due to some exceptional property of nomina tive case. If nominative case is not subject to more general licensing conditions covering other types of sentence elements, this might provide an explanation for why no ergative languages are SVO (Anderson 1 976; Trask 1 979; Dixon 1 994; Nash 1 996, 1998), since then ergative languages would have to license both arguments in the same direction, unlike accusative languages. 4 At yet another remove from Deep Structure, it is sometimes argued that ordering is strictly a P[honetic] F[orm] phenomenon (e.g. Chomsky 1 994; Uriagereka 1 999). Takano 1 996 and Fukui & Takano 1 998 propose a specific process of linearization of syntactic terminals occurring at PF. They identify two mechanisms operating on syntactic trees to derive pronounceable phonetic
6
PETER SVENONIUS
structures, Demerge and Concatenate. Demerge breaks down syntactic trees, operating top-down and removing maximal projections, defined as non-projecting nodes, and passing them to Concatenate, which linearizes them. A specifier is always a maximal projection, while its sister is not; this leads to specifiers consistently being linearized before their sisters. A complement XP will also be linearized before a head Y, since the head projects Y' and is therefore not maximal. Fukui & Takano assimilate head movement to phrasal movement in taking head movement to attach to phrasal projections; from the derived position, a moved head does not project, while its sister does. Therefore, a moved head will be linearized before other material in the phrase it has attached to. The difference between VO and OV structures, on this view, is due to the fact that in VO structures, the verb has moved to a higher position. Fukui & Takano provide morpho syntactic evidence that this is correct, drawing for example on the connection between OV languages and case noted by Greenberg, as well as various other features. In this volume, Fukui and Takano extend their analysis to noun phrases, arguing that English has N to D raising, resulting in N-initial structures, while Japanese does not. They connect the lack of N to D raising in Japanese to the presence of a classifier system there. They also present a novel analysis of relative clauses and connect the placement of the relative clause (prenominal in Japanese) to the presence or absence of a relative complementizer (absent in Japanese).
3.
Asymmetry
Though most of the typological traits mentioned so far in association with the VO/OV distinction have involved symmetry, neither Haider's theory nor Fukui & Takano's leads to symmetrical structures for VO and OV languages. This is no accident; asymmetrical properties of languages have been apparent from the start. For example, in all of Greenberg's original basic categories, SVO, SOY, and VSO, the subject precedes the object (Universal 1). B ach 1 97 1 noted that wh-movement, when it occurs, is invariably to the left, and that SOY languages don't usually have wh-movement at all (cf. Greenberg's Universal 1 2). It has also been noted that clausal dependents have a tendency to appear to the right of the elements they are dependent on. For example, many OV languages have clausal complements to the right of V (Dryer 1980; Givon 1 984). Such observations often invite explanations in terms of processing (cf. Hawkins 1 994 and references there, e.g. on pp. 5-9). Frequently, processing accounts can fruitfully be seen as interacting with formal syntactic accounts. For
INTRODUCTION
7
example, NRel (relative clause after noun) is vastly more likely than ReIN in va languages, but OV languages are mixed (cf. Hawkins 1983). Two tendencies can thus be seen as interacting: the tendency for clausal elements to be to the right, plausibly due to processing factors, and the headedness parameter, placing depen dents on the same side of their heads, arguably not a function of processing. In this light, consider the fact that the subject almost always precedes the verb in OV languages, but va languages are more mixed; the two tendencies, then, might be for the subject to appear on the same side of the verb as the other dependent elements versus the tendency for the subject to be to the left. In this case as well, the force counteracting the headedness parameter may be amenable to a functional interpretation: subjects may appear to the left because they tend to be topics, and the tendency of topics to be initial may stem from information packaging strategies (Halliday 1 967; Li & Thompson 1 976; Vallduvf 1990). On the other hand, grammar has a tendency to steer the individual language; for example, in many languages, subjects appear in a fixed position, whether they are topics or not, and in others, there is a distinct position for subjects which are not topics (cf. Svenonius in press for some examples from Germanic languages). Thus, the functional explanation for the relative order of the subject and the object interacts with a formalist explanation for the properties of positions of case assignment and other types of formal licensing. In the papers in this volume, the formal constraints are of central interest. Thus, given that Dutch and English are identical in the relative order of indirect object and direct object, or of direct object and PP, as seen in (1) (from Haider 1992), it can be assumed that the licensing positions for the various elements are identical (presumably universally), as in the analyses of Haider 1 992 and Zwart 1 993, without investi gating why exactly the order of the licensing positions is the way it is. Attention focuses then on the position of the verb, which precedes the VP-internal elements in English and follows them in Dutch, in an embedded clause. (1)
a.
b.
. .. omdat ze iedereen een pakje naar hem thuis zullen because they everybody a package to his home should opsturen (Dutch) send . . . because they would send everybody a package to his home address
Haider's proposal, referred to above, is designed to account for the parallel word order. On Haider's account, OV order is basic: VP-internal elements are base generated in the projection of V. The verb starts low and stays there in a strict OV language, but in a va language, it climbs by head movement to a higher
8
PETER SVENONIUS
position. Thus, a clause like the English one in ( 1b) is derived by head move ment, and the VP-intemal arguments are in situ, much as in Larson 1 988. (2)
[AuxP would [yp send1 [yp everybody t1 [yp a package [v' t1 to his house]]]]]
But the structure for the corresponding German clause is not identical; the B asic Branching Conjecture that Haider proposes coupled with the licensing parameter gives the same hierarchical ranking for the sentence elements, but with the verbal heads in substantially different locations, as sketched in (3) (see Haider's paper for details). (3)
[yp ein Paket [v' an seine Privatadresse [ AuxP [yp jedem everyone a package to his private. addres s
schenken]]] werden] send
would
In his contribution to this volume, Haider extends and refines his proposal, showing how it makes better predictions in several arenas than competing analyses (for which see below).
4.
The Universal Base Hypothesis
Bach 1 968 proposed the Universal Base Hypothesis (UBH) , that all languages have identical Deep Structures, which are permuted by language-specific transformations. This completely shifts the weight of cross-linguistic correlations from the Deep Structure to other parts of the grammar. In fact, Peters & Ritchie 1 969 demonstrated that the UBH is vacuous in the absence of meaningful constraints on transformations. However, once transformations are better understood, the UBH invites a different kind of explanation for asymmetries of the type noted in the previous section. For example, consider the fact that in languages in which the auxiliary precedes the main verb, or Aux-V languages (generally VO languages), material can generally appear between the auxiliary and the main verb, whereas in V-Aux languages, this is not the case. Furthermore, in languages in which both orders are possible, Aux-V order tends to allow elements to intervene, while V-Aux order does not. This can be straightforwardly illustrated using German. 5 (4)
a.
Ich habe (schwer) gearbeitet. I have hard worked 'I have worked (hard). '
(German)
9
INTRODUCTION
b.
. . . daj3 ich gearbeitet (*schwer) habe that I worked hard have ' . . . that I have worked'
The obligatory adjacency of V-Aux order, as opposed to Aux-V order, can be seen as a function of the one being derived from the other (various specific implementations of this are discussed below). If the two orders were simply a matter of the auxiliary taking its complement to the left or the right, this systematic difference would be unexpected. The UBH has been revived in a new form by Kayne 1 994, who proposes, among other things, that all branching is binary, all complements are to the right, all adjuncts or specifiers are to the left, and all movement is to the left. On this view, whatever differences there are between Japanese and English must be a function of the derivation of their respective surface structures. Zwart 1 993, 1 997, and Koster 1 994 demonstrated how OV order in Dutch could be derived from a basic VO order by movement of VP-intemal material to the left, into functional specifier positions (making good the promise of Ross 1 967a, which took languages like Dutch not to be 'Deep' OV languages). A Dutch embedded clause is analyzed much as in (5) (cf. Zwart 1993, Ch. IV). (5)
. . . dat [AgrSP Jan 1 [AgrOP het boek2 [PredP op de tajel3 [yp wi! tl that
John
leggen t2 t3 ]]]]
the book
on the table
will
put
In English, on this view, the movements to licensing positions such as SpecAgrOP
occur covertly. However, such accounts raise various questions. For example, if such movements as those in (5) are triggered by strong features, why are the features generally strong (in consistent OV languages) or generally weak (in consistent VO languages)? In mixed cases, what determines what stays behind (e.g. CP) and what moves (e.g. DP)? (For one mixed case, Icelandic, see my contribution to this volume.) If the word order in OV languages is the result of movement, what ensures that the derived order of constituents is identical to the order of those same constituents in a VO language, in cases like those discussed in §3? The same-order cases are handled straightforwardly by Haider's account, as they are actually in essentially the same positions. In addition, the fact that everything shows up on the same side of the verb is handled by Haider's account, based on a general licensing principle. Two different types of solution have been proposed for same-order effects in VO and OV languages. One solution is to assume that the individual elements
10
PETER SVENONIUS
do not move separately, but together. For example, if the Dutch clause begins somewhat as in (6a) (after subject movement to a high functional projection), then if the main verb moves to a low functional head as in (6b), followed by movement of the VP remnant to a specifier position above the auxiliary as in (6c), the correct surface order is established. (6)
a.
dat Jan [AuxP wil hp [vp leggen het boek op de tafel]]]
b.
dat Jan [AuxP wi[ [IF leggen [vp tv het boek op de tafel]]]
c.
that Jan
will
that Jan
will
put
put
the book on the table
the book on the table
dat Jan [AuxP [vp tv het boek op de tafel] wi[ [IF leggen tvp]]
the book on the table will that Jan ' . . . that John will put the book on the table'
put
This type of movement is proposed by Hinterholzl (cf. his 1 997a: 1 98/1 997b: 1 4), though only in conjunction with various other movements. It is also advocated in the contributions to this volume by Pearson and Haegeman. As Haegeman points out, since such an analysis requires the verb to move to a higher functional head, it is consistent with the assumption that verbs in languages with agreement inflection (including German and Dutch) move overtly to a relatively high functional position, bringing the OV languages more into line with facts about VO languages. Haegeman also shows that this type of analysis can preserve the generalizations of Holmberg 1 986 and Diesing 1 992 which are otherwise compromised by the Zwartian analysis. Haegeman furthermore demonstrates in her contribution to this volume that well-motivated assumptions about negation support the claim that the verb in West Flemish occupies a relatively high position overtly (even though West Flemish exhibits OV structures like Dutch). For example, she argues that the two negative markers in West Flemish, en and nie, correspond to the two negative markers in Piedmontese, pa and nen respectively (discussed in Zanuttini 1 997). Piedmontese pa precedes the adverbs pi 'no longer' and sempre 'always' , as shown in (7a) (from Zanuttini 1 997: 70). However, controlling for V2, West Flemish en follows the corresponding adverbs nie meer and atent, as shown in (7b) (cf. Haegeman's (17), from this volume). (7)
a.
Da
'ntlura, a l'ha
since then
invito
pa pi
sempre acetla
i
nost
he he-has not more always accepted the our (Piedmontese)
invitations 'Since then, he hasn't any longer always accepted our invitations.'
INTRODUCTION
b.
11
. .. da Valere nie meer atent no Gent en-goat (West Flemish) that Valere no more always to Gent EN-goes , . . . that Valere doesn't any longer always go to Gent'
In contrast, adverbial negation (West Flemish nie, Piedmontese nen) follows both
adverbs in both languages (and precedes West Flemish en). Haegeman shows that the clause structures for the two languages can be made harmonious if an extended projection of the VP fronts across the negative head en in West Flemish. The moved constituent carries with it the various adverbial elements, preserving their relative order. The second type of UBH approach to identical sequences in VO and OV structures is to assume that the individual elements in a given clause do occupy the same positions but that the verb moves, roughly as in Haider's approach but with a universal base. B arbiers, in this volume, argues for identical base positions in English and Dutch for a number of different sentence elements, arguing that differences in word order are due primarily to the obligatoriness of 'verb intraposition' (see below) in English coupled with the existence of a short verb movement in English. Verb intraposition, on Barbiers' analysis, is optional in Dutch, and Dutch has no short verb movement. An important variation on this type of approach is to assume that the surface positions of sentence elements are derived: the individual elements in a given clause move to distinct specifier positions in OV languages, as on the Kayne and Zwart proposals, but the same also occurs in VO languages, followed by movement of the verb across those elements roughly as in the Haider and B arbiers proposals. In other words, languages like English would have (5) or (6c) above as an intermediate structure, followed by raising of a constituent contain ing the auxiliary and the main verb to a position below the subject. Such proposals have an ancestor in Larson's 1 988 analysis of Heavy NP Shift phenomena as VP-movement to the left of a heavy object (cf. also Kayne 1994). A more direct influence is Hinterh61zl's 1 997a analysis of Verb Raising in German as remnant VP movement, as sketched in (8): first, arguments move out of VP into higher positions, then the functional projection (FP in (8)) containing the verb moves to a specifier to the left of the higher verb (the specifier of PredP in (8), cf. Hinterh6lzl 1 997a). (8)
a. b.
er sie1 nicht [PredP wagte [FP zu kiissen t1]]
he her not
dared
to kiss
er sie1 nicht [PredP [FP ZU kiissen trJ wagte tFPJ
to kiss dared he her not ' . . . (because) he didn't dare to kiss her'
12
PETER SVENONIUS
c. d.
er sie 1 [PredP wagte [pp nicht zu kUssen td]
he her
dared
not to kiss
er sie 1 [PredP [pp nicht zu kUssen td wagte tpp]
he her not to kiss dared , . . . (because) he dared to not kiss her'
On a VP-fronting analysis of VO languages, the remnant moving to the left is larger (it contains the auxiliaries) and moves further (it moves to the left of the licensing position for the object). Thus, Kayne 1 998 argues that in English, there are in fact overt movements in some cases of VP-internal material to the left of the VP, followed by VP-remnant fronting past that material again, resulting in what superficially looks like the original word order. This brings VO languages much closer to Hinterh61zl's vision for the OV ones. For instance, to account for wide scope of negation in a sentence like I will force you to marry noone, Kayne postulates the derivation in (9), where 'W' and 'Neg' are heads in the extended projection of every verb, but shown only for the superordinate clause. For the narrow scope reading, the derivation in (10) is used, and here the functional heads are only shown for the embedded clause. (9)
a. b. c.
(10)
a. b. c.
I will [wp W [NegP Neg [yp force you to marry noone]]] I will [wp W [NegP noone 1 Neg [yp force you to marry td]] I will [wp [yp force you to marry td W [NegP noone 1 Neg typ]]]
I will force you to [wp W [NegP Neg [yp marry noone]]] I will force you to [wp W [NegP noone 1 Neg [yp marry td]] I will force you to [wp [yp marry td W [NegP noone 1 Neg typ]]]
Note that the surface position of noone in each case matches the scope of negation. Kayne notes that Icelandic and archaic Norwegian manifest the order in the (b) examples; this is discussed more fully in my article in this volume. By taking the English VO order to be derived in this way, Kayne proposes to eliminate covert movement from the system of grammar. Note that the claim that objects in VO languages like English leave their base positions overtly has been around for a while; for example, Johnson 1 99 1 , B orer 1 994, and Runner 1995 all take this position. In each of those proposals, however, the verb moves to the left by head movement, as on Haider's account, rather than by VP-remnant movement, as in (9)-(1 0). In this volume, Taraldsen and Hr6arsd6ttir argue for a much expanded role for VP movement in VO languages. Taraldsen reasons that if VO order is derived by VP movement, then in some cases, it should be possible to observe material other than the verb being pied-piped along with the VP. Drawing on
INTRODUCTION
13
evidence from a range of Germanic languages, he argues that prepositional particles provide exactly this kind of evidence for VP movement. For example, Taraldsen proposes that the particle-noun phrase order illustrated in ( l l a) is derived in stages, one stage being similar to theav structure in ( l lb). (11)
a. b.
at han slapp inn katten.
that he let in the. cat . .. dat hij de kat binnenliet. that he the cat in.let , . . .that he let in the cat'
(Norwegian) (Dutch)
Taraldsen argues that the particle originates to the left of the verb, with the object first moving across both verb and particle, leading to the Dutch order. In valanguages, a series of VP movements results in the Norwegian order. Thus, all languages are alike in requiring various VP-internal elements to move overtly to licensing positions, and va languages in addition have the property that there is successive leftward movement of verbal projections to specifier positions. Taraldsen suggests that object shift, too, may profitably be seen as a result of leftward VP movement. Hr6arsd6ttir, drawing on her previous work (see references in her paper), examines historical data from Icelandic, demonstrating that various types ofav structures disappeared simultaneously from Icelandic. In particular, before the nineteenth century, objects frequently appeared to the left of non-finite verbs (finite verbs appeared in second position, as in modern Icelandic). With non finite auxiliaries preceding the main verb, various options are observed: the object might precede both verbs, follow both verbs, or appear between the two verbs. In addition, the auxiliary could precede or follow the main verb. The following word orders are then attested. ( 1 2)
a. Aux Va b. Auxa v c. aAux V d. V Auxa e. a v Aux
(The final logical possibility, VaAux, is unattested; cf. the discussion of V-Aux adjacency above.) Hr6arsd6ttir shows that av orders (including VAux orders) disappeared simultaneously from Icelandic in the nineteenth century, with the result that only the order in ( 12a) is grammatical (with the exception of the examples with quantified objects discussed in my contribution to this volume). Pintzuk 1 997 has also documented the same observation for the history of
14
PETER SVENONIUS
English, showing that various classes of elements ceased to appear to the left of their selecting verbs at the same time. This is puzzling on an approach to OV structures like that in (5) above, because if distinct features move the object to the different positions above and below the auxiliary and cause the reversal, then there is no reason to expect them all to have become weak simultaneously. Hr6arsd6ttir demonstrates how an analysis of VO order based on VP movement yields the right results: if what happened in the history of Icelandic is that VP movement of the type illustrated in the derivations in (9)-(10) went from being optional to being obligatory, then only one change need be postulated, instead of a series of independent but simultaneous ones. Deriving VO order through phrasal movement rather than head movement is one of the main themes running through this volume. Holmberg's and Pearson's papers (discussed below) also take up this issue, as does Brody's, in a way, and HinterhOlzl's, briefly. Aside from this difference, the main distinction between the type of account presented by Taraldsen and Hr6arsd6ttir and Haider's is whether the licensing positions are derived positions or not. Haider argues in this volume that they are not, noting in particular that there are no island effects as might be expected from moved elements; in his paper, Hinter hOlzl argues that they are. HinterhOlzl examines several pieces of evidence from West Germanic languages, including the position of the infinitival marker, word order in complex verbal clusters, and facts about postverbal CPs and PPs. He concludes that there is substantial evidence for an account in which licensing positions are reached through overt movement in OV languages. He provides an analysis of so-called extraposition, such as that exhibited in ( 1 3). ( 1 3)
Hans hat Peters Buch gelesen fiber Chomsky.
Hans has Peters book read about Chomsky 'Hans has read Peter's book about Chomsky'
(German)
Hinterh6lz1 notes that a rightward movement analysis is incompatible with Antisymmetric principles. What is more, the construction does not bear the usual characteristics of extraction structures: note that the noun phrase is specified, a situation which ordinarily prevents extraction (Ross 1 967b). Hinterh6lz1 presents an analysis in terms of copying with partial deletion which is fully consistent with the Kaynean treatment of OV languages.
15
INTRODUCTION
5.
Mirrors
Several asymmetries were pointed out in the previous section. A particular type of asymmetry not discussed yet is mirror effects, or what is sometimes called 'centripetal' organization (cf. Dik 1 989: 342). For example, Greenberg's Universal 20 states that if any or all of the demonstrative, numeral, and descriptive adjective precede N, then they do so in that order; whereas if they follow N, they appear in the same or the opposite order. As Hawkins 1 983 observes, the mirror-image order is by far the most common, for elements following N. 6 The example in ( 14b) is from Koromfe (Niger-Congo, Burkina Faso; after Rennison 1997: 85-86). (14)
a. b.
those three black cats
lug<mi bfniA tiiii
cats
(Koromfe)
hElJ
black three those
Hetzron 1 978 notes that adjective orders also show a mirror effect; in AN languages like English (those in which the attribute adjective precedes the noun it modifies), adjectives expressing evaluation precede those expressing size, which in turn precede color adjectives, while in NA languages the order is exactly the reverse. (15)
a. b.
a beautiful big red ball
(Indonesian)
bola merah besar jang tjantik
ball red
big
REL beautiful
Koster 1 974 noted a mirror effect for prepositional phrases in Dutch. PPs following the verb have the opposite relative order compared with their sequence preceding the verb (these examples are taken from Barbiers 1 995: 103). ( 1 6)
a.
Hij is door
'n stuuifout
met een knal op het hek
he is through a steering. error with a
gestrand. b.
stranded
Hij is gestrand op het hek
met een knal door
he is stranded on the fence with a
stuuifout.
bang on the fence (Dutch)
een
bang through a
steering.error 'He got stranded on the fence with a bang because of a steering error'
The order in ( 1 6b) is basically the same as in English, while the preverbal order in ( 16a) is the reverse.
16
PETER SVENONIUS
Mirror-image structures are familiar from morphology; Baker's 1985 Mirror Principle observed that the relative closeness of morphemes to a stem reflects the order in which the corresponding syntactic transformations take place; for example, if a benefactive argument is realized as a direct object when an applicative morpheme is attached to the verb, and if passive promotes a direct object to subject position, then in order for a benefactive argument to appear as a subject under passive, the applicative morpheme must be closer to the stem than the passive morpheme. This can be seen in the Chichewa example in (17) (from Baker 1988: 307; SP is subject agreement, ASP is aspect). (17)
Mbewa zi-na-sem-er-edw-a
mitondo.
mice SP-PAST-carve-APPL-PASS-ASP mortars 'The mice were carved mortars' (i.e. somebody carved mortars for the mice)
(Chichewa)
Baker derives the order of morphemes through strictly local head movement (following Travis 1984). In the case of the applicative passive, the applicative suffix is argued to originate as a preposition in a PP complement to the verb, and the passive morpheme is taken to occupy a higher head. The derivation of the applicative passive in (17) proceeds as sketched in ( 1 8), ignoring the prefixes and the aspect suffix. ( 1 8)
a. b. c. d.
PASS [vp carve [pp APPL mice] mortars] PASS [vp carve-APPL 1 [pp tl mice] mortars] [carve-APPLlh-PASS [vp t2 [pp tl mice] mortars] mice3 [carve-APPLlh-PASS [v p t2 [pp tl t3] mortars] In (1 8a), the verb takes two complements, a PP and an NP. In (1 8b), the preposition has incorporated into the verb, appearing as a suffix (cf. Baker op. cit. p. 23 1). In (1 8c), the verb has moved to a higher head, attaching to the left of the passive suffix (op. cit. p. 309). (1 8d) simply shows the NP movement of the derived subject. In subsequent work, however, it has often been assumed that head-movement necessarily leads to adjunction to the left, with the result that suffixes on a stem must reflect successively higher functional heads above that stem (cf. Kayne 1994); predictions about prefixes are less clear; see Julien 2000 for some discussion. Furthermore, it has been common to take these mirroring effects to cover non-grammatical function changing affixes as well; thus, Belletti 1990 could argue that AgrP dominates TenseP, contra Pollock 1989, on the grounds that agreement morphemes occur further from the stem than tense morphemes, cross-linguistical ly. The order of functional elements appears to support this view. For example,
INTRODUCTION
17
the order of suffixes in the Northern Sfuni verb form in (19a), where the potential mood suffix is closer to the stem than the subject agreement suffix, is taken to mirror the order of those heads in the syntactic tree. This view is strongly supported by structures like that in (19b), where the negative and perfective auxiliary verbs take the same suffixes, but where the first auxiliary takes the agreement suffix and the second the mood suffix (cf. Mitchell 1 99 1 ; Holmberg et al. 1993 on Finnish). (19)
a. b.
Veahkeh-earra-n.
help-POT-1 SG 'I could help' J-n
NEG-lsG
(Northern Sami)
l-earra veahkeh-an.
be-pOT help-PERFECT 'I couldn't have helped'
Cinque 1999 develops a strong argument for this extended mirror principle, identifying a number of categories of adverbial meaning and connecting different morphemes with those categories. He demonstrates that cross-linguistically, the ordering of the adverbs mirrors the ordering of the morphemes. The adverbs are taken to occupy specifier positions in phrases projected by functional heads, and the morphemes are taken to be associated with those selfsame heads; morphology then mirrors the syntactic tree in a striking way. Brody 1997 takes these observations as fundamental and proposes the Mirror Theory, by which morphological structure and complement structure are isomorphic: a syntactic complement is realized with its selecting head as a suffix; this yields a result similar to that of a theory in which the head of a complement always raises to its selecting head (cf. Svenonius 1 994), but only if head to head movement happens at the surface: Brody points out, in his contribu tion to this volume, that in his theory there is no room for covert incorporation. He examines the case of Romance restructuring, and demonstrates the advantages of his analysis over one involving covert head movement. He also examines the case of Hungarian verbal clusters, which exhibit an interesting property: certain sequences of Hungarian verbs can appear in either order, e.g. 'begin-swim' or 'swim-begin' , with the same meaning ('begin to swim'). He argues that this reflects an option, either the dependent (main) verb is identified as the specifier or the complement of the superordinate (auxiliary) verb. If the dependent verb is a complement, then by Mirror Theory, the superordinate verb must immediately follow it, as a suffix; if the dependent verb is a specifier, then it is a distinct phrase, and whether it precedes or follows the superordinate verb depends on where that verb is realized (in situ, following the
18
PETER SVENONIUS
subordinate verb, or in a higher functional posltIon preceding it). This is strikingly like the cross-linguistic observation noted above that while Aux-V order does not necessarily require adjacency, V-Aux order does. If V-Aux order is generally derived by head movement or a closely related mechanism, as in Brody's theory, then this cross-linguistic fact may be captured. However, there are other ways in which mirror images may be derived, and head-movement is often reserved for cases in which the elements so combined are clearly wordlike. Kayne (1994: 52ff.; cf, also Julian 2000, Ch. 5) has suggested a derivation for mirror-image ordering based on a phrasal analogue of head movement: if an element Z moves from the right to the left of an element Y, then the order of Z and Y is reversed. If a node dominating both Y and Z then moves leftward across a node X, the order Z-Y-X results, the opposite of the original order. This is illustrated in (20) for the (gloss of the) Indonesian noun phrase in (I5b) above. Brody (1997) refers to structures produced by movement of a node dominating the target of previous movement as 'roll-up' structures. (20)
a. b. c. d.
[xp beautiful [yp big [zp red [NP ball]]]] [xp beautiful [yp big [AP [NP ball] red tNP]]] [xp beautiful [yp [zp [NP ball] red tNPJ big tzp]] [xp [yp [zp [NP ball] red tNP] big tzp] beautiful typ]
If beautiful is the head of XP, big the head of YP, and red the head of ZP, then no functional projections need be assumed. This is consistent with the view of Abney 1987 that adjectives are heads in the extended projection of N. However, if the elements in the reverse order are phrasal, some more complicated structure is necessary. This is how Barbiers 1995 analyzes postverbal PPs in Dutch. According to Barbiers, the various PPs are adjoined to VP, as shown in (21 a) (compare (16», where the different segments of VP are marked with different lower-case letters. The lowest segment of VP moves into the specifier position of the lowest PP, in order to enter into a subject-predicate relation with it, as shown in (21b) (this is the 'intraposition' alluded to above). The next two segments (marked 'b' and 'c' ) move into the next two PP specifiers in (21c-d), resulting in reverse order (cf. Barbiers 1995: 1 15).
(21)
a. b.
[VPd [pp by mistake] [VPc [pp with a bang] [VPb [pp on the fence] [VPa stuck]]]] [VPd [pp by mistake] [VPc [pp with a bang] [VPb [pp [VPa stuck] on the fence] tv pa]]]
INTRODUCTION
c. d.
19
[YPd [pp by mistake] [YPe [pp [YPb [pp [YPa stuck] on the fence] typaJ with a bang] typb]] [YPd [pp [VPe [pp [YPb [pp [YPa stuck] on the fence] tYPaJ with a bang] tYPbJ by mistake] type]
These movements, according to Barbiers, are necessary in order to establish the Spec-Head relation necessary for a PP to predicate over the relevant VP, and must occur covertly in languages in which they do not occur overtly. Holmberg proposes a kind of roll-up structure for Finnish verbal clusters in his contribution to this volume. He notes that in Finnish, V-Aux order requires strict adjacency. He proposes that V-Aux order is derived by movement of VP into the specifier of a functional head which selects the auxiliary. This leads to derivations like that in (22). (22)
a. b. c. d. e.
hp Tense [AuxP have [AspP Aspect [yp write novel]]]] [TP Tense [AuxP have [AspP Aspect [yp novell write t I ]]]] [TP Tense [AuxP have [AspP [yp novel 1 write ttl Aspect typ]]] [TP Tense [AuxP [AspP [yp novel 1 write ttl Aspect typJ have tAspP]] [TP [AuxP [AspP [yp novel 1 write ttl Aspect typ] have tAspP] Tense tAuxP]
In (22a), the base order of heads is shown. In (22b), the object has moved to the left of the verb (here, to SpecVP). In (22c), the VP has moved into the specifier of AspectP, in order for the verb to check its aspectual features against the functional head of that phrase. The same occurs in the next two steps. It can be seen that as long as adjunctions to X' are not allowed, and adjunctions to the right of XP are not allowed, no material will intervene between the verb and its auxiliary in the resulting head-final structure. In the final string, novel write Aspect have Tense, the functional elements Aspect and Tense are suffixal, and are pronounced as a word with the immediately preceding stems, hence some thing like novel writ-ten ha-s in pseudo-English, 'has written a novel' . This affixation is subject to adjacency, according to Holmberg, so that for example the object could not have remained in situ in the first step, leading to an AspP of the form *write novel -Aspect. Pearson, in his paper in this volume, applies the roll-up technology to several additional kinds of mirror-image effects. He notes that although same order patterns of the type discussed above are common, they are not universal. There are languages, such as Malagasy, Palauan, and Zapotec, in which the unmarked order for indirect object and direct object is the reverse of that in
20
PETER SVENONIUS
English. Similarly, some of these languages have the reverse relative order for certain series of adverbs, and show scrambling effects to the right, rather than to the left. But Pearson notes that the mirror effects do not warrant a fully symmetric analysis; each of these languages also shows some same-order effects, for example they are consistently VO, and never OV. Pearson develops an analysis that is consistent with Kayne's LeA, and proposes a novel kind of rationale for why roll-up structures occur; in short, they occur in languages in which verb movement is systematically unavailable. In this way, the proposal bears a relation to Holmberg's, in which V movement and VP movement are alternatives for checking V features; but unlike Holmberg's proposal, in which the two possibili ties are manifested within a single language, Pearson casts the option as a point of cross-linguistic variation.
6.
Conclusion
The analyses in the different contributions to this volume are not completely compatible with each other, though there are several harmonious strands running through most of them. Papers like those of Hinterh61zl, Haegeman, Hr6arsd6ttir, and Taraldsen largely share the same theoretical assumptions and come to highly compatible conclusions. Others of the papers are parts of broader programs initiated by the author or authors; those of Fukui & Takano and Brody stand out in particular in this respect, distinguishing themselves more sharply in their basic theoretical assumptions. Another example is the paper by Haider, which defies what is to some extent the majority view here, deriving from Kayne and earlier work by Hinterh61zl, which envisions leftward movement on a large scale to derive OV order. Hinterh61zl's paper is in part a response to some of the challenges posed by Haider. Another interesting tension is that between the two papers focusing on Icelandic: Hr6arsd6ttir's paper holds that there is general VP movement in modern Icelandic, of the sort proposed by Kayne 1998; but part of Kayne's argument is based on the OV patterns discussed in my paper, which are an intermediate stage in the derivation of a VO language like English, for Kayne. In order for my and Hr6arsd6ttir's analyses to be reconciled, either the landing site of quantified noun phrases in Icelandic must be higher than their landing site in English (i.e. higher than the landing site of the VP), or else the landing site of VP must be lower. Yet another point of tension, noted above, is that between Pearson, who would have head movement versus phrasal movement be a major parameter
INTRODUCTION
21
distinguishing language types, and Holmberg, who makes use of a very similar distinction within Finnish, arguing that both head and phrase movement are options. For these two to be reconciled, either something would have to be said about what makes Finnish unusual in this respect, or else, if Finnish turns out not to be unusual, about what is special about this parameter that allows both settings to be manifested as options in a single language. A third perspective on this issue is held by B arbiers, who uses a combination of head and phrasal movements to derive surface word order in Dutch. The harmonies among the various papers represent a certain kind of forward momentum, while the tensions hint at different possible directions for that movement.
Notes 1.
Thanks to Anders Holmberg, Hubert Haider, and Matthew Pearson for feedback on this introduction.
2.
On the other hand, it seems that some languages have both prepositions and postpositions; e.g. English has the postposition ago, and Finnish has a variety of postpositions and prepositions cf. Holmberg this volume.
3.
Preliminary investigations provide some cross-linguistic support. For example, possessive suffixes in Finnish bear some resemblance to subject agreement suffixes, suggesting that genitives are subjectlike; and Finnish is SVO and GN; possessive pronouns in Moroccan Arabic are identical to object pronouns, meaning that genitives are objectlike, and Moroccan Arabic is SVO and NG. Dryer 1 999 discusses the different motivations for GN versus NG in SVO languages.
4.
Of course, this explanation presupposes that absolutive case is different from nominative.
5.
The generalization, as stated here, holds for very many languages, though perhaps not always for the same reasons. For example, it holds of languages with second-position auxiliaries such as various Slavic, Australian, and Uto-Aztecan languages, but there the obligatory adjacency of V-Aux order may be of a different nature from the obligatory adjacency of V-Aux in for example Japanese or Persian. (i) Aux-V order may be interrupted (ii) V-Aux order may not be interrupted The generalization can be read on two levels: as a claim about unmarked orders and as a claim about any order. As a claim about unmarked orders, counterexamples to (i) are Spanish and Rumanian, and Dutch in embedded clauses. Counterexamples to (ii) include two Bongo-Bagirmi languages, Ngambay and Mbaye. These two languages were the only ones listed in Dryer 1 998 as having base order SVOAux (with verbal auxiliary), while very many languages turned up with base order SAuxOV. As a claim about derived orders, various VP-topicalization structures violate (ii), e.g. in Scandinavian languages.
6.
There are many mixed cases, in particular numbers very frequently precede the noun (cf. Greenberg 1978: 29). In mixed cases, the relative orders of the individual elements remain true
22
PETER SVENONIUS
to the generalization; thus we frequently find for example Num-N-A-Dem (Basque, Indonesian, Vietnamese, Jacaltec, Welsh, and many others) but not Num-N-Dem-A. Greenberg mentions Kikuyu as having N-Dem-Num-A, i.e. the N-final order but with N initial; Hawkins 1983 points out that all such non-mirror image examples seem to come from the Bantu family (though Old Norse allows N-Dem-A, Nygaard 1906, it is not the usual order). See Rijkhoff 1992: 223ff. for discussion.
References Abney, Steven. 1 987. The English Noun Phrase in its Sentential Aspect. Ph.D. disserta tion,
MIT.
Anderson, Stephen. 1976. "On the notion of subject in ergative languages."
In Charles N.
Li (ed.), Subject and Topic. New York: Academic Press. 1-24. B ach, Emmon. 1962. "The order of elements in a transformational grammar of German."
Language 38: 263-269. B ach, Emmon. 1968. "Nouns and noun phrases."
In Emmon Bach and Robert T. Harms
(eds), Universals in Linguistic Theory. New York: Holt, Rinehart, and Winston.
9 1-124. B ach, Emmon. 197 1 . "Questions." Linguistic Inquiry 2: 153-166. B aker, Mark. 1985. "The Mirror Principle and morphosyntactic explanation." Linguistic
Inquiry 16: 373-416. B aker, Mark C. 1988. Incorporation. Chicago: The University of Chicago Press. B arbiers, Sjef. 1995. The Syntax of Interpretation. The Hague: Holland Academic Graphics [HIL dissertations 14] . Belletti, Adriana. 1990. Generalized Verb Movement: Aspects of verb syntax. Turin: Rosenberg and Sellier. Borer, Hagit. 1994. "On the projection of arguments."
In Elena B enedicto and Jeff
Runner (eds), Functional Projections. Amherst: GLSA, University of Massachusetts.
19-47. Brody, MichaeL 1 997. "Mirror theory." London: University College London [DCL Working Papers in Linguistics 9]. Chomsky, Noam. 1970. "Remarks on nominalization."
In Roderick Jacobs and Peter
Rosenbaum (eds), Readings in English Transformational Grammar. Waltham: Ginn and Co. 1 84-221 . Chomsky, Noam. 1980. Rules and Representations. New York: Columbia University Press. Chomsky, Noam. 1 98 1 . Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. 1994. "Bare phrase structure." MIT Occasional Papers in Linguistics 5. Cinque, Guglielmo. 1994. "On the evidence for partial N-movement i n the Romance DP."
In Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi, and Raffaella Zanuttini (eds), Paths Towards Universal Grammar: Studies in honor of Richard S.
Kayne. Washington, D.C.: Georgetown University Press. 85-1 10.
23
INTRODUCTION
Cinque, Guglielmo. 1 999. Adverbs and Functional Heads: A cross-linguistic perspective. Oxford: OUP. Diesing, Molly. 1992. Indefinites. Cambridge,
MA: MIT Press.
Dik, Simon C. 1 989. The Theory of Functional Grammar: Part I: The structure of the
clause. Dordrecht: Foris. Dixon, R.M.W. 1994. Ergativity. Cambridge: CUP. Dryer, Matthew S . 1980. "The positional tendencies of sentential noun phrases in universal grammar." Canadian Journal of Linguistics 25: 123-195. Dryer, Matthew S . 199 1 . "SVO languages and the OV: VO typology." Journal of
Linguistics 27: 443-482. Dryer, Matthew S . 1992. "The Greenbergian word order correlations." Language 68:
8 1-138.
Dryer, Matthew S . 1998. Typological Database. SUNY, Buffalo NY. Dryer, Matthew S . 1 999. Explaining the Order of Genitive and Noun
in SVO Languages.
Ms. SUNY Buffalo. Emonds, Joseph. 1976. A transformational approach to English syntax: Root, structure
preserving, and local transformations. New York: Academic Press. Frazier, Lyn. 1979. "On comprehending sentences: Syntactic parsing strategies." Bloom ington: Indiana University Linguistics Club. Fukui, Naoki, and Takano, Yuji. 1998. "Symmetry in syntax: Merge and demerge."
Journal of East Asian Linguistics 7: 27-86. Fukui, Naoki. 1 993. "Parameters and optionality." Linguistic Inquiry 24.3 : 399-420. Gazdar, Gerald, Pullum, Geoffrey, and Sag, Ivan. 1982. "Auxiliaries and related phenom ena in a restrictive theory of grammar." Language 58: 59 1-638. Givan, Talmy. 1984. Syntax: A functional-typological introduction, Volume 1 . Amsterdam and Philadelphia: John Benjamins. Greenberg, Joseph H. 1963. "Some universals of grammar with particular reference to the order of meaningful elements." In Joseph H. Greenberg (ed.), Universals of Lan guage. Cambridge, MA: MIT Press. 73-1 1 3 . Greenberg, Joseph H. 1978. "Numerals." I n Joseph H. Greenberg (ed.), Universals of Human Language, vol. 3 . Stanford: Stanford University Press. 249-295. Haider, Hubert. 1992. "Branching & discharge." Arbeitspapiere des SFB 340 23, 1-23. Universities of Stuttgart and Tlibingen; revised version to appear in Lexical Specifi cation and Insertion, Peter Coopmans and Jane Grimshaw (eds.), 135-164. The Hague: Holland Academic Graphics. Halliday, M.A.K. 1967. "Notes on transitivity and theme in English, part II." Journal of
Linguistics 3 : 199-244. Hawkins, John A. 1 983. Word Order Universals. New York: Academic Press. Hawkins, John A. 1 988 (ed.). Explaining Language Universals. Oxford: Basil Blackwell. Hawkins, John A. 1994. A Performance Theory of Order and Constituency. Cambridge: CUP.
24
PETER SVENONIUS
Hetzron, Robert. 1978. "On the relative order of adjectives." In Hansjakob Seiler (ed.), Language Universals. Tiibingen: Gunter Narr. 165-1 84. Hinterh61zl, Roland. 1997a. "A VO-based approach to verb raising." In Kiyomi Kusumoto (ed.), Proceedings of the North East Linguistic Society 27. Amherst: GLSA. 1 87-202. Hinterh61zl, Roland. 1 997b. An XP-movement account of restructuring. Ms. University of Southern California. Holmberg, Anders. 1986. Word Order and Syntactic Features in the Scandinavian Languages and English. Ph.D. dissertation, University of Stockholm. Holmberg, Anders, Urpo Nikanne, Irmeli Oraviita, Hannu Reime, and Trond Trosterud. 1993 . "The structure of !NFL and the finite clause in Finnish." In Anders Holmberg and Urpo Nikanne (eds), Case and Other Functional Categories in Finnish Syntax. Berlin: Mouton de Gruyter. 177-206. Huang, C.-T. James. 1982. Logical Relations in Chinese and the Theory of Grammar. Ph.D. dissertation, MIT. Jackendoff, Ray. 1977. X-bar Syntax: A study of phrase structure. Cambridge, MA.: MIT Press. Johnson, Kyle. 199 1 . "Object positions." Natural Language & Linguistic Theory 9: 577-636. Julien, Marit. 2000. Syntactic Heads and Word Formation: A study of verbal inflection. Ph.D. dissertation, University of Tromssz>. Kayne, Richard S. 1 983. "Connectedness." Linguistic Inquiry 14: 223-249. Reprinted in 1984 in Connectedness and Binary Branching, Dordrecht: Foris. Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kayne, Richard S . 1 998. "Overt vs. covert movement." Syntax 1: 128-19 1 . Koopman, Hilda. 1984. Th e Syntax of Verbs: From verb movement rules in the Kru languages to universal grammar. Dordrecht: Foris. Koptjevskaja-Tamm, Maria. 1993. Nominalizations. London: Routledge. Koster, Jan. 1974. "Het werkwoord als spiegelcentrum." Spektator 3 : 601-6 1 8 . Koster, Jan. 1 994. "Predication incorporation an d the word order o f Dutch." In Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi, and Raffaella Zanuttini (eds) , Paths Toward Universal Grammar: Studies in honor of Richard S. Kayne. Washing ton, D.C.: Georgetown University Press. 255-276. Larson, Richard. 1988. "On the double object construction." Linguistic Inquiry 19: 335-39 1 . Lees, Robert B . 1959. The Grammar o f English Nominalizations. Ph.D. dissertation, MIT Published 1963 in revised form by Indiana University Press, Bloomington, with Mouton, The Hague. Lehmann, W. P. 1 973. "A structural principle of language and its implications." Lan guage 49: 47-66. Li, Charles N., and Thompson, Sandra A. 1976. "Subject and topic: A new typology of language." In Charles N. Li (ed.), Subject and Topic. New York: Academic Press. 457-490. .
INTRODUCTION
25
Li, Yen-Hui Audrey. 1 990. Order and Constituency in Mandarin Chinese. Dordrecht: Reidel. Mitchell, Erika. 199 1 . "Evidence from Finnish for Pollock's theory of IP." Linguistic Inquiry 22: 373-379. Nash, Lea. 1996. "The internal ergative subject hypothesis." Proceedings ofNELS 26. Nash, Lea. 1998. On the absence of SVO ergative languages: The case of Georgian. Ms. University of Paris 8. Nygaard, Marius. 1 906. Norr¢n Syntax. Oslo: H. Aschehoug & Co. Peters, P. Stanley Jr. and Ritchie, RW. 1969. "Context-sensitive immediate constituent analysis: Context-free languages revisited." ACM Symposium on Computing, 1-8, ACM. Reprinted 1 973 in Mathematical Systems Theory 6: 324-333. Pintzuk, Susan. 1997. From OV to VO in the history of English. Ms. University of York. Pollock, Jean-Yves. 1989. "Verb-movement, universal grammar, and the structure of IP." Linguistic Inquiry 20: 365-424. Pullum, Geoffrey and Wilson, Deirdre. 1 977. "Autonomous syntax and the analysis of auxiliaries." Language 53.4: 741-788. Rennison, John R 1 997. Koromfe. London: Routledge. Rijkhoff, Jan. 1992. The Noun Phrase: A typological study of its form and structure. Ph.D. dissertation, Universiteit van Amsterdam. Ross, John Robert. 1967a. "Gapping and the order of constituents" Paper presented at the 10th International Congress of Linguistics. Published 1970 in Progress in Linguis tics, ed. by Manfred Bierwisch and Karl Erich Heidolph. The Hague: Mouton. 249-259. Ross, John Robert. 1967b. Constraints on Variables in Syntax. Ph.D. dissertation, MIT. Published 1986 as Infinite Syntax! Norwood: Ablex. Runner, Jeffrey. 1995. Noun Phrase Licensing and Interpretation. Ph.D. dissertation, University of Massachusetts, Amherst. Steele, Susan. 1978. "Word order variation: A typological study." In Joseph H. Greenberg (ed.), Universals ofHuman Language, vol. 4: Syntax. Stanford: Stanford University Press. 585-624. Stowell, Tim 198 1 . Origins of Phrase Structure. Ph.D. dissertation, MIT. Svenonius, Peter. 1994. "C-selection as feature-checking." Studia Linguistica 48: 133-155. Svenonius, Peter. 1 998. "Conference report: Workshop on VO and Ov." Glot International 3.9/10: 25-28. Svenonius, Peter. In press. "Subject positions and the placement of adverbials." To appear in Peter Svenonius (ed.), Subjects, Expletives, and the EPP. Oxford: OUP. Takano, Yuji. 1996. Movement and Parametric Variation in Syntax. Ph.D. dissertation, University of California, Irvine. Trask, R L. 1979. "On the origins of ergativity." In Franz Plank (ed.), Ergativity: Towards a theory of grammatical relations. New York: Academic Press. 3 85-404. Travis, Lisa. 1984. Parameters and Effects of Word Order Variation. Ph.D. dissertation, MIT.
26
PETER SVENONIUS
Uriagereka, Juan. 1999. "Multiple spell-out." In Samuel David Epstein and Norbert Hornstein (eds.), Working Minimalism. Cambridge MA: MIT Press. 251-282. Vallduvf, Enric. 1990. The Informational Component. Ph.D. dissertation, University of Pennsylvania. Vennemann, Theo. 1974. "Topics, subjects, and word order: From S:XV to SVX via TVX." In J.M. Anderson and C. Jones (eds), Historical linguistics 1 . Amsterdam: North Holland. 339-376. Zanuttini, Raffaella. 1 997. Negation and Clausal Structure: A comparative study of Romance languages. Oxford: OUP. Zwart, Jan Wouter. 1993 . Dutch Syntax: A minimalist approach. Ph.D. dissertation, University of Groningen. Groningen: Grodil. Zwart, Jan Wouter. 1997. Morphosyntax of Verb Movement. A minimalist approach to the syntax of Dutch. Dordrecht: Kluwer.
Restructuring
Mirror
Michael Brody University College London and Linguistic Institute, HAS
Abstract Sections 1 and 2 briefly outline some central features of mirror theory (Brody 1997) and discuss consequences with respect to 'basic' word order. In Sections 3 and 4 I note that mirror theory is incompatible with covert roll-up head chain type relations and argue that contrary to recent claims the analysis of Romance restructuring need not involve such structures. In Section 5 I note that Kayne's correlation between null subjects and clitic climbing may be better captured under the proposed analysis than it has been in earlier approaches. I argue that both phenomena involve licensing of a Spec by an element of Infl. In Sec tion 6 I discuss some aspects of the behavior of Hungarian restructuring infinitives and their treatment in mirror theory. Section 7 looks at some similarities and differences between the 'climbing' options of Hungarian verbal modifiers and Romance clitics. Finally in Section 8 I argue that to understand Hungarian restructuring constructions it is necessary to distinguish (strictly local) head chain and (successive step) phrasal chain type relations, - a fact that constitutes further evidence for some core assumptions of mirror theory.
1.
Mirror theory, a brief sketch
In mirror theory (Brody 1997), the morphological structure of words is expressed syntactically as complementation structure. The mirror principle of this theory ensures that if x is the complement of y then x is taken to be the morphological specifier of y. For a large set of morphemes (generally suffixes) it is also typically (though probably not always) true that if x is the morphological specifier of y, then x is the complement of y syntactically. As a simplified example, V is (part of the) the morphological Spec of Infl and V is also typically (part of the) the syntactic complement of Infl.
28
MICHAEL BRODY
In mirror theory, complementation structure is taken to be the default expression of the morphological structure - the mirror hypothesis. According to this hypothesis the syntactic head-complement relation expresses the morpho logical specifier-head relation in inverse topological order: complements follow while specifiers (whether syntactic or morphological) precede the head. Thus no separate :xo-internal representation needs to be assumed that matches and duplicates the complement series. Consider for example the simplified structure in (1) of, say, Mary loves John. Here (Mary) is the trace of the subject in Specv, so Mary and (Mary) form a chain; the object John is taken for the sake of presentation, probably counterfactually, not to form a chain with a position higher than Specv; and Infl, v and V represent, again in a simplified fashion, the morphemes (some null) from which the word loves is composed.
(1)
�
�
Mary
v @ (loves)
�
(Mary)
V
/
John
The complement series in (1), Infl+ v+ V serves as the syntactic representation of the morphological word (MW) V+ v+ Infl. Morphology spells out the syntactic representation of an MW (a) in one of the head positions, here in that of v (as indicated by "@"), from which the MW is composed and (b) in inverse order, due to (the appropriate version of) the mirror hypothesis, an axiom of the system. All this gives the correct morpheme order both word-internally (love+ s) and also word-externally if the Spec (and whatever it dominates) precedes the head, and the complement (and whatever it dominates) follows it. The order of head and complement follows from the order of Spec-Head, given the assumption that specifiers uniformly precede the head both in syntax and in morphology and that syntactic complement relations are inverse order morphological Spec-Head relations - the mirror hypothesis again. Thus in (1) Mary precedes Infl, Infl precedes v where loves is spelt out and v precedes V and whatever V dominates, i.e. in particular John. Under mirror theory the syntactic complementation relation entails morpho logical specifierhood: if x is the complement of y then x is the morphological specifier of y. Thus first of all the arguments of the verb that are morphological ly independent of it, - (non-incorporated, non-clitic) subject or object, or clausal
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
29
complement etc. - must be specifiers. Secondly if the subparts of what in standard systems are extended projections do not form an MW then these parts must also be in the specifier-head rather than in the complement-head relation. For example in "John has come" come cannot be (part of) the complement of has. The auxiliary is an element that is part of the extended word but not of the morphological word of the main verb. It must therefore be a specifier as in (2).
Infl @ (has)
(2)
�
John
Infl
/
come
To ensure the correct word order, it must be assumed that come is the specifier of a head that is lower than the head in which has is spelt out in morphology. The representations in (1) and (2) are simplified in several ways for presentational purposes, but there is a particular simplification that is not presentational, but is meant as a substantive restrictive hypothesis. This has to do with the elimination of phrasal nodes. The structure in (3) for example that (1) replaces is obviously redundant.
(3)
InflP
�
Su
Infl'
�
Infl
vP
�v'
(Su)
�
v
VP
�
Obj
V'
/
V
Call the claim that the set of phrasal and XO-intemal projections of a head can be systematically collapsed the telescope hypothesis. XO-intemal projections are unnecessary given the treatment of MWs as (inverse order) complement lines. As for phrasal projections, the telescope hypothesis embodies the expectation that
30
MICHAEL BRODY
given the accumulating evidence for multiple additional heads of various types in the structure, phrasal projections will invariably be unnecessary. For example the major evidence for the V' level, based on the hierarchical subject-object asymmetry, disappears when the subject is taken to be Specv. This ensures without the intermediate bar level that the subject is higher than the object. Any category can be interpreted as either a phrase or a head in mirror theory - a head by itself and a phrase together with all categories it dominates. (Notice also a terminological point: since categorial projection is eliminated, extended projections are better referred to as extended words.)
2.
Mirror theory and word order
As we have seen, under mirror theory all non-clitic arguments must be specifiers. Does mirror theory therefore entail a strict "underlying" (i.e. chain-root) head final "sov" order? While such a statement would be partly true, it would also be in part incorrect and in part misleading. There are three main reasons. First, a morphologically dependent object, like an incorporated noun for example may be the complement of the verb. Given the mirror hypothesis, it will then be spelt out preverbally as required. Secondly, as noted above, a head like a verb for example, may form an MW with one or more higher heads like v and Infl, each of which is the syntactic complement of the next. The MW may be spelt out in any one of the complement positions that its component members (the morphemes) occupy. Now the notion of MW corresponds to the concept of head chain in standard frameworks, but it is different from this notion in that it provides no natural way of talking about "underlying" or chain root position. The set of heads, each a complement of the previous one, is a decomposed representation of the MW So there is no clear sense in which the lowest head in the series would be an "underlying" or chain root position. Thus mirror theory consistently entails head final structures in the sense that arguments must be specifiers, but there is no sense in which it can be said to entail underlying or chain root SOY, because the positions in which the verb may be spelt out, and which may precede or follow the object and also the subject, form an MW and not a chain. The standard ways of creating non chain-root word orders involve not only V-raising but also VP shift. The third reason why mirror theory cannot be taken to entail chain root SOy is that it provides a treatment for the relevant VP shift operations in which the V-phrase's chain root position corresponds to what is taken to be its shifted, non-root position in other approaches. Recall the discussion .
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
31
preceding (2) above: a V-phrase that does not form an MW with the rest of what in standard terms is its extended projection (in mirror-theoretical terms its extended word, EW), must be in a Spec position. So in general EWs do not necessarily form a series of complements, they can also continue via specifiers. Thus we have a potential solution to the triggering problem of VP shift. When the V-phrase is in a Spec position, this need not be because it forms a chain with a root position where the V-phrase is a complement, part of its extended word. The V-phrase may simply continue its EW via the Spec position, a configuration that must be available in general if the mirror hypothesis is on the right track. Thus at least in some of the cases where non V-final order is achieved in standard terms by VP shift, under mirro r theory the "underlying" chain root V(-phrase)-final order may not exist.
3.
Restructuring and (covert) roll-up
Let us refer to a series of chains as a roll-up structure if it meets the following condition: each chain (except the last) takes the top of the previous chain together with the host of this top member (where this host includes the root of the previous chain) to be the root of the next chain. The term "cascade" has sometimes been used but it I shall avoid it here, since it is often employed also in a different sense. Roberts ( 1 997) has recently suggested analysing Romance type restructuring as (in these terms) a roll-up structure that involves covert head chains. He proposed that restructuring between two verbs V I and V2 involves head movement of V2 up to V I . The V I V2 order in Romance is due to a filter that prevents spelling out V2 in the higher position in its chain. This filter would distinguish between morphemes and words: V can be spelt out on the left of its Inff host since both elements are morphemes, but in restructuring V2 cannot be spelt out on the left of its host VI because both verbs are full words. V2 therefore has to surface lower, and so the VI V2 surface order remains. Roberts assumes (a) that head movement of V2 to V I creates an extended projection that includes both verbs and (b) that the locality/relativized minimality requirement of XP chains makes use of a principle of equidistance (in Chomsky's 1 995 sense), for which positions internal to an extended projection count as equidistant. Following Sportiche 1 995, Roberts assumes that clitic climbing involves XP chains. In the clitic climbing structure in (4), for example, Roberts takes the clitic to move as XP via the lower SpecAgrO and the higher SpecAgrO. That the former position is involved is suggested by the well known participle agreement phenomenon. The involvement of the higher AgrO, as he points out, is suggested by the participial agreement in the matrix in (4).
32
MICHAEL BRODY
(4)
Maria li
ha voluti
prendere.
Maria them (MASC.PL) has wanted (MASC.PL) take
(It)
(This evidence for phrasal chains is strong only on the assumption that the notion of 'checking domain' should be eliminated. Otherwise, as has been noted, agreement of the participle with a nonphrasal element adjoined to the head, i.e. still in its checking domain, is an obvious alternative. See Sportiche 1 992 and Cardinaletti & Starke 1 999 for additional evidence for a phrasal clitic chain.) So for Roberts, invisible movement of the verb prendere to a position hosted by voluti creates an extended projection and thus makes the SpecAgrS of the lower head and the SpecAgrO of the higher one equidistant from the lower AgrO - all three positions are in the same extended projection. In mirror theory, roll-up structures involving heads are analyzed in terms of MWs (as opposed to 'phrases' , i.e. categories taken together with their constitu ents). Elements of MWs are morphemes and the whole MW is a word. Given this restrictive notion, it would make no sense to distinguish component elements of MWs as being either word-level or morpheme-level elements. Another consideration that may be taken to indicate that it may be worth while to look for an alternative treatment of restructuring is the following. Roberts points out that his approach accounts for the possibility of (XP-)move ment across restructuring predicates "without any operation deleting structure in the lower clause" (p. 432). It is not clear however if a solution based on the notion of equidistance is a priori more desirable than one based on structure deletion. The hypothesis that in restructuring contexts intervening position B is deemed not to intervene between positions A and C (whether this is due to A and B being deemed equidistant from C or to some other reason) says nothing about how B will behave under conditions or processes other than movement. On the other hand the hypothesis that in the same contexts B does not intervene because B is not present (either not present at all in the structure or present but in fact occupies a non-intervening position) entails that no principle or operation can make use of B (at all or in the intervening position). These empirical consequences are missing in the weaker equidistance approach. (For the same reason, the layered VP analyses in which AgrO is lower than the chain-root position of the subject in Specv (e.g. Koizumi 1 993; Bobaljik 1 995) would appear to be a priori more desirable than Chomsky' s ( 1995) equidistance solution, where the paths of subject and object cross.)
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
4.
33
Restructuring as 'I in C'
As is well known, there is direct empirical evidence that the complementizer level of the lower infinitive is present in restructuring. (5) is a case with si-passive, (6) clitic climbing and (7) an easy to please construction (cf. Rizzi 1 982; Sportiche 1 995; Kayne 1 987). (5)
? Certe riposte non si
(6)
?Mario, non 10
(7)
sanna mai
come dare.
(It)
certain answers not REFL knows never how give 'One never knows how to give certain answers. '
saprei a chi
affidare.
Mario not him know to whom entrust 'Mario, I would not know to whom to entrust him
? Ce genre d'article est difficile
a
savoir au
(It) .'
classer.
(Fr)
this kind of.article is difficult to know where file 'This kind of article is hard to know where to file. ' Roberts combines Sportiche's phrasal chain analysis of cliticization with the Kaynean approach to restructuring as involving movement of the lower Inil to the higher one. Apart from the suggestion already discussed that this operation creates an extended projection spanning the two clauses, he also suggests that Inil movement is coextensive here with verb movement. Following B elletti ( 1990) he assumes that the infinitival verb in Italian raises to Inil. If this is so, then raising of Inil will involve raising of the verb, resulting in the incorrect word order. Hence the necessity of the * V V filter he proposes, discussed above. The evidence that Roberts quotes from B elletti ( 1990) for the infinitival being in Inil comes from the observation that these must precede elements like mai and piu, naturally associated with negation: (8)
Gianni ha deciso di non tornare mai/piu.
(It)
Gianni has decided to not return ever/more 'Gianni has decided not to come back ever/anymore. ' (9)
* Gianni ha decisio di non mai/piu tornare.
(It)
However even if the polarity elements are in SpecNeg, it does not strictly follow that the verb has raised to the highest Infl. position: it may still be in principle the case that Inil raising in restructuring involves a higher head in the Inil domain than the one to which the verb raises. (Cf. e.g. Cinque 1 999 on multiple Neg positions.) So this evidence in fact does not necessarily prevent a return to Kayne's Inil
34
MICHAEL BRODY
raising analysis. Kayne ( 1 989) assumed that in clitic climbing the clitic raised to the lower Inti, Inti+clitic to C and the whole complex then moved to the higher Inti. It is, however, not clear why the infinitival Infl needs to move to the higher Inti of the restructuring V if clitics form XP-chains. For Kayne, movement to the higher Inti is necessary to carry the clitic up into the higher clause, which is now achieved via the XP-chain. Additionally questions arise concerning the fact that Inti movement to the higher clause appears to cross the higher V position. Let us then dispense with the now apparently unnecessary and problematic part of Kayne' s "I to C to I", namely with "c to I". The residue, "I to C", is motivated by Kayne primarily by the contrast between the restructuring construc tions like in (5)-(7) and those with an overt C as in ( 10) and ( 1 1) (his example is ( 1 1 ) : (10)
*Certe riposte non s i
(11)
*Non li
sanna mai s e dare.
certain answers not REFL knows never if give 'One never knows whether to give certain answers. '
so
(It)
s efare.
not them know if do 'I don't know whether to do them. ' As he notes it is natural to assume that "I to C", hence restructuring, is possible only where C is otherwise empty. (Currently standard assumptions about head movement as head adjunction do indeed provide a basis for expecting some correlation, although they provide no grounds for expecting exactly the observed correlation. If heads do not move through other heads, it remains accidental that only those heads to which Inti cannot adjoin happen to have overt phonological realization. ) Notice however, that the evidence pertains only to the claim that (elements of) Inti are present in C, and not to the assumption that a chain has been formed. Thus we could equally assume that restructuring involves a special type of infinitival C that is in some closer than usual relation with its Inti domain. Suppose Inti can merge with the C selected by restructuring verbs in the sense that Inti (and its Spec, if any) becomes part of the C-domain and thus will not qualify as an A-type intervener. It will thus not interfere with A-chains construct ed across it. The Inti in C analysis automatically covers the case of long si passive in restructuring. Lack of an A-type (Spec-)Inti in the lower clause entails that there will be no (relativized) minimality violation in long si-passives like (5) either. (Like others, I assume that auxiliary selection phenomena in restructuring involves long movement and thus in the relevant respect also falls under the same generalization.)
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
5.
35
Null subjects and ditic climbing
The correlation conjectured by Kayne between the existence of null subjects and the option of clitic climbing is only weakly captured in Kayne 1989. His proposal was that an Infl strong enough to licence null subjects has the ability to void the barrierhood of VP, thereby enabling clitics to escape higher. It is not clear however why these two Infl properties should go together - there is no obvious intrinsic connection between licensing a particular type of Spec and voiding the barrierhood of the complement. Under the phrasal chain analysis of cliticization, the correlation can be captured more directly. Consider Cardinaletti & Starke's ( 1 999) hypothesis that strong and weak pronouns systematically differ from clitics in being XPs, while clitics are heads. In mirror theory terms the only natural way to translate this is to say that non-clitic pronouns are specifiers of the verb's extended word, while clitics are heads, members of the verb's extended word. We can leave it open here whether the clitic starts out as a member of this extended word (essentially as in Sportiche 1 995) or as seems more likely (thanks to Michal Starke for helpful discussion) a head in the Infl domain inherits the features of its (weak) pronominal specifier via Spec-Head agreement. On these approaches syntactic cliticization in the core cases would correspond to a special type of Spec-Head agreement, possible only where the head has the ability or expressive power to carry the referential, anaphoric etc. functions of its designated Spec. Presumably the 'designated' Spec is one whose every feature participates in the Spec-Head agreement relation. With the clitic option taken (the default case where possible, cf. Cardinaletti & Starke's "minimize structure"), the Spec will be typically (apart from clitic doubling) empty or null. The corresponding heads on the other hand are typically non-null, phonologically. The similarity of VP-external, or Infl domain, cliticisation to null subjects is now conspicuous. Null subjects are also licensed by a typically non-null head in the Infl domain that is able to carry the semantic functions of the subject. Thus both VP-external high cliticization and null subjects are licensed in the same way. Both will be possible in a language where clitics exist and Infl domain heads can carry the semantic functions normally carried by Specs as in Italian. In French, as Kayne ( 1 989) points out, even though "easy to please" constructions show restructuring effects, neither null subjects nor VP-external clitics (hence no clitic climbing) are licensed. The analysis outlined in the previous two sections assumes that the clitic's 'phrasal' chain (Spec-to-Spec constituent chain in mirror theory terms) extends to a SpecX position higher than the matrix restructuring verb V*. In mirror
36
MICHAEL BRODY
theory terms this entails that x (which expresses the clitic) and V* cannot form an MW If they did, this would result in the enclitic order: V*-x. To achieve the proclitic order, x must have a complement y with the MW of V* in the Spec of y. (The alternative on which xlclitic and V* in fact form an MW also seems worth exploring, although I will not do so here. If x-V* is an MW then x must originate lower than V*. Hence the Spec-to-Spec constituent chain of the clitic must end lower than V*, which often (as in (5) and (6)) but not always (e.g. (4) above) means lower than the restructuring head, i.e. presumably within the Infl domain of the embedded clause. Such an analysis would be more in the spirit of Kayne's I-to-C-to-I rule, but it would not inherit the problem of I(+C) crossing the matrix V.) .
,
6.
Hungarian verbal dusters
Verbal clusters in Hungarian involving typical restructuring infinitives may appear in two orders: what we might call the straight order as in Romance (and elsewhere) and the inverted roll-up order: ( 1 2)
Utalok [kezdeni jami uszni] .
(Hu)
hate.I begin.INF go.INF swim.INF 'I hate to begin to go swimming (regularly)' ( 1 3)
Utalok [uszni
jami kezdeni] .
(Hu)
hate.I swim.INF go.INF begin.INF - same The infinitives in ( 1 3) are behaving as dependent bound morphemes while those in ( 1 2) behave as words. Adverbials (or other material like the matrix subject for example) may intervene between the infinitivals in ( 1 2) but not between those in ( 1 3) cf. Koopman and Szabolcsi 1 999. Thus, Hungarian restructuring verbs must apparently be intrinsically underspecified (or dually specified) for word hood/morphemehood. The infinitives in ( 1 3) in standard terms appear to form an XO roll-up structure and thus in mirror theory they must be analysed as constitut ing a single MW syntactically a series of complements: -
-
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
( 1 4)
37
-ok@
�
utal-
/
-ni@
�
kezde-
�
-ni
�
jar-
�
-ru
�
usz-
The correct word order results in ( 14) if the MW utalok is spelt out higher than the position of the verb utal, i.e. in some head in the Inil domain represented here by Tns+Agr -ok. Utalok will then precede the (specifier of) its complement the MW uszni jarni kezdeni, a complement series spelt "backwards". In ( 1 2) the morphologically independent infinitives cannot be each other' s complements, they must therefore each b e specifiers: ( 15)
-ok@
�
utal-
/
-ni@
�
kezde-
/
-ni@
�
jar-
/
-ni@
�
usz-
MICHAEL BRODY
38
High spellout of each infinitive ensures the correct word order: each verb in (15) spelt out in the relevant inflectional head position IH will precede the (specifier of the) complement of IH. If we did not have evidence for the infinitives in (13)-(14) constituting a single MW the word order there could also correspond to what in standard terms we could only analyze as a phrasal roll-up structure. While that can be repro duced in mirror theory, this framework appears to provide also an additional possibility. The analysis could be the same as that of ( 1 2)/( 1 5), but with the spellout positions of the infinitives being those of the verbs instead of the higher functional heads. Then each infinitive would be preceded by (the complement of) its specifier resulting in the roll-up order. Low spellout is apparently not an option in Hungarian, a fact perhaps connected to the language allowing null subjects (cf. Roberts 1 997). (Note that in ( 14) and ( 1 5) ' @ ' indicates the spellout position and not that of the strong features. These correspond to the spellout positions in ( 15) and may be the same in ( 14) on the assumption that the spellout position of an MW is its highest strong position.) ,
7.
Verbal modifiers and ditics
As discussed in Koopman & Szabo1csi ( 1998) and also in Brody ( 1997), E. Kiss ( 1 998), Hungarian verbal modifiers (VMs: particles, small clause predicates, bare nouns etc. that can form a single MW with the associated verb, e.g. szet-szed, 'apart-take' ) can apparently also form long-distance chains across a set of restruct uring infinitives (the trace of the VM is indicated by the copy in parentheses): (16)
Szet fogom akami
kezdeni
szedni
(szit) a
radi6t.
(Hu)
apart will.I want.INF begin.INF take.INF apart the radio 'I will want to begin to take apart the radio. '
E. Kiss ( 1998) argues that the VMs i n structures like ( 1 6) relate the verb they semantically belong to and their spell out position via a head chain type relation on the basis of the fact that the string following the VM does not appear to form a constituent and thus cannot be co-ordinated. (17)
*Szet [akarom probalni valogatni a apart will.I
szedni
a
try.INF
radi6t].
sort.INF
babot] es [fogom kezdeni
the beans and will.I begin.INF (Hu)
take.INF the radio 'I will try to sort ( ' apart' ) the beans and will begin to take apart the radio. '
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
39
The co-ordination facts however can pertain only to the question of whether the spellout position of the VM is a head or a Spec position, they do not tell us how the relation between the VM in the matrix and the associated embedded verb is mediated by a head-chain or phrasal chain type relation. Typically the same coordination facts obtain with clitics. ( 1 8)
*Je l 'ai 1
vu
et ai
aime.
(Fr)
him. have seen and have liked
But as we have seen above there is reason to think that clitics can be involved in phrasal chain type relations. Thus Hungarian VMs can be analyzed like clitics: they involve a phrasal chain but at the top position of this phrasal chain a Spec-Head relation is established. The head participating in this Spec-Head relation is of the type that is able to express the content of its VM spec. There is however a real difference between Romance clitics and Hungarian VMs in restructuring. While clitics can cross a filled C-Spec as (6) above shows, reproduced here as ( 1 9), VMs cannot. The infinitival with a +wh SpecC is grammatical in Hungarian as a complement of tud ('know') and van ( 'is' / 'exists' ), but no VM can cross this spec. ( 1 9)
?Mario, non 10
saprei a chi
affidare.
(It)
Mario not him know.I to whom entrust 'Mario, 1 would not know to whom to entrust him. ' (20)
Tudok mit szetszedni.
(Hu)
know.I what apart.take 'I know what to take apart.' (2 1 )
*Szet tudok mit szedni. apart know.I what take.INF - same as (20)
Given the assumption adopted here that the clitic in (19) forms a phrasal chain that spans the matrix and the embedded clause across SpecC, the solution that would attribute the contrast between (19) and (2 1 ) to the difference between the type of interveners to which head chain and phrasal chain type relations are sensitive to is not available. But a different account, still in the spirit of relativized minimality, could be adequate even if both the clitic and the VM form phrasal chains. The VM is an adverbial type element, hence it cannot cross a filled SpecC, - the ungrammaticality of (2 1 ) would then be on a par with that of (22) : (22)
a. b.
*How much did Mary wonder why John weighed? *60 kilos, Mary wondered why John weighed.
40
MICHAEL BRODY
The clitic in (19) on the other hand is an argument. If the top of its phrasal chain is in an A'-position then the structure is like other A'-argument extraction, like e.g. (23). (23)
?Who did Mary wonder why John weighed?
If the clitic is in an A-position, then A'-Specs must be taken to be irrelevant as A-chain interveners, just like A-Specs are irrelevant for A'-chains (cf. Rizzi 1990). As Anna Cardinaletti points out (p.c.), the present approach in terms of an argument-adjunct contrast is supported by the complete lack of idiomatic clitic climbing across filled SpecC:
(24)
a.
Ce
la devo fare.
(It)
there it must.I do 'I have to succeed.' b. *Non ce la so come fare. not there it know.! how do 'I do not know how to succeed. '
8.
Straight and roll-up orders again
There is also Hungarian-internal evidence for the claim that the chain of the VM in straight order restructuring constructions is of the phrasal type (cf. Brody 1 997). This is based on Szabolcsi's ( 1996) observation concerning the interpreta tion of the focussed infinitive in restructuring constructions with more than one layer of clausal embedding. (25)
a. b. c. d. e.
AKARNI fogok kezdeni
(Hu) uszni. want.INF will.I begin.INF swim.INF 'I will indeed want t o begin t o swim.' 'I will WANT to begin to swim. ' (and not, say, TRY to begin to swim) 'I will begin to WANT to swim.' (and not, say, TRY to swim) * 'I will indeed begin to want to swim.'
If the focussed infinitive in (25) has scope over the others, i.e. if it is taken to be associated with the highest infinitival position, then the interpretation of the structure is ambiguous between an emphatic and an 'exhaustive list' reading, as indicated in (25b) and (25c). If however the focussed infinitive has lower scope, i.e. a lower chain-root position, then the emphatic reading (2Se) disappears and only the 'exhaustive list' reading (25d) remains.
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
41
If head-chain and phrasal chain type relations are distinguished, then this state of affairs is straightforward to explain. There are independent reasons to assume that focussing in Hungarian involves a dedicated head, F (cf. Brody 1 990, 1 995) and that the emphatic reading is associated directly with this head rather than with its Spec. Only focussed heads but not focussed phrases can receive the emphatic reading. It is natural to complement this with the assump tion that the exhaustive list reading is associated with the Spec position of F. If head-chain type relations are strictly local but phrasal chains can span larger distances (whether this involves the combination of strictly local links is not relevant here), only the exhaustive list reading (the phrasal chain construction) will be compatible with the lower scope reading. The explanation is contingent on interpreting strictly the standard distinction between head chain and phrasal chain type relations: the former but not the latter are strictly local; head-chains cannot (even apparently, by combining several local steps) cross nonlocal distances. Thus, if the explanation of the correlation between scope and focus interpretation is on the right track, then head chains and phrasal chains must have different locality properties. Whatever way syntax expresses the distinction between these two types of relations, the long-distance chain of the VM in ( 1 6) must belong to the phrasal type. Additional evidence for the existence of the need to distinguish within syntax the head chain and the phrasal chain relations is provided by the fact that the scope of the infinitives in the roll-up structure ( 1 3), reproduced here as (26), is fixed. Thus in (26) kezdeni ( 'begin' ) has scope over jami ('go regularly'), the opposite interpretation is impossible:
(26)
Utalok [uszni
jami kezdeni] .
(Hu)
hate.I swim.INF gO.INF begin.INF 'I hate to begin to go (regularly) swimming.' This is as expected if the roll-up structure involves head chain type relations, that is in mirror theory a series of complements expressing an (inverse order) MW, as in (14) above. Suppose, however that the roll-up structure was created by phrasal chains, by the lowest infinitive (here: uszni) moving to a Spec above the next (here: jami) and then a phrase that includes both this Spec with the lower infinitive in it and the next higher infinitive (i.e. uszni jami) moved in front of the highest infinitive (kezdeni). Clearly there is nothing intrinsic in the concept of phrasal chain that would prevent the lowest infinitive phrase (uszni) from moving in front of the highest one (kezdeni), either in one or in several steps. But this would result in scope relations between the crossed infinitives that correspond to
42
MICHAEL BRODY
their surface order. Under such an analysis uszni would have crossed jami kezdeni and (26) would have the interpretation 'I hate to go (regularly) to begin to swim' which it in fact cannot have. (Admittedly the relevant reading is slightly strange, but this is exactly what (26) would mean if the order of the last two infinitives was reversed, as in Utalok uszni kezdeni jami.) The analyses reviewed in this section make it necessary to take the locality requirement on the relevant head chains to be strict and inviolable. This provides additional evidence for a basic assumption of mirror theory according to which the head chain relation corresponds simply to a local syntactic relation, typically the head-complement relation. Since in this framework no head-chains are formed either in syntax or in morphology and head chains correspond to elements in local relations spelt out in inverse order (MWs), there are no means provided to violate HMC type locality.
Acknowledgments I am grateful to Anna Cardinaletti and Peter Svenonius for helpful correspondence relating to this material.
References Belletti, Adriana. 1990. Generalized verb movement: Aspects of verb syntax. Turin: Rosenberg and Sellier. Bobaljik, Jonathan. 1 995. Morpho syntax: The syntax of verbal inflection. Ph.D. disserta tion, MIT. Brody, Michael. 1990. "Some remarks on the focus field in Hungarian." UCL Working Papers in Linguistics Vo1.2, 201-226. University College London. Brody, Michael. 1 995. "Hungarian focus and bare checking theory." In Arbeitspapiere des Sondeiforschungsbereichs 340, University of Tiibingen. Brody, Michael. 1997. Mirror Theory. ms. University College London. Cardinaletti, Anna and Starke, Michal. 1999. "The typology of structural deficiency: A Case Study of three classes of pronouns." In Henk van Riemsdijk (ed.), Clitics in the Languages of Europe. Berlin: Mouton de Gruyter. 145-29 1 . Chomsky, Noam. 1 995. The Minimalist Program. Cambridge, MA : MIT Press. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A cross-linguistic perspective. Oxford: OUP. Kayne, Richard S. 1989. "Null subjects and clitic climbing." In Osvaldo Jaeggli and Ken Safir (eds), The Null Subject Parameter. Dordrecht: Kluwer. 239-261 .
WORD ORDER, RESTRUCTURING AND MIRROR THEORY
43
Kayne, Richard S. 1993. "Toward a modular theory of auxiliary selection." Studia Linguistica 47: 3-3 1 . E. Kiss, Katalin. 1998. The Hungarian verbal complex revisited. ms. Linguistic Institute, HAS. Koizumi, Masatoshi. 1993. "Object agreement phrases and the split VP hypothesis." In Jonathan D. Bobaljik: and Colin Philips (eds.), Papers on Case and Agreement I [MITWPL vol. 18]. Cambridge, MA: ITWPL. 99-149. Koopman, Hilda, and Anna Szabo1csi. 1998. Verbal Complexes. Ms. UCLA. In press, Cambridge, MA: MIT Press. Koopman, Hilda, and Anna Szabo1csi. 1999. "Hungarian complex verbs and XP-move ment." In Istvan Kenesei (ed.), Crossing Boundaries: Current Issues in Linguistic Theory. Amsterdam and Philadelphia: John Benjamins. 1 1 5-137. Rizzi, Luigi. 1982. Issues in Italian Syntax. Dordrecht: Foris. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, MA: MIT Press. Roberts, Ian. 1997. "Restructuring, head movement and locality." Linguistic Inquiry 28: 423--460. Sportiche, Dominique. 1995. "Clitic constructions." In Laurie Zaring and Johan Rooryck (eds), Phrase Structure and the Lexicon. Dordrecht: Kluwer. Szabo1csi, Anna. 1996. Verb and particle movement in Hungarian. Ms. UCLA.
OV is More Basic than VO Hubert Haider University Salzburg
Abstract This paper defends the claim that head-final structures are principally less complex than head-initial structures. The additional complexity of head-initial structures is the result of head movement within the lexical projection. Head movement guarantees the convergence of a head-initial phrase structure, given a universal constraint that admits only a right-branching projection structure. On the theoretical level, this position is in competition with the hypothe sis that OV structures are derived from va structures by evacuation of the VP. In a detailed comparison of the empirical implications, this paper attempts to demonstrate that on the empirical level an evacuation theory is less successful than a head movement theory.
1.
Background
Is OV more basic than VO (or vice versa) is just another way of asking a crucial question in the theory of phrase structure: Is one of the two phrase types head-final-phrases and head-inital ones - a derivative of the other one? In terms of the familiar OV-VO distinction, the question reads: Is the head-initial ( 'VO language') or the head-final structure COV language' ) more basic in terms of the basic X'-structure and its derivational continuation? In Kayne's ( 1994) LCA theory and work building on it, head-initial projections are head-initial underlyingly. This is a consequence of the basic axiom of this theory, the linear correspondence axiom (LCA). 1 It postulates that phrase-structures are built in such a way that the linear order of terminals is a function of asymmetric c-command relations. In the specific implementation of Kayne ( 1994), asymmetric c-command is mapped on precedence. So, in particular,
46
HUBERT HAIDER
the head precedes the complement. Head-final structures are taken to be the result of XP-movement applied to the complements of head-initial projections. So, the OV structure for a VP in particular is the result of evacuating all phrasal elements out of the head-initial VP. The result is a VP that contains nothing but the verb and traces, with everything else situated in functional projections to the left. In a more recent proposal (Kayne 1 998), it is suggested that even in a VO-system like English there are movements of VP-internal phrases in some cases (e.g. negative quantifiers) to the left of VP, followed by fronting the remnant VP past that material again, resulting in what superficially looks like the original word order. This remnant VP-fronting must be ruled out for OV languages. The theory developed in Haider ( 1992) is an alternative to the LCA-based theory of the OVNO-relation. The basic axiom, the branching constraint (BC) postulates a universal right-branching structure for (extended) projections: The projecting node follows the non-projecting sister node. The linear aspect of the head-complement relation is determined by the parametric direction of structural licensing. Licensing to the left triggers the OV structure, licensing to the right the VO structure. In OV languages, with the licensing direction to the left, specifiers and complements precede their respective sister node, that is, the node on the projec tion line of the head of the projection. So, the licensing direction harmonizes with the right-branching structure. Complex head-initial structures, however, cannot meet the directionality of the licensing requirement unless the head is raised (see Section 2 for the details). Thus, head-chains (heads raised to the left) must be formed for the licensing of complements in the (extended) head-initial projections. Head-final structures, on the other hand, will not require head-chains, since the heads follow their complements as well as every left branch (including specifiers) in the (extended) projection. Repercussions of these differences, in particular with respect to the grammar of extraction and the distribution of VP-internal non-arguments, will tum out to be an epiphenomenon of the licensing require ment that forces VO languages, but not OV languages, to apply V chaining. Given this alternative - OV is derived from VO by phrasal movement to the left versus VO is derived from OV by head movement to the left - it should not be too difficult to find decisive empirical evidence. After all, the implica tions of the two accounts are diverse enough. The paper is organized as follows: Section 3 sketches the basics of the head movement account (BC theory) of VO and OV structures. Section 4 presents five areas of data that are of relevance for the evaluation of the competing approach es. It is argued that central implications of the LCA system are in conflict with empirical evidence both on the level of observational as well as on the level of
OV IS MORE BASIC THAN va
47
descriptive adequacy. It is argued that the alternative account provides a more satisfactory modelling of the pertinent facts. Section 5 discusses the trigger problem for each account.
2.
A BC account of VO and OV
The central axiom2 of the approach presented in Haider (1992) and developed further in Haider (1994) is the branching constraint formulated in (1) and illustrated in (2) with a V projection embedded under a functional projection that conforms to (1). It requires that lexical projections and their functional extensions are right-branching: (1)
Branching Constraint (BC): Projection-internal branching nodes on the (extended) projection line follow their sister node. 3
The qualification 'projection-internal' singles out the top node. Without this proviso, the BC would not admit basic head-final structures with a phrasal complement preceding the head, because in this case the top node of the complement phrase would have a right sister, namely the complement-selecting head. What the BC is to capture is the rigid right-branching structure internal to (functionally extended) projections of a lexical head. For the V projection, for instance, this is the V projection proper (as the lexical projection) plus its functional extensions up to CPo (2)
In (2), all non-top branching nodes on the V projection line, that is, all the V' nodes, follow their sister node. On the F-projection line, there is only one
48
HUBERT HAIDER
relevant node, namely F', and it follows its sister node, namely SpecF. The crucial structural difference between a head-inital and a head-final lexical projection will be explained below. (3a) violates the B e, because F' precedes its sister node SpecF. In (3b), V' precedes a VP-intemal node and therefore violates the Be. The Be applies to lexical projections and their functional extensions in general. VP in (2) and (3) is just representative for any complex lexical, head-initial projection. (3)
a. *
FP
�
F'
�O
D
VP
�
V'
� O
V
F
FP
b. *
D
D
The Be forbids a left branch in a binary projection which is a node on the projection line: For any two nodes immediately dominated by different nodes of the (extended) projection line, the node that precedes c-commands the node that follows, and the node that precedes is not a node on the (extended) projection line. This constraint eliminates right-associative projection structures as in (3). As a consequence of (1), precedence and c-command coincide. For all nodes that are immediately dominated by a node on the same (extended) projection line, the preceding nodes c-command the nodes that follow. The Be rules out base generated right-associative structures such as (3) as well as structures derived by movement, that is, as a result of adjunction to the right.4 If a functional projection is a functional extension of the projection of a lexical category, the Be rules out functional heads to the right in general. In this case the projection of the lexical category would be a left sister of the functional head and at the same time a node on the extended projection line as a left branch. This is illustrated in (4): (4)
*
FP
� D
F'
�O
VP
D
F
I
The VP node in (4) is a top node in the V projection, but simultaneously a node on the extended V projection, whose top node is the FP. At this point there is
OV IS MORE BASIC THAN va
49
need for a precise definition of functionally extended lexical projection: A functional projection is a functionally extended lexical projection if and only if the lexical head moves to the functional head position overtly. The BC, for instance, does not rule out clause structures with a clause-final complementizer particle. In this case, the CP is not a functional extension of the complement. It is just a functional projection with a complement selected by a CO-head. Rowev er, the BC would rule out a clause structure if a verb overtly moves to the position of a clause-final functional head that is alternatively realized by a lexical complementizer. In sum, the BC forbids any kind of movement to the right: It forbids head movement to the right, as in (4). It forbids movement into Spec-positions to the right, resulting in structures like (3a). And it forbids adjunction to the right, because this produces structures of the kind (3b). Let us now proceed to head-initial projections. Read-final and head-initial lexical projections have the same kind of branching structure but different head positions. The head position is a function of the licensing directionality.s In (5), the licensing direction is indicated by an arrow. The combination of the BC with the parametric options of licensing (either to the left or to the right) provides the grammatical reason that explains why complex head-initial projections involve shell-structures with a head-chain (cf. Raider 1 993: 28-30). The lexical projection in (2) is a head-final projection. The head and each node on the projection line follows the position to be licensed. So, the structure is simultaneously in harmony with the directionality of licensing and the BC. If, however, the directionality of licensing is progressive, that is, the position to be licensed has to follow, the BC structure provides a single position that meets the licensing requirement. This is the sister of the foot position of the head in (5). The other positions cannot be licensed unless the head is raised. In terms of Chomsky's (1995) framework, the structure (5) is a result of the merge function constrained by the BC: first a Spec-Read-complement configura tion is projected in accordance with the BC. This structure is merged with a head position in order to project another Spec-Read-complement structure. The result is a shell structure of the Larsonian kind (cf. Larson 1 988):
HUBERT HAIDER
50 (5)
The licensing requirement for a phrasal position in a projection is twofold. Licensing combines formal licensing plus identification: The phrase must be assigned to a possible phrase structure position and its dependency relation must be identified. Positions are formally licensed. The parametric direction of formal licensing is the trigger for the shell structure in head-initial lexical projections. Let us now compare the two alternative approaches: In both systems, movement to the right is blocked. The reason is straightforward: The structure presupposed or generated by movement to the right is characterized as ill-formed. In both systems, asymmetric c-command equals precedence. Since movement targets commanding positions, movement is to the left. The differences between the two approaches are easy to trace: In the LCA system, both in the version of Kayne (1994) and ( 1 998), OV is a derivative of a basic VO structure. In the BC system, the OV structure is a potential base structure (cf. structure (2) above). A complex head-initial projection of a lexical head is a shell-structure with a head-chain, as in (5). In order to check the empirical validity of each of these proposals, it is sufficient to derive distinctive, testable implications and check them. (6) lists a selection of areas of grammar in which the two approaches evidently differ. The relevant phenomena will be discussed in the following section.
(6) a. b. c. d. e.
V-head positions per VP VP-internal DP objects VP-internal = postverbal FP topicalization with Vfin universal Aux-V base order
LCA
BC
VO = OV *OV yes yes yes
VO > OV ,/ov
no no no
OV IS MORE BASIC THAN VO
51
(6a) refers to the difference illustrated by (2) and (5). In the LeA system, the V projection of a head-final VP is derived from a head-initial VP structure. Therefore the VP-internal structure is identical for both. In the Be system, however, a head-final V projection is well-formed with a single V position, that is, the verbal head in the foot-position of the projection. In a head-initial VP, the structure is potentially complex, since V chaining is necessary in order to license multiple argument positions. It will be argued below that the distribution of particles in English and Norwegian is a direct reflex of V chaining. (6b): According to the LeA system, the preverbal position of objects in an OV language is always a derived position (i.e. Spec-position of a functional head). In the Be system, the base positions of DP or PP arguments in a head final VP are preverbal. The differentiating prediction is evident: According to the LeA system, any argument is subject to the restrictions on derived positions. In the Be system, preverbal arguments in base positions are expected to behave like arguments in base-positions. In other words: The LeA system predicts that the arguments in an OV clause structure are subject to the kind of restrictions that apply to elements in Spec-positions (e.g. opacity restrictions). The Be system predicts that preverbal arguments in a head-final VP behave just like VP-internal arguments in a head-inital VP. (6c) addresses obligatorily VP-internal elements like selected adverbials or secondary predicates. According to the LeA theory, an obligatorily VP-internal element will surface in a postverbal position in an OV language. Obligatorily VP-internal just means that the element cannot and must not be fronted to a pre VP position. In the Be account, obligatorily VP-internal elements of a head-final VP are necessarily preverbal. (6d) concerns what is traditionally referred to as VP topicalization. In the Be system, it is VP topicalization after all. In the LeA system, the position of a preverbal object is in a higher functional projection. So, traditional VP topicalization must be reanalyzed as the topicalization of a functional projection. If the functional projection contains a link of the head-chain of the finite verb, crossing violations are expected in the LeA system, but not in the Be system. (6e) focusses on the order of auxiliaries. Since the universal base order in the LeA system is head-initial, the auxiliary precedes its complement VP. The VP is head-initial, too. So, the resulting base linearization is: auxiliary before main verb. In German, the linearization is in most cases the mirror image of the English order. In the Be perspective, this is a possible base order.
52 3.
HUBERT HAIDER Data and generalizations: against [OV]
=
[Oi [V ei]]
For ease of reference and for the sake of clarification, the issues listed in (6) will be rephrased as generalizations that ought to be captured by a descriptively adequate model of grammar. Generalization 1 : a. If a result predicate (particles included) may occur in more than one VP-internal position (i.e. adjacent or non-adjacent to the verb), the given language has a head-initial VP. b. In languages with V-final VPs, there is only one particle position for particle verbs, and it is preverbal and V-adjacent. 6 (cf. Haider 1 997 a for details) In English and in Norwegian, verb-particle combinations as well as combinations of a verb plus a resultative predicate occur in one of two serialization patterns: either adjacent or split. In Haider ( 1997a), the non-adjacent pattern is analyzed as the result of optional stranding.
(7)
a. b. c. d.
She cut the tree down carefully. The joggers ran the pavement thin. She cut down the tree carefully. The joggers ran thin the pavement.
(8)
a.
at han kastet matten ut that he threw carpet.the out 'that he threw out the carpet' at de skar kjrjJttet i biter that he cut meat. the to pieces 'that he cut the meat to pieces' at han kastet ut matten at de skar i biter kjrjJtet
b.
c. d.
(Nor)
If the stranding analysis is correct, the particle distribution in English and Norwegian is a direct reflex of verb positions on the V-chain in the V projection. In a stranding language there are as many particle positions as there are links on the VP-internal V-chain. 7
(9)
a. b.
[cu� [the tree [[ei down] carefully]]] [[cut down]i [the tree [ei carefully]]]
In German, which is representative for the Germanic OV languages in this respect, the particle position for verb-particle combinations is preverbal ( IOa-b vs. IOc-d), and adjacent (lOe).
OV IS MORE BASIC THAN va
( 1 0)
a. *daj3 er den Teppich waif weg that he the carpet threw out b. *daj3 er das Fleisch schnitt in (3 gleiche) StUcke that he the meat cut in 3 equal pieces c. daj3 er den Teppich weg waif d. daj3 er das Fleisch in (3 gleiche) StUcke schnitt e. *daj3 er den Teppich weg nichtlachtlos waif that he the carpet out not/carelessly threw
53 (Ger)
It is important to honour two facts: First, this pattern is not confined to particles but applies also to phrasal result predicates, as in ( l Ob, d). Particle incorporation, therefore, would not be an adequate solution. Second, in VO languages, particles and result predicates cannot move to clause internal positions preceding the VP: ( 1 1)
a. *The pavement was thi� [yp run ei] b. *The tree was dow� [yp cut ei]
The problematic aspect for the LCA account is this: If there are elements that do not obligatorily move out of the VP-internal position proper in a VO language, they are - ceteris paribus - expected to do so in an OV language: Given that particles and result predicates are VP-internal elements, they are incorrectly predicted to occur as postverbal elements in OV languages like German, which admit postverbal elements. In the BC account, the facts follow straightforwardly. Since there is only the basic V position in the head-final V projection, there is only one position for particles (cf. ( l Oc)) or result predicates (cf. ( lOd)). Generalization 2: Preverbal argument positions in VO clause structures are VP-external, pre-verbal argument positions in OV languages, however, display VP-internal properties. Phrases in uncontroversial Spec-positions of functional heads are opaque domains for extraction in general. But, as a matter of fact, preverbal positions in OV languages like German, scrambled or unscrambled, are not opaque. 8 So, they cannot be analyzed as Spec-positions of functional heads (without non-standard amendments to the theory, which would need to be independently justified). The contrast in (12), discussed in Kayne ( 1 983: 224), is representative for this generalization. VP-external positions in VO structures are opaque for wh extraction. Preverbal argument clauses in German, however, are not opaque, as (13a) illustrates. (13a) contains a subject clause. ( 1 2)
a. the book that it became difficult [to talk about e] b. *the book that [talking about e] became difficult
HUBERT HAIDER
54 ( 1 3)
Welches Buchi hat [ei zu lesen] dir mehr Spaj3 gemacht? which book has [to read] you more fun.Acc made 'Which book did you enjoy reading?' b. *Welches Buch sagte sie [cp [ei zu lesen] [habe [ihr which book said [she [to read] [has her a.
Spaj3 gemacht]]] ? fun made]]] 'Which book did she say she enjoyed reading?' Although extraction starts from within a subject clause in ( 13a), wh-extraction is possible in this case in German. The contrast between English and German becomes straightforward once it is realized that a German subject stays in its VP-internal position. 9 For a recent survey on this and related issues of German clause structure see Haider ( 1997b), Haider & Rosengren ( 1 998). In ( 1 3b), an embedded V2-clause, the extraction site is contained in a constituent fronted to SpecCP. In this case extraction is ill-formed, as expected. This shows that the standard opacity domains are operative in German. The various movement operations that produce the apparent OV order in the LCA system target functional Spec-positions. Thus, the preverbal XPs in an OV language are predicted to be as opaque for extraction as XPs in uncontroversia1 functional Spec-positions (as for instance the functional subject position in a VO language). These counterfactual implications for VO-to-OV do not arise, of course, if the structure of a head-final projection is identified as an underived, base-generated structure. Generalization 3 : Selected adverbials and secondary predicates are VP-internal in V O languages and preverbal in OV languages. The fact that an element can surface in a VP-internal position in a head-initial VP proves that there is no universal necessity for moving this element to a VP-exter nal position. For OV languages, VP-internal elements would appear postverbally, because VP-internal in OV is, according to the LCA approach (Kayne 1994), identical with VP-internal in a head-initial V projection. Elements that are obligatorily VP-internal in VO languages are particles, result-predicates, object related depictive predicates and selected manner adverbials. All these elements are ungrammatical in preverbal positions in uncontroversial VO structures in English, Romance or Scandinavian languages (cf. ( 14a'-d')). This proves at least that there is no universal trigger for moving these elements out of the VP.
OV IS MORE BASIC
( 14)
a. a'. b. b'. c. , c. d. d'.
THAN
55
VO
They handed it out *They out handed it They drank him under the table *They under the table drank him The meati was served ei raw *The meat was raw served They have lived frugally *They have frugally lived
In the LeA-geared VO-to-OV scenario, the postverbal elements in (15) would have failed to move. In the Be system they could not have been generated in a postverbal position from the beginning. The crucial problem for the LeA approach is that there is no grammatical reason at all for moving these elements: Particles must not move out of the VP (14a). Result predicates do not leave the VP (14b). Depictive predicates are obligatorily VP-internal (14c). Particularly instructive is the distribution of selected manner adverbials. Although a manner adverbial may be placed before the VP, this is not grammatical for selected manner adverbials. All these elements cannot be moved out of the VP in VO languages, so they ought to stay in their VP-internal position in OV languages as well, just like in the regular VO languages. The resulting pattern would be ( 1 5). ( 1 5)
a. *Sie haben es gehiindigt aus ( 14a) they have it handed out b. *Sie haben ihn getrunken unter den Tisch ( they have him drunken under the table c. *Das Fleisch wurde serviert roh ( 14c) the meat was served raw d. *Sie haben gelebt geniigsam ( 14d) they have lived frugally =
=
14b)
=
=
The examples in (15a-d) correspond to (14a-d), respectively. They are ungram matical just because the postverbal elements in ( 1 5) cannot appear postverbally in a head-final projection. The grammatical order is preverbal. To claim that they have to move obligatorily to the left would amount to merely begging the question. There is no OV language known in which these elements appear postverbally, 1O and what is crucial, there is no VO language in which these elements could move to the left, out of the VP. In fact, given the LeA system, one would expect that there are some VO languages in which these elements have to move out of the VP, because the triggers for the evacuation of the VP are in principle independent for each class of items. Why this is trigger obligatorily
HUBERT HAIDER
56
absent in VO languages but obligatorily present in OV languages is a crucial question without an answer at hand. So, we conclude with the insight that there is no straightforward way to derive the OV order from a VO structure for these elements. Generalization 4: Topicalized projections must not contain the trace of the finite verb. In the LCA approach to OV, preverbal elements are assigned to functional Spec positions above VP. What used to be analyzed as VP topicalization must now be analyzed as the topicalization of a VP contained in a functional projection that hosts the preverbal object. In this case, it is important to note that the F-project ion must be lower than the starting point of the head chain that leads to the V2-position of the finite verb. That this is so is easy to demonstrate with the contrast between ( 16a) and ( 16c).
( 1 6)
a.
b. c.
[Einen Fehler nachgewiesen] hat er ihm noch nie.
[a mistake PRT.proven] has he him ever never 'Given him proof of a mistake, he never ever has'
Er wiesj
ihm noch nie
einen Fehler nach-ej .
he prove� him ever never a
mistake PRT-ei
* [Einen Fehler nach-ej ] wiesj er ihm noch nie. [a mistake PRT- ei] prove� he him ever never 'Given him proof of a mistake, he never ever has'
The examples in ( 1 6) feature a particle verb. If the finite verb moves to SpecC, the particle is stranded ( 1 6b). ( 16c) shows that the stranded particle cannot be part of the fronted constituent. In other words, the trace of the finite verb must not be part of the topicalized constituent: ( 16c) demonstrates that the topicalized constituent cannot contain the trace of the finite verb. The particle in ( 16b-c) is an indicator for the position of the trace of the finite verb since the verb is one with a so-called separable particle (cf. ( 1 6b)) that is stranded by V2. Topicalizing the constituent that contains the trace of the verb incurs a crossing violation: The trace of the verb in (16c) fails to be in the c-command domain of the moved verb. With this in mind, let us proceed to the argument: If the topicalized constituent in ( 16a) contained the trace of the finite verb, it would be predicted to be ungrammatical, contrary to the facts. The conclusion must be, therefore, that the topicalized phrase cannot contain the trace of the finite verb. For ( 16a) this means that the topicalized constituent must be a complement of the auxiliary verb. It is only in this case that the finite auxiliary would not have to pass through the functional head of the functional projection that hosts the argument of the main verb (cf. (17)).
OV IS MORE BASIC
THAN
57
va
Unavoidable though this conclusion is, it is an unwanted conclusion, and so is the premise, therefore, which it is derived from: It is unwanted at least for the following reason: This solution is bound to overgenerate. The serialization of the structure (17) is ungrammatical, as the examples ( 1 8a-b) illustrate. The finite auxiliary follows the main verb ( 1 8c). (17)
a. b.
[Aux-VP VAux [FP XPj FO [vp VO . . . ej VP
• • •
]]]
�
VO-aux
FP
�
)Opi
F'
�
FO
VP
�
VO . . . � .
. .
According to this structure, the auxiliary verb in its base position licenses a functional projection as its complement. The Spec of the functional head hosts an element that is moved out of the head-initial VP. If there are more elements, there must be a cascade of functional projections, each of which gives room for one phrase. The result is the surface OV order within the FP. As the examples ( 1 8a-b) illustrate, the order in (17) results in an ungrammatical serialization. ( 1 8)
a. *daj3 er hat (ihm) einen Fehler nachgewiesen that he has him a mistake given.proof.of b. *daj3 er (ihm) einen Fehler hat nachgewiesen that he him a mistake has given.proof.of c. daj3 er (ihm) einen Fehler nachgewiesen hat that he him a mistake given.proof.of has
So, either the structure ( 17) is inadequate or the whole FP must move to a higher position. In the latter case, the FP would tum into an opaque extraction domain. ( 19a) would be predicted to be ungrammatical for the same reason that accounts for the ungrammaticality of extraction out of a fronted constituent in English, as illustrated in (19b): ( 1 9)
WOriiberj [hatj [er [ej gesprochen] ej]] ? what.about has he talked 'What has he talked about?' b. *the book [that [talking about e] he liked] a.
58
HUBERT HAIDER
On the other hand, structure ( 17) could be replaced by a structure in which the auxiliary selects a VP, not an FP (cf. Zwart 1 993), with the surface verb order derived by verb-raising (head-head adjunction). But in this case, the functional projection targeted by VP-intemal elements would be higher than the auxiliary position. So the finite auxiliary would have to pass through the corresponding functional head position. The result would be that (16a) is parallel to ( 16c). In both cases the fronted constituents would contain the trace of the finite verb, and both are predicted to be ungrammatical. There is still another but equally unsuccessful derivational trail that starts with (17) and leads to the word order in ( 1 8c). This is the combination of VP evacuation and the fronting of the remnant VP, as suggested by Kayne ( 1998) for overt quantifier movement in English. The problems remain, however. Since the remnant VP is placed between the evacuated material and the auxiliary, the evacuated material necessarily c-commands the auxiliary, and so do the function al heads that host the material. If the auxiliary moves to the top F-head in the V2-clause, 1 1 it will pass through these head-positions and the ensuing crossing violations cannot be avoided when the minimal constituent that contains the object and the main verb is fronted. Generalization 5: Universal #16 and its extension to auxiliary sequences 12 A final piece of evidence is the order of auxiliaries. The German order, which is representative of OV systems in that respect, is the mirror image of the English sequence. 13 If the OV order is the derivational result of a basic VO order, the base orders of auxiliaries in a VO structure and in an OV structure are predicted to be identical. The evacuation of VP, which produces the OV order, leaves the serialization of verbs unaffected (cf. (20)). If they differ nevertheless, this must be attributed to an additional move in the derivation (20c): (20)
b.
[Vaux [vpVO XP ]] [XPi [Vaux [vpVO ei]]]
Evacuation
c.
XP VO Vaux
Aux-V-reordering (details below)
a.
VO order
In German, the grammatical order (21b) is the mirror image of the English order (21 a). The order parallel to the English one (2Ic) is ungrammatical. In the LCA system, (21b) must be derived from (21 c). In the BC approach, (21b) is a possible base order: Each VP complement is licensed by an auxiliary as the head of a V projection and the VP complement precedes since a verbal head licenses the complement to the left.
OV IS MORE BASIC THAN va
(21)
a.
They [must [have [been [persuaded by him.]]]]
b.
daB [[[[sie von ihm uberredet vp] worden yp] sein vp] that they by him persuaded been have mussen yp] ( 2 1 a) must
59
=
c. *dafi sie miissen sein worden iiberredet von ihm that they must have been persuaded by him The following discussion is meant to highlight the difficulties of integrating this word order property into the LeA system. Of course, there cannot be a demon strative proof that it could not be implemented somehow. The crucial point is just this: In the Be account the German pattern falls out as the expected pattern for head-final languages. This is in accordance with what is known about OV languages. In the LeA account, additional derivational machinery is necessary. Dutch has been claimed to clearly support the LeA approach towards Ov. That this is not so, and that the V-raising phenomenon is a neutral ground at best for a decision between the LeA system and the headedness approach towards OV NO, is what the following paragraphs are meant to indicate. Two possibilities come to mind that could lead to (21b), namely either V-raising or iterative VP movement to the left. Zwart (1993), who primarily analyzed Dutch, advocates a V-raising approach. The auxiliary order is derived by iterative head-to-head adjunction (cf. Zwart 1993 Sect. 2.4). It is possible to derive the order in (21b) from the order in (21 a, c) by means of the devices suggested by Zwart, but the theoretical and empirical soundness remains to be demonstrated. There are several problem areas that undermine this approach as a viable analysis in a Kaynean framework. First, multiple head-to-head adjunction cannot be assumed under the premises of the LeA system, because multiple head-to-head adjunctions are explicitly ruled out by Kayne ( 1994: 8 , 1 9). Secondly, Zwart (1993: 335f.) is forced to admit non-local adjunctions to heads in a derived position in order to capture optional, alternative serializations in the verb cluster in Dutch and German IPP-constructions in which the finite verb is sandwiched by two non finite verbs: (22)
a.
dat Jan verteld zal hebben, dat hij Marie gekust heeft (Dut) that John told
b.
will have
that he Marie kissed has
dafi er sie grufien hiitte mussen that he her greet had must 'that he should have greeted her'
(Ger)
60
HUBERT HAIDER
In terms of the solution proposed by Zwart ( 1 993 : 337), the past participle has moved across the non-finite auxiliary in (22a) and the modal in (22b), respective ly, in order to adjoin to the finite verb in the position of the functional head. But this is in conflict with two standard assumptions: First, head movement is subject to minimality (i.e. the head movement constraint), so the intervening auxiliary should block this movement. Second, if it left-adjoins to the finite verb in the position of an F-head, V-second should transport both verbal elements to SpecC, and not just the finite verb (cf. the movement of finite verbs with adj oined clitics). The result would be ill-formed in Dutch and German, however. If the adjunction approach is hard to maintain, VP movement seems to be the only option left: Starting with the hypothetical base order (2Ic), the order (2 Ib) should be the result of evacuating the VP plus moving every VP out of the VP that contains it to a position preceding the mother VP. The intermediate derivational steps are illustrated in (23). (23a) is the outcome of the evacuation of the proj ec tion of the main verb. The next steps are less straightforward. Let us assume that the finite verb moves to a functional head position (23b), and the remnant VP is fronted (23c). How do we get the inverse order within the fronted VP? (23)
a.
*daj3 siei von ihmj [VP l miissen [VP2 sein [VP3 worden [VP4 ei that she by him
must
have
been
iiberredet ej]]]]
persuaded
b. *daj3 sie von ihm miisseni [VP l ei [VP2 sein [VP3 worden [VP4 e iiberredet e]]]] c.
*daj3 sie von ihm [VP l ei [VP2 sein [VP3 worden [VP4 e iiberredet e]]]] miisseni
Since adjunction is banned in the LCA system, the desired order cannot be achieved by VP fronting. More structure is needed. The mirror order can be obtained if each auxiliary projects at least one separate functional projection, whose Spec will then host the fronted complement. (24) illustrates the derivation for the most deeply embedded VP. (24)
a.
[FP Spec [Fo [VP3 worden [VP4 e iiberredet e]]]] been
b.
persuaded
[FP [VP4 e iiberredet e]j [wordeni [VP3 ei ej]]]]
The FP in (24b) will have to move into the Spec of the functional projection whose complement is the VP of the auxiliary sein, as illustrated in (25).
OV IS MORE BASIC THAN VO (25)
a. b.
61
[FP- l Spec [Fo [vp sein [FP-2-]]]] [FP-l [FP-2 -]j [seini [vp ei ej ]]]
The result would be a complex of stacked F-projections in the Spec o f an F-projection. But the complexity of the derivation is not the problem. The real problem is an empirical one. The assumption that the VP and the higher auxiliary-verb projection are each stacked in separate F-Specs overgenerates: (26a) shows that the extraposed object clause can be part of the topicalized phrase that contains only the main verb. But, independent of topicalization (cf. (26b, d)), the extraposed clause cannot intervene. This would be possible, however, if in (26b, d) there were a VP or a higher projection in a SpecF position. In this case, (26b, d) would be expected to pattern parallel to (26a). (26)
a.
[Sagen, [wo
sie wohnt]], wurde er dir nicht konnen
[tell where she lives] would he you not be.able 'To tell you where she lives he would not be able' b. * [ [Sagen, wo sie wohnt] konnen] wurde er dir nicht [tell where she lives be. able] would he you not c.
[Sagen konnen, wo sie wohnt], wurde er dir nicht [tell be.able where she lives] would he you not
d. *Er wurde dir nicht [sagen [wo sie wohnt]] konnen he would you not tell [where she lives] be.able The VP-movement analysis misses an important property. It does not capture the 'verb cluster property' : The clause-final sequence of non-finite verbs in a simple clause in German must not be interrupted by nonverbal material. It would be unfair to withhold that the Be-based account needs a separate condition for capturing the verb-cluster property. But the crucial difference is that the LeA approach has no consistent way of implementing this: In the Be system, the order restrictions for (26a) and (26d) are different, because (26d) must be a possible base order. In the LeA system, however, (26d) and (26a) are instances of orders derived by the same type of movement, that is, movement of V-projec tions into functional Spec-positions. In sum, there are central data areas that bear on the empirical validity of the LeA account and the Be approach. If put to the test, it turns out that central implications of the LeA system with respect to the analysis of OV structures are not compatible with the full range of the empirical evidence.
62
4.
HUBERT HAIDER
Structure and linearization: What triggers what?
The LeA account is too strong and too weak at the same time: It is too strong because it employs more derivational machinery than justified by independent evidence and it is too weak because it does not provide a general enough trigger theory for the OVNO-parametrization. As for the additional machinery, it should be noted that the LeA account and the Be account of complex head-initial projections share one essential device, namely head chaining: In double object constructions, for instance, the verb in the VP-initial position is the head of a head chain. In addition to head chains, the LeA system must invoke phrasal movement in order to derive the OV structure from the VO structure. In the Be system, head movement is the only device. Head movement is sufficient to capture both, the principal differ ence between the OV and VO structures and the concomitant difference in the internal organization. The evidence discussed above points to the conclusion that the more economical approach is at least empirically adequate. The lack of independent empirical evidence for the need of phrasal movement in view of systematic counterevidence is a non-trivial challenge for the LeA strategy. The lack of a trigger theory is a serious deficit. In the absence of a trigger theory, the LeA theory predicts that natural languages are head-initial languages : The O V order i s the result o f evacuating the basic head-initial projection, but this evacuation must be triggered. In the LeA system, the trigger must be a global one because it indiscriminately affects all VP-internal elements, independent of their category and grammatical function: arguments, attributes (including relative clauses), adverbials, particles, (secondary) predicates. Holmberg (this volume) proposes c-selection (i.e. strict subcategorization) as the parametric locus for the global trigger of the respective movement types: The basic difference between OV and VO languages is in his view reducible to a difference in terms of checking c-selection features: In OV languages, c-selec tion features are to be checked by phrasal movement, but in VO languages by head movement. As he acknowledges, this does not account for the serialization of non-subcategorized elements, such as PP adverbials, adverbial clauses, extraposed relative clauses, etc. without some non-standard assumptions. What is more problematic however, is the existence of mixed language types, such as for instance German: N-, P- and functional projections are head-initial, V- and A-projections are head-final. Since the trigger is framed as a computational property of the different implementation of a feature checking mechanism, parametrization according to category subclasses would amount to a parametri zation on the computational level. In addition, this account is open to the issues
OV IS MORE BASIC THAN va
63
raised against an account that takes head-initial structures as a starting point for deriving head-final ones. The trigger for a head movement structure (i.e. a complex head initial lexical projection) in the Be-based approach is the conflicting set of demands for a convergent structure assignment posed by a serialization with an initial head: A convergent structure must be binary branching and it must comply with the Be and the endocentricity requirement, that is, there must be a head position at the bottom of the projection in a right-branching structure. Given these demands, what is the convergent structure for a serialization like (27a)? The solution is (27b): The convergent structure for (27a) is a structure with head movement: This is the only way to simultaneously have a head-position at the foot of the projection and a lexical head element in initial position without violating the Be. Each head position in (27b) is the local directional licenser of its complement, and for each phrase in the projection there is a local head. In a head-final structure (27c), each phrase has a sister that is either the head or a projection of the head in the canonical licensing configuration and, what distinguishes (27c) and (27b), (27c) meets the Be with the head in the base position. (27)
a. b. c.
[HP hO XP YP ZP] [hOi [XP [ei [YP [ei ZP]]]]] [XP [YP [ZP hOH'] H'] HP]
This way of describing the triggering situation may appear strange because it characterizes the problem from a representational point of view: Given an array of terminals, what is the minimal convergent structure? In this perspective, UG is a complex cognitive capacity of symbol processing recruited for representa tional, that is, projective robustness. It enables the learner to assemble the knowledge system called core grammar. The system of representations and principles of core grammar is the recursive solution of the projection problem for a given natural language L, that is, the function from one-dimensional expres sions ( a string of terminals of L) to an at least two dimensional expression ( the grammatical structure of the string). The core grammar determines the projection of a syntactic structure onto a given string of terminals of L. The solution of the projection problem is the criterion of empirical adequacy for grammar theory. The solution is the algorithm that maps strings of L onto well formed structures. From the point of view of the Be hypothesis, the primary question is 'What is the structure for a given serialization?' and not 'What is the serialization for a given structure?' The Be is an attempt to model a principle of structure =
=
64
HUBERT HAIDER
assignment provided by UG that narrowly restricts the construction space for projections, given an array of terminals: Precedence in linear order corresponds to c-command in structure. The LeA perspective is a perspective of structure disambiguation: UG is supposed to provide a principle that organizes the structure in such a way that the terminals of a given phrase structure can be mapped in a unique way on a linear ordering. Kayne ( 1 994) chose to implement this as a function that maps asymmetric c-command relations on precedence relations. This is only one possible implementation out of many possible alternatives, however. Asymmetric c-command could be mapped on succession in linear order rather than on precedence. The result would be a grammar with consistently left branching structures. 14 Moreover, a grammar in which each head is parametrize able for either precedence (i.e. Spec-head-complement) or succession (i.e. complement-head-spec) would also provide a well-formed system of structure-to string mapping. In other words, the branching directionality is not an integral part of Kayne's system. It is first of all an empirical issue, and secondly, a fact about grammars to be modelled in a theory of UG. A final area of difference in a meta-theoretic perspective is the level of applicability of the respective constraints: The Be is a universal local constraint on structure projection or merger. The LeA, however, is a global constraint. It is a prespecified theorem rather than an axiom. What it amounts to is this: The LeA does not constrain structures but first of all gramma rs. The LeA is a criterion of success for grammars. Grammars must be organized such that they obey the LeA. They obey LeA if the structures they admit fulfill the mapping requirements spelled out by the LeA. So, a grammar has to conspire to meet the LeA: The conspiracy combines constraints on structures (Spec before head before complement), distinctions on the projection level (between categories and segments), constraints on merger (at most one adjunct), and constraints on movements (e.g. no adjunction to the right). A local constraint on admissible data structures, as the Be, is necessarily less far-reaching in its theory-internal implications but it is as strong a claim on the level of the cognitively adequate modelling of a UG property as the global benchmarking of grammars is.
Acknowledgments I want to thank Anders Holmberg and Peter Svenonius for many helpful comments. Remaining shortcomings are of course to be blamed on the author.
OV IS MORE BASIC THAN va
65
Notes 1.
Linear correspondence axiom (LCA): d(A) is a linear order of T, for a given phrase marker P, with T the set of terminals and A the maximal set of ordered pairs (�, Yj), such that for each j , � asymmetrically c-commands Yj. (Kayne 1994: 5f.)
2.
The BC is axiomatic in the following sense: If correct, it describes a universal property of data structures of human grammars. The property modelled by the BC is considered to be a primitive element of UG. The axiomatic element of the LCA system is the postulate that asymmetric c-command is mapped on precedence. The system itself does not break the symmetry.
3.
There are at least two areas of facts that are not yet satisfactorily integrated, namely, the order and structure of postverbal adjuncts in VO languages, and the structure of mirror-image VO languages (cf. Pearson, this volume). The order of postverbal adjuncts in English (V-respect > process > space > time [cf. Quirk et al. 1 986: §8.87]) is the inverse of the order of adjuncts in German, which are preverbal. If it turns out that the traditional account of right-adjunction is correct, the scope of the BC would have to be narrowed in to the projection of the A-structure and the functional extension of a lexical projection.
4.
So, scrambling by movement to the right is ruled out. The phenomena discussed under the heading 'Scrambling to the right' are better analyzed as the result of V movement to the left (cf. Haider 1997c; Mahajan 1997). Extraposition by movement to the right is ruled out by the same reason. In Haider (1994, 1997c), extraposition is analyzed in terms of V movement to the left. The type of V movement in these constructions is an instance of the kind of V movement found in head-initial lexical proj ections.
5.
Licensing in this context is to be understood as a structural relation of node-admissibility in the overt projection structure: For a given projection P with the head pO, the sister node of pO and the sister node of each intermediate projection of pO is subject to a directionality requirement. Depending on the parametrization, the head either precedes or follows the nodes to be licensed.
6.
If the verb is finite, this particle is stranded by V-to-C. Adjacent is but a descriptive character ization: The only elements that may intervene between the particle and the verb proper are elements of the inflectional morphology: The ge-prefix of the participial morphology and the zu-prefix of the infinitival morphology (cf. Hinterh61z1, this voL). These prefixes cannot be stranded.
7.
As noted by Den Dikken (1992: 38) and others (e.g. Fraser 1974; Svenonius 1992), particles in the medial or V-adjacent position cannot be modified (cf. b, d). In these positions particles are XO_ elements that combine with the verb. Only in the final position can a particle be treated as phrasal: a. Mike tossed me the wrench (right) up Mike tossed me (*right) up the wrench b. c. He threw the ball rightlstraight up/back/down He threw (*rightlstraight) up/back/down the ball d.
8.
Local scrambling, as in German, is assumed to be a VP-intemal phenomenon, whence the transparency for extraction. A detailed account of this approach in comparison to competing accounts is presented in Haider & Rosengren (1998).
9.
Since subjects remain in their VP-intemal position and since scrambling is confined to VP-intemal positions as well, neither extraction out of scrambled elements nor out of subjects
66
HUBERT HAIDER is affected by an opacity restriction that applies to Spec-positions.
10.
There seem to exist languages in which arguments precede the verb and the rest follows, for instance Bambara (cf. Diallo 1 987). However, this does not contradict the claim just made. These languages are VO languages in which all arguments, not only the subject, are moved to functional Spec-positions overtly. Non-arguments remain in their VP-internal positions.
11.
Movement of an AuxP rather than the auxiliary itself would not help, since the auxiliary would be confined to the moved phrase. If it were allowed to move out, the ungrammatical German patterns discussed above would reappear.
12.
In languages with dominant order VSO, an inflected auxiliary always precedes the main verb. In languages with dominant order SOV, an inflected auxiliary always follows the main verb (Greenberg 1963).
13.
A systematic exception is the so-called IPP construction (infinitive instead of participle) with modals, some perception verbs, and the causative verb lassen (= let, make): (i) daB er es nicht lesen gekonnt hat that he it not read can has 'that he has not been able to read it' (ii) daB er es nicht hat lesen konnen Normally, the finite verb occurs at the left side of the verb cluster or in some cases even further to the left. But it is not ungrammatical if it it just changes place with the modal, i.e. in the order lesen hat konnen (see below in the text; see also Hinterholzl, this volume).
14.
The time-slot hypothesis (Kayne 1994: 37) does not provide the independent empirical motivation that is necessary for the breaking of symmetry in his system. This hypothesis is itself in urgent need of empirical justification.
References Chomsky, Noam. 1995 . The Minimalist Program. Cambridge, MA: MIT Press. Dikken, Marcel den. 1992. Particles. Leiden: Holland Institute of Generative Linguistics. Diallo, Mohamed L. 1987. Zur Verbalstruktur und Syntax des Bambara. Ph.D. dissertation, Universitat Bayreuth. Fraser, Bruce. 1974. The Verb-Particle Combination in English. Tokyo: Taishukan. Greenberg, Joseph. 1963. "Some universals of grammar with particular reference to the order of meaningful elements." In Joseph Greenberg (ed.), Universals ofLanguage. Cambridge, MA: MIT Press. 73-1 1 3 Haider, Hubert. 1992. "Branching & discharge." In Working Papers of the Sonderfor schungsbereich 340 (Universities of Stuttgart and Tlibingen) 23, 1-3 1 . (In press in Peter Coopmans & Martin Everaert & Jane Grimshaw (eds), Lexical Specification and Insertion. Amsterdam and Philadelphia: John Benjamins. 1 35-164. Haider, Hubert. 1993. Deutsche Syntax Generativ. Tlibingen: Niemeyer. Haider, Hubert. 1994. "Detachment - the later, the deeper." Working Papers of the Sonderforschungsbereich 340 (Universities of Stuttgart and Tlibingen) # 4 1 . Haider, Hubert. 1997a. "Precedence among predicates." The Journal of Comparative Germanic Linguistics 1, 3-4 1 .
OV IS MORE BASIC THAN VO
67
Haider, Hubert. 1997b. "Projective economy. On the minimal functional structure of the German clause." In Abraham, Werner & EUy van Gelderen (eds), German: Syntactic problems problematic syntax. Tlibingen: Niemeyer. 83-103 . Haider, Hubert. 1997c. "Extraposition." In Beerman, Dorothee, David LeBlanc & Henk van Riemsdijk (eds), RightwardMovement. Amsterdam: John Benjamins [Linguistics Today 17]. l I S-lS I . Haider, Hubert and Rosengren, Inger. 1998. "Scrambling." Sprache & Pragmatic #49. Lund University. Kayne, Richard. 1983. "Connectedness." Linguistic Inquiry 14: 223-249. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kayne, Richard. 1998. "Overt vs. covert movement." Syntax 1: 219-19I. Larson, Richard. 1988. "On the double object construction." Linguistic Inquiry 19: 335-39 I. Mahajan, Anoop. 1997. "Against a rightward movement analysis of extraposition and rightward scrambling in Hindi." In Tonoike, Shigeo. (ed.), Scrambling. Tokio [Linguistics Workshop Series #5]. 93-124. Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan 1986. A Compre hensive Grammar of the English Language. London: Longman. Svenonius, Peter. 1992. "Movement of pO in the English verb-particle construction." In Black, Andrew & James McCloskey (eds.), Syntax at Santa Cruz, vol. 1 . Santa Cruz: Syntax Research Center, UCSc. 93-1 13. Zwart, Jan-Wouter. 1993 . Dutch Syntax. A minimalist approach. Ph.D. dissertation, Groningen University.
Remnant Movement Liliane Haegeman University of Lille III
Abstract This paper focuses on the derivation of the verb-final pattern in the West Germanic OV languages. The paper concentrates entirely on West Flemish. An antisymmetric approach which assumes short V-movement or no V-movement at all is compared with more recent antisymmetric approaches which postulate both V-to-I movement and remnant XP-movement. I will show that an analysis which postulates absence of V-to-I movement and one-by-one leftward feature driven movement of the V-complements seems to entail a number of funda mental syntactic differences between the West Germanic OV languages on the one hand and the Scandinavian languages, English and the Romance languages on the other hand. These differences have to be stipulated. The analysis elaborated below is in line with Kayne's ( 1994) LeA. I propose that the West Germanic OV order be derived by (i) morphology-driven V-movement to a functional head and (ii) remnant movement of the extended projection of V. This analysis brings the West Germanic OV languages in line with the other languages examined.
1.
Introduction: Antisymmetry and OV order
Embedded clauses in the West Gennanic OV languages such as Dutch, Gennan, and West Flemish (WF), the language focussed on here, have a surface verb-final pattern, as illustrated in WF ( 1 a). (1)
a.
da Valere nen hoek no Gent stiert that Valere a book to Ghent sends 'that Valere sends a book to Ghent'
LILIANE HAEGEMAN
70
In universal base approaches based on Zwart ( 1 993), it is assumed that the com plement-verb order is derived by one-by-one feature-driven leftward movement of the complements of the verb. For instance, in ( 1 a) the indefinite direct object nen boek ('a book' ) moves to SpecAgrOP; the predicate PP no Gent ('to Ghent' ) moves to SpecPredP (Hinterh61zl 1 999; Koster 1 994; Zwart 1 993). Since the finite verb follows the object in SpecAgrOP and the constituent in SpecPredP, the sentence-final position of the verb is taken to be relatively low. Either the verb remains in its base position ( 1 b), or it moves to a low functional head, such as Pred, for instance, in terms of Koster's ( 1 994) analysis ( l c). (1)
b. c.
da [IP Val. [IO] [AgrOP nen boeko [PredP no Gentpp [yp [vo stiert] to tpp]]]] da [IP Val. [IO] [AgrOP nen boeko [PredP no Gentpp [PredO stiert] [vp [vo tv] to tpp]]]]
The consensus is then that the West Germanic OV languages do not display morphology-driven V-to-I movement in the sense of Pollock ( 1 989) and later work. While in Haegeman ( 1 998a, 1 998b) I myself did argue for morphology driven finite V-movement to T and to Neg in WF I continued to assume that the landing site of V was not the 'traditional' high T-head, but rather a second, lower T-head. I will reconsider this analysis below. The present paper re-evaluates the antisymmetric approach illustrated in ( lb-c) in comparative terms. I concentrate on WF but I assume that the analysis carries over to the other West Germanic OV languages. I will show that an analysis which postulates absence of V-to-I movement and one-by-one leftward feature-driven movement of the V-complements implies a number of fundamental syntactic differences between the West Germanic OV languages on the one hand and the Scandinavian languages, English and the Romance languages on the other hand. These differences have to be stipulated. The analysis elaborated below is in line with Kayne's ( 1 994) LeA and parallel to Pearson's analysis (this volume). I propose that the West Germanic OV order be derived by (i) morphol ogy-driven V-movement to a functional head and (ii) remnant movement of the extended projection of V. As we will see, this analysis brings the West Germanic OV languages in line with the other languages examined. The paper is organised as follows. Section 2 summarises a set of problems raised for the analysis outlined above in the light of comparative syntax. These problems have been signalled in the literature. Section 3 introduces a set of additional problems concerning the (comparative) syntax of negation. Section 4 provides an alternative analysis which derives OV orders in terms of double movement. Section 5 shows that the problems raised in Section 2 are solved by ,
REMNANT MOVEMENT AND OV ORDER
71
this analysis. Section 6 briefly examines another implementation o f the anti symmetry approach elaborated by Zwart (1 996). The major problem with this analysis concerns the syntax of infinitival clauses. Section 7 is a brief conclusion to the paper.
2.
Problems with the analysis
This section will show that the analysis with short or no V-movement and one by-one fronting of the complements of V gives rise to paradoxes for a compara tive approach to syntax. Two broad areas will be considered: constructions involving argument positions and constructions involving sentential negation. 2. 1
V-morphology and V-to-/
Based mainly on comparative data drawn, among other places, from the German ic VO languages and from the diachronic development of English, it is generally assumed that there is a correlation between V-movement and inflectional morphology (Roberts 1 985; Pollock 1 989; Rohrbacher 1 994; Holmberg & Platzack 1 995; Vikner 1 997; Ackema 1 999; etc.). The West Germanic OV languages Dutch, German and their dialects, have a fair amount of overt inflectional morphology, which would lead one to expect V-to-I movement (see B obaljik & Thrainsson 1 997: 66; Vikner 1 997: 1 92). Based on the agreement morphology of the relevant languages, Kayne ( 1 994) himself suggests that V-to-I movement must be assumed for the West Germanic OV languages: Since German and Dutch show person agreement, it would follow that in those languages, too, V raises to AgrS . . . . this raising must be thought of as leftward, just as for Icelandic, French, and Italian. The specificity of German and Dutch lies in their requiring most of their complements to move leftward past V. Since the finite V in German and Dutch must raise to AgrS, by the preceding argument, I am led to conclude that in those languages the leftward-moved complements must end up to the left of the AgrS position. (Kayne 1994: 52) Concerning the derivation of the embedded OV order Kayne adds: That leftward movement of complements in German and Dutch does not simply take place within the V projection is reinforced by two observations. First, when the verb is infinitival, the complements must precede not only it but also the infinitival marker zulte, which I take to be heading a separate projection, as with English to. Second, the leftward moved complements in
72
LILIANE HAEGEMAN West Flemish precede not only the verb but also the preverbal negative clitic, which is almost certainly as high as NegP. (Kayne 1994: 52)
While Kayne himself advocates V-to-I movement, this is not the position adopted in Zwart (1993) and those following his approach. As mentioned, if V is taken to move overtly at all, it is taken to move to PredO (cf Koster 1 994) and no link is established between the overt movement of the verb and its morphology. In antisymmetric approaches there also seems to be a tacit agreement that sentence final finite verbs occupy the same position as infinitives. The fact that infinitival morphology is poorer than the finite morphology is not taken into consideration. 2.2 Diesing 's generalisation: syntax and interpretation of indefinite objects A second problem for the antisymmetric approaches to the West Germanic OV order concerns DiesiIlg's generalisation, which postulates a correlation between i a.! �10 ���e aIlc! ,VP-inteJ:n.Cl.1 positiQns. In the more traditional OV ap i , , " ,�� �,!�12� proaches, indefinite pre-verbal objects such as nen boek in ( 1 a) were taken to occupy their VP-internal base positiop"s , ,!() t.l1� left of the verb. Such an analysis ' was in line with Diesing; s scope-related analysi� o{ihe distribution of object DPs ( 1992, 1 996, 1 997). On the basis of German data such as those in (2), Diesing proposed that indefinite objects with existential readings relIl,aip withill yg ,' where they are suojecfto e'-xisteiiiliil closure. '
(2)
a.
weil Elly immer Lieder singt
, . ._,.
(Diesing 1 996: 5)
since Elly always songs sings 'since Elly always sings songs' b.
weil Elly Lieder immer singt since Elly songs always sings 'since Elly always sings songs'
In (2a) indefinite Lieder ('songs' ) is VP-internal and receives an existential reading. In (2b), it has moved out of the VP (hence it precedes the adverbial immer ('always'» and it receives a strong reading. 1 See also de Hoop 1992. In Diesing's approach, the distribution of indefinite object DPs in the West Germanic OV languages was analogous to that in Scandinavian. For reasons of space I will not elaborate the data here, and refer the reader to Diesing's own discussions. Diesing's analysis also accounts for the asymmetries in was for extraction illustrated in (3). In a GB approach, extraction from the object DP in the VP-internal base position is grammatical (3a); extraction from the object in a moved position (3b) is interpreted as an ECP-viohitioll.
REMNANT MOVEMENT AND OV ORDER (3)
a.
73
Wasi hat Otto immer [yp [ti flir Bucher] gelesen ?]
what has Otto always for books read 'What kind of books did Otto always read?'
b. *Wasi hat Otto [ti flir Bucher] immer [yp [t] gelesen ?] (Diesing 1 992: 1 1 8-1 20) According to the antisymmetric view the object in OV languages always moves to SpecAgrOP to check its accusative case (den Dikken 1 996; Van den Wyn gaerd 1 989; Zwart 1 993, 1 996). It is hence not possible to make the leftward movement of the object dependent on features such as [±specific] . This means that the distribution of the indefinite object in the West Germanic OV languages would have to differ substantively from the Scandinavian patterns discussed by Diesing, where movement to SpecAgrOP forces a strong reading, as also expressed by the following citation from Hinterh61zl: nominal complements of the verb always have to leave the VP before Spell-out (independently of whether they are definite or indefinite) and are licensed in functional projections (HinterhOlzl 1999: Ch. 4, p. 42; this volume, p. 304).
There is, of course, a solution to this problem but one that is unsatisfactory in that it creates further problems for Germanic comparative syntax. If all nominal complements move leftward in OV languages, we might restate Diesing's generalisation in terms of the degree of leftward movement and propose that existential indefinites move somewhat less than strong indefinites. The asymme try in the was flir extraction data would then also have to be accounted for in terms of the 4egree of movement of the object. This solution is not satisfactory because it leacis to new problems with respect to the syntax of the Germanic VO languages. ()ne w0l..lld have to assllllle that either indefinite object,s.,�!th. existen tial readings" do" not move ln Scandinavian, or else tlieY UIldergo . similar short movement. . In the" former option, the syntax �f S'candi�a�ian indefirrlte objects �wouId'-remain different from its West Germanic counterpart, an unexpected result. In the latter case, the syntax of indefinites in Scandinavian would receive the same analysis as that in West Germanic, but the postverbal position of Scandinavian existential indefinites will imply that the verb itself must now be taken to move leftward. This conclusion again would lead to a number of modifications in our assumptions concerning the grammar of the Scandinavian languages. In particular, the question would arise as to the trigger of V-move ment. (But see Hr6arsd6ttir 1 999, this volume for a different analysis.) The analysis below will allow us to maintain Diesing's generalisation in full and to assume that the syntax of the Scandinavian indefinite object is identical
74
LILIANE HAEGEMAN
to that of its West Germanic counterpart, i.e. the existential indefinite object remains VP-internal. 2.3 Holmberg 's generalisation The syntactic operation responsible for the movement of the object out of the VP in ( 1 a) is generally taken to be object shift (Van den Wyngaerd 1989), which is also at the basis of the leftward object movement in the Scandinavian languages. For the latter, object shift tends to be correlated to V-movement (Holmberg 1986: 176, Chomsky 1 993: 1 8, Zwart 1996: 237-243) . Zwart ( 1996) reformulates the generalisation as in (4): (4)
Holmberg's generalisation Object shift of an element ex from the complement position of a verb 13 to the specifier position of 'Y, the AgrOP associated with 13, is possible only if 13 moves to the head of 'Y (Zwart 1 996: 238)
An approach i� which tll,ere jS )1Q , Qr yepj )J�stricte.d. �Y�to�I movelIlel1t in the , . Germanic OV· langu �ges implie s. that Holmberg's , ��p�
TP1 > NegPl > . . . NegP2 AspP Vfin pas ne nen pi nen
b.
WF
SU >
NegPl AspP
nie
nie meer
> TP2 > NegP2 > . . . Vfin en
This hypothesis is not satisfactory, in that it postulates an unexplained difference in the hierarchical organisation of the negative projections. The semantic correlates of the hypothesis are also undesirable: th� _ st:1:l?:f:lQ!iS,JEP99-Q]}$-8t-fu� ne�a,tiye_ ._p:roj�cti()Ils . Negg� .,.:wd NegP2 .become'-reversed. ,;inc,the · two -language �g91lp�. :Zanuttini sh�w-� that' in Romance NegPl is associated with what she calls -----·�' p�e��pp ositional negation, i.e. where the negation in fact reverses the presup posed positive polarity of the clause. As a matter of fact, WF en- in the lower Neg2, also expresses presuppositional negation (see Haegeman 1 999 for discus sion). Typically spelling out en- emphasises the polarity of the clause and contrasts it with a presupposed polarity. In some well-defined contexts, en- may ..
82
LILIANE HAEGEMAN
be even used without an accompanying negative constituent in which case it purely functions as a polarity reversing element. (21)
a. b.
Valere ee nog nie betoald vu zen gezette.
for his newspaper Valere has yet not paid 'Valere has not paid for his newspaper yet.'
J' en-doet.
he EN-does 'He has.'
Semantically then, the low WF NegP2 in (20b) would have to have the function of the high Romance NegPl in (20a). A further drawback of the hypothesis in (20) is that temporal and aspectual adverbs would also have rather different distributions in the two language groups. In Romance, they are c-commanded by the presuppositional negation; in Germanic they c-command the presuppositional NegP. Such oppositions go against current attempts to propose a uniform analysis of adverbial positions related to their semantics (Cinque 1 999). I consider the parametric variation in the hierarchical organisation of projections in (20) to be undesirable. It would be preferable to assume that such specialised projections correlate with some semantic property and have a universal hierarchy. To solve the paradoxes in the syntax of negative sentences, I continue to follow Zanuttini ( l997a, 1 997b) in postulating at least two NegPs in the clausal domain. One NegP is associated with negative markers and adverbials and a second NegP is associated with the negative head and relates to sentential polarity. The latter NegP could plausibly be relabelled PolP (Cormack & Smith 1998). I assume that the hierarchy of projections is that found in Romance (and in English cf. Cormack & Smith 1 998), with the polar NegP, NegP1 , c-com manding the adverbial NegP, NegP2. On the basis of the semantic functions associated with the two projections and following Zanuttini's analysis of the distribution of the temporal and aspectual adverbials, I propose that the WF IP displays the hierarchy in (22). (22)
TPI NegPl FP TP2 NegP2 AspP
polarity
en
adverbial
a
niet
AspP
nie meer atent
N�gP l,-, head�d , PY , gn:: . in , WE, is a, Polarity projectioJ;l. . ,It . is . dominated . . by a .'
fLJn�tjonal projectiQIl" .w,h.ose hea� �ttract� the fiI?te v�rb, let us continue to call r -it TP1': "A�" � fir�t app��ximaf1on� ('assume th"at 'TFf is 'the familiar 'Pollockian' TP. In the next section I will examine the word order in WF in terms of the structure in (22). FP is the projection whose head hosts the infinitival verb.
83
REMNANT MOVEMENT AND OV ORDER
4.
Double movement and. OV ord.ers
4. 1 Overt V-to-/ movement I have shown (Haegeman 1 998a, 1 998b) that WF displays morphology-driven V-to-I-movement. The data in (8) above illustrate the patterns. The finiteinon finite asymmetry in the distribution of en- and in IPP constructions reflects the degree of movement. Past tense Vs and verbs with en- move to T l via F (8c-d). A present tense V without en- may remain in F, or it may move to T l (8c-d). An infinitive must remain in F (8e-g). Some degree of infinitival movement would not be unexpected since the WF infinitive is clearly morphologically identified by its -en ending. That the infinitive does not move as high as the pnite v:�r.!?.jrt ,�9-mparal?1�.J2 t1ie sI��Iati61i" �iri' 'prench- (p6116ck" 1989: 1997). A deiaii;d discussion of the s��tureis- f���((ln'Haegeinan (1998a, 1998b, 1999). As a result of the head-to-head movement of the verb, the head position of the VP, and all V-related head positions between VO and FO or TO (as the case may be) contain a trace of the moved V. My analysis is different from the . antisymmetric proposal discussed in Section 1 which assumes either very restricted V-movement or no V-movement at all and is in line with the analysis suggested by Kayne ( 1994) and by Pearson (this volume). 4.2 Deriving the West Germanic OV order Without further modifications, the hierarchy in (22) obviously derives the wrong word order. V moves to Tl ; en-, which heads NegP l , left-adjoins to T l ; negative, temporal and aspectual adverbials and negative quantifiers are associated with NegP2. Without additional movements, we expect to find the surface order in (23). But these examples are ungrammatical. The grammatical counterparts are given in (24). (23)
a. *da Valere en-komt a nie meer atent nor us that Valere EN-comes already no more always to house b. *da Valere en-goat a nie no Gent that Valere EN-goes already not to Ghent c. *mee Valere te kommen nie meer atent nor us with Valere to come no more always to home
(24)
a.
da Valere a
nie meer atent
nor us
en-komt
that Valere already no more always to house EN-comes 'that Valere already doesn't always come home any more.'
84
LILIANE HAEGEMAN
b.
da Valere a
c.
mee Valere nie meer atent
nie no Gent en-goat
that Valere already not to Ghent EN-goes 'that Valere is already not going to Ghent.'
nor us
te kommen
with Valere no more always to home to come 'as Valere does not always come home any more'
In order to derive the grammatical patterns in (24), we need to reorder all the constituents found to the right of the verb in (23) with respect to the verb, while preserving their relative order. Crucially, the reordering is. r�qull:�d, both,for firgte . that 'ill '��h�negative sentences as w�iC'" ye�lJs alJ.d for non-finite verbs:1:tecaii also the � ��piemetits' pre��de-ilie verb, again both in finite and non-finite clauses. (25)
a. b.
da Valere neu atent
nor us
komt
mee Valere neu atent
nor us
te kommen
that Valere now always to house comes 'that Valere now always comes home.'
with Valere now always to home to come 'as Valere now always comes home'
4.3 Remnant movement
In earlier antisymmetric analyses the reordering of the verb and its complements
was achieved by one-by-one movement of individual constituents, such as object DPs and small clause predicates. VP-complements such as the direct object and the small clause predicate which precede the finite verb in the West Germanic OV-Ianguages follow the finite V in Romance. Like Pearson ( 1999), I propose that the West Germanic OV order is not achieved by one-by-one feature-driven leftward movement of the individual constituents. Rather, thLQY�,,,,,Q!der is E�E��,�,g�1Jx,i!l,y;,!g;,.l;lJlQ.;�t, and by (lD",,!':�' ll111�t�Q.x�, m�p.t(cf. Mtille;T996) _QL�,£�1endedpmj ., �ctiQll-q,(""Y".m_�£�£.!!:t Further research into the relative positions of adjuncts and into the mechanisms that derive extraposition effects (see Haegeman 1998c) will be needed to determine precisely which constituent undergoes this remnant movement. I will provisionally label this projection XP. (26) is a schematic representation.
85
REMNANT MOVEMENT AND OV ORDER
CP
(26)
�
Spec
C'
�
C
AgrSP
I
�
dat Spec
I
SU
AgrS'
�
AgrS
TP1
�
Spec
I
XP
T1'
�
T1
I
�
en-Vv Spec XP NegP2
niets
ten
�
I
tv
�
I
Neg2'
Neg2
Neg1'
Neg1
�
Spec
NegP1
VP
�
etc.
FP
�
Spec
F'
I
F
�
Tb,e." remnant movem�DCi�.jl1de12el:l,g�Ilt 9f th� m�v�l11ent. ()f V }0 l" 1 alld of the .. . v�i-b p�ese��'-oi�th�'����gative h���i " 'It t���" pi�ce" 'also!'"�h�� ' 'th�"\�fi�ti��(' "-'� ....-"" .-
,
j"�--I.�.:::-t ""'.r�."�h-"""-'";'-";.';::,\.'�h,J ''-��'''t,.t.>'_'''' ,_",,"\:-i';'�� XP
Remains
Solved
REMNANT MOVEMENT AND OV ORDER
89
complements move individually to a pre-verbal position. With respect to the problems raised by sentential negation, the non-adjacency of a preposed negative constituent and the negative head as such can be solved if one postulates two NegPs; the remaining problem concerns the hierarchical relation between these NegPIlPolP and NegP2. Various implementations are possible here but for reasons of space I will not develop them. See Haegeman 1999.
6.
An alternative antisymmetric account: Zwart's (1996) F-feature movement analysis
6 . 1 Introduction Zwart's (1996) analysis of the embedded V-final position captures the V-to-I movement effects without postulating overt V-movement to a high inflectional position. Following Minimalist traditions (Chomsky 1995), lexical items are taken to be bundles of features of two types: lexical-categorial features (LC-fea tures) and formal features (F-features). Formal features are involved in feature checking operations (tense, agreement, Case, wh, etc. see Zwart 1 996: 170). To account for the sentence-final position of the verb while attaining the movement effects, Zwart proposes that in both main and embedded finite clauses, the V-features of AgrS are strong and attract the F-features of the verb. The F-features of the verb move successive cyclically via AgrO and T to AgrS. In embedded finite clauses, AgrS, with the F-features of the verb, moves to C. As C is lexically filled, the F-features of the verb can be united with the LC-features of the complementizer and there is no need for movement of the LC-features of the verb to C. As a result the verb is spelled out in V (see Zwart 1996: 244). Zwart's 1996 proposal thus achieves the effects of V-movement without overt reflex. As a result, it overcomes some of the problems encountered by other anti symmetric views. However, the analysis has a number of drawbacks compared to the double movement analysis. The major issue that needs clarifying in this approach is to what extent the movement of F-features proposed for finite verbs extends to infinitivals. Zwart himself does not address the issue, but it is crucial in the light of the discussion here. I will therefore attempt to sketch an analysis in the spirit of Zwart's own proposal and point out its drawbacks. The absence of a root/embedded asymmetry in infinitivals might be taken to mean that there is no F-feature movement. However, infinitival clauses exhibit object shift and TEC. To maintain the correlation between V-movement on the one hand and object shift or TECs on the other, F-feature movement therefore
90
LILIANE HAEGEMAN
has to generalise to infinitives. Let us try to implement this idea. If F-features may move separately from the LC features of the verb as long as they can be spelt out in association with LC-features, we might propose that whenever the infinitival complementiser is overt, the F-features of the infinitival verb can be spelt out by associating with the LC-features of the complementiser while the LC-features of the verb remain stranded. This applies straightforwardly to examples like WF (3 1 a) in which van might be said to realise the embedded C: a.
(3 1 )
Jan ee beloofd
[van dienen boek te lezen] .
Jan has promised [of this book to read] 'Jan has promised to read this book.'
One problem with an F-feature analysis for infinitives is that it is not obvious how to analyse the position of te in such an approach. If te is a functional head of the I-system (cf. Kayne 1 994; Haegeman 1996b, 1 998a, 1 998b) and if the F-features of V always move to C, then F-feature movement will have to be able to skip or strand teo This movement would seem to lead to a violation of the Head Move ment Constraint or its Minimalist equivalent (cf. Zwart 1996: 203, note 1 0). In WF it is not necessary to spell out the complementiser in complement clauses of beloven. This means that no LC-features are available in the embedded C. This in tum might be taken to imply that the conditions for stranding the LC features of the verb are not met and that pied-piping is required, contrary to fact. ,
(3 1 )
b.
Jan ee beloofd
[dienen boek te lezen].
Jan has promised that
book to read
To solve this problem, one might argue5 that the F-features of infinitival lezen move into the matrix domain and associate with the LC-features of the matrix V, giving rise to some form of reanalysis. However, if IPP is an indicator of reanalysis, the absence of the IPP effect in (3 1a-b) suggests there is no reanaly sis (cf. Haegeman 1998b). Further support for the absence of reanalysis is that WF disallows the so-called third construction which involves long distance scrambling of a constituent out of complement clauses as that in (3 1 c) (cf. den Besten and Rutten 1989; Rutten 199 1 on the Dutch third construction). (3 1 )
c. *Jan ee dienen boek beloofd te lezen. Jan has that book promised to read
It is not clear then how the F-features of the complement of beloofd in (3 1b) can be lexicalised without pied-piping the LC-features. A similar question is raised by examples of so-called VP-topicalisation, illustrated in (3 1 d), in which an extended projection of an infinitival V is fronted.
REMNANT MOVEMENT AND OV ORDER
91
For arguments that the fronting involves more than VP see Haegeman 1 995. (3 1 )
d.
Dienen boek an Marie tuogen ee Valere niet gewild.
that book to Marie show has Valere not wanted 'What Valere did not want was to show Marie that book. '
If the F-features of infinitival tuogen move t o the C node of the fronted projec tion, then it is not obvious why the LC-features are not pied-piped, since there are no LC-features available in the embedded C for the moved F-features of the infinitive to associate with. Similarly, in independently used infinitival clauses, the infinitive remains sentence-final, even if there is no lexical host for the F-features: (3 1)
e.
Jenen boek morgen
nie vergeten!
your book tomorrow not forget 'Don't forget your book tomorrow! '
In the next section, I return to the problems discussed above with respect to the earlier anti symmetry approach. 6 For finite clauses, Zwart's (1 996) F-feature movement analysis solves some of the problems, but as shown above, the account does not naturally generalise to infinitives. In contrast, the double movement analysis elaborated in this paper offers an account that extends naturally to infinitival clauses and is therefore superior. 6.2 F-feature movement and the problems discussed above Zwart's (1 996) F-feature movement analysis will not alleviate the problem raised for Diesing's generalisation, since OV orders in his approach continue to be derived by feature-driven one-by-one movement of constituents. The problems raised by absence of morphology-triggered V-to-I movement are solved, if we re-interpret V-to-I movement as F-feature movement. However, on such an account, the observed finite/non-finite asymmetries in WF are hard to capture. Moreover, in Section 6. 1 . I have shown that infinitival F-feature movement is problematic. However, if infinitives are taken not to display F-feature movement, the OV order and TEe remain problematic. It is not clear if the problems raised with respect to the syntax of negation can be solved in the F-feature analysis. On the positive side, the non-adjacency of the negative head en- and the moved negative quantifier finds a natural explanation if we assume that F-feature movement pied-pipes the feature Neg. Assuming successive cyclic movement of the F-features, a fronted negative constituent which occupies the specifier of say AgrO, will be able to check its
92
LILIANE HAEGEMAN
Neg feature with the matching feature which is among the F-features in AgrO. On the negative side, in terms of Zwart's ( 1 996) account, the inflected V with its overt negative head en- remains in a low position. Since the presence of en triggers V-movement to that particular position, the most plausible hypothesis is that the lower projection whose head attracts the verb with the negative head is PolPlNegPl . This hypothesis, though, implies that there is a difference between the hierarchical organisation of Romance (and English, see Cormack & Smith 1 999) and WF. Table 2 summarises the discussion. For the F-feature movement analysis, I consider the variant with infinitival F-feature movement and without such movement separately. Table 2. Survey ofproblems (ii) No V-movement
Double movement
F-feature movement (also in infinitive)
F-feature movement (not in infinitive)
Morphology and V-movement
Remains
Solved
Solved
Solved
Finite/non-finite asymmetry
Solved
Solved
Remains
Solved
Diesing's generalisation
Remains
Solved
Remains
Remains
Holmberg's generalisation Remains
Solved
Solved
Remains
Transitive expletives
Remains
Solved
Solved
Remains
PolP > XP
Remains
Solved
Remains
Remains
Analysis
Problem
7.
Summary
In one type of antisymmetric analyses of the Germanic OV languages, the embedded verb-final order is derived by feature-driven leftward movement of lower constituents. The sentence-final position of the embedded verb is taken as evidence for the absence of V-to-I movement or for short V-movement. Such analyses create a number of problems, concerning the correlation between (i) V-morphology and V-movement, (ii) V-movement and object shift, (iii) V-movement and transitive expletives, (iv) the interpretation and position of indefinite objects. A further paradox arises with respect to the syntax of negation. In the Germanic OV languages, as well as in the Scandinavian VO languages, negative
93
REMNANT MOVEMENT AND OV ORDER
quantifiers obligatorily move leftward in order to establish a specifier head relation with a Neg head feature (Haegeman 1 995). On the basis of the WF IPP construction, Haegeman ( 1998a, 1 998b) shows that the presence of the negative head en- on the verb triggers overt V-movement. This suggests that V-movement is induced by Neg. If (i) Neg-movement of XPs targets the specifier of a head with the Neg feature, and if (ii) en-V-movement is triggered by a Neg head feature, the moved negative XP and the en-affixed verb will be expected to be adjacent, contrary to fact. A first analysis proposes that there are two NegPs (cf. Zanuttini 1 997a, 1 997b) and that movement of the negative XP targets the specifier of a higher NegP (NegP1 ) than movement of the en-affixed V (NegP2). Under this hypothe sis a problem is shown to arise w.r.t. the hierarchical relations of the two NegPs. To solve these problems, it is proposed that, like in Romance, the WF higher NegP1 (plausibly PoIP) hosts the negative head en; a Spec feature of the lower Neg2 triggers Neg-movement (32a). (32)
a. b.
SU > TP1 > PolP (en) > . . . FP . . . NegP2 (nie) SU > [TP 1-.lXPk. . . [NegP2 n iel . . . l [Tl en-V] > PoIP > [XP tk .J] .
I propose that Germanic embedded OV orders are derived by double movement: (i) generalised V-to-I movement, and (ii) generalised remnant extended VP movement to SpecTP1 (32b). This derivation applies to both negative and non negative clauses, and to both finite and non-finite structures. The finite/non-finite asymmetries in WF are captured in terms of the degree of V-to-I movement. Essentially based on a set of empirical arguments from WF my analysis is similar to that of Pearson ( 1 999). In the last section of the paper I review Zwart's 1 996 F-feature movement analysis. His analysis fares better in accounting for some of the issues raised, but it can only do so if there is taken to be no finite/non-finite asymmetry in F-feature movement. This is problematic, also in view of the observed finite/non finite asymmetries in WF Moreover, the F-feature movement analysis cannot capture Diesing's generalisation and the problem of the hierarchical relations of the two negative projections remains. ,
.
Notes 1.
Definite DPs may also move leftward: (i) wei! Elly diese Lieder immer singt since EUy these songs always sings 'since EUy always sings these songs'
LILIANE HAEGEMAN
94 2.
Cf. the conditions on domain extension in Chomsky's 1993 analysis of object shift.
3.
For reasons of space I ignore the difference between eet in (8a) and ee in (8b) here. I refer to Haegeman 1998c.
4.
It is not possible to assume that the indefinite subject of the TEC occupies SpecTPl all by itself, given the structure I assume. Following Pearson (1999), we may assume that after the removal of the V-head from the remnant XP, the features of the indefinite subject become accessible for attraction by T and that remnant movement is the result of pied piping. In such a view, the indefinite subject of the TEC would be licensed in SpecTP.
5.
This would be the option assumed by Zwart (p.c.).
6.
It is sometimes claimed that the transparency of extraposed CP complements in the West Germanic OV languages is accounted for in the antisymmetry approach because, it is claimed, under this view so-called extraposed clauses may be said to occupy their base positions. Obviously this would not be compatible with the remnant movement theory espoused here. However, in independent work I have demonstrated that it is not the case that all transparent sentence-final CP complements can be argued to be in their base position. For reasons of space I will not elaborate this point here and refer to HinterhOlzl 1999, this volume, and to Haegeman 1998c, 1999.
References Ackema, Peter. 1999. On the relation between V-to-I and inflectional features. Ms. University of Groningen. Besten, Hans den and Rutten, Jean. 1989. "On verb raising, extraposition and free word order in Dutch." In Danny Jaspers, Wim Klooster, Yvan Putseys, and Pieter Seuren (eds), Sentential Complementation and the Lexicon: Studies in honour of Wim de Geest. Dordrecht: Foris. 4 1-56. Bobaljik, Jonathan, and Jonas, Diane. 1996. "Subject positions and the role of TP." Linguistic Inquiry 27: 1 95-236. Bobaljik, Jonathan, and Hoskuldur Thrainsson. 1997. "Two heads aren't always better than one." Syntax 1 : 37-7 l . Chomsky, Noam. 1993. " A minimalist program for linguistic theory." In Hale, Kenneth and Samuel J. Keyser (eds), The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger. Cambridge, MA: MIT press. 1-52. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Christensen, Kristi Koch. 1986. "Norwegian ingen: A case of post-syntactic lexicalization in Scandinavian dialects." In Paola Beninca (ed.), Dialect Variation and the Theory of Grammar. Dordrecht: Foris. 53-83. Cinque, Guglielmo. 1 998. Adverbs and Functional heads. A cross-linguistic perspective. Oxford: OUP. Cormack, Annabel and Smith, Neil. 1998. "Negation, polarity and V positions in English." UCL Working Papers in Linguistics 10, 285-322. Diesing, Molly. 1992. Indefinites. Cambridge, MA: MIT Press.
REMNANT MOVEMENT AND OV ORDER
95
Diesing, Molly. 1 996. "NP types and conditions on interpretation." In Don, J., B . Schouten and W. Zonneve1d (eds), OTS Yearbook 1995. Utrecht: OTS. 1-12. Diesing, Molly. 1997. "Yiddish VP order and the typology of object movement in Germanic." Natural Language and Linguistic Theory 15: 369-427. Dikken, Marcel den. 1996. "The minimal links of verb (projection) raising." In Werner Abraham, Sam. Epstein, Hosku1dur Thnlinsson, and Jan-Wouter Zwart (eds), Minimal ldeas. Amsterdam and Philadelphia: John Benjamins. 7-96. Haegeman, Li1iane. 1995. The Syntax ofNegation. Cambridge: CUP. Haegeman, Liliane. 1996a. "Object clitics in West Flemish." In Aaron Halpern and Arnold Zwicky (eds), Second Position Clitics and Related phenomena. Stanford: CSLI Press. 135-164. Haegeman, Liliane 1996b. "The typology of syntactic positions: L-relatedness and the AlA' distinction." In Werner Abraham, Samuel Epstein, Hoskuldur Thnlinsson and Jan-Wouter Zwart (eds), Minimal Ideas. New York and Philadelphia: John Benja mins. 141-165. Haegeman, Liliane. 1998a. "V-positions and the middle field in West Flemish." Syntax. 1 : 259-99. Haegeman, Liliane. 1998b. "Verb Movement in embedded clauses in West Flemish." Linguistic Inquiry 29: 63 1-56. Haegeman, Liliane. 1998c. Extraposed clauses in the Germanic SOY languages. Ms. University of Geneva. Chomsky Celebration Website. MIT Press. Haegeman, Liliane. 1999. Embedded verb positions and remnant movement in the West Germanic OV languages. Ms. University of Geneva. Haegeman, Liliane and Zanuttini, Raffaella. 1 99 1 . "Negative heads and the Neg criteri on." The Linguistic Review 8 : 233-25 1 . Haegeman, Liliane and Zanuttini, Raffaella. 1 996. "Negative concord in West Flemish." In Adriana Belletti and Luigi Rizzi (eds), Parameters and Functional Heads: Essays in comparative syntax. Oxford: OUP. 1 17-1 80. Hamann, Cornelia. 1 993. Notes on negation in German. Ms. University of Geneva HinterhOlzl, Roland. 1999. Restructuring Infinitives and the Theory of Complementation. Ph.D. dissertation, USc. Hoekstra, Teun. 1995. "To have to be dative." In Hubert Haider, Susan Olsen, and Sten Vikner (eds), Studies in Comparative Germanic syntax. Kluwer: Dordrecht. 1 1 9-137. Holmberg, Anders. 1986. Word Order and Syntactic Features in the Scandinavian Languages and English. Ph.D. dissertation, University of Stockholm. Holmberg, Anders and Platzack, Christer. 1 995. The Role of Inflection in Scandinavian syntax. Oxford: OUP. Hoop, Helen de. 1 992. Case Configuration and NP Interpretation. Ph.D. dissertation, University of Groningen. Hr6arsd6ttir, porbjorg. 1999. Verb Phrase Syntax in the history of Icelandic. Ph.D. dissertation, University of Troms¢.
96
LILIANE HAEGEMAN
Jonas, Diane. 1996. "Clause structure, expletives and verb movement." In Werner Abraham, Samuel Epstein, Hoskuldur Thrainsson, and Jan-Wouter Zwart (eds), Minimal Ideas. Amsterdam and Philadelphia: John Benjamins. 1 67-1 88. J6nsson, J6hannes Gfsli. 1 996. Clausal Architecture and Case in Icelandic. Ph.D. dissertation, University of Massachusetts, Amherst. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kayne, Richard. 1998. "Overt vs. covert movement." Syntax 1: 128-1 9 1 . Koster, Jan. 1 994. "Predicate incorporation and th e word order o f Dutch." In Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi and Raffaella Zanuttini (eds), Paths towards Universal Grammar. Georgetown: Georgetown University Press. 255-276 Mliller, Gereon. 1 996. Incomplete Category Fronting. Habilitation thesis. Tlibingen. Pearson, Matthew, 1999. "Feature inheritance and remnant movement: Deriving SOY order under the LCA." In Gianluca Storto (ed.), Syntax at Sunset 2, vol. 3. Los Angeles: UCLA Department of Linguistics. 133-173 . Pollock, Jean-Yves. 1989. "Verb-movement, universal grammar, and the structure of !P." Linguistic Inquiry 20: 365-424. Pollock, Jean-Yves. 1 997. "Notes on clause structure." In Liliane Haegeman (ed.), Elements of Grammar. Dordrecht: Kluwer. 237-279. Roberts, Ian. 1 985. "Agreement parameters and the development of English modal auxiliaries." Natural Language and Linguistic Theory 3 : 21-58. Rohrbacher, Bernard. 1994. The Germanic VO Languages and the Full Paradigm: A theory of V-to-I raising. Ph.D. dissertation, University of Massachusetts, Amherst. Rutten, Jean. 199 1 . Infinitival Complements and Auxiliaries. Ph.D. dissertation, Universi ty of Amsterdam. Vikner, Sten. 1 997. "V-to-I and inflection for person in all tenses", In Liliane Haegeman (ed.), The New Comparative Syntax. London: Longman. 237-279. Watanabe, Akira. 1 998. Absorption: Interpretability and feature strength. Ms. Department of English, University of Tokyo. Wyngaerd, Guido van den. 1 989. "Object shift as an A-movement rule." MIT Working Papers in Linguistics 1 1 : 256-27 1 . Zanuttini, Raffaella. 1997a. Negation and Clausal Structure: A comparative study of Romance languages. Oxford: OUP. Zanuttini, Raffaella. 1 997b. "Negation and V-movement." In Liliane Haegeman (ed.), The New Comparative Syntax. London: Longman. 214-245. Zinsmeister, Heike. 1995. Die Lizenzierung von kein. Ein Vergleich der syntaktischen Vertaelung von N-Phrasen im Danischen und Deutschen. MA dissertation, Universi ty of Stuttgart. Zwart, Jan-Wouter. 1993. Dutch Syntax: A Minimalist Approach. Ph.D. dissertation, University of Groningen. Zwart, Jan-Wouter. 1996. Morphosyntax of Verh Movement. A Minimalist Approach to the Syntax of Dutch. Dordrecht: Kluwer.
V-movement and VP-movement in Derivations Leading to VO order Knut Tarald Taraldsen University of Troms¢
Abstract In this article, I assume that the Gennanic VO languages are just like the Gennanic OV languages except that a raising operation not found in Gennan, Dutch or Afrikaans has the effect of placing the V to the left of its comple ments in English and Scandinavian. The main purpose of the article is to investigate the nature of this operation: Is it an instance of head-movement (V-raising) or phrasal movement ((remnant) VP-raising)? What is the trigger? Using the distributional properties of verb particles as a diagnostic, I conclude that the VO order of English and Scandinavian must be the result of remnant VP-preposing. The analysis developed below can be seen as an attempt to extend the proposals of Kayne ( 1998), integrating them with certain insights due to Haider ( 1992), (1997).
1.
The options
Assuming the correctness of Kayne's ( 1994) conclusion that all constituents must have Spec-Head complement order, Koster (1994), Zwart ( 1993) and others have analyzed German and Dutch OV order as the result of verbal complements raising out of the VP to check "strong" features in the functional layer above the VP. In the Germanic VO languages, the complements of the verb would remain inside the VP until LF, because the relevant features are "weak". Disregarding criticism that could be directed against the theoretical frame work underlying this analysis, e.g. the absence of any independent motivation for the "strong" vs. "weak" distinction, I tend to consider a simple observation made
KNUT TARALD TARALDSEN
98
by Haider ( 1 992) as a decisive argument against this approach: The observation is that the complements of the verb are ordered with respect to one another as in ( 1 ) both in the Germanic OV languages and in the Germanic VO languages: 1 (1)
(V) - 10 - DO - PP - (V)
This is illustrated by the following examples from German and English (both from Haider ( 1 992), see also Barbiers (this volume)): (2)
dass sie jedem
ein Paket
that they everybody a
an seine Privatadresse schicken
package to his
home. address send
werden will
(3)
(Ger)
that they will send everybody a package to his home address
If the German sentence has the three verbal complements in SpecFP positions above the VP, while the complements of the verb remain inside the VP in its English counterpart, the system must add stipulations about the ordering of the relevant licensing positions in the functional layer to guarantee that extraction of the complements from the VP must preserve the initial ordering. So, I agree with Haider that deriving the VO order of English and Scandina vian from OV by shifting the V leftward across its complements is preferable to deriving the OV order of German or Dutch by moving complements across the V. Haider ( 1 992) proposes the following initial structure common to all German ic languages: (4)
[10 [DO [PP V]]]
OV languages are characterized as those which allow the V to license its complements from right to left ("regressive licensing"), taking the domain of licensing to be the maximal projection of the head. Hence, in German or Dutch, the V licenses its complements in (4) without leaving its base position. English and Scandinavian, on the other hand, only allow the V to license its complements from left to right ("progressive licensing"). Hence, licensing requires that the V raises (successively) across the structure in (4): 2 (5)
[V [10 tv [DO tv [PP tv]]]
Haider attributes the universality of (4) to his "basic branching condition" (the BBe) requiring the projecting node to follow its (unique) sister. However, (4) obvious ly becomes consistent with Kayne's LCA when viewed as a VP-shell structure as in (6) below, and the BBC would follow from the LCA plus the assumption that all complements of V must be analyzed as specifiers of VP-shells: 3
DERIVATIONS LEADING TO VO-ORDER (6)
99
[10 V [DO V [PP V]]J
There is, however, a different way of implementing the idea that VO order results from placing the V to the left of its complement. Suppose that the complements of the V raise overtly to SpecFP positions above the VP in all 4 Germanic languages, giving rise to (intermediate) representations like (7):
(7)
[10 Fio [DO Fdo [PP Fpp [yp V tlO tDO tpp]]]
The fact that the sequence IO-DO-PP shows the same internal order in all Germanic languages will then reflect the universal ordering of the heads Fio' Fdo and Fpp . The VO order characteristic of English and Scandinavian can then be attributed to the remnant VP of (7) raising to a Spec-position above FioP: (8)
[[yp V tro tDO tpp] [IO Fio [DO Fdo [PP Fpp typJ]]
Remnant VP-raising will of course not provide any directly visible clues enabling one to distinguish the two analyses on empirical grounds unless the remnant VP sometimes contains material that cannot be part of the V. In the following sections, however, I will argue that verb particles cannot be part of the V, but may raise across a direct object as part of a remnant VP.
2.
The distribution of particles correlates with VO vs. OV
In the Germanic VO languages, verb particles have two properties not shared by verb particles in the Germanic OV languages: 5 they follow the verb even outside V2-contexts, and whereas some Germanic VO languages (e.g. English) allow a verb particle to precede the direct object, no Germanic OV language has this property. The first of these two contrasts is illustrated in (9)-( 1 0), and the second one in ( 1 1 )-( 14): (9)
that they *outwentlwent out
( 1 0)
dass sie ausgingen/*gingen aus
(11)
They will let the cat in.
(12)
They will let in the cat.
(13) ( 1 4)
that they out.wentl*went
out
(Ger)
Sie werden deine Freunde ein laden.
(Ger)
*Sie werden ein deine Freunde laden.
(Ger)
they will they will
your friends in.invite
in your friends invite
1 00
KNUT
TARALD TARALDSEN
Leaving aside the first contrast, 6 I will briefly show how the second one would be explained on Haider's V-raising account of Germanic VO order. 7 Taking the V and the particle to be one complex V, Haider takes the position of the particle in ( 1 2) as a side-effect of V-raising, as indicated in ( 1 5): (15)
[[vV Prt] [DO tv] ] ]
In the derivation of ( 1 1), the V excorporates from the complex V, stranding the particle in its base position:
( 1 6)
[V [DO [v tv Prt]]]
If particles never move on their own, the ungrammaticality of ( 14) follows straightforwardly from the lack of V-raising in languages with "regressive licensing". In general, no Germanic OV language can have "particle-shift". It is obvious, however, that Haider's V-raising analysis establishes a link between particle-shift and VO order only if a particle is in fact part of a complex V. I will now present a number of reasons to reject that assumption.
3.
The particle is not part of the V
First, we have the familiar problem with V-raising to the V2-position. In no Germanic language does a particle accompany the host V in this situation: 8
Hvoifor slapp hun ut katten ?
(Nor)
( 1 8) *Hvoifor slapp ut hun katten ? why let out she cat. the
(Nor)
( 17)
why
let
she out cat.the
If the particle forms a complex V together with the verb, why must it always be stranded under V2-movement? The problem seems particularly serious, since it is the host of the "incorporated" particle that must excorporate, an option taken not to be available in much recent work (e.g. Kayne 1 994). Notice also that even though one might get around the excorporation problem for Norwegian by saying that the particle may (e.g. in the Norwegian counterparts of ( 1 2)) or may not (e.g. in the Norwegian counterpart of ( 1 1 )) be part of a complex V and assuming that somehow only a simple V is allowed in the V2-position, this would not extend to a language like Swedish, where only the word order in ( 1 2) is gram matical (see Section 5), forcing the conclusion that in Swedish, the particle must be part of a complex V, and yet, the particle is stranded in V2-contexts. 9 Another problem with the assumption that the particle is part of a complex
101
DERIVATIONS LEADING TO VO-ORDER
V i n sentences like (12) i s that the structure [v V Prt] i s inconsistent with Kayne's ( 1994) LeA and would in any event be the opposite of the [x Y X] structure typical of compound words in Germanic. To circumvent this objection, the V-raising account of particle-shift would have to posit obligatory leftward movement excorporating the V even in the derivation of (12):
( 1 9)
They will let [vin tv] the cat.
This extra movement is not motivated by anything in Haider's analysis, however. Nor is there any obvious reason why it would have to strand the particle. Moreover, it undermines an otherwise promising account of the contrast between (20) and (21): 1 0
(20) She let the cat in and the dog out. (21) *She let i n the cat and out the dog. In (20), we apparently witness a successful conjunction of non-constituents. However, (20) lends itself to an analysis in terms of across-the-board (ATB) V-movement, in the sense of Williams (1 978) out of a conjunction of FPsY
(22)
... X [[the cat in [vp let tDP trrJ] and [the dog out [vp let tDP tPrt]]] I
I
I
ATE
The ATB-derivation of (21) in (23) is illicit, if the V cannot excorporate from complex V containing the particle, since ATB-movement only applies to identical tokens of a constituent:
(23)
... Y [[[let in] [the cat [vp tv tDP tPrt]]] and [[let out] I
I
ATE
I
[the dog [vp tv tDP tPrt]]]]
But if the verb actually always excorporates from the complex V in the deriva tions leading to particle-shift, this account of (20) vs. (21 ) clearly becomes unavailable. That the particle cannot in general be taken as part of a complex V is also shown rather straightforwardly by examples like the Afrikaans sentence in (24) (from Donaldson 1 993): 1 2
(24)
Sy hoe! nie die Zig
af te geskakel het nie.
she need not the light off to turned have not 'She needn' t have turned the light off.'
(Afr)
1 02
KNUT TARALD TARALDSEN
Here, the particle is separated from the potential host verb by the infinitival marker te, which itself cannot be taken as part of that verb, a past participle. A similar argument can be made on the basis of the distribution of weak pronouns in English and Norwegian. Both of these languages allow the word order seen in ( 1 2). Yet, when the direct object is a weak pronoun, it must precede the particle: (25)
(Nor)
Han har sluppet n ut. he
has let
it out
(26) *Han har sluppet ut n. he has let out it
(Nor)
Thus, the sequence containing the verb and particle shifting over the direct object position in the derivation of particle-shift constructions like ( 1 2) must also contain a weak pronoun, sandwiched in between the verb and the particle. But the weak pronoun cannot easily be analyzed as forming a complex verb either 3 with the verb or the particle, since it separates from both in V2-contexts: 1 (27)
Deifor
slapp han n ikke ut.
therefore let
he it not out
(Nor)
(28) *Deifor slapp n han ikke ut. therefore let it he not out
(Nor)
(29) *Deifor slapp han ikke n ut. therefore let he not it out
(Nor)
Finally, notice that even sequences that are indisputably phrasal may behave like particles, as in the following examples from Norwegian: (30)
Hun sendte pengene
(3 1 )
Hun sendte med lens pengene.
she sent
she sent
med lens.
money. the with Jens
with Jens money.the
(Nor) (Nor)
The contention that the PP behaves like a particle in (30)-(3 1 ) is strengthened by the observation that Danish, which never allows particles preceding the direct object, also does not allow (3 1), while Swedish, which never allows particles following the direct object, rejects (30).
DERIVATIONS LEADING TO VO-ORDER
4.
103
Particle-shift via remnant VP-raising
If VO order is derived by raising a remnant VP to a Spec-position preceding the complements previously extracted from the VP, as suggested at the end of Section 1 , we can take particle-shift constructions like (12) to have the derivation indicated in (32), where the particle is inside the raised remnant VP: (32)
[[vp V Prt] [DO Fdo tvp]]
The structures with the particle left behind to the right of the direct object, like ( 1 1), must then have the particle extracted from the VP, on a par with the other complements of the V: (33)
[[vp V] [DO Fdo [Prt Fprt tvp]]]
The problems listed in Section 3 do not arise on this analysis. Since the particle is not taken to be part of the V, it is not expected that it will carried along under V-movement to the V2-position. 14 The V-Prt order seen in ( 1 2) is consistent with the LeA, since the particle is analyzed as (the head of) a VP-intemal comple ment rather than a head adjoined to V. The fact that the particle can be separated from the V by an infinitival marker or a weak pronoun is also unproblematic. This analysis of particle-shift generalizes Kayne's ( 1 998) analysis of particle-shift in focus constructions. Kayne argues that a sentence like (34) is to be analyzed as in (35), where the particle shifts as part of a remnant VP raised to a Spec-position immediately above the the focus phrase (FocP): (34)
She will let in only one student.
(35)
she [wp[vp will [let in tDP]] Foc+W [FoeP [DP only one student] tFoe tvp]]
(36), then, must have the particle outside the VP: 1 5 (36) (37)
?She will let only one student in. she [wp[vp will [let tPrt tDP]] Foc+W [FoeP [DP only one student] tFoe [in Fprt tvp]]]
Kayne does not take remnant VP-raising to be the source of particle-shift except in focus constructions like (34). But certain observations concerning the distribu tion of verb particles across the Mainland Scandinavian languages suggest that an analysis of particle-shift as a by-product of remnant VP-raising in (34) cannot co-exist happily with a V-raising analysis of particle-shift in (12). As already mentioned, Norwegian is the only Mainland Scandinavian language allowing both options illustrated by (1 1 )-(12) :
104
KNUT TARALD TARALDSEN
(38) (39)
Hun slapp katten inn.
(Nor)
Hun slapp inn katten.
(Nor)
�
let
she let
cat.the in
in cat. the
Swedish allows only the word order in (39), while Danish only allows the one in (38) 16 :
(40)
Norwegian Swedish Danish
DP Prt ok *
ok
Prt DP ok ok *
Significantly, this pattern is replicated with focussed direct objects:
(41)
Hun slapp bare en katt inn.
(Nor)
(42)
Hun slapp inn bare en katt.
(Nor)
(43)
she let she let
only one cat in
in only one cat
Norwegian Swedish Danish
only DP Prt ok *
ok
Prt only DP ok ok *
No Mainland Scandinavian dialect has only the DO-Prt order with non-focussed direct objects (like Danish) while allowing the Prt-DO order with focussed direct objects (like Norwegian and Swedish). If particle-shift across non-focussed direct objects is an effect of V-raising, while particle-shift across focussed direct objects reflects remnant VP-raising, we must assume that a particle is necessarily pied-piped under V-raising unless it is outside the VP. But there is no obvious reason why this assumption should hold, unless one takes a VP-internal particle to be necessarily adjoined to the V, an assumption which is inconsistent with the evidence discussed in Section 3 . If all instances of particle-shift are attributed to the raising of a remnant VP containing a particle, however, the identity of the patterns in (40) and (43) follows directly. On the analysis of particle-shift as a side-effect of remnant VP-raising, Swedish must only allow VP-internal particles, 17 Danish must only have particles outside the VP and Norwegian (like English) must have both options. Interesting ly, this view allows us to see a parallelism between the NorwegianiSwedish contrast and a superficially different contrast between Dutch and Afrikaans. In both of the latter two languages, complements of the main verb appear separated
DERIVATIONS LEADING TO VO-ORDER
1 05
from it, preceding the modal, in sentences like (44)-(45) (Dutch) or (46)-(47) (Mrikaans) : (44)
dat Jan Marie wi! opbellen
(Dut)
(45)
dat Jan Marie op wi! bellen
(Dut)
dat Jan Marie sal opbel
(Afr)
*dat Jan Marie op sal bel
(Afr)
(46) (47)
that Jan Marie will up.call
that Jan Marie up will call that Jan Marie will up.call that Jan Marie up will call
Particles, however, show a different behavior from that of DP complements like the direct object Marie in (44)-(47). In Dutch, a verb particle either immediately precedes the main verb, as in (44), or moves in front of the modal, like the direct object, as in (45). In Mrikaans, the particle must remain next to the main verb, as in (46). Assuming an analysis of "verb raising" constructions along the lines of HinterhOlzl ( 1 997) or Koopman & Szabolcsi ( 1 997), only complements that raise out of the VP, 1 8 to some SpecFP above it, can end up preceding the modal. Hence, we take both structures in (48)-(49) to be available to Dutch, while Afrikaans only has access to (48) : (48) (49)
[DO Fdo [vp Prt V]]] [DO Fdo [Prt Fprt [vP V]]]
Thus, Dutch is just like Norwegian (and English), while Afrikaans is like Swedish, although this underlying parallelism has different reflexes in West Germanic as opposed to Mainland Scandinavian, since only the latter are VO languages, i.e. have remnant VP-raising. 5.
An excursus on Kayne (1998)
According to Kayne ( 1 998), scope cannot be assigned via covert movement. 1 9 Thus, the wide scope reading of bare en student "only one student" in Norwegian example (50) comes from a structure in which this DP has raised to a position above the matrix VP, with subsequent remnant VP-raising to a still higher Spec position masking the effect of the first movement (and stranding the second object previously extracted from the VP, as will be discussed below):
106
KNUT TARALD TARALDSEN (50)
I de siste arene har hun ¢nsket a matte gi
bare en
in the last years has she wished to have.to give only one
student en A. (5 1)
student an A
(Nor)
. . Foc+W [PocP only one student tFoc [an A [vp has wished ... .
I
I
Kayne proposes that the FocP must be the complement of a head W which attracts a VP to its Spec, as indicated in (5 1) . He makes no commitment as to how the raising of the remnant VP to SpecWP is to be related to the processes determining the position of the V in general. Yet, there seem to be good reasons to believe that if the analysis of (50) in (5 1) is correct, VP-raising to the pre FocP SpecWP should be assimilated to the process inducing VO order. First of all, the property of attracting a VP to its Spec cannot be a property of W in general. Even in (52), where the subject is focussed by only, the focussed DP must be preceded by W, taken to attract only from the head position of the FocP: (52)
Only John chose the marinated salmon.
But this W does not attract the VP to its Spec. In general, W attracts a VP to its Spec just in case the following focussed constituent is such that it is generally postverbal, e.g. a direct object rather than a subject. This fact seems to force one to view the VP-raising to SpecWP in (5 1) as a special case of whatever process makes complements of the verb postverbal in VO languages. Assuming the correctness of Kayne's analysis of (50), this process should therefore involve VP-raising, although not to the Spec of a head (W) appearing only in front of a FocP, but rather to the Spec of some FP always projected in between the subject position and the various complements extracted from the VP (according to the analysis sketched at the end of Section 1) . 20,21 Another straightforward observation supporting this view is that not even focussed direct objects have the VP shifting to their left in languages which, like German and Dutch, are not VO languages, even though the fact that the focus particle (e.g. German nur "only") precedes the focussed phrase would require the presence of a W even here. Finally, notice that the focussed indirect object in (50) is followed by a non focussed direct object. On the wide scope reading of the focussed DP, this direct object must also be in a position outside the matrix VP, as indicated in (5 1), i.e. it shows the distribution characteristic of DP complements in the West Germanic "verb raising" construction. The unavailability of the wide scope reading in (53),
DERIVATIONS LEADING TO VO-ORDER
107
then, would be correlated with the fact that hate is not a trigger for "verb raising" in West Gennanic either:
(53)
I de siste arene har hun hatet a matte gi
bare en student
in the last years has she hated to have.to give only one student (Nor) en A. an A
But if the direct object of the infinitive can raise across the matrix V (provided the matrix V is a "verb raiser"), and VO order is derived by V-raising, (54) would be derived by applying V-raising to the intennediate structure in (55):
(54) *Hun har en A ¢nsket a fa.
she has an A wished to get
(55)
(Nor)
. . . [an A [vp has wished . . .
If VO order is the result of remnant VP-preposing, on the other hand, only the grammatical (56) can be derived from (55) : 22
(56)
Hun har ¢nsket a fa en A. she has wished to get an A
(Nor)
I conclude that if Kayne's ( 1998) account of the wide scope reading of (50) is correct, Gennanic VO order should be derived by remnant VP-preposing, a conclusion which obviously dovetails with the analysis of particle-shift argued for in preceding sections. 6.
Two problems
At this point, it seems that the weight of the evidence favors analyzing Gennanic VO order as the result of remnant VP-raising. But particle constructions raise two important problems that remain to be dealt with. First, we must return to a question raised in Section 1 : Why do we always find the order Prt-V in the Gennanic OV languages (modulo V2), but always V-Prt in the VO languages? In the analysis developed so far, the particle is VP-internal in particle-shift constructions like (12) (repeated below), and it is likewise VP-internal in Dutch and Afrikaans sentences like (44)1(46) (also repeated below): (12)
(44)
They will let in the cat.
dat Jan Marie wil opbellen
that Jan Marie will up.call
(Dut)
KNUT
108 (46)
TARALD TARALDSEN
dat Jan Marie sal opbel
that Jan Marie will up.call
(Afr)
Yet, we have the order V-Prt in (12), but Prt-V in (44)/(46). In view of the previous discussion of the contrast (20) vs. (21) (repeated below), we do not want to derive ( 1 2) by having V-raising applying to the output of VP-raising (assuming the VP-internal order to be always Prt-V, perhaps with Prt(P) in the Spec-position of a VP-shell):
(20) She let the cat in and the dog out. (21) *She let in the cat and out the dog. But the only available alternative would seem to be postulating VP-internal V-Prt order for the VO languages and VP-internal Prt-V order for the OV languages, a devastating conclusion in the context of an attempt to reduce the word order differences between the Germanic VO languages and their OV cousins to the effects of remnant VP-raising. The second problem concerns the claim that the particle is never part of a complex V. Although we have seen good evidence supporting this conclusion, there is also a well-known fact which at first sight seems inconsistent with it. A particle modified by an element like right doesn't shift, as illustrated by the following examples from English and Norwegian:
(57) They will let the dog right in. (58) *They will let right in the dog. (59) Hun sendte katten rett ut.
(Nor)
(60) *Hun sendte rett ut katten.
(Nor)
she sent she sent
cat.the right out
right out cat.the
Strikingly, even Swedish, which allows simple particles only to the left of the direct object, can only have a modified particle after the direct object:
Dam sliippte hunden ratt in.
(Swe)
(62) *Dam sliippte ratt in hunden. right in dog.the they let
(Swe)
(61)
they let
dog.the right in
This fact would find a straightforward explanation if the theory only allowed a particle to shift as part of the V, since presumably only a head can incorporate into the V. Thus, we seem to be led towards a paradox: A shifted particle both must and must not be part of the V.
DERIVATIONS LEADING TO VO-ORDER
1 09
The same paradox may also arise in a somewhat different guise in Mri kaans. According to Donaldson ( 1993), bare nouns optionally behave like particles, in the sense that they may stay next to the infinitives in "verb raising contexts", while modified nouns must shift across the matrix V:23 (63 )
dat die olifante
kom water drink
(Afr)
dat die olifante
water kom drink
(Afr)
(65) *dat die olifante
kom vars water drink
(Afr)
dat die olifante
vars water kom drink
(Afr)
(64)
that the elephants come water drink
that the elephants water come drink
that the elephants come fresh water drink
(66)
that the elephants fresh water come drink
This strongly suggests that in Mrikaans, nouns (and particles) may remain next to the selecting V just in case they are incorporated into the V. But, as pointed out in Section 3, the construction exemplified in (24) (repeated below) shows that particles are not incorporated into the V, and (67) leads to the same conclusion for bare nouns:
(24)
(67)
Sy hoe! nie die Zig
af te geskakel het
nie.
she need not the light off to turned have not 'She needn't have turned the light off. '
Die olifante
hoe! water te gedrink het.
the elephants need water to drunk have 'The elephants need to have drunk water.'
(Afr)
(Afr)
In the next section, I present an implementation of the basic idea underlying the VP-raising account of VO order which seems capable of resolving this paradox and of eliminating the VP-internal Prt-V vS. V-Prt problem as well. 7.
Generalized VP-to-SpecWP raising
The fact that modified particles don't shift means that a shifted particle must be incorporated into the head of the constituent that raises across the direct object: We can then prevent modified particles from shifting by saying that all comple ments of that head, including a PrtP containing a modified particle, are extracted out of its maximal projection:
1 10
KNUT TARALD TARALDSEN (68)
... FeompI ... Feompn (x:p Spec Prt+X compI ... compn]
I'
I
I
'
I
XP should contain the V, since particle-shift only occurs in languages where the V as well ends up preceding the direct object. But this does not entail that X is the V, since we can take the V to be contained in a remnant VP in SpecXP, i.e. the VP originates as the complement of the particle and subsequently raises to SpecXP above it. By making that assumption, we avoid the undesirable conclu sion that a shifted particle must be part of the V. In other words, the paradox that the facts about modified particles apparently led to is not a real one: It dissolves under the scenario depicted in (68). Likewise, the problem raised by Prt-V order in the OV languages vs. V-Prt in VO languages is sidestepped: According to (68), the particle is outside the VP, and V-Prt order is the result of placing a remnant VP in SpecXP, preceding the head X to which the particle is adjoined. We may take it that the VO languages always place the VP in SpecXP, 24 while the OV languages never do, leaving it instead in its base-position below X. The analysis embodied in (63) bears an obvious resemblance to Kayne's ( 1 998) analysis of focus constructions. As we saw in Section 4, Kayne analyzes a sentence like (34) (repeated below) as in (35): (34)
She will let in only one student.
(35)
she [wp[yp will [let in tDP]] Foc+W [FoeP [Dpon1y one student] tFoe typ]]
The focussed DP is raised to SpecFocP, the head Foc adjoins to the immediately higher head W which also attracts the VP to its Spec. The parallelism between this analysis and the analysis of particle constructions in (68) is brought out more clearly if we adopt the more elaborated version of (68) in (69), which is the analysis I propose for the Norwegian particle-shift sentence in (39) (repeated below):
(39)
Hun slapp inn katten.
(69)
. . . [wp [yp slapp ts d [Prt inn]] +W [PrtP [s c katten P] tPrt typ]]
she let
in cat.the
(Nor)
According to this analysis, a shifted particle originates as the head of a particle phrase (PrtP) whose Spec is filled by a small clause (SC) with a null predicate identified under Spec-Read-agreement by the Prt head. 25 The Prt head incorpo rates into the immediately higher W (corresponding to the X of (68)), and the remnant VP is raised to SpecWP. The "stationary" particles of Dutch and Afrikaans will be analyzed like the shifted particle of (39), except that the VP doesn't raise to SpecWP:
DERIVATIONS LEADING TO VO-ORDER
(44) (46) (70)
111
dat Jan Marie wi! opbellen
(Dut)
dat Jan Marie sal opbel
(Afr)
that Jan Marie will up.call
that Jan Marie will up.call . . . [wp [Prt op] ]+W [PrtP [s c Marie P] tPrt [yp bel(len) ts cJ]]
Before we consider the details of the derivation of (39) from (69), the derivation of (44)/(46) from (70) and the treatment of non-shifting particles, it is appropri ate to clarify certain aspects of the general analysis of the "extended VP" which my analysis of particles must fit into. Here, I will outline the approach which seems most appealing to me for general conceptual reasons. In the final section, however, I will discuss an empirical problem that may force one to adopt an alternative formulation. The approach I shall pursue takes it that the analysis in (69)-(70) is a special instance of the general structure (7 1), which replicates itself at all levels of the extended VP above the VP itself up to and including the F hosting the indirect object DP in its Spec:
(7 1)
. . . Fxp+W [FXPP XP tFxp [
. . •
[yp V . . .
W attracts the VP to its Spec in some languages, which for that reason come out as VO languages. The OV languages are those where W does not attract VP.
Thus, this extension of Kayne's analysis of focus constructions is motivated by the empirical observations made in Section 5: The VP-raising to SpecWP in focus constructions in English and Scandinavian must be identified with the process creating VO order in general. But how does the theory enforce the presence of the W heads interspersed between the functional projections of the extended VP? The functional heads instantiating the Fxp of (7 1) select a phrase and are themselves selected by the lexical head in whose extended projection they appear. Adapting a proposal by Fukui & Speas ( 1986), we may say that a functional head only establishes a selection relation with a phrase in its Spec, while a lexical head only establishes the required selection relationship with a projection whose head is incorporated into it, hence only with its complement. The second half of this assumption entails that each functional projection above the VP in the extended projection of the V must be the complement of a copy of the V, i.e. it must be the complement of W now taken to be essentially an abstract V: 26
KNUT TARALD TARALDSEN
1 12
We can now say that the remnant VP raises to SpecWP in the Gennanic VO languages in order to identify the +V feature of W via Spec-Head-agreement, arriving at an analysis which in certain respects is rather close to Haider's ( 1992) proposal: The W heads correspond to the V-positions filled by successive V-raising in Haider's (5) (repeated below):
(5)
[V [10 tv [DO tv [PP tv]]]
They indirectly license complements of the verb by licensing (under selection) the functional heads that license them. There are two important differences, however. In (72), V-raising to the W-positions is not an option because of the intervening functional heads. Hence, the identification requirement on the +V feature can only be satisfied by VP-raising to SpecVP. Moreover, the general assumption that selection by a lexical head only can be established via incorpora tion into it forces the presence of [w +V] heads even in the OV languages, raising the question why VP-raising is not forced. However, we might adapt Haider's assumption about a basic asymmetry between OV and VO languages with respect to the relationship between a V and its complement, by taking this asymmetry to be relevant to the identification of +V features on the W heads: In an OV lan guage, the overt V may identify the Ws in its extended projection from right to left, but in the VO languages, identification must proceed from left to right. The idea that W heads are V s also implies that a WP is really itself a VP. This will become important when we examine the subsequent steps in the derivation of (39) from (69).
8.
The derivation of Prt-DO order
The next steps in the derivation from (69) are driven by the need for the DP subject of the SC to be licensed: 27
(73) (74)
. . . Fdp +W I [PdpP katten tpdp [WP2 [vp slapp ts cJ [Prt inn]]+W2 [PrtP [SC tDP P] tPrt tvp]]] . . . [WP I [WP2 [vp slapp ts cJ [Prt inn]+W2 [PrtP [s c tDP P] tPrt tvp]]Fdp +W I [PdpP katten tpdp twp2]]
In (73), the DP subject of the SC has been raised to the appropriate SpecFP above the WP having the remnant VP in its Spec. The head Fdp is selected by V, and so there is an occurrence of W immediately above FdpP to which F dp adjoins. (74) is fonned by raising the lower WP (WP2) into the Spec of the higher WP (WP I ) ·
DERIVATIONS LEADING TO VO-ORDER
1 13
The raising of WP2 to SpecWP 1 must be motivated by the need for the +V feature of W2 to be identified. Obviously, the latter requirement would have been satisfied by raising the VP from SpecWP I stranding the particle adjoined to W2. However, the VP and WP2 both have the +V feature attracted by W I . 28 Assum ing a version of the A-over-A principle, it follows that SpecWP I must be filled by WP2' and we get the effect of particle-shift. The analysis correctly predicts that particle-shift won't occur in the German ic OV languages, since particles shift only as part of a WP and the W above FdpP does not attract VPIWP to its Spec in these languages. The only relevant further step in the derivation of (44)/(46) from (70) is the one driven by the licensing requirements of the small clause subject, corresponding to (73) above: (75)
9.
. . . Fdp +W l [FdpP Marie tFdp [WP2 [Prt oP]] +W2 [PrtP [sc tDP P] tPrt [yp
bel(len) tsd JJ
A resolution of two problems for the V-raising account of particle shift
The analysis also provides an account of the particle-like behavior of full PPs seen in (30)-(3 1) (repeated below): (30)
Hun sendte pengene
(Nor)
(3 1)
Hun sendte med lens pengene.
(Nor)
she sent she sent
med lens. money.the with Jens
with Jens money. the
Assuming that the SC-predicate can contain the complement of a null head, we can give the particle-shifted version in (3 1 ) a representation like (76): 29 (76)
. . . [WPI [WP2 [yp sendte tsd [Prt med] +W2 [PrtP [s c tDP P lens] tPrt typ]] Fdp +WI [FdpP pengene tFdp twp2]]
Pairs like (30)-(3 1) are problematic for the V-raising account of particle shift, since that account must analyze a shifting particle as part of the V undergoing raising, and therefore seems to predict that full PPs shouldn't shift. For similar reasons, the fact that a weak pronoun always interrupts the V Prt sequence in English and Norwegian is also unexpected on the V-raising approach, since, as we saw in Section 3, the weak pronoun does not incorporate either into the verb or into the particle: (25)
Han har sluppet n ut.
he has let
it out
(Nor)
1 14
KNUT TARALD TARALDSEN
(26) *Han har sluppet ut n. he has let out it
(Nor)
The analysis we are now exploring has an account of (25)-(26), provided weak pronouns must be inside the VP moving to the Spec of the W to which the particle adjoins, as suggested by the analogy with the partially homophonous definiteness markers in DPs, which seem to originate in a position immediately above the NP: (77)
den svarte katten i nabohuset
the black cat.n in the.neighbors .house r
In particular, we might take weak pronouns to originate as functional heads in the extended V-projection immediately above the VP, perhaps with a null DP raising to its Spec, as in Sportiche (1992):
(78)
. . . W [FelP DP [Fc1 n ] [vp V tDP]] When (78) is embedded under a particle head, we end up getting (79): (79)
. . . [WP I [WP2 [vp sluppet tsd [Fc1 n ] +W2 [FelP DP tFc1] ] [Prt inn]] +W 1 [PrtP [sc tDP P] tPrt twp]]
At first sight, this account of weak pronouns seems desperately at odds with the well-known fact that Scandinavian weak pronouns sometimes clearly occur in positions higher than those occupied by other DPs. This is the case in V2 contexts: (80) (81)
Vi sa
n
ikke (*n).
Vi sa
(* katten) ikke katten.
we saw him not we saw
him
cat.the not catthe
(Nor) (Nor)
But in fact, the analysis under consideration suggests an analysis that has certain advantages over the traditional "object shift" analysis of (80). Suppose V2 has remnant VP-raising across negation and adverbials as a necessary subcompo nent. 3D Assuming, as before, that VP-raising applies to WPs as well (since these are essentially VPs), we can then take both of (80)-(81) to have a WP corre sponding to (78) (with the VP shifted to SpecWP) raised across the negation. This correctly only allows for the pre-negation position of the weak pronoun in (80): (82)
. . . [wp [vp sa tDP] [Fc1 n ] +W [FelP DP tFc1]] [NegP ikke . . . twp In (8 1), Fc1 is null, and the object DP is extracted from the relevant WP to a licensing SpecFdpP, so that it necessarily follows the negation:
DERIVATIONS LEADING TO VO-ORDER
(83)
1 15
. . , [wp [vp sa tDP] [Fcl 0]+W [FelP DP tFcl]] [NegP ikke . , . [FdpP katten Fdp . . . twp]]
so, this approach would take object shift to be a side-effect of remnant VP-rais ing, made possible precisely by the property that weak pronouns themselves remain very low in the tree. 3 1 Its main appeal comes from the fact that Holm berg's generalization is now automatically accounted for: A weak pronoun may appear in a position preceding negation or sentence adverbials only when the governing V moves. 32 Unfortunately, there are serious problems as well. One of them is a special case of a more general problem discussed in the next section. In structures containing both a weak pronoun and a particle, successive application of VP-rais ing applying to VP and WP will assemble WPs like (79) (repeated below): (79)
. . , [WP I [WP2 [VP sluppet tsd [Fcl n] +W2 [FelP DP tFcl] ] [Prt inn]] +WI [PrtP Esc tDP P] tPrt twp] ]
Since (27)-(29) show that the particle and the weak pronoun must eventually end up on opposite sides of the negation, we must ensure that remnant VP-movement in this case can only apply to the smallest WP, i.e. WP2 in (79): ' (Nor) (27) Derfor slapp han n ikke ut. therefore let he it not out (Nor) (28) *Derfor slapp n han ikke ut. therefore let it he not out (Nor) (29) *Derfor slapp han ikke n uf. therefore let he not it out And apart from the consequences with respect to object shift, the analysis based on (78) raises a question concerning the distribution of the contrast (25) vs. (26) across the Scandinavian languages. In fact, the contrast is reversed in Swedish, i.e. (26) is grammatical, and (25) isn't. The fact that even weak pronouns follow particles only in the one Scandinavian language which also only has ordinary DP objects following particles, is unexpected on the analysis suggested above, as Anders Holmberg (p.c.) points out. But even if this observation should lead one to reject the analysis of weak pronouns in (78), we still have the more fundamen tal conclusion that while the pattern in (25)-(26) is inconsistent with attributing particle shift to V-movement, it is in principle compatible with seeing particle shift as a reflex of remnant VP-movement. 33
KNUT TARALD TARALDSEN
116
10. The derivation of DO-Pri order
On the analysis developed above, a particle that fails to shift in a VO language cannot be the head of PrtP. Thus, I will take it that particles, in English or Norwegian, either are functional heads, occurring in the Prt-positions, or else are part of the SC in SpecPrtP. 34 A sentence like (38) will have a representation like (84) at an early stage of its derivation (with Prt an empty functional head): (38) (84)
Hun slapp katten inn.
she let cat.the in . . . [wp [yp slapp tpp] Prt+W [[pp katten [pp inn]] tPrt typ]]
(Nor)
Unlike an SC with an empty head (licensed by a "functional" particle), the SC in (84) must raise to a designated licensing position outside the WP, just like the small clause subject: (85)
. . . [WPI [WP2 [yp slapp tpp] Prt+W2 [PrtP tpp tPrt typ]] Fdp +W I [FdpP katten tFdp [FscP [pp tDP [pp inn] Fsc twp2]]]]
I also hold movement to SpecFscP responsible for the appearance of the particle to the left of the modal in Dutch "verb raising" sentences like (45): 35 (44) (45)
dat Jan Marie wil opbellen
(Dut)
dat Jan Marie op wil bellen
(Dut)
that Jan Marie will up.call
Taking the "verb raising" effect to be automatic, as is indeed the case for DPs in (Standard) Dutch, we conclude, as before, that Dutch, like English and Norwe gian, allows a particle to be analyzed either as a functional head outside the corresponding small clause, or as the lexical head of the small clause itself. In the latter case, (45), but not (44), is derived. The derivation of (38) as reflected in (85) has two instances of VP-raising to SpecWP, neither of which has a counterpart in Dutch, an OV language. (Also, Norwegian shows no "verb raising" effects. 36) First, the VP raises (string vacuously) to the SpecWP immediately above PrtP, and then, that WP raises to the SpecWP immediately above FdpP, crossing over both the lexical particle and the object DP. According to the scenario described in Section 7, however, one should also expect an intermediate step at which the lowest WP (WP2 in (85)) raises to a SpecWP immediately above FscP, yielding (86): (86)
. . . [FdpP katten Fdp [WPI [WP2 [yp slapp tpp] Prt+W2 [PrtP tpp tPrt typ]]
Fsc+W 1 [FscP [pp tDP [pp inn] tFsc twp2]]]]
DERIVATIONS LEADING TO VO-ORDER
1 17
This is because every functional projection in the extended projection of the VP should be immediately dominated by an occurrence of W attracting VP to its Spec. But if (86) is a stage in the derivation of (38), VP-raising to SpecWP above FdpP must target WP 1 rather than the larger WP2 including it. Otherwise, the particle would incorrectly end up preceding the object DP. But we have previously seen that when VP-raising reapplies to the output of VP-raising to the SpecWP immediately above the functional particle projection (PrtP), the larger VP (WP) must be targeted. Thus, the pied-piping effect for VP-raising must somehow be made to take force selectively. 37 It seems impossible to achieve this in a way entirely consistent with the general ideas about the extended VP-projection put forth in Section 7. In particular, it now seems impossible to maintain that all functional projections are governed by Ws each of which is a replica of the lexical head V. Rather, one must introduce some formal distinction between the Ws cooccurring with the PrtP (and FelP, if the analysis of (25) in Section 9 is correct) on the one hand, and all higher occurrences of W on the other, in a way reminiscent of Hinter holzl's distinction between his PredP level and the higher functional projections. The least costly implementation of this with respect to the proposals of Section 7 might be to say that the + V feature of a higher W is erased once a licensing VP has been raised to its Spec, while the + V feature of the W associat ed with PrtP (or FelP) remains visible. This move might be justified, if it can be shown that the latter plays a role at LF, while the + V features of the higher Ws do not. Alternatively, one must fall back on a different scheme according to which there is a W attracting VP just above the PrtP and a different W' just above FdpP attracting WP, but no Ws in between, leaving the nature of these two heads un determined for now. Unlike the analysis adopted above, this alternative analysis would also seem to predict that there could be languages where VP doesn't raise above a functional particle, but WP (i.e. Prt VP) raises above the object DP, or vice versa. Although I know of no Germanic language verifying the second expectation, Yiddish appears to have VO order (resulting from WP raising above the direct object), but Prt V order (indicating failure of the VP to raise past the functional particle), judging from Diesing ( 1997) ((87) Diesing's (23)): =
(87)
Maks vet avekshikn a briv.
Max will away. send a letter
(Yid)
118
KNUT
TARALD TARALDSEN
11. Conclusion
The principal claim made in this article is that VO order in Germanic (outside V2 contexts) arises as the result of remnant VP-raising across a functional layer inside which various complements of the V have previously been placed. From this point of view, the complements of the V behave the same way in all the Germanic languages, but the VP is forced to raise only in the Germanic VO lan guages. The arguments in favor of this view have largely centered on the syntax of verb particles. In particular, I have argued that the particle shift phenomenon characteristic of the Germanic VO languages cannot be successfully analyzed as a by-product of V-raising, while it can be understood as a side-effect of remnant VP-raising. On the other hand, the special behavior of modified particles (and of modified nouns vs. bare nouns in Afrikaans) turns out to be reconcilable with this view only if certain assumptions are made about the nature of the extended V-projection. In Section 7, I presented a set of assumptions generalizing recent ideas of Kayne's about focus constructions and suggested a partial integration of these with the basic ideas of Haider (1992). Whether this interpretation can be maintained, will depend in part on whether it allows for a satisfactory solution of the problem discussed in Section 10. If not, alternative conceptualizations compatible with the basic idea that Germanic VO order reflects remnant VP-rai sing remain to be explored. Notes 1.
Even disregarding the effects of Scrambling, the text statement must perhaps be somewhat modified. The basic (e.g. non-scrambled) order of double objects has been claimed to be DO-IO in German; cf. den Dikken (1995). The relative ordering of postverbal PPs in Dutch may not correspond to the one found in English or Scandinavian; cf. Barbiers (1995), Nilsen (1998).
2.
Actually, Haider has the lowest complement to the right of the V in VO languages. Given the assumption that heads license their complements from left to right in VO languages, this allows the lowest complement of the V to be licensed without V-raising.
3.
See Brody (1997) for discussion.
4.
The idea that direct objects raise out of the VP in overt syntax even in English and Scandina vian was already argued for by Johnson (1991).
5.
Yiddish has Prt-V order. Swedish and some dialects of Norwegian have Prt-V order where the V is a past participle not combining with auxiliary ha 'have' . Danish never allows a particle to the left of the direct object. See Taraldsen ( 1983), Vikner (1987), Svenonius (1994), Haider (1997).
6.
Haider actually attributes the V-Prt order of English and Scandinavian to the V having to precede its lowest complement in the VO languages (cf. footnote 2).
DERIVATIONS LEADING TO VO-ORDER
1 19
7.
If *(14) is explained as a result of further DP-movement across a (shifted) particle, (i) still calls for further assumptions (about Q-stranding): (i) * . . . deine Freunde ein aUe laden
8.
Except, perhaps, in Yiddish, in which case Prt+V in Yiddish should probably always be analyzed as a lexical compound similar to lexical Prt+V compounds in Scandinavian and the suggestion made about Yiddish at the end of Section 10 is incorrect.
9.
In the following paradigm from Afrikaans, a particle associated with the second V in what appears to be a conjunction of Vs moving to the V2 position, is stranded exactly as in the cases where a single V moves: dat sy hom sit en aanstaar (i) that she him sits and at.stares (ii) *dat sy hom aan sit en staar that she him at sits and stares (iii) *Sy sit en aanstaar hom. she sits and at.stares him (iv) Sy sit en staar hom aan. she sits and stares him at It is not immediately obvious how this is to be integrated into the standard analysis of the effect of V2 on verbs.
10.
The basic observation is due to Svenonius (1 994), and is also discussed by den Dikken (1995).
11.
However, this predicts that there should be no OV counterpart of (20), if the V never raises in (Germanic) OV languages except in V2 contexts. The prediction may be wrong. Wilma ter Haar (p.c.) finds a Dutch embedded sentence modeled on (20) relatively acceptable.
12.
See also Hinterh6lzl (1 997, this volume), who constructs a similar argument on the basis of West Flemish.
13.
Notice that a weak pronoun precedes a particle even in contexts where the weak pronoun does not undergo "object shift", i.e. in contexts where the governing verb has not moved, as in (25)-(26).
14.
V2 as V-raising incorrectly predicts (i) to be grammatical as an instance of ATE V-raising to C, unless the complement of C for some reason cannot be a conjoined structure: (i) *Han slapp inn hunden og ut katten. he let in dog.the and out cat.the Anders Holmberg (p.c.) suggests that the ungrammaticality of (i) in Norwegian may be due to the particle being unable to bear contrastive focus when it precedes the object. If focused constituents must be phrases (licensed in SpecFocP), this approach fits in with the analysis in Sections 7-8 where a particle that precedes the object must be a bare head. However, Holmberg also notes that (i) is marginally acceptable in Swedish.
15.
The fact that (37) is less than perfect, in contrast to ( 1 1), raises a problem for the generalization of the remnant VP-raising approach proposed here.
16.
See Taraldsen (1 983), Vtkner (1987), Svenonius (1994), Haider (1997).
17.
Except in the special case discussed in Section 6: When modified by (the Swedish equivalent of) right, the particle must follow the direct object even in Swedish.
18.
Actually, a particle will raise across the modal just in case it is outside the PredP, a constituent that properly contains the VP, on Hinterh6lzl's analysis. This distinction can be overlooked here, but will be important in Section 7.
120
KNUT TARALD TARALDSEN
19.
This conclusion is virtually forced by HinterhOlzl's (1997) observations about scope in German.
20.
The assumption of a fixed target for VP-raising may not be necessary. The central claim is rather that the SpecWP appearing in conjunction with a FocP cannot be the (only) target.
21.
A reason for dissociating VP-raising to SpecWP from the process leading to VO order in general might come from the fact that in the Scandinavian languages, all of them VO languages, the VP is not raised to a position preceding an internal complement raised to SpecNegP.
22.
Sentences with the word order of (54) remain derivable to the extent that the embedded VP can also be extracted from the matrix vP, an option which I take (modem) Scandinavian languages and English not to have, but which may have been available to Middle Icelandic; cf. Hr6arsd6ttir (1999, this volume).
23 .
But Robbers (1997) disputes the facts. In particular, she claims that some speakers allow even the phrasal complements to stay low. The grammar of these speakers might have the "verb projection raising" option found in varieties of Dutch and German.
24.
Yiddish is perhaps an exception (see footnote 8).
25.
From the perspective of Kayne's (to appear) analysis of (prepositional) complementizers, Prt may be taken to correspond to the C of the SC, and the empty SC-predicate would correspond to the trace of the head of that predicate, an "intransitive" P, raised to C.
26.
It is as yet not clear how T and Agr are best fitted into this analysis.
27.
For clarity of exposition, I sidestep the issue of the licensing of the SC-predicate, taking it that a SC-predicate with a null head needs no further licensing. The issue will come up again below, however.
28.
Because the +V of W2 is itself identified by the vP, we may assume that it can itself be used to identify a second occurrence of +V.
29.
Here, the assumption that a SC with a null head needs no further licensing is crucial.
30.
That V2 movement should involve remnant VP-raising is independently supported by the prop erties of negation in Norwegian. Many Norwegian speakers can license an inherently negative phrase like ingen "nobody" by raising it to SpecNegP (according to Kayne's (1997) analysis of negation) only in V2 contexts, i.e. they have the contrast in (i-ii) (cf. Svenonius, this volume): (i) *fordi VI zngen m¢tte because we nobody met (ii) Vi m¢tte ingen. we met nobody Assuming Kayne's account, the ungrammaticality of (i) might be attributed to the Neg head needing to adjoin to W which in tum requires VP-movement to its Spec. But on this assump tion, the derivation of the V2 sentence in (ii) should involve VP-raising.
31.
If this account of object shift is correct, precedes the subject, may suggest that VP-raising rather than V-raising: (i) Diirfor gomde sig therefore hid themselves
32.
Swedish sentences like (i), where the weak pronoun even the final step of V2 (sometimes) is remnant
bamen. the. children
Holmberg's generalization has been interpreted in a variety of ways in the syntactic literature, e.g. as a reflex of Chomsky's (1992) "minimal link condition" in conjunction with his notion of "equidistance", an account which Holmberg (1997) shows to be incapable of handling cases like (i), where object shift is licensed by Yep) topicalization:
DERIVATIONS LEADING TO VO-ORDER (i)
121
Kysset har jeg henne ikke. kissed have I her not
33.
Anders Holmberg (p.c.) suggests the following: Weak pronouns are generated in the same positions as other DPs, but have the special property that they, rather than Prt, are attracted to W in W [PrtP [sc n Pred] Prt VP], when Prt is null. Subsequently, VP is raised to SpecWP and derivations proceed as in the text. Since, by assumption, a particle can only be introduced under the functional Prt head in Swedish, this correctly predicts that even weak pronouns follow verb particles in Swedish, but must precede particles in Danish, where particles are only introduced in the SC-predicate. However, Norwegian, which may introduce particles either in the SC-predicate or under Prt, is incorrectly predicted to allow weak pronouns preceding particles.
34.
Corresponding to heads of SC-predicates that have not raised to C, on the interpretation in footnote 25.
35.
See Cinque (1997) and HinterhCilzl (1997) for two different approaches to restructuring, both of which seem compatible with the text proposal.
36.
Among the Germanic VO languages, at least Icelandic seems to have "verb raising" effects at an earlier stage of its development; cf. footnote 22. Hr6arsd6ttir (1999) discusses the change that led to their loss.
37.
This is the "more general problem" alluded to at the end of Section 9.
References Barbiers, Sjef. 1995. The Syntax of Interpretation. Ph.D. dissertation, University of Leiden. Brody, Michael. 1997. Mirror theory. Ms. University College London. Chomsky, Noam. 1992. "A minimalist program for linguistic theory." MIT Occasional Papers in Linguistics 1 . Cinque, Guglielmo. 1997. O n clitic climbing and other transparency effects. Paper presented at MIT. Diesing, Molly. 1997. "Yiddish VP order and the typology of object movement in Germanic." Natural Language & Linguistic Theory 15: 369-427. Dikken, Marcel den. 1995. Particles: On the syntax o/verb-particles, triadic, and causative constructions. Oxford: OUP. Donaldson, Bruce. 1993 . A Grammar 0/Afrikaans. Berlin: Mouton de Gruyter. Fukui, Naoki and Speas, Margaret. 1986. "Specifiers and projection." MIT Working Papers in Linguistics 8. Haider, Hubert. 1992. "Branching and discharge." Arbeitspapiere des SFB 340 23. University of Stuttgart. Haider, Hubert. 1997. "Precedence among predicates." Journal o/Comparative Germanic Linguistics 1 : 3-41 . Hinterh6lzl, Roland. 1997. A XP-movement account of restructuring. Ms. University of Southern California.
122
KNUT TARALD TARALDSEN
Holmberg, Anders. 1 997. "The true nature of Holmberg's Generalization." In Kiyomo Kusumoto (ed.), Proceedings of the North East Linguistic Society 27. Amherst: GLSA, University of Massachusetts. 203-218. Hr6arsd6ttir, porbjorg. 1999. VP Syntax in the History of Icelandic. Ph.D. dissertation, University of TromS!i>. Johnson, Kyle. 199 1 . "Object positions." Natural Language & Linguistic Theory 9: 577-636. Kayne, Richard S. 1 994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kayne, Richard S. 1 998. "Covert vs overt movement." Syntax 1: 128-19 1 . Kayne, Richard S . To appear. "A note o n prepositions and complementizers." In Parame ters and Universals. Oxford: OUP. Koster, Jan. 1994. "Predicate incorporation and the word order of Dutch." In Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi, and Raffaella Zanuttini (eds.), Paths Toward Universal Grammar: Studies in honor of Richard S. Kayne. Washing ton, D.C.: Georgetown University Press. 255-276. Nilsen, 0ystein. 1998. The Syntax of Circumstantial Adverbs. M.A. thesis, University of Troms!i>. Robbers, Karen. 1997. Non-Finite Verbal Complements in Afrikaans. Ph.D. dissertation, University of Amsterdam. Sportiche, Dominique. 1995. "Clitics constructions." In Laurie Zaring and Johan Rooryck (eds.), Phrase Structure and the Lexicon. Dordrecht: Kluwer. Svenonius, Peter. 1994. Dependent Nexus: Subordinate predication structures in English and the Scandinavian languages. Ph.D. dissertation, University of California, Santa Cruz. Taraldsen, Knut Tarald. 1983. Parametric Variation in Phrase Structure. Ph.D. disserta tion, University of TromS!i>. Vikner, Sten. 1987. Case assignment differences between Danish and Swedish. In Robin Allan and Michael Bames (eds), Proceedings of the 7th Conference of Scandinavian Studies in Great Britain. University College London. 262-28 1 . Williams, Edwin. 1978. "Across-the-board rule application." Linguistic Inquiry 9 : 3 1-43 Zwart, Jan-Wouter. 1993. Dutch Syntax: A minimalist approach. Ph.D. dissertation, University of Groningen.
.. ...... ....·. "" .....
in
Anders Holmberg University of Troms¢
Abstract It
is claimed that head-final order is derived by movement of the complement of a head H to the Spec of H in a right-branching structure. The movement is triggered by selection, expressed syntactically as feature-movement from the selectee to the selector, pied-piping the selectee head or pied-piping nothing in the case of head-initial phrases, pied-piping the entire phrase hosting the selectee feature in the case of head-final phrases. In mixed order structures a head-final phrase may be embedded as a complement of a head-initial phrase, but not vice versa, universally. This generalization is explained by postulating that selection requires adjacency between selector and selectee, either the relevant features or the heads hosting these features. The theory is exemplified mainly by facts from Finnish, a mixed order language. The special condition on OV order in Finnish, that the sentence must have sentence-initial focus, is accounted for by postulating a focus head in IP, obligatory except when the sentence has initial focus, which takes VP as complement and (as a language particular property) attracts the verb, creating a head-initial phrase, blocking the derivation of higher head-final phrases.
1.
Introduction
Finnish is standardly classified as an SVO language, based on the fact that SVO is the unmarked order in declarative sentences. Yet SOY order occurs as an optional variant in certain contexts. This word order variation is interesting for several reasons. To begin with, the conditions on SOY order are intriguing: SOY order occurs typically when the sentence has initial focus, as in a wh-question or a declarative sentence with a sentence-initial contrastively focused constituent.
ANDERS HOLMBERG
124
The preverbal object cannot itself "convey the main news of the sentence", in the words of Vilkuna ( 1989). Another interesting fact about the mix of VO and OV which Finnish exhibits is the following: A verb, auxiliary or main, may precede or follow its complement whether it is a nonfinite clause, a DP, a PP, or a VP. However, if an auxiliary verb selects a VP headed by a verb V which itself takes a complement, call it 0, then the selected VP can precede the auxiliary only if o precedes V. In other words, the orders Aux-V-O, Aux-O-V, O-V-Aux are permitted, but the order V-O-Aux is excluded. The same phenomenon has been observed in other mixed order languages; see Hr6arsd6ttir ( 1999, this volume) on Old Icelandic. This is a special case of a more general pattern: In general, a language which is head-initial in VP is head-initial all the way up to CP, while a language which is head-final in VP need not be head-final all the way up. As will be shown, this is supported by observations concerning word order universals among the languages of the world, based on Dryer ( 1992). Cf. also Sigursson's ( 1988) suggestion that change of word order type from head-final to head-initial is always top-down: first IP-C changes to C-IP, then VP-I to I-VP, then OV to VO. The generalization which captures these facts can be expressed as follows: (1)
If a phrase Ci
Ci is head-initial, then the phrase � immediately dominating is head initial. If Ci is head-final, � can be head-final or head-initial.
In this paper I will present the pertinent facts concerning the Finnish SOY construction, and propose a formal explanation of them. This explanation is based on a particular explanation of generalization ( 1), which in turn is based a theory of OV order suggested in Kayne ( 1994) and elaborated in Julien (2000). 2.
Properties of the Finnish OV construction
The unmarked word order in declarative finite sentences in Finnish is SVO, as in (2). (2)
a.
Jussi kirjoitti
romaanin.
Jussi wrote (a) novel
b. *Jussi romaanin kirjoitti. Jussi novel wrote SOY order occurs quite frequently, though, as a marked alternative. The condition can be formulated as follows; see Vilkuna ( 1989: 1 3 1). 1 (3)
OV order in finite sentences is possible if and only if the 0 and V are embedded in a sentence headed by a focus-marked C.
DERIVING OV ORDER IN FINNIS H
125
So the word order in (2b) is, in fact, well formed under one condition, that is if the subject is (contrastive) focus. The analysis of (4a) is shown in (4a'): the focused subject is in the Spec of a C with a focus feature. (4b) is a question marked by a question particle on the focused, initial constituent, which in this case is the subject. (4c) is a wh-question, and (4d) has a focused, initial verb, with a focus particle attached to it. The analyses are given in (4b', c', d'), respectively. (4)
a. b. c. d.
(4')
a'.
fUSSl romaanin kirjoitti.
wrote Jussi novel 'It was Jussi who wrote a novel.'
fussiko romaanin kirjoitti?
wrote Jussi.Q novel 'Was it Jussi who wrote a novel?'
Milloin fussi romaanin kirjoitti?
when Jussi novel wrote 'When did Jussi write a novel?'
Onpas fussi romaanin kirjoittanut.
has.Foc Jussi novel written 'Jussi HAS written a novel.' [cp ruSSI [c' C [IP t . . . ]]] [FOC]
b'.
[cp Jussi [c' [c -ko] [IP t . . . ]]] [Q]
c'.
[cp Milloin [c' C [IP . . . t . . . ]]] [Q]
d'.
[cp [c' [c on [c -pas] [IP . . . t . . . ]]]] [FOC]
Taking Q to be a kind of focus feature, the generalization is that the sentence in each case has a focus feature in C which attracts a category to SpecCP (if the category is a phrase) or to C (if the category is a head). The analysis (4c') is uncontroversial; Finnish has a 'standard' form of wh-movement, with obligatory movement of one and only one wh-phrase to SpecCP. The analyses (4b', d'), too, are relatively uncontroversial: The question or focus particle attaches to the first constituent of the sentence, with the result that that constituent is focused, except in unmarked yes/no-questions, where the finite verb is moved to initial position to serve as host of the question particle. Compare (4b) with (Sa-c), where (Sc) shows that the question particle must attach to the first constituent.
126
ANDERS HOLMBERG (5)
a. b.
Romaaninko Jussi kirjoitti?
Jussi wrote novel.Q "Was it a novel that Jussi wrote?"
Kirjoittiko Jussi romaanin ?
wrote.Q Jussi novel "Did Jussi write a novel?"
c. *Jussi kirjoittiko romaanin ? Jussi wrote.Q novel This is accounted for if the question or focus particle is a realization of an affixal feature in C, attracting a focus able category to SpecCP (if it is a phrase) or C (if it is a head), with attraction of the closest head, namely the finite head (verb, auxiliary, or negation) as a last resort.2 See Holmberg et al. (1993) on the syntax of IP in Finnish. The analysis (4a') is less obviously correct. How can we tell that the initial contrastively focused subject is in SpecCP rather than in the usual subject position, say SpecIP? There are, in fact, strong indications that the subject has moved to SpecCP in this case: As discussed by Holmberg (1997), an epistemic adverb cannot precede a phrase in SpecCP, for instance a fronted wh-phrase or a fronted, contrastively focused object (in Finnish and a number of other languages). (6)
a.
(*Oikeastaan) milloin sina (oikeastaan) tulet takaisin ?
b.
(*Ilmeisesti) taman romaanin (ilmeisesti) Jussi (ilmeisesti)
actually when you actually come back 'When are you actually coming back?' apparently this
novel
kirjoitti (ilmeisesti) jo
apparently Jussi apparently
nuorena.
wrote apparently already as.young 'This novel Jussi apparently wrote already as a young man.' As shown by (6b), the epistemic adverb can precede the subject. As shown by (7), an epistemic adverb cannot precede a contrastively focused subject, though, most clearly so when the sentence exhibits OV order. (7)
(*Ilmeisesti) JUSSI (ilmeisesti) taman romaanin (ilmeisesti) on
apparently Jussi
apparently this
(ilmeisesti) kirjoittanut.
novel
apparently has
apparently written
This follows if the contrastively focused initial subject is in SpecCP. In addition,
127
DERIVING OV ORDER IN FINNIS H
note that we can insert a resumptive pronoun between the contrastively focused initial subject and the finite verb. 3 (8)
JUSSl se
taman romaanin on kirjoittanut.
Jussi he/it this
novel
has written
The pronoun is presumably in SpecIP; see Holmberg & Nikanne (to appear). I conclude that the cases of OV order in (4) all have a phrase or word moved to the checking domain of C. See also Vilkuna (1989, 1995) for a discussion of initial contrastive focus in Finnish. The examples under (9) show that the focus licensing OV order has to be sentence-initial. (9a) shows that contrastive focus can be indicated by intonation alone, without (overt) movement.4 (9b) shows that this does not license OV order between the direct object and the verb. (9c), finally, shows that movement of the focused indirect object to initial position (SpecCP) does license OV order. (9)
a.
Jussi sai sen kellon ANNALTA.
Jussi got that watch from.Anna 'It was from Anna that Jussi got that watch. '
b. *Jussi sen kellon sai ANNALTA. Jussi that watch got from.Anna c. ANNALTA Jussi sen kellon saL from.Anna Jussi that watch got 'It was from Anna Jussi got that watch.'
The examples under (10) show that the focused category and the verb undergo ing the VO-OV switch need not be in the same simplex clause. In ( lOa) the wh phrase fronted from a position in the main clause licenses OV in the embedded clause, shown by the fact that the same structure without the wh-phrase is ill formed with OV order. (10)
a.
Mista
sina kuulit [etta Jussi romaanin kirjoittaa] ?
from.where you heard that Jussi novel
b. *Mina kuulin [etta Jussi romaanin kirjoittaa]. I heard that Jussi novel is.writing
is.writing
As shown in (11) the complement in the OV construction can scramble. In this example the object precedes the main verb, but the auxiliary ole (here in the conditional form olisi) and the modal verb voi (here in the participial form voinut) each precede their complement. In this case the object can occur in any of the indicated positions, with no discernible effect on the interpretation.
128
ANDERS HOLMBERG
( 1 1)
Miten Jussi (romaanin) olisi
how Jussi novel
(romaanin) voinut (romaanin)
would.have a novel
could a novel
kirjoittaa ?
write 'How could Jussi have written a novel?' If there are several verb complements, they may scramble to different positions. (12a) is a consistent VO structure. In (12b) both of the bold-face constituents of the small clause have moved, each to its own position. (12)
a.
Mina olen kuullut lussin puhuvan ranskaa.
b.
Milloin sina lussin olet ranskaa kuullut puhuvan ?
I
have heard Jussi speak
French
when you Jussi have French heard speak 'When have you heard Jussi speaking French?'
If there are several verb complements, some may precede and others follow the verb.
(13)
a.
Minka takia sina pullat panitjiiiikaappiin ?
b. c.
Minka takia sina panit pullatjiiiikaappiin ? Minka takia sina pullatjiiiikaappiin panit?
what reason you buns put in. fridge 'Why did you put the buns in the fridge?'
The examples under (14) show that the complement of the auxiliary ole 'be, have' and other verbs taking nonfinite verbal complements may precede their selecting head. ( 14a) is a consistent VO structure. In (14b) the DP object precedes the main verb. In (14c) the VP headed by the main verb precedes the auxiliary verb (here in the conditional form olisO. However, as shown by (14d), when the VP precedes the auxiliary, the object must precede the main verb. (14)
a.
Milloin Jussi olisi
kirjoittanut romaanin ?
b.
Milloin Jussi olisi
romaanin kirjoittanut?
c.
Milloin Jussi romaanin kirjoittanut olisi.
when Jussi would.have written
when Jussi would.have a novel
a novel
written
when Jussi a novel written would.have 'When would Jussi have written a novel?'
d. *Milloin Jussi kirjoittanut romaanin olisi? when Jussi written a novel would.have
Thus, even though Finnish in general appears to have OV order optionally, OV order
DERIVING OV ORDER IN FINNIS H
129
in the VP is obligatory in this case, that is when the VP is itself embedded as a preverbal complement. This is an instantiation of generalization (1), repeated here: (1)
If a phrase ex is head-initial, then the phrase � immediately domi nating ex is head-initial. If ex is head-final, � can be head-final or head-initial.
Furthermore, when there are two or more verbs taking verbal complements, each verbal complement may precede its selecting head. (15a) is consistently va, while (15b) is consistently OV, with a sequence of sentence-final verbs mirroring the order in the va sentence. (15)
a.
Milloin Jussi oUsi
b.
Milloin Jussi romaanin kirjoittaa ehtinyt oUsi?
c.
Milloin Jussi romaanin kirjoittaa olisi ehtinyt?
ehtinyt
kirjoittaa romaanin ?
when Jussi would.have had.time write novel 'When would Jussi have had time to write a novel?' when Jussi a novel
write
had.time would.have
when Jussi a novel write would.have had.time
As shown in (15c) the order of the sentence final verb cluster need not be strictly OV in the sense that the highest verb is furthest to the right. The contrast between (15b-c) looks, on the face of it, like the contrast between German and Dutch: In (standard) German the sentence final verb cluster has OV order (except for the IPP construction). In (standard) Dutch, although nonverbal objects all precede their head, the order of the verbs is "Va", i.e. the higher the verb, the further to the left. 5 In most theories of German and Dutch this variation is accounted for under the heading of Verb Raising, an operation (or series of operations) which has the effect of creating a sentence final verbal cluster which cannot be broken up by intervening constituents; in this sense the verb cluster is wordlike. In Finnish, the sentence final verb cluster in (15b) is similar to its German counterpart in that it cannot be broken up by intervening material. As I will discuss in more detail below the claim is that it is derived essentially by the same process that derives the German verb cluster, that is successive movement of complement phrases leftwards: simplifying somewhat, first the infinitival clause moves to the Spec of the governing head ehtinyt 'had time' , then the resulting constituent moves to the Spec of the auxiliary. The final verb sequence in (15c), on the other hand, does not form an opaque verb cluster. Consider the fact that an adverbial may intervene between the two final verbs in (15c), as shown in (16a), but not between the two final verbs in (15b), as shown in (16b).
130
ANDERS HOLMBERG
(16)
a.
(siina
Milloin Jussi romaanin kirjoittaa olisi
when Jussi novel
write
would.have in.that
tapauksessa) ehtinyt?
b.
case
had.time
Milloin Jussi romaanin kirjoittaa ehtinyt (* siina
when Jussi novel
write
had.time
in.that
tapauksessa) olisi ?
would.have case 'When would Jussi, in that case, have had time to write a novel?' I therefore assume that the word order in (15c) is the result of long movement of the complement of ehtinyt, across both that verb and the auxiliary. The order of Neg and its complement is strictly VO. 6 (17)
a.
Miksi Jussi ei romaania kirjoittaisi ?
why Jussi not novel would. write 'Why would Jussi not write a novel?'
b. *Miksi Jussi romaania kirjoittaisi ei.
The order of (finite) C and its complement is also strictly YO. (18)
a.
Milloin Jussi ilmoitti etta han tuZee ?
when Jussi reported that he comes 'When did Jussi report that he would come?'
b. *Milloin Jussi ilmoitti han tuZee etta.
Finally, a complement CP can never precede the verb (as in the Germanic OV languages) : (19) *Milloin Jussi [etta han tuZee] ilmoitti ? when Jussi that he comes reported 3.
Finnish as an OV language
3 . 1 The head-chain hypothesis of OV order The properties of the Finnish OV construction, in particular its discourse semantic properties, are discussed in Vilkuna (1989, 1995), who observe� the connection between sentence initial focus and OV order. She does not actually propose any explanation of this connection, though. A kind of formal explanation is suggested in Holmberg (1997): The idea is that OV order is the result when V
131
DERIVING OV ORDER IN FINNIS H
i s not moved to I . Movement of V to I i s otherwise obligatory, at least in finite clauses, but is optional when C contains a focus feature, because such a C somehow licenses an empty I. Let us refine this hypothesis, which I will refer to as the head-chain hypothesis, to see how far it will take us. First, the hypothesis that OV order is the result of not moving the verb presupposes that Finnish is underlyingly an OV language, either in the sense that OV is a base-generated order, or in the sense that movement of 0 to a position preceding V in VP is an obligatory rule. Second, assume that the feature in I which triggers V-movement is a finiteness feature (cf. Holmberg & Platzack 1995). Assume the finiteness feature needs to be checked either by movement of the properly inflected verb to I, or by movement of the finiteness feature to C, which is possible if C contains a focus feature. That V-movement is associated with finiteness is a familiar observation from the study of V2 languages. That there is a close connection between focus and finiteness is less often noted but probably true nonetheless; for example, there are languages where a focus particle serves as the only marker of finiteness (see Testelec 1997). So the present hypothesis has a certain initial plausibility. If V-movement in Finnish is triggered by the need to check a finiteness feature in I, the prediction is that we should find OV order in nonfinite clauses, independent of focus. This is partly true: OV order occurs not infrequently in infinitival clauses, even in the absence of any sentence-initial focus, as in (20a), in contrast with the finite clause (20b). (20)
a.
Mina en osaa viela [siihen kysymykseen vastata]. I 'I
not can yet that question can't answer that question, yet.'
answer
b. *Mina jo siihen kysymykseen vastasin. I already that question answered Matters are complicated by the fact that VO order is always an option, and often preferred or even obligatory in infinitival clauses, unless the matrix clause has initial focus. We would have to postulate other, optional, triggers for verb movement in addition to finiteness. However, there are other, stronger reasons for rejecting the head-chain hypothesis. To begin with, clearly OV order is not only the result of not moving the verb, since 0 moves as well in various constructions; cf. the example of scrambling shown in (12b), repeated here: (12)
b.
Milloin sina oiet lussin
kuullut ranskaa puhuvan?
when you have Jussi.GEN heard French speak
1 32
ANDERS HOLMBERG
Here Jussin is the subject of a small clause (a nominal participial small clause, hence the genitive case; see Koskinen 1998) which is selected by the verb kuullut 'heard' . Thus, excluding the possibility that the verb kuullut is lowered into the small clause, there is no way that the order in (l2b) can be derived without movement of Jussin leftwards, or (less plausibly) movement of the small clause predicate rightwards. This movement is conditional on the sentence initial focus. Second, the relation between the triggering C and the verb's potential landing site is non-local. (21)
a.
Miksi sina pelkaiit [etta se palamaan syttyy] ?
why you fear that it burning starts 'Why are you afraid that it will catch fire?'
(Vilkuna 1989)
b. *Mina pelkaan [etta se palamaan syttyy] . I fear that it burning starts
Here the licensing focus feature is in the matrix clause C, but the finite I licensed by the focus feature is in the embedded clause. (21b) shows that the order in the embedded clause is, indeed, conditional on the focus feature in the matrix clause. Under the head-chain hypothesis, and given standard assumptions about head head relations, we would have to assume some kind of head-chain relation, linking the initial focus-marked C with the embedded finite I, via a chain of intervening heads. But this analysis is compromised by the fact that the structural restrictions on the intervening heads are not the ones we expect from a head chain. Consider (22), which shows that the finite I licensed by the matrix focused C can be embedded in a relative clause. In (22a) the order in the relative clause is VO. (22b) , with OV order, is clearly not well formed. (22c) is well formed (although somewhat stilted), due to the focused matrix clause subject, which, as discussed, is in SpecCP, checking a focus feature in C. (22)
a.
Olen jo
nahnyt sen elokuvan josta
Lhave already seen
the movie
ne puhui
about.which they talked
uutisissa.
on.news
b. *Olen jo nahnyt sen elokuvanjosta ne uutisissa Lhave already seen the movie about. which they on.news puhui.
c.
talked
Minakin olen
Ltoo
sen elokuvan nahnyt josta
Lhave that movie
uutisissa puhui.
on.news talked
seen
ne
about.which they
DERIVING OV ORDER IN FINNIS H
1 33
Postulating a head chain relation linking a head in a relative clause with a head outside the DP containing the relative clause would entail giving up virtually all known locality conditions on head-chain relations, except perhaps the c-command condition. 3.2 OV order and information structure
There is another condition on the OV construction, observed and discussed by Vilkuna ( 1989, 1 995), which does not follow from the head-chain hypothesis. In Vilkuna's words, the preverbal V-complements cannot convey the main news of the sentence. In other words, 0 in the OV construction should be old informa tion. Initial focus does not help if 0 is not old. Consider, for example, the following conversation between A and B : (22)
A: What are you going to give the kids for Christmas? i. JUSSILLE minii annan KIRJAN. to.Jussi I (will) give book ii. #JUSSILLE minii KIRJAN annan.
B:
Here the alternative (ii) is sharply inappropriate even though the sentence has a focus feature in C, checked by the focused subject. The reason is that the object conveys new information. 7 Note that 0 in the OV construction can be contrastive (hence in that sense focus), as long as it is old, in the sense of being already a topic of the discourse, as in (23), assuming a discourse where Jussi's novel is under discussion. (23)
Miksi Jussi ROMAANIN kirjoitti? Miksei VAITOSKIRJAA ?
why Jussi novel wrote why.not dissertation 'Why did Jussi write a NOVEL? Why not a dissertation?'
In some cases it is almost sufficient that a verb complement is old information, for OV order to be acceptable. In (24) and (25) the embedded finite clauses are factive, conveying just old information. OV order may be unpreferred to VO order in this case in the absence of initial focus, but not inconceivable. (24) ?Se pelotti
minua [ettii se niin helposti palamaan syttyi].
it frightened me that it so easily burning 'It frightened me that it caught fire so easily.'
started
1 34
ANDERS HOLMBERG
(25)
? [Se etta tala
palamaan syttyi] ei merkitse ettei
it that house burning
enaa
asua.
slzna VOlSl
started not means that.not in.it you can
anymore live 'The fact that the house caught fire doesn't mean that you can't live in it anymore.' 4.
The universal validity of Generalization 1
One of the goals of the present paper is to propose an explanation of Generaliza tion 1 , repeated here: (1)
If a phrase ex is head-initial, then the phrase 13 immediately dominat ing ex is head-initial. If ex is head-final, 13 can be head-final or head initial.
The idea is that once we have an account of this generalization, the Finnish word order facts discussed above will fall out, given some additional parameter distinguishing between Finnish and other languages which are also mixed VO/OV but do not exhibit exactly the same pattern as Finnish, with OV dependent on a focus-marked C. Besides, whatever we think about Finnish, a theory of word order which has no explanation of Generalization 1 cannot be the right theory. How general is Generalization I ? We can get an idea by looking at the results of a large-scale cross-linguistic survey of head-complement order, such as Dryer (1992), the largest and most authoritative one to date. In this work Dryer tests a number of the word order universals proposed by Greenberg ( 1966) (and other generalizations in the same vein) against a much larger sample of languag es than Greenberg's, carefully sampled so as to avoid genetic and areal biases. Dryer's total sample was 625 languages, or more correctly "genera", where a genus is a group of closely related languages roughly of the order of the Germanic languages. 8 What Dryer investigated was the correlation between V and 0 order and a the order of a number of other pairs of categories that have been claimed to be analogues of V and 0, such as C and IP, Aux and VP, Article and NP, etc. One of the pairs he looked at is Want (a volition verb) and the main verb, the head of the complement of Want. (I use capitalized 'Want' to refer to a lexical item expressing volition.) On the basis of Greenberg's Universal 15 the expectation is that VO languages should strongly favour the order Want V, while OV languages should favour V Want. Dryer's (1992) results are shown in (26). The table can be read as follows: Dryer found data concerning Want
DERIVING OV ORDER IN FINNIS H
135
from 8 5 genera in his sample. Out of 46 va languages/genera 42 had the expected order Want V, and only 4 the order V Want. , Out of 39 OV languages/ genera 29 had the expected order V Want, while 10 had the order Want V. (26)
Want V & va Want V & OV V Want & va V Want & OV
42 10 4 29
genera genera genera genera
Dryer (1992) concludes that Want and V pattern like V and 0, as Greenberg ( 1966) had claimed. Consider Generalization 1 : If it is correct, we should find the orders Want-v-a, Want-a-V, 0-V-Want, but not V-a-Want. This is clearly the tendency in the table under (26). There are four potential conterexamples, though, the four V want & va languages. Do they, in fact, exhibit the order V-O- Want? Matthew Dryer has informed me that he now has seven languages/ genera in his database exhibiting the pattern V Want & va. I have so far been able to check his sources for four of those languages (thanks to Matthew Dryer's information). None of them contained any example of the word order V-O- Want. Two of them (Luvale and Pero) 9 did not, in fact, contain any examples of postverbal Want. Two of them did: Yindjibamdi and Guajajara. l O However, in these examples the main verb was intransitive. Both of these languages allow va as well as OV order. We are therefore not surprised to find that they allow the main verb, presumably heading a VP, preceding Want. What we expect, given Generalization 1, is that when the main verb precedes Want, the object, if there is one, will precede the main verb (0- V-Want) or possibly, will be found to the right of Want (V-Want-a), depending on other properties of the construction. ll In fact, mixed order languages like Finnish and Old Icelandic, if they had been in Dryer's (1992) sample, and if he had had access to rich enough data, would have appeared in the category V Want & va, since they are generally classified as va, but do allow the main verb to precede Want, along with other head-final constructions. However, they do not allow V-O- Want. A word order generalization which apparently is universal is that no va languages have sentence-final complementizers (see Dryer 1 992). This tallies well with Generalization 1 : A VP with the order v-a cannot be embedded in a VP-I but only in an I-VP structure, and an I-VP structure can only be embedded in a C-IP structure. Internal to Finnish we can test the validity of Generalization (1) in nominal as well as verbal projections. Finnish has the order N-Gen as well as Gen-N. (27)
a.
kuva
Marjasta
picture of.Marja
136
ANDERS HOLMBERG
b.
Marjan kuva
Marja's picture
Finnish also has postpositions as well as prepositions. There are even some adpositions which can be either. (28)
a. b.
ilman
kuvaa
without picture kuvaa ilman
picture without
If an NP containing a Gen is embedded in such a PP, the permitted orders are
P-N-Gen, P-Gen-N, Gen-N-P, but not N-Gen-P, in accordance with Generali zation 1 . (29)
a. b. c.
ilman
[kuvaa Marjasta]
ilman
[Marjan kuvaa]
without picture of.Marja without Marja's picture [Marjan kuvaa] ilman
Marja's picture without d. * [kuvaa Marjasta] ilman picture of.Marja without We conclude that Generalization 1 is valid over a wide spectrum of constructions across a wide spectrum of languages. 5.
How to derive OV order
The theory is based on a set of more or less controversial assumptions. First, I assume that inflected forms are derived by movement in the syntax, as used to be widely assumed within GB theory; see for example Baker ( 1988). More recently this theory has fallen in disrepute, following Chomsky's ( 1 995) revival of the lexicalist theory of inflection. However, forthcoming work by Marit Julien shows that the incorporation theory of inflection is well motivated both empiri cally and theoretically. 1 2 Another assumption is that all phrase structure is right branching (following Kayne 1 994; Haider 1 992, 1 997, this volume). I further more assume that the lower section of a transitive VP has the order OV univer sally, where I do not take a stand on whether the OV order is base-generated or derived by an obligatory O-movement operation.
DERIVING OV ORDER IN FINNIS H
1 37
Second, following a suggestion by Kayne ( 1 994), developed by Julien (2000), I assume that head-final structures are derived by movement of the complement of a head H to SpecHP. The derivation of a structure with three heads is shown schematically in (30): (30)
a. b. c. d.
Y [xp 0 X] [yp [0 X] [Y t]] W [yp [0 X] [Y t]] [wp [[0 X] [Y tn [W t]]
In (30a) Y is merged with XP (where the complement 0 already precedes X). In (30b) the XP [OX] has moved to the Spec of YP. In (30c) W is merged with
and in (30d) yP has moved to the Spec of WP. Successive application of complement movement in this fashion yields a structure which is effectively left branching except for the trace of movement left in each complement position. The theoretical status of the trace is unclear. It is presumably not a site of reconstruction for example, so it may well be that the movement depicted in (30) leaves no trace in the sense of a phonetically empty copy or any kind of empty category - in which case the movement derives a truly left-branching structure. However, the way this theory is set up, it is crucial that the structure is in fact derived by movement. Third, the trigger of complement movement is c-selection (Pesetsky 1982) or strict subcategorization (Chomsky 1 965). Following Svenonius ( 1994) I assume that this relation always involves movement, as follows: Every head selecting a complement has a feature specifying some syntactic feature(s) of its complement, essentially Chomsky's ( 1 965) subcategorization feature, here called a selectionfeature. The selection feature is an uninterpretable feature in the sense of Chomsky ( 1 995: Ch. 4), and therefore has to be checked and erased in the course of the syntactic derivation. It does so by attracting the relevant feature(s) of the complement. That is to say, just Merge of a head with its complement is not sufficient to check the selection feature: the feature selected has to move to the checking domain of the selection feature. This movement can take three forms, listed under (3 1): YP,
(3 1)
a. b. c.
Pure f(eature)-movement (i.e. covert movement; see Chomsky 1 995: Ch. 4), head movement, i.e. f-movement pied-piping the minimal word containing the relevant feature, XP-movement, i.e. f-movement pied-piping the minimal maxi mal category containing the relevant feature.
ANDERS HOLMBERG
138
Svenonius ( 1 994) recognizes (3 1 a) and (b) but not (c), but then Svenonius does not take OV languages into account. Following an idea suggested by Kayne ( 1 994) I will assume that so-called VO languages are languages which prefer (a) and (b), while so-called OV languages are languages which prefer (c). (32) is an example of the strategy (3 1 a) : (32)
[[c that] [IP John has+I . ]] .
.
There i s a selection feature in C attracting a feature o f L We may represent the feature as [_I] . There appears to be no pied-piping/overt movement involved in this case, so LI] appears to be checked by pure feature movement. 1 3 The strategy (3 1b) may be exemplified by (33a), and the strategy (3 1 c) by (33b) . (33)
a.
Milloin Jussi olisi when
b.
Milloin Jussi olisi when
kirjoittanut romaanin?
Jussi would-have written
a novel
romaanin kirjoittanut?
Jussi would-have a novel
written
Consider the derivations in detail: In both constructions the first step is forming the VP by merging the verb and its object, followed by movement of the object to a specifier position, let us say Spec VP; I omit taking the subject into account at this point, as well as any functional heads below the participle head labelled Prc. The next step is that the participle head -nut is merged with the VP (the trace of the object will be represented as to the trace of the verb as tv' etc.). ' (34) [PreP -nut [vp romaanin kirjoitta- to]] This head has a selection feature LV] attracting the feature [V] of the head of the VP. Since -nut is an affix the feature [V] has to pied-pipe (at least) the whole verb stem. At this point there are two options: (a) pied-pipe the minimal word dominating [V] , and adjoin it to Prc, or (b) pied-pipe the minimal maximal phrase dominating [V], and place it in SpecPrc. The result of applying option (a), the head movement option, is (35). ,
(35)
[PreP [Pre kirjoitta-nut] [vp romaanin 1y to]]
The result of applying (b), the XP-movement option, is (36): (36)
[PreP [vp romaanin kirjoitta- to] [Pre' -nut 1yp]]
A s a result o f the movement the verb stem and the participle affix are morpho logically merged into a word. This is uncontroversial in the case of (35) . Following Kayne ( 1 994) and Julien (2000) I assume that the verb stem and the
DERIVING OV ORDER IN FINNIS H
139
affix can be interpreted as a word in (36) as well, even though they do not fonn a syntactic constituent. The necessary and, by assumption, sufficient condition for the principles of the morphological component (henceforth called Morphological Structure, or MS, as in Halle and Marantz 1 993) to interpret the stem and the affix as a word is that they are phonologically adjacent, which they are in (36), being separated only by a trace (which may in fact not be there). Thus (36) as well as (35) satisfies the requirements of both Syntax (feature checking) and Morphology (word fonnation). See Julien (2000) for plenty of supporting evidence that MS is sensitive strictly to adjacency, not to syntactic constituency. 1 4 In tenns of economy of derivation the head-movement and the XP-movement option can be assumed to be equal: The head-movement option is more economi cal in the sense that it moves fewer features (cf. Chomsky 1995: Ch. 4 for an elaboration of the Last Resort condition in tenns of feature movement). On the other hand the XP that is a complement of a head H is structurally closer to H than the head XO of XP, so XP-movement is technically shorter than head-movement. Note that checking the selection feature of the participle head by VP-move ment presupposes that the VP have OV fonnat. If it had VO fonnat, the object would block the morphological merger of the verb stem and the participle affix. The head movement strategy, on the other hand, is compatible with either a VO or an OV complement. This is the beginning of an explanation of Generalization (1). The generalization is that structures of the fonn [H2P [HI X] H2] where X is overt are not found. When the selecting head H2 is an affix, this structure is ruled out by conditions on word fonnation: the selector affix and the selectee head have to be phonologically adjacent. What happens when H2 is not an affix? It seems that the adjacency condition between the selector head and the selectee head is a condition even then. Consider for instance the Gennan verb cluster, as in (37): (37)
dass er dieses Buch lesen konnen muss
that he this book read can must 'that he must be able to read this book'
Given that the verb cluster is a result of successive selection-triggered XP-move ment, the structure is as in (3 8 ): (3 8 )
[AuxP [AuxP [yp dieses Buch lesen to] konnen 1:"p] muss tauxp]
As well known, there is a strict adjacency condition holding among the verbs in the verb cluster. No phonologically visible category is allowed to intervene. In this sense the verb cluster is word-like. Yet the constituent parts are not affixes, judging by the fact that they do occur as independent words in other contexts, in
1 40
ANDERS HOLMBERG
particular, they undergo movement as independent heads under V2. In the case of Finnish, the effect of the adjacency condition can be seen in (39) : In (39b) the object intervenes between the fronted V and the auxiliary. (39)
a.
Milloin Jussi romaanin kirjoittanut olisi?
when Jussi novel written would.have b. *Milloin Jussi kirjoittanut romaanin olisi? The examples in (40) and (4 1 ) show that while an adjunct may intervene at various places in a multi-verb structure, it may never intervene between two verbs in an an OV structure. (40)
a.
Milloin Jussi (sinun when
mukaan) olisi
Jussi according to you
mukaan) kirjoittanut (sinun
(sinun
would.have according
mukaan) romaanin ?
to.you written according to.you novel 'When would Jussi, according to you, have written a novel?' b.
Milloin Jussi (sinun
mukaan) romaanin (sinun
when
Jussi according to.you a.novel according mukaan) kirjoittanut (*sinun mukaan) olisi? to.you written according to.you would.have (41 )
a.
Ei Jussi tule
(hiinkiiiin) hyviiksymiiiin (hiinkiiiin) sitii
not Jussi come he.neither accept
he. neither that
ehdotusta. proposal 'Jussi will surely not accept that proposal, either.' b.
Ei Jussi (hiinkiiiin) sitii ehdotusta (hiinkiiiin) hyviiksymiiiin not Jussi he.neither that proposal (*hiinkiiiin) tule. he.neither come
he. neither accept
What this suggests is that Generalization 1 may be a special case of something more general: In a head-final structure the selector and selectee heads must be adjacent, regardless of structure. I 5 When the selector is an affix , the movement to check the selection feature must be overt. By assumption they have a feature [affix] which is checked by a phonological feature matrix, forcing pied-piping of a head or a phrase. Assume that not only "real affixes", but various independent words as well have a feature which forces pied-piping of a head or a phrase with the selected fea ture(s). In terms of Chomsky ( 1 995), the selector feature can be weak or strong, its strength being fixed independently for each head. Now stipulate (42).
DERIVING OV ORDER IN FINNIS H (42)
141
The selection feature F of a head ex is checked by the selected feature F' of the head 13 if and only if either F' or 13 is adjacent to ex .
If the selection feature F is weak, it is checked by pure feature movement of F'. F' ends up adjoined to ex, satisfying (42). If F is strong, it is checked by overt move ment. If it is head-movement, then 13 will end up adjoined to ex, satisfying (42). If it is XP-movement, the head of 13 must end up adjacent to ex, to satisfy (42).
A consequence of this is that while selection by XP-movement at step s in the derivation is compatible with selection by XP-movement or by f-movement or head-movement at step s + 1 , selection by f-movement or head-movement at step s blocks selection by XP-movement at step s + 1 . To illustrate, consider the derivations of (44a-b), where (44a) is a well formed alternative to (33a-b) while (44b) is ill formed, violating (42): (44)
a.
Milloin Jussi romaanin kirjoittanut olisi.
when Jussi novel written would.have b. *Milloin Jussi kirjoittanut romaanin olisi. when Jussi written novel would.have Take (36) and merge the auxiliary stem ol- 'be' with it: (45)
[AuxP ol- [PreP [yp romaanin kirjoitta- to] [Pre' -nut �]]]
The auxiliary has a selection feature checked by a participle feature. There are two options: Pure f-movement or XP-movement. Head movement is ruled out in this case since (a) the participle head cannot move on its own since it is a suffix, and (b) the verb stem and participle suffix cannot move together since they do not form a constituent. Assume first that the selection feature of ol- is checked by pure f-movement. The next step in the derivation is Merge of T, here in the form of the conditional suffix -isi (see Holmberg et aL 1 993 on the properties of the Finnish tense and mood system). (46)
[TP -isi [AuxP ol- [PreP [yp romaanin kirjoitta- to] [Pre' -nut �]]]]
The T-affix has a selection feature checked by [V] . There is only one option in this case, namely head movement, which yields the word order in (33b). Pure f movement is not sufficient since T is an affix . XP-movement of AuxP, the result of which would be (47), is ruled out as a violation of the adjacency condition (42). (47)
[TP [AuxP ol- [PreP [yp romaanin kirjoitta- to] [Pre' -nut �]]][T' -isi tauxp]]
142
ANDERS HOLMBERG
The selector head and the selectee head do not end up adjacent to each other. The result in this case, where neither the selector nor the selectee head are independent words, is virtually a morpheme salad. Now consider the option when XP-movement is applied to (45). The result is (48), satisfying (42), since the participle head and the selecting auxiliary end up adjacent.
[AuxP [PreP [yp romaanin kirjoitta- to] [Pre' -nut �]] [Aux' ol- tprep]] The next step is merging T, in the form o f -isi. (48)
(49)
[TP -isi [AuxP [PreP [yp romaanin kirjoitta- to] [Pre' -nut �p]] [Aux' oltprep]]]
There are two options for checking the selection feature of T: head-movement of the auxiliary stem, which would give the same word order (though not exactly the same structure) as head movement applied to (46) . The other option is XP-movement, the result of which will be (50): (50)
[TP [AuxP [PreP [VP romaanin kirjoitta- to] [Pre' -nut �]] ol- �rep] [T' -isi tauxp]]
The adj acency condition (42) i s respected at every step, and the resulting string is (44a). To derive the ungrammatical (44b) we start from (5 I a), merge the auxiliary as in (S Ib), move the participle head with the adjoined verb to SpecAuxP, as in (S Ic). Then merge T, as in (S I d), and finally move the AuxP to SpecTP. (5 1 )
a. b. c. d. e.
[PreP [kirjoitta [-nut]] [yp romaanin [v' �]]] [AuxP ol- [PreP [kirjoitta [-nut]] [yp romaanin [v' �]]]] [AuxP [PreP [kirjoitta [-nut]] [yp romaanin [v' �]] [Aux' ol- �rep]] [TP -isi [AuxP [PreP [kirjoitta [-nut]] [yp romaanin [v' tV]] [Aux' ol tprep]]] hp [AuxP [PreP [kirjoitta [-nut]] [yp romaanin [v' �] ] [Aux' ol�rep]] [T' -isi]]]
The forbidden step in this derivation is the XP-movement resulting in (S Ic), forbidden because prior to it, at (5 I a), a selection feature was checked by head movement, resulting in a head-initial phrase, and hence a violation of (42). The only way the derivation can proceed to a convergent result after (5 I a) is by checking selection by head-movement or f-movement.
DERIVING OV ORDER IN FINNIS H
6.
143
The initial focus condition
So the derivation of an OV structure presupposes selection by XP-movement all the way. If at any point in the derivation selection is accomplished by pure f-movement or head movement, selection by XP-movement cannot be applied after that. The explanation I will now propose of the initial focus condition on OV order in Finnish makes use of this generalization. The idea is that any finite clause except those where focus is initial has verb movement at an early stage of the derivation, blocking any later applications of selection by XP-movement. It is a common observation that the focus in the sense of 'the news' of a finite sentence in the unmarked case is contained in VP, while the old or presup posed information is found outside, preceding VP. A prominent exception is when the focus is sentence-initial, as in wh-questions (in languages with wh-movement) or clefts, or declaratives with a contrastively focused initial constituent. I propose the following formal account: Among the categories provided by UG there is a functional head which I label New, hosting a [focus] feature, projecting NewP and taking VP as its complement. The semantic contribution of New is that of defining VP as the domain of new information. I leave open a number of questions concerning the formal properties of New. The properties which are most directly relevant for Finnish sentential word order are listed under (52): (52)
a.
b.
c.
Universally every independent (or root) sentence must contain at least one instance of the feature [focus] ; an independent sentence without focus is filtered out at LF as uninterpretable. That is to say, an independent sentence cannot be all presuppo sition; some part of the sentence must be intepretable as focus/ new information. The focus features which appear in C, including Q and Con trast, count as instances of the feature [focus] . Therefore a wh question or a sentence with a fronted contrastively focused constituent need not contain the head New, although they may do so, in which case the sentence has two foci. A parametrized property of Finnish is that New always attracts the verb. In present terms, the selection feature of New can only be checked by head movement.
Property (52c) explains why only sentences with initial focus permit OV order in Finnish. All independent sentences which do not have a focus feature in C have the head New next to VP, attracting V. Thus all such sentences go through a stage like (53) in the derivation:
1 44
ANDERS HOLMBERG (53)
[NewP V+New [yp Obj tv]]
A s discussed i n the previous section, once head movement i s applied at any stage in the derivation, this blocks subsequent selection by XP-movement. Consequent ly OV sentences cannot be derived in Finnish, in the presence of New. So the reason why (2b), repeated here as (54), is out is that New attracts V, which precludes any subsequent selection by XP-movement. There is no way to derive (54) from (55), provided that the object cannot scramble from its position in VP to a position preceding T (a point which I will return to below). (54)
*Jussi romaanin kirjoitti. Jussi novel
(55)
wrote
[TP -i [NewP kirjoitta+ New [yp romaanin iy]]]
The string Jussi romaanin kirjoitti is, however, well formed if the sentence is analyzed as having a C containing a focus feature attracting the subject, endowed with a feature [contrastive] , to SpecCP. In this case there need be no head New in IP attracting V. Consequently nothing prevents checking the selection feature of T by VP-movement. The result is (56a), analyzed as (56b) (still ignoring the IP-internal position(s) of the subject). (56)
a.
JUSSI romaanin kirjoitti.
b.
Jussi novel wrote 'It was Jussi who wrote a novel. ' [cP Jussi [c' C [TP [yp romaanin kirjoitta-] [T' -i �]]]]
This is consistent with the facts discussed in Section 3 .2: If the object conveys new information, OV order is out even if the sentence has initial focus. One of the examples was the sentence (57) ( (22Bii)), uttered as an answer to the question 'What are you going to give the kids for Christmas? ' , in which context the sentence is totally inappropriate. =
(57)
#JUSSILLE minii KIRJAN annan. to.Jussi
I
a.book give
In this context the expected answer has a VP which contains new information. This entails presence of New taking VP as complement. This triggers V-move ment to New, and this rules out the possibility of a preverbal object. The (somewhat marginal) possibility of OV order in factive clauses follows if they do not have New; presumably a defining characteristic of factive clauses. That OV order in those cases is somewhat marginal falls under the more general observation that OV order is never obligatory: even in cases where VP is clearly
DERIVING OV ORDER IN FINNIS H
1 45
not new information VO is always an option. This follows if the choice between head-movement, f-movement, or XP-movement to check selection is always optional in Finnish, except where other properties of the derivation disallow one option or the other. One of the problems posed by the Finnish SOY construction is the apparent nonlocality of the "ov rule". In Vilkuna's words, on the face of it "a switch at the end of the sentence has an effect on the first constituent" (Vilkuna 1 989: 1 3 1 ), the two positions being in principle infinitely far apart. The problem disappears in the present theory as no direct relation is postulated between the initial focus and the OV structure. OV order is made possible by the absence of sentence internal focus, in the form of the category New, which has two effects: (a) in the absence of a functional head triggering early head movement, selection of VP by means of XP-movement is possible, resulting in OV order, and (b) it makes a higher focus mandatory, since the sentence must have a focus somewhere. I6
7.
Typological predictions
In general, in any language where selection by head movement or pure f-move ment applies at stage s in a derivation D, OV order (or more generally, head final order) cannot be derived in D at any stage subsequent to s. In these terms, what is characteristic of rigid or consistent VO languages is that they have a head merged at an early stage of the derivation which can only select by head movement or pure f-movement. This is all it takes. Characteristic of a mixed language such as Finnish is that the head specified to select by head movement is an optional one (in this case optional in the sense that the feature it hosts can be realized higher up in the sentence). The present theory predicts that languages with [IP C] order have [VP I] order in IP (particles or auxiliary verbs encoding tense, mood, aspect etc. occurring to the right of the verb), and that languages with [VP I] order have OV order. Consequently languages with sentence-final complementizers have OV order. 1 7 The converse implication is not predicted: [VP I] does not imply [IP C] , and OV does not imply [VP I] or [IP C] . The predictions are by and large correct. As mentioned in Section 4 there are said to be no VO languages with sentence-final complementizers. As also discussed in Section 4 there are very few if any languages with the order V-O-Want, while the other 'mixed order' Want-O-V appeared to be fairly common. I 8 With regard to other sentential functional categories such as tense and aspect particles and the negation the situation is less clear. For instance, apparently there are languages of the VO type
ANDERS HOLMBERG
146
with sentence-final tense or aspect particles, presumably indicative of the position of I relative to VP, or with a sentence-final negation particle; see Dryer ( 1 992). If these languages actually allow the order VO in construction with a sentence final tense or aspect particle they are potentially counterexamples to the present theory. Certain other well known word order correlations, observed by Greenberg ( 1 966) and corroborated in Dryer ( 1 992) also follow from the theory outlined here, at least in part. Consider for example the correlation between postpositions . and OV order: Typically a VP containing a PP complement has the form [vp [DP P] V] in an OV language. In the case where the PP is selected by the verb, this is a consequence of the adjacency condition on selection expressed in (42) above. The fact that adjuncts have the same order in the PP does not follow directly, though. Furthermore, there are well known cases of OV languages which have prepositions (for instance German). I leave it an open question here how to account for such cases. 19 Sigursson ( 1 988) suggested, partly on the basis of a conceptual argument, partly on the basis of the history of Germanic, that diachronic change of sentential word order from head-final to head-initial (OV to YO) and vice versa, always proceeds top-down. 2o We can now conclude that Sigursson is right with regard to change from OV to VO but wrong with regard to change from VO to OV, as evidenced, for example, by the absence of SVO languages with sentence final complementizers, and more generally by facts discussed here in support of generalization ( 1 ) . If this generalization holds universally, change from VO to OV must be bottom-up.
8.
Remaining problems: Scrambling and mixed order
So far the theory accounts for the cases of "pure OV" and "pure VO" order, where all complements of a head H are on the same side of H. More has to be said about constructions where some complements follow and others precede H. More needs to be said as well about scrambling (also called object shift), where the complement of H turns up in the specifier position of a higher head H'. Consider first scrambling; (59) is a case in point.
(59)
JUSSI romaanin on kirjoittanut. Jussi a.novel has written 'It's Jussi who has written a novel'
The object of the verb has moved to the specifier position of the auxiliary. The
DERIVING OV ORDER IN FINNIS H
147
construction is possible only when the sentence has initial focus. A priori the object could have moved to its surface position directly from the complement-of V position. However, this would undermine the connection we have argued for between OV order and initial focus. Instead we must assume roughly the analysis (60) of (59) : (60)
[cp JUSSl C [IP ts romaanin on [PreP [yp to kirjoitta-] [Pre' -nut �J]J]
In the absence of New, Prc has selected VP by XP-movement. The sentence focus is realized in C, attracting the subject in this case. Following VP-move ment to specPrcP, the object has moved to a higher Spec or adjunct position. What we have to say, in fact, is that such scrambling or object shift is possible only from a phrase which itself has moved to a specifier position, since other wise we have no explanation for the absence of scrambling when the sentence lacks initial focus (and thus must have New generated outside VP). This is at odds with standard GB-theoretic assumptions about subject-object asymmetries, according to which complements (especially of lexical categories) are transparent while specifiers are islands: cf. Huang's ( 1 982) CED, and Chomsky's ( 1 986) Barriers theory. See also Haider (this volume). What we have to say is that the island conditions were wrongly formulated in those works. While movement out of certain specifiers is impossible or highly restricted, for example out of sentential subjects, movement is by no means excluded out of specifiers across the board. In fact, any theory which assumes that head-final structures are derived by movement of complements must assume that various types of movement are possible out of these moved complements, including wh-move ment and verb movement (for instance in V2 languages). 2 1 Note that if scram bling can only take place from specifiers, we have the beginning of an explana tion of the observation that scrambling is typically found in OV languages rather than VO languages. Note also that the condition may fall under some version of 'Holmberg's Generalization' (cf. Holmberg 1 999). Consider (6 1 a), where the verb has moved and adjoined to the participial head: Here the verb will block scrambling/object shift. (6 1 )
a. b.
Jussi on [PreP [Pre kirjoitta -nut] [yp romaanin 1y]] Jussi on [PreP [yp romaanin kirjoitta] [-nut 1"pJ]
In (6 1b), where the VP has moved to check the selection feature of the participi al head, the object precedes the verb, and is free to scramble/object shift. As for 'mixed' constructions, where some complements of H precede, while others follow H, there are at least three cases to consider. One case can be exemplified by (62).
ANDERS HOLMBERG
148 (62)
Milloin Jussi sinulle kertoi etta han on lahdiissa
Pariisiin ?
when Jussi to.you told that he is on.his.way to.Paris 'When did Jussi tell you that he is on his way to Paris?' Here the conditions are met for OV order, as shown by the fact that the goal object is preverbal, yet the CP complement is postverbal, and must be postverbal, since finite clauses are excluded in preverbal position. A possible analysis is that the finite clause has first been extracted from the complement XP to the Spec of an abstract functional head prior to selection of XP (now containing only the goal object) by the verb. Assuming only leftwards movement, (62) would have an intermediate structure like (63), where F is the abstract functional category attracting the finite embedded clause and PP is the goal object. (63)
[yp V [FP CP [F' F [xp PP X tep] ] ] ]
This presupposes that the mystical category FP is transparent for selection, since V selects XP, not FP. This may not be an unproblematic assumption. Another case can be exemplified by (64) ( 1 3 a) : =
(64)
Minka takia sina pullat panitjaakaappiin (mikset leipalaatikkoon) ? what for you buns put in. fridge why.not in.bread.box 'Why did you put the buns in the fridge (why not in the bread box)?'
This word order is natural with contrastive focus on the postverbal constituent, prompted in the example by the continuation in parentheses. Again a possible analysis, along the lines of Kayne' s ( 1 998) analysis of postverbal focus, is that the focused constituent jaakaappiin has first moved out of the complement of the verb, by assumption a form of small clause (SC), to the specifier of a Focus head. Subsequently the SC, selected by the verb, has moved to specVP. (64)
[yp [s c pullat X �p] [panit [PoeP jaakaappiin [Poe tse] ] ]
Acknowledgments Thanks to the audiences at the Tromsfli workshop on VO and OV, ConSOLE 7, and the Linglunch seminar at MIT, where parts of this paper have been presented. Special thanks to Matthew Dryer, Liina Pylkkanen, and Peter Svenonius.
Notes 1.
See also Vainikka (1989: 91ff.).
DERIVING OV ORDER IN FINNIS H
1 49
2.
A complication which I will ignore here is that the question particle attaches either to the first constituent of the sentence or to the first word of that constituent, as shown in (i): (i) a. Sinun autonko ne varasti? your car.Q they stole b. Sinunko auton ne varasti ? your.Q car they stole "Was it your car that they stole?" This means that simple phonological cliticization of a particle in C to a constituent in SpecCP is insufficient as an account of the morphology of question and focus particles in Finnish. See Halpern 1 995.
3.
The 3rd person pronoun se can refer to human as well as nonhuman entities.
4.
Alternatively, in the framework of Kayne ( 1998), it is overt movement of the focused constituent to a lower focus position, followed by remnant VP-movement to a higher position.
5.
See for example Kroch and Santorini (1991). The following examples exemplify this contrast in verb sequencing: (German) dass er das Buch lesen k6nnen muss. (i) that he the book read can.INF must (Dutch) (ii) dat ie het boek moet kunnen lesen. that he the book must can.INF read 'that he must be able to read the book'
6.
See Holmberg et al. (1993) on the properties of the Finnish negation; it is an auxiliary-like element in the sense that it is inflected for subject agreement
7.
Anne Vainikka (1989: 94f.) has observed the contrast shown in (i). (i) a. Kuka tahtoo kahvia ? who wants coffee Kuka kahvia tahtoo ? b. who coffee wants While (ia) is a conventional way of offering coffee to a group of visitors, (ia) , although grammatical, is totally inappropriate in that context, being most readily read as a rhetorical question with 'nobody' as implied answer. Why that should be the case is an interesting question which I have no answer for.
8.
Counting genera instead of languages is a way of avoiding genetic bias: The sample contains no closely related languages.
9.
See Horton (1949: 190, 193), Frajzyngier (1989: 256).
10.
See Bendor-Samuel (1972: 95), Wordick (1982: 1 63 , 204).
11.
In Wordick (1 982) on Yindjibamdi there is only one example (p. 1 63), where the verb is intransitive. Apparently the word meaning 'want' is a noun rather than a verb (see footnote on p. 204), so its inclusion in the table (26) is questionable (Dryer, personal communication). In Guajajara there are several morphemes expressing volition. One belongs to a class of 'post verbals' , including a causative, a future, and a negation particle, among other elements (pp. 92-95). The example with Want (p. 95) has an intransitive verb. There is only one example with a post-verbal which also contains an object, the causative construction (i). As seen, the object(s) follow(s) the causative particle. a-zuka kar kure awa pe (i) I-kill CADS pig man to 'I made the man kill a pig.'
150
ANDERS HOLMBERG This suggests that the main verb is moved across the causative particle either by head movement or by remnant VP-movement. Speculating, this may be how the post-verbal Want is derived as welL There is also a class of suffixes called 'verbalizers' , containing two items expressing volition. Again the examples (p. 99) do not contain any transitive verb.
12.
The theoretical advantage that the incorporation theory holds over the lexicalist theory is that it makes a stronger claim about the relation between syntactic structure, hence scope relations, and linear morphological order. In the overwhelming majority of cases the stronger claim holds true, according to Julien (2000).
13.
In the spirit of Kayne (1 998) one may also consider the possibility that selection in (32) is the result of two overt movements rather than one covert: one selection-driven movement pied piping IP to SpecCP, followed by another (specific to languages with clause-initial comple mentizers) moving the complementizer to a higher head position. I will not discuss this possibility further in this paper.
14.
Kayne ( 1994) suggests that a derivation like the one in (36), including word formation of nonconstituents, may be characteristic of languages with agglutinating morphology. Finnish is consistent with this suggestion; cf. Holmberg et al. 1993 on Finnish verbal morphology.
15.
Cf. van Riemsdijk (1995), who concludes, on the basis of evidence quite different from that discussed here, that "there is a tendency to organize the Mittelfeld [in Dutch: AH] in such a way that it consists of a sequence of phrases followed by a sequence of heads, regardless of whether these heads are simultaneously XPs as well".
1 6.
Vilkuna ( 1989), too, does not postulate any causal relation between the initial focus and the OV structure. The present theory can be viewed as a particular formalization of Vilkuna's theory.
17.
The languages Vata and Gbadi, described in great detail in Koopman (1984), are potential counterexamples: They are OV but I VP, and they have a sentence-final complementizer. However, that complementizer introduces specifically infinitival clauses, where I is not realized in pre-VP position, as far as I can see. Finite clauses have a sentence-initial complementizer. See Koopman (1984: 45ff.). If so, these languages provide interesting support for the general ization that I VP precludes IP C.
18.
I have not checked whether the languages listed as Want V & OV in Dryer (1992) do in fact allow a transitive VP with the order OV as complement of Want.
19.
There are other triggers for movement, though, than selection. For example, the movement of DP to preverbal position is not triggered (only) by selection, but also by something like Case. This may explain why there is at best a weak correlation between VOIOV and D NPINP D (see Dryer 1992).
20.
The conceptual argument is as follows: Assume a language, VO or Ov, which has V-movement to I, as many languages are known to have, especially in finite clauses. In such a language a change in the relative order of V and 0 in VP will not show in simplex (finite) sentences, while a change in the relative order of I and VP will show. Hence word order change within VP is unlikely to precede change in IP.
21.
This is most clearly true of derivational theories. In a derivational theory observing something like Chomsky's (1 995) Extension Condition we cannot, for example, first apply wh-movement out of a category C, a complement of a head H, and subsequently move C to spec-of-H; the shorter movement must apply first.
DERIVING OV ORDER IN FINNIS H
151
References Baker, Mark. 1 988. Incorporation: A theory of grammaticalfunction changing. Chicago: University of Chicago Press. Bendor-Samuel, David. 1972. Hierarchical Structures in Guajajara. Norman, OK: Summer Institute of Linguistics. Bobaljik, Jonathan. 1995. Morphosyntax: The syntax of verbal inflection. Ph.D. dissertation, MIT. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press. Chomsky, Noam. 1995. Minimalist Theory. Cambridge, MA: MIT Press. Cinque, Guglielmo. 1998. Adverbs and Functional Heads. A cross-linguistic perspective. Oxford: Oxford University Press. Dikken, Marcel den. 1994. "Minimalist verb (projection) raising." Groninger Arbeiten zur
germanistischen Linguistik 37: 7 1-88. Dryer, Matthew. 1992. "The Greenbergian word order correlations." Language 68:
8 1-138. Frajzyngier, Zygmunt. 1989. A Grammar of Pero. Berlin: Dietrich Riemer Verlag. Greenberg, Joseph H. 1966. "Some universals of grammar with particular reference to the order of meaningful elements." In Joseph Greenberg (ed.), Universals ofLanguage. Cambridge, MA: MIT Press. 73-1 13. Haider, Hubert. 1992. "Branching and discharge." Arbeitspapiere des Sonderforschungs bereichs 340-23, UniversWit Stuttgart. Haider, Hubert. 1997. "Precedence among predicates." Journal of Comparative Germanic
Linguistics 1 : 3-41. Halle, Morris and Marantz, Alec. 1993. "Distributed morphology." In Kenneth Hale & S . Jay Keyser (eds), Th e Viewfrom Building 20. Cambridge, MA: MIT Press. 1 1 1-176. Halpern, Aaron. 1995. On the Placement and Morphology of Clitics. Stanford: CSLI Publications. Hr6arsd6ttir, Thorbjorg. 1999. Verb Phrase Syntax in the History of Icelandic. Ph.D. dissertation, University of Troms¢. Holmberg, Anders. 1997. "Word order variation in some European SVO languages: A parametric approach." In Anna Siewierska (ed.), Constituent Order in the Languages of Europe. Berlin: Mouton de Gruyter. 553-598. Holmberg, Anders. 1999. "Remarks on Holmberg's Generalization." Studia Linguistica
63 : 1-39. Holmberg, Anders, Urpo Nikanne, Irmeli Oraviita, Hannu Reime & Trond Trosterud. 1 993. "The structure of !NFL and the finite clause in Finnish." In Anders Holmberg and Urpo Nikanne (eds), Case and Other Functional Categories in Finnish. Berlin: Mouton de Gruyter. 177-206. Holmberg, Anders and Platzack, Christer 1995. The Role of Inflection in Scandinavian Syntax. Oxford: OUP.
152
ANDERS HOLMBERG
Holmberg, Anders, and Nikanne, Urpo. To appear. "Subjects, topics, and the EPP in Finnish." In Peter Svenonius (ed.), Subjects, Expletives, and the EPP. Oxford: OUP. Horton, A. E. 1949. A Grammar ofLuvale. Johannesburg: Witwatersrand University Press. Huang, James. 1982. "Move WH in a language without WH movement." The Linguistic
Review 1 : 369-4 16. Julien, Marit. 2000. Syntactic Heads and Word Formation: A Study of Verbal Inflection. Ph.D. dissertation, University of Troms!ZS. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kayne, Richard. 1 998. "Overt vs. covert movement." Syntax 1 : 1 28-1 9 1 . Koopman, Hilda. 1 984. The Syntax of Verbs. Dordrecht: Foris. Koskinen, Paivi. 1998. Features and Categories: Non-finite constructions in Finnish. Ph.D. dissertation, University of Toronto. Kroch, Antony and Santorini, Beatrice. 199 1 . "The derived constituent structure of the West Germanic verb-raising construction." In Robert Freidin (ed.), Principles and Parameters in Comparative Grammar. Cambridge MA: MIT Press. 269-338. Pesetsky, David. 1 982. Paths and Categories. Ph.D. dissertation, MIT. Riemsdijk, Henk van. 1995. Push chains and drag chains: Complex predicate split in Dutch. Ms. CLS, Tilburg University. Sigursson, Halld6r A.. 1988. "From OV to VO." Working Papers in Scandinavian Syntax 34, Department of Scandinavian Languages, University of Lund. Svenonius, Peter. 1994. "C-selection as feature-checking." Studia Linguistica 58: 133-155. Testelec, Yakov. 1997. "Word order variation in some SOY languages of Europe." In Anna Siewierska (ed.), Constituent Order in the Languages of Europe. Berlin: Mouton de Gruyter. 649-680. Vainikka, Anne. 1 989. Deriving Syntactic Representations in Finnish. Ph.D. dissertation, University of Massachusetts at Amherst. Vilkuna, Maria. 1989. Free Word Order in Finnish. Helsinki: SKS. Vilkuna, Maria. 1995. "Discourse configurationality in Finnish." In Katalin E. Kiss (ed.), Discourse Configurational Languages. Oxford: OUP. 244-268. Vilkuna, Maria. 1998. "Word order in European Uralic." In Anna Siewierska (ed.) Constituent Order in the Languages ofEurope. Berlin: Mouton de Gruyter. 173-234. Wordick, F. J. F. 1982. The Yindjibamdi Language. Canberra: Australian National University [Pacific Linguistics, Series C, No. 71].
Parameter Change
Icelandic
porbjorg Hr6arsd6ttir University of Troms¢
Abstract In this paper, we will discuss various ways to account for the loss of the attested OV word order patterns in the history of Icelandic in line with a uniform VO base (SHe) hypothesis. First, by means of loss of overt leftward movements of the objects, leading to covert movements only in the modern language. The second approach to the loss of OV word order is in terms of interacting (leftward) movements of both the object(s) and the VP. According to this hypothesis, the cross-linguistic (superficial) word order differences observed depend on the number of both VP movements and object movements. It will be argued that although the uniform VO base hypothesis with overt versus covert leftward movement of objects yields positive results in the analysis of the Old(er) Icelandic VP, it cannot easily handle the diachronic aspect. Therefore, we will revise the uniform VO base hypothesis in the direction of a theory of remnant VP-preposing. We will focus on the fact that several different features of OV order all disappeared from Icelandic at the same time, arguing that this is not accounted for in earlier proposed analyses of the derivation of OVNO word order. We will propose that VO order is derived by remnant VP-preposing, an account which successfully links all features of OV order and correctly captures the fact that they disappeared simultaneously.
1.
Introduction
There has been much discussion of the possible base pOSItIOns of nominal objects, and base word order in general, in the Germanic languages in recent years. In particular, leftward movements of objects have received much attention
I>ORBJORG HROARSDOTTIR
1 54
in recent syntactic works. In the checking theory of Chomsky ( 1 993, 1 995), leftward movement is driven by the requirement that morphological features are 'checked off' in inflectional positions in the functional domain, external to the VP. Leftward object movements also play an important role in Kayne's ( 1994) anti-symmetry proposal. These developments raise a number of interesting questions; is it, for instance, plausible to assume that movement can always be reduced to morphological feature-checking? Moreover, how should languages with different surface word order be described and contrasted; especially, languages that have traditionally been taken to have an initial OV structure (such as German, Dutch, Yiddish and Old English), in comparison to languages that traditionally have been analyzed as VO languages (such as Modern English and the Modern Scandinavian languages). Modern Icelandic has pure VO order within the VP, as shown in example (1). The word order in ( 1 ) with [yp [auxiliary verb - main verb - object]] is the only possible order of these elements in Modern Icelandic (abstracting away from topicalization and stylistic fronting). The position of the sentence adverb marks the left edge of the VP. (1)
Peir munu aldrei hafa euo hdkarlinn . they will never have eaten shark.the 'They will never have eaten the shark. '
Unlike in Modern Icelandic, several other orders of the VP-internal arguments were possible at earlier stages in the history of Icelandic, including both pure and mixed OV word order patterns, in addition to VO word order (cf. Sigurosson 1 988; R6gnvaldsson 1 99411 995, 1 996; Indrioason 1 987; Hr6arsd6ttir 1 995, 1 996, 1 998, 1 999a, 1 999b). Icelandic has had rich subject-verb agreement morphology and case morphology throughout its history. It is generally assumed that both Old and Modern Icelandic have obligatory overt movement of the finite verb to Infl, in both main and subordinate clauses, since the difference with regard to the position of the finite verb in main and subordinate clauses in Icelandic is insignificant, contra most of the other Germanic verb-second languages, as, for instance, the Mainland Scandinavian languages, German and Dutch. This, of course, can have interesting consequences for the analysis of the argument movements necessary to derive all the various word order patterns in Old(er) Icelandic, especially in comparison to languages with SOY word order, or other mixed word order languages, lacking this overt movement of the finite verb.
PARAMETER CHANGE IN ICELANDIC 2.
1 55
The decline of OV word order
The frequency of the different word order patterns in the VP was studied in various texts dating from the fourteenth to the nineteenth century, in addition to personal letters dating from the nineteenth century. The frequency of OV order was counted according to the equation in (2), (2)
f(OV) =
OV
--
OV + VO
where OV order is taken to include sentences where not only the nominal object but also other complements of the main verb precede it (prepositional phrases, adverbials, adjectives, and other non-finite verbs). All of the complements included in the main study here are ungrammatical in a preverbal position in Modern Icelandic, except for negative and quantified complements. In (3) are shown some simple sentences with OV word order to exemplify the classification into the various types of the complements. Example (3a) shows a sentence containing a preverbal complement that is a nominal object, (3b) shows an example of a preverbal prepositional phrase, (3c) is an example of a preverbal adverb, (3d) shows a preverbal adjective, example (3e) shows a preverbal particle, and finally, examples (3f-g) show sentences containing a preverbal non-finite verb; that is, where the non-finite main verb occurs to the left of the non-finite auxiliary. (3)
a.
b.
c.
d.
e.
so Porsteinn skyldi l{jinu tapa so I>orsteinn should life.the lose 'so that I>orsteinn should die' ao pu mcettir hjd mer vera nokkra daga that you could with me stay few days 'that you could stay with me for a few days' pv[ var riddarinn hingao sendur therefore was knight.the here sent 'Therefore, the knight was sent here' at pio munit nu satt segja that you will now true say 'that you will now tell the truth' dour sol var niOur runnin before sun was down slid 'before the sun had gone down'
I>ORBJORG HROARSDOTTIR
156
f.
g.
at vitinn hefir brendr verit that light.house.the has burned been 'that the light-house has been burned' Pu muntfrett hafa, ao . . . you will heard have that 'You will have heard, that . . . '
All sentences that showed signs of OV order were counted as OV (either 'pure' or 'mixed' order). A few examples of this classification are in (4) through (6). (4) illustrates pure OV order, and (5) and (6) show examples of the possible mixed OV orders. (4)
Pure OV word order ao hann hafi hana drepio a. that he had her killed 'that he had killed her' b. ao jJu . . . hafir pao bref fengio that you have that letter received 'that you have received that letter' ao eg skal per pao allvel launa c. that I shall you it well reward 'that I shall reward you well for it' d. at ek skylldi eigi fleiri born upp ala that I should not more children up bring 'that I should not bring up any more children' ao eg mundi hann sigrao geta e. that I would him defeat could 'that I would be able to defeat him'
(5)
Mixed word order: one non-finite verb plus two or more objects a. hafer jJu pinu lidi jatat peim have you your assistance promised them 'if you have promised them your assistance' b. jJa uilldi hann nu giarna hialp weita leoninum then wanted he now readily help give lion-the 'Then, he readily wanted to help the lion' c. Hafdi jJa huorgi sari komit a annann had then neither wound got on other 'Neither had been able to wound the other'
PARAMETER CHANGE IN ICELANDIC (6)
1 57
Mixed word order: at least two non-finite verbs plus at least one object a. ao hann skyldi aldrei mega sol sja that he should never be. allowed sun to.see 'that he should never be allowed to see the sun' og ekki skal Jaoir minn geta per hjalpao b. and not shall father mine can you helped 'And my father will not be able to help you' og hvor mundi pat hafa gjort c. and who would it have done 'And who would have done it' d. peir quaouz eigi pat mundu gera they said not it would do 'They claimed they would not do it' e. at hann mun raoa vilja feroum sinum that he will decide want journeys his 'that he will want to decide his own journeys' hann kuaz fundit hafa barn nyfrett f. he said found have baby newborn 'He claimed to have found a newborn baby'
In all the texts and letters, 4875 sentences were studied, of which 3497 were counted as VO and 1 378 sentences as Ov. The main results for the frequency of OV order are shown in Tables 1 and 2. There is no distinction made here between pure and mixed OV orders. Table 1 shows the rate of the frequency of OV order in each of the texts and letters from each age group studied, and Table 2 contains similar information for each century. There is no distinction made between main and embedded clauses here. OV orders occurred most frequently in texts dating from the fourteenth through the seventeenth century (from the average of 58.2% to 50.0%) and decreased to the average of 37.0% in texts from the eighteenth century. OV word order then gradually disappeared in texts and letters dating from the nineteenth century. These results are in accordance with previous studies on OV order in Old(er) Icelandic (cf. Rognvaldsson 1 99411995, 1 996; Indrioason 1 987). The OV orders therefore showed a remarkable stability for at least five or six centuries. The first important decline seems to occur in the language of writers in the eighteenth century, until OV word order has almost disappeared from the nine teenth century texts. This rapid disappearance of OV orders is especially interest ing since the frequency seems to have been quite stable in the preceding centuries.
I>ORBJORG HROARSDOTTIR
158
Table 1 . Number of sentences with OV and VO orders (in each text) Texts
All sentences OV
early 14th mid. 14th
late 14th late 14th early 15th late 15th early 1 6th late 1 6th late 17th late 1 7th ca 1700 early 1 8th early 1 8th late 1 8th early 19th early 19th 1. 2. 3. 4. 5. 6. 7.
age group age group age group age group age group age group age group
Finn Guam Am
Din
Sig Vikt Afs Mora Skill Ann Munn J.Ey Bisk Pr6f Alf Esp 1730-1750 1750-1770 1770-1790 1790-1 810 1 8 10-1830 1 830-1 850 1 850-1870
VO
Total
% OV
78 43 41 68 81 59 29 100 43 1 82 73 15 25 48 138 18
46 54 32 33 47 65 24 79 103 1 08 87 15 50 85 123 1 16
1 24 97 73 101 128 124 53 179 146 290 1 60 30 75 133 26 1 1 34
62.9% 44.3% 56.2% 67.3% 63.3% 47.6% 54.7% 55.9% 29.5% 62.8% 45.6% 50.0% 33.3% 36. 1 % 52.9% 13.4%
88 47 65 45 36 27 29
250 203 345 342 425 418 447
338 250 410 387 461 445 476
26.0% 1 8 .8% 15.9% 1 1.6% 7.8% 6. 1 % 6. 1 %
1 378
3497
4875
Table 2. Number of sentences with OV and VO orders (in each century) Texts
All sentences OV
14th 1 5th 1 6th 1 7th 1 8th 19th
century century century century century century
VO
Total
% OV
230 140 129 298 88 493
165 1 12 103 298 150 2669
395 252 232 596 238 3 1 62
58.2% 55.6% 55.6% 50.0% 37.0% 15.6%
1 378
3497
4875
PARAMETER CHANGE IN ICELANDIC
3.
1 59
Theoretical implications
In the following, some theoretical implications of the Old(er) Icelandic facts will be introduced. We will begin to discuss the possibility of a uniform VO base for both Old(er) and Modem Icelandic, in accordance with Kayne's ( 1 994) theory of phrase structure, where derivations from the initial stage to later stages are claimed to invariably involve movements of syntactic heads and phrases to the left. On that hypothesis, the OV orders found in the Old(er) Icelandic corpus would be derived by optional object movements, regulated by feature strength. This leads to both ' short' (within the VP), and 'long' (out of the VP) object movements, as well as adjunction of the main verb to the auxiliary. We will show that even though this approach yields positive results in the analysis of the Old(er) Icelandic VP, it does not easily handle the diachronic aspect. According to this proposal, we are forced to account for the loss of the attested OV word order patterns in the history of Icelandic by means of loss of overt leftward movements of objects, leading to covert movements only in Modem Icelandic. Hence, the parameter change in question would have to do with the loss of the relevant strong N-features; the N-features of both AgrOP and PredP. We will show that this has certain unwanted theoretical consequences. Therefore, we want to revise the uniform VO base hypothesis in the direction of a generalized remnant VP-preposing hypothesis, which has properties in common with Kayne's ( 1 998) analysis of negative/focus constructions. 3.1
Uniform VO base
Let us begin by assuming a uniform VO base hypothesis, with overt versus covert leftward object movements. According to Kayne ( 1 994), all languages have S-H-C order. Chomsky ( 1 995 : Ch. 4) adopts a similar position. Besides, Zwart ( 1 993, 1 997), Koster ( 1994) and Lattewitz ( 1 996) have extended these insights by showing that this approach yields positive results in the analysis of Dutch and German, and Roberts ( 1997) and van der Wurff ( 1997) also argue that Old and Middle English can plausibly be analyzed as head-initial. Kayne's system can also account nicely for the word order variation in the Old(er) Icelandic VP, and the argument movements necessary to derive the existing patterns. A Kayne-style analysis for the VP in Old(er) Icelandic is therefore plausi ble; assuming that the object could move either 'short' (within the VP) or 'long' (out of the VP), and that the main verb could move individually and adjoin to the auxiliary. In (7) is an example of the possible argument movements for some common word order patterns in Old(er) Icelandic.
I>ORBJORG HROARSDOTTIR
1 60 (7)
a. b. c. d.
Vfin . . . [vp [Vaux - Vmain - object]] (no movement) Vfin . . . [yp [Vmain - V aux - object] ] (V-raising) Vfin . . . [object - [yp [Vaux - Vmain]]] (object moves long scrambling) Vfin . . . [object - [yp [Vmain - Vaux]]] (V-raising; object moves long scrambling) Vfin . . . [yp [Vaux - object - Vmain] ] (object moves short)
=
=
e.
According to this hypothesis, Old(er) Icelandic had two optional movements which resemble obligatory movements postulated for GermaniDutch by Zwart ( 1 993). First, the main verb can optionally adjoin to the left of a (non-finite) auxiliary, as in (7b). Second, the object may optionally move to a specifier position to the left of the verbs, as in (7c-d) (with and without V-raising). In addition, Old(er) Icelandic had one more option not observed in Standard German, namely short object movement, as in (7e), where the landing site of the object precedes the main verb but not the auxiliary (that is, movement of the object internal to the VP). More exactly, OV word order would involve raising of the object to SpecAgrOP for Case-checking reasons, both inside and out of the VP. The object movement would be triggered by the strong N-feature of AgrOP, in agreement with the assumption that overt movement is triggered by the necessity to eliminate the strong features in the Spec-Head configuration in the functional domain (Chomsky 1 993). This is, for instance, supported by the fact that the word order pattern [Vmain - object - Vaux] is absent from the Old(er) Icelandic corpus while [Vaux - object - Vmain] exists. Not even a sentence adverb seems to have been able to intervene between the main verb and the auxiliary in the [Vmain - Vaux] order. Therefore, we conclude that Kayne's ( 1 994) system can account for the word order in the Old(er) Icelandic VP, and the argument movements necessary to derive the existing patterns, but at the price of optionality. The other option is variable parameter-settings, namely, that the feature triggering the complement movement can be of variable strength; if we follow Kayne's ( 1 994) anti-symme try proposal, according to which all languages have S-H-C order, and combine it with Chomsky's ( 1 993) hypothesis that objects always have their case checked in SpecAgrOP, we have to assume that Old(er) Icelandic had a choice between overt and covert object movement to SpecAgrOP, whereas Modem Icelandic only has this movement in LF. We still have to explain the possible positions of the other complements in Old(er) Icelandic. One possibility is to follow Zwart's ( 1 993) and Koster' s ( 1 994) analysis o f Dutch a s an S-H-C language, where they assume that nominal
PARAMETER CHANGE IN ICELANDIC
161
objects move to SpecAgrOP but that PPs and (other) small clause predicates must move to a separate Predicate-Phrase position, SpecPredP, where they can check off their predicative features . That is, we have a movement into a special functional projection (the Predicate Phrase), located between AgrOP and the VP. Hence, there existed two object positions in Old(er) Icelandic, as well as in Modern Dutch, the SpecAgrOP position for nominal objects, and SpecPredP for directional PPs and small clause predicates. As previously mentioned, the optionality of these complement movements in Old(er) Icelandic seems to violate Chomsky's ( 1 993) principle of Procrasti nate, which says that LF movements are preferred to movements before Spell Out. If movement can be procrastinated, we expect it always to be so. It may be that the optionality in complement-movement concerns the strength of features of functional heads that trigger movement. One possible way to get around the economy problems is then to simply assume that while AgrOP and PredP have weak N-features in Modern Icelandic (postponing the movements of nominal objects and small clause predicates until LF), they have had optionally (weak or) strong features in Old(er) Icelandic, forming both VO and OV orders. 3.2 Problems So far, we hope to have shown that it is possible to assume a uniform VO base throughout the history of the Icelandic VP. In other words, it is possible to derive all the existing word order patterns, and exclude the non-attested ones, by assuming SVO word order base with overt versus covert object movements. However, there is one problem with this hypothesis, and that is the dia chronic aspect. With regard to the word order within (and out of) the VP, Old Icelandic differs from Modern Icelandic in three major aspects, as shown in (8). (8)
a. b. c.
Obje ct - Vaux - Vmain Vfin Vfin Vaux - Object - Vmain Vmain - Vaux • . .
• . .
All these three word order patterns are ungrammatical in Modern Icelandic (abstracting away from negative phrase constructions), namely, the short object movement (within the VP), the long object movement (scrambling out of the VP, to the left of non-finite auxiliary verbs and sentential adverbs), and the possibility of placing the non-finite main verb to the left of the non-finite auxiliary (V-raising). The striking fact is that all these three leftward movement processes seem to have disappeared from the language at the same time in the history of
1 62
I>ORBJORG HROARSnOTTIR
Icelandic. Furthermore, the possibility of placing nominal objects, on the one hand, and PPs and small clause predicates, on the other hand, to the left of the main verb also seems to have disappeared from the language at the same time. Given a uniform VO base for all stages in the history of Icelandic, together with leftward movements of the complements, the difference noted between Old and Modem Icelandic - that is, the language change in question - must be explained away by use of feature strength, as already mentioned: strong versus weak N-features in both AgrOP and PredP. This is, of course, a possible account. However, it leaves some important questions unanswered. First, if the decline of the OV word order was subject to the disappearance of the strong features in question, it would be necessary to assume that the N-features in AgrOP and PredP both became obligatorily weak at exactly the same time in the history of Icelandic. This is not easily accounted for within this framework. There is nothing in the theory that predicts or explains this fact. Second, the hypothesis does not connect the decline of the three operations shown in (8). It is possible, though, to connect the first two operations, the long and the short object movement, claiming that the short object movement was a necessary step in the derivation of the long object movement. Thus, when the short object movement disappeared, so did the long object movement. However, the third leftward movement, the movement of the non-finite main verb to the left of the non-finite auxiliary verb, would always be independent of the object movements (this is, for instance, supported by the difference between German and Dutch). Hence, there is nothing in the theory that leads us to expect this process to have disappeared from the language at the same time as the other two leftward movement processes. Third, there is the question of morphology. More exactly, the question of why the relevant features triggering the overt leftward movements in question have been weakened (or lost their possibility of being strong) in the history of Icelandic. According to our proposal, the old language contained optional leftward movements of the complements in overt syntax, while these are ungrammatical in overt syntax in the modem language. Hence, the parameter change in question has to do with the weakening of the relevant N-features, leading to the impossibility of movements due to UG-internal economy condition (the Procrastinate Principle). More exactly, the change consists of the disappear ance of the strong alternative in question, since the possibility of having strong or weak features must have existed at least from the time of the earliest pre served Icelandic manuscripts. This leads to the question of why the relevant strong N-features disappeared from the language, and why they disappeared at the same time.
PARAMETER CHANGE IN ICELANDIC
163
It is a well known fact that languages with rich case morphology have freer word order than languages without rich case morphology. Roberts (1997), for instance, has recently linked the cause of word order changes in the history of English to inflection, more precisely, the loss of morphological case marking. In this light, it seems plausible that some morphological changes have occurred in late Icelandic, causing the decline of OV word order. However, this does not have any empirical support since the agreement system and case in Modem Icelandic are as rich as they were in Old Icelandic. In fact, Icelandic has the richest overt inflectional system of any modem Germanic language.
3.3 Interacting movements Due to these shortcomings of the uniform va base hypothesis, as proposed earlier, with regard to the diachronic aspect, it might be desirable to revise the proposal in the direction of a theory of interacting movements. Because of the fact that there has not been any overt weakening of the status of the morphological system in the history of Icelandic, we will propose that the morphological features relevant for the triggering of leftward object movements in Old(er) Icelandic were indeed not weakened. Instead, the objects still have to move to the left of the main verb (say, to SpecAgrOP) in overt syntax in Modem Icelandic.
3.3 . 1 Non-finite verb movement There are actually at least two different ways to approach this diachronic view, assuming interacting movements. First, assuming that the difference between Old and Modem Icelandic has to do not with the position of arguments, but with the position of the verb. Hence, the non-finite verbs began to move to a higher position, to the left of the moved objects, resulting in va word order only. On this proposal, languages can differ minimally or maximally, depending on the number of both verb-movements and object-movements. Hence, any language that ends up moving the object further left than the verb results in having surface OV word order, and vice versa. This, of course, is similar to saying that languag es differ in the number of positions/strong features that attract both object and verb movements. The main advantage with this proposal is that it can account for the fact that many of the various OV word order patterns disappeared from the language at exactly the same time in the history of Icelandic. If the only difference between the word order in Old and Modem Icelandic has to do with the movement of the non-finite verbs higher up in the structure than all the various complements, then it would indeed be expected that all the various OV word order patterns disappeared at the same time in the history of Icelandic,
1 64
I>ORBJORG HROARSDOTTIR
independent of the type of the object. Hence, the disappearance of the OV orders would be due to a change in the setting of only one parameter: the relevant (non tense) features of the non-finite verbs became obligatory strong. Note, however, that this hypothesis has to face high empirical hurdles, since first, it can only account for the loss of OV word order patterns that were derived by short object movement (within the VP) and not for the more restricted long object movement (scrambling out of the VP). Second, the relative position of non-finite auxiliary verbs, especially with regard to the long object movement, and all the various mixed word order patterns in Old(er) Icelandic, complicate the picture tremendously; the mixed/split word order patterns cannot be account ed for on a proposal where the crucial difference between OV and VO languag es has to do not with the position of arguments, but only with the position of the verb (whether it has moved or not). Third, it does not explain the loss of the word order pattern [Vmain Vaux] nor the fact that this pattern declined at the ' same time as the other attested OV word order patterns. -
3.3.2 Remnant VP-preposing Therefore, we will revise our proposal in the direction of a hypothesis of remnant VP-preposing. More exactly, we will assume obligatory leftward movement of objects (nominal objects to SpecAgrOP and PPs and (other) small clause predicates to SpecPredP as before) at all stages in the history of Icelandic, together with optional extraction of the embedded VP from the matrix VP in the old language only, followed by obligatory remnant VP-preposing in VO lan guages, including all stages of (written) Icelandic. Our proposal has properties in common with Kayne's ( 1 998) analysis of negative/focus constructions. In this paper, we want to generalize Kayne's proposal. In his paper, Kayne did not propose that every single instance of VO in English involves VP-preposing. In particular, he did not propose there that ordinary non-quantified, non-negative, non-focused objects involve a derivation with VP-preposing. More precisely, Kayne ( 1 998) claims that there are no covert movements, so that some overt movements must be involved in the derivation of some VO structures in English, namely, overt movement of (quantified, negative or focused) arguments to the left of the VP, followed by remnant VP-fronting to the left of the already extracted arguments, deriving VO word order again. One of the aims of Kayne's paper is to account for negative phrase constructions; the difference between English and the Scandinavian languages. The Scandinavian negative phrase construction has some similarity with the English construction 'I have seen nobody' , which Kayne argues to involve a step parallel to the Icelan dic preposing, with the difference that in English there is a further operation
PARAMETER CHANGE IN ICELANDIC
1 65
(lacking in Icelandic) that 'undoes' the change in word order effected by the first preposing step. In other words, negative as well as quantified and focused arguments must exit the VP. However, the second step in the derivation, the VP-preposing, does only take place in English but not in Icelandic nor the Mainland Scandinavian languages. More generally, Kayne ( 1 998) makes a distinction between short and long movement, where the movement operations in question are always movement of a phonetically realized phrase (overt (pre Spell-Out) movement only) . This proposal requires that the object (the direct object) necessarily raises in overt syntax to SpecAgrOP. This is illustrated in (9) below. (9)
a. b. c.
[verb - object] [[object] [vp verb tobject]] [[vp verb tobject] object tvp ]
Generalizing this idea, we propose that the crucial difference between VO lan guages (like English) and OV languages (like Dutch) is that the OV languages lack the last step in the derivation, namely the VP-preposing. In VO languages, on the other hand, the overt movement of the objects is masked by the subse quent movement of the VP. Thus, the landing site of the objects in OV lan guages is not obscured by VP-preposing the way it is in VO languages (see also Hinterh61z1 1 997, this volume, and Johnson 1 996). In sum, the proposal makes it possible to dispense with covert movements and replace them with a combina tion of overt movements. The picture is a bit more complicated for mixed OVNO languages, such as Old(er) Icelandic, especially since it follows from the generalized remnant VP-preposing analysis that Icelandic must have been a VO language throughout its (written) history (since the finite verb always precedes its complements) . However, we will show that it is indeed possible to derive all the attested OV word order patterns in Old(er) Icelandic with the tools of this framework, by means of three main derivations: (a) obligatory movement of the direct object out of the VP (to SpecAgrOP in the functional domain) at all stages in the history of Icelandic, together with (b) optional extraction of the embedded VP from the matrix VP in the old language, followed by (c) obligatory VP-preposing (pre posing of the remnant VP, containing the finite auxiliary verb) in VO languages, including all stages of (written) Icelandic. The main advantage with this proposal is that, in addition to deriving the attested word order patterns, we can account for the decline of the various leftward movement processes. More exactly, this proposal can give answers to the diachronic aspect, addressed earlier; it can account for the fact that all the
I>ORBJORG HROARSDOTTIR
1 66
various OV word order patterns disappeared from the language at the same time in the history of Icelandic. As mentioned, Old(er) Icelandic mainly differs from Modem Icelandic in that the older stage has the (superficial) patterns in ( 1 0), while the modem language does not. Namely, given the previously proposed uniform VO base hypothesis, short object movement, long object movement, and V-raising (movement of the non-finite main verb to the left of the non-finite auxiliary). ( 1 0)
a. b. c.
Vfin . Object - Vaux - Vmain Vfin . Vaux - Object - Vmain Vm ain - Vaux . .
. .
The claim we want to make is that the two first movements still take place in Modem Icelandic, but that the third possibility was lost. On this proposal, the third possibility shown in ( 10) alone (or, that is, the loss of it) can account for the loss of the superficial existence of the two other patterns. In other words, after the loss of the possibility of extracting VPs out of VPs, the obligatory preposing of the remnant VP will always mask the two first operations, leading to VO word order only in Modem Icelandic. The three word order patterns in ( 1 0) are typical examples of restructuring in the West Germanic languages : ( l Oa) is the typical pattern of restructuring (long DP-movement) in Dutch, ( l Ob) is a case of verb-projection raising typical for West Flemish, Swiss German, and Tyrolean, and ( 1 0c) is the standard pattern of verb-raising in Standard German. Furthermore, these word order patterns all occurred in Older German until about the eighteenth century, where most of them suddenly disappeared (HinterhOlzl, p.c.). Hence, German seems to also have had VP-preposing, but lost it, while Icelandic preserved VP-preposing and lost verb-raising (VP-out-of-VP), and Yiddish still has both. This leads to the interesting comparative and historical question of what factors preserved a process in one language and led to its loss in the other.
4.
VP-preposing
4. 1
Outline
The diachronic problem, as mentioned, is that we do not want three unrelated changes, hence, it is necessary to unify (a), (b) and (c) in example ( 1 0). On the earlier proposed DPIPP-movement approach (the uniform VO base hypothesis), no unification (by implicational relationship) is possible regarding ( l Oa-c):
PARAMETER CHANGE IN ICELANDIC
1 67
namely, there is no reason to expect the loss of the verb cluster [Vmain - VauJ to have caused the loss of leftward movement of objects, nor vice versa. Finally, we could have the loss of ( l Ob) causing the loss of ( l Oa) if OV versus va is the result of more object-movement in OV languages, but if OV versus va is the result of more verb-movement in va, we would still have to explain why the word order pattern [Vaux - Object - tvaux - V main] (the va counterpart of ( l Ob)) is ungrammatical in Modern Icelandic. Therefore, we will try unification by reversing the implicational relationship, namely, we will show that the loss of the three word order patterns in ( 1 0) holds on two plausible assumptions. The first assumption is that (Germanic) va derives from OV by raising the verb across the object overtly extracted from the VP, and that Old Icelandic was a va language. In fact, even Old Icelandic is in a sense less of an OV language than German or Dutch, since the finite verb always preceded its complements (on our analysis as a result of raising a constituent (VP) containing (at least) the finite verb). In other words, we cannot say that Old Icelandic is just like Dutch, only that the VP is preposed in Icelandic. Rather, Old Icelandic was a va lan guage; it has the preposing of the finite VP (typical for va languages), but it also has the VP-out-of-VP phenomenon, characteristic of OV languages like German, namely [Vmain - Vaux] (unlike Dutch which does not have this VP-out of-VP option). So far, we have remained vague as to exactly how the finite verb (Vfin) comes to end up preceding its complements. As suggested in recent work by Kayne ( 1 998), Hinterh61z1 ( 1 997), Koopman & Szabolcsi ( 1 997), we will now assume that the verb can only be raised as a part of a (remnant) VP raising to SpecFP above the positions of the complements extracted from it. On this view, the intermediate structure [Object [Vfin [Vmain tObj ect]]] would still only yield the va order [Vfin - Vmain - Object] (as a result of raising the matrix VP). In order to get [Vfin - Object - Vmain] ' we must have a further step leading to [Object [Vmain tObj eca [Vfin tvp]] , namely, the embedded VP must extract from the matrix VP. This extraction is arguably also what leads to the [Vmain - Vaux] word order in general. Hence OV order as the result of 'long' object-movement depends on ( 1 0c). Therefore, if the [Vmain - Vaux] option is lost, the Modern Icelandic strict va order is an automatic consequence, that is, only one parametric change is needed to account for the change from Old to Modern Icelandic.
4.2 Additional support So far, we have been discussing the diachronic aspect, which clearly supports our
PORBJORG HROARSDOTTIR
1 68
analysis. However, there exist some interesting synchronic aspects in support of this analysis as well. Before moving on to more detailed implementation, we will briefly mention some additional arguments in favor of the preceding analysis of Old Icelandic. First, the optionality problem is greatly reduced. OV word order now results from taking two options: (a) 'long' object movement (as opposed to movement to SpecAgrOP in the extended projection of the selecting verb), and (b) extract ing the embedded VP from the matrix VP. Second, the analysis correctly predicts that we should not find any OV word order in contexts where the object could not have raised across a higher verb, as for instance in the sentence shown in ( 1 1 ) below. What excludes ( 1 1 ) is that it has failed to have VP-preposing apply within the subject phrase. ( 1 1)
*Bcekur ao lesa er pao skemmtilegasta
sem hann veit.
books to read is the most.entertaining that he 'Books to read is his greatest pleasure.'
(Icelandic)
knows
This is ungrammatical since there is no higher verb in this sentence, providing the AgrOP-position of the object. Hence, this correctly predicts the non-existence of this construction in (Old) Icelandic; we never expect to find an object without a preceding verb in Icelandic. In German and Dutch, on the other hand, this construction is fine because these languages lack the VP-preposing Icelandic has always had. Third, this approach captures the fact that the clause structure seems to have been the same throughout the history of Icelandic, by the absence of mirror image effects of OVNO word order patterns. Within this approach, the VP-pre posing makes sure that the order of the constituents within that VP comes out in the same way in OV and VO word order. The only way to obscure this order is to extract out of the VP, prior to the VP-preposing. Moreover, it follows from this hypothesis that within each clause in VO languages, the main verb moves in front of the complement. Therefore, the object must have moved long in order to derive OV order in Old Icelandic, as illustrated in ( 1 2) below. ( 1 2)
DO [VPaux Vaux [VPmain Vmain tDO]]
Such long movement is exemplified by a Dutch example in ( 1 3). (13)
dat Marie het boek wil proberen te lezen tDO that Mary the book will try to read 'that Mary will try to read the book'
PARAMETER CHANGE IN ICELANDIC
1 69
Our claim is that this is what has happened whenever we find OV order in Old Ice landic, namely, long object movement. Consider the Old Icelandic example in (14).
(14)
a.
En stulkan neitaoi pao . . . ao piggja. but girl.the refused it to accept 'But the girl refused to accept it'
b.
[Vfin object tVPfin Vmain tobj ect]
c.
En stulkan [VPfin neitaoi tvpmain] [A crOP pao [VPmain ao piggja tDO] tVPfin b
Hence, (14) is a case of long object movement (where the object has moved into the matrix VP). If the preposing of the finite VP would not take place, the result would be a perfect German (embedded) sentence. The prediction is further that we should only find this long object movement with (matrix) verbs that are so-called verb-raising verbs. This seems to be born out in the Old Icelandic data. Now, consider the Old Icelandic clause in (15) below.
(15)
a.
hafer pu pinu lidi jatat peim have you your help offered them 'if you have offered them your help'
b.
[Vfin DPex Vmain DP f3]
Since no examples of this kind are found in German or Dutch, this should be an effect from Old Icelandic being a VO language. Hence, we have the derivation in ( 1 6).
( 1 6)
a. b. c. d.
Initial order:
[VPfin Vfin [VPmain Vmain 10 D O ]] VPmain moves to SpecVPfin : [VPmain Vmain 10 D O ] [VPfin Vfin [tVPmain]] DO moves long:
D O [VPmain Vmain 10 tDO] [VPfin Viin [tVPmain]] VPfin moves to F:
[VPfin Vfin [tVPmain]] D O [VPmain Vmain 10 tDO] tVPfin
In other words, an extraction of VPmain out of VPfin followed by a movement of ' the direct object to SpecAgrOP. Finally, the finite VP is preposed across the moved complements. Note that we are assuming here that the direct object always necessarily exits the VP (it must move to SpecAgrOP for case-checking
I>ORBJORG HROARSDOTTIR
170
reasons), while the indirect object never can; the only way for the indirect object to move is within the VP, when it is preposed. 4.3 Implementation We assume that in Old and Modem Icelandic (in fact, in all Germanic VO lan guages), a complement raised to a licensing position within the extended projection of some verb activates the entire functional layer of the verb, which in the VO languages contains F attracting VP to its Spec, across intervening complements. This happens whether the verb is finite or not, an auxiliary verb or a main verb. Thus, 'short' object movement always leads to VO order. In order to obtain successive cyclic application of VP-extraction resulting in intermediate structures of the form [[Vmain V aux] [Vfin . . . ]], we take VP-extraction to be PredP-extraction, that is, VP-extraction is to be implemented as movement to SpecPredP, where PredP is immediately above the VP. Assuming that only the VP, and not the PredP, raises to SpecFP (across the complements), the final step in the derivation always puts the finite verb in front of its complements. 1 Let us now compare these derivations to the earlier proposed argument movements, where ( 1 8) through (26) illustrate the possible derivations for some common word order patterns in Old(er) Icelandic, given the tools of our propos al. Example (17) shows the initial structure, and the derivations shown in ( 1 8) and (19) would be the analysis of a (Modem Icelandic) VO sentence.
(17)
Initial order: F [PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain object]]]]]]
There are actually two options to derive pure VO word order, depending on whether the complement is coherent/transparent or incoherent. 2 In ( 1 8), the object moves long, and if nothing else can be extracted, the finite VP preposes, containing not only the finite verb, but also the non-finite ones, thus, they end up to the left of the moved object. In ( 1 9), the object has moved short, and since this is a VO language, the verb (the main verb) must move in front of it, deriving VO word order again.
( 1 8)
[Vfin - Vaux - Vmain - object]
a.
Initial order:
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain object]]]]]]
PARAMETER CHANGE IN ICELANDIC b.
171
Object moves long (coherent complement) :
object [PredPfin [VPfin Vfi n [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain tobject3]]]]] c.
VPfin moves to F:
[VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain tobject]]]]] object [PredPfin tvpfin]
( 1 9)
[Vfin - Vaux - Vmain - object] a. Initial order:
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain object]]]]]] b.
Object moves short (incoherent complement) :
c.
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [object] [PredPmain [VPmain Vmain tobject]]]]]] VPmain moves to F: [PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [Ie [VPmain Vmain tobject]] [ object] [PredPmain tvpmain]]]]]
Example (20) shows the derivation of a simple OV word order, derived by the possibility of extracting VPs out of VPs (or moving PredPs into Spec of PredPs by cyclic application).
(20)
[Vfin - obj ect - Vmain] a. Initial order: b.
[PredPfin [VPfin Vfin [PredPmain [VPmain Vmain object]]]] PredPmain moves to SpecPredPfin : (the German option) [PredPfin [PredPmain [VPmain Vmain object]] [VPfin Vfin [tpredPmain]]]
c.
Object moves long:
object [PredPfin [PredPmain [VPmain Vmain tobject]] [tpredPmain]]] d.
[VPfin Vfin
VPfin moves to F:
[VPfin Vfin [tpredPmain]] object [PredPfin [PredPmain [VPmain Vmain tobject]] [tVPfin]] Example (21) illustrates the derivation of the [Vmain - Vaux] order, previously called V-raising. This order now simply reflects the VPs-out-of-VPs option. Alternatively, it is possible to derive this order in a more complex way, as shown in (22), depending on the relative position of the object in this respect.
172
I>ORBJORG HROARSDOTTIR (21)
[Vfi n - Vmain - VauxJ Initial order: a. b.
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain]]]]]]
PredPmain moves t o SpecPredPaux : (the German option)
[PredPfin [VPfin Vfin [PredPaux [PredPmain [VPrnain Vmain]]] [VPaux Vaux]] [ tPredPmain]]
c.
(22)
VPfin moves to F:
[VPfin Vfin [PredPaux [PredPmain [VPmain Vmain]]] [VPaux Vaux] [ tPredPmain]] [PredPfin tvpfin] [Vfin Vmain - Vauxl -
a.
Initial order:
b.
PredPmain moves to SpecPredPaux: (the German option)
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain]]]]]] [PredPfin [VPfin Vfin [PredPaux [PredPmain [VPmain Vmain]]] [VPaux Vaux]] [tPredPmain]]
c.
PredPaux moves to SpecPredPfin : (the German option)
[PredPfin [PredPaux [PredPmain [VPmain Vmain]] [VPaux Vaux] [tPredPmain]] [VPfin Vfin tPredPaux] d.
VPfin moves to F:
[VPfin Vfin tPredPaux] [PredPfin [PredPaux [PredPmain [VPmain Vmain]] [VPaux Vaux [tPredPmain]] [tvpfin]]
In (23) through (26), there are examples showing that by combining these three derivations, VPs-out-of-VPs, object movement and remnant VP-preposing, it is possible to derive all the various OV (pure and mixed) word order patterns. These are the only three main derivational possibilities (two of them are obliga tory), and the option is simply whether or not we choose to take VPs out of VPs and how often.
(23)
[Vfin - Vmain - Vaux - obje ct] a. Initial order:
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain object]]]]]] b.
PredPmain moves to SpecPredPaux : (the German option)
[PredPfin [VPfin Vfin [PredPaux [PredPmain [VPmain Vmain object]] [VPaux Vaux [tPredPmain]]]]]
c.
Object moves long:
object [PredPfin [VPfin Vfin [PredPaux [PredPmain [VPmain Vmain tobject]] [VPaux Vaux [tPredPmain]]]]]
PARAMETER CHANGE IN ICELANDIC
173
VPfin moves to F: [VPfin Vfin [PredPaux [PredPmain [VPmain Vmain tobject]] [VPaux Vaux [tpredPmain]]]] object [PredPfin tvpfin] [Vfin - object - Vaux - Vmain] a. Initial order: d.
(24)
(25)
(26)
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain object]]]]]] b. PredPaux moves to SpecPredPfin : [PredPfin [PredPaux [VPaux Vaux [PredPmain [VPrnain Vmain object]]]] [VPfin Vfin tpredPaux]] c. Object moves long : object [PredPfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain tobject]]]] [VPfin Vfin tPredPaux]] d. VPfin moves to F: [VPfin Vfin tpredPaux] object [PredPfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain tobject]]]] [tvpfin]] [Vfin - Vaux - object - Vmain] a. Initial order: [PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain object]]]]]] b. PredPmain moves to SpecPredPaux : [PredPfin [VPfin Vfin [PredPaux [PredPmain [VPmain Vmain object]] [yPaux Vaux tpredPmain]] c. Object moves short (incoherent complement): [PredPfin [VPfin Vfin] object [PredPaux [PredPmain [VPmain Vmain tobject]] [VPaux Vaux tPredPmain]]] d. VPaux moves to F: [PredPfin [VPfin Vfin]] [Ie [VPaux Vaux tPredPmain]] object [PredPaux [PredPmain [VPmain Vmain tobject]] [tVPaux]] V [ fin - object - Vmain - Vaux] a. Initial order:
b.
[PredPfin [VPfin Vfin [PredPaux [VPaux Vaux [PredPmain [VPmain Vmain object]]]]]] PredPmain moves to SpecPredPaux : [PredPfin [VPfin Vfin [PredPaux [PredPmain [VPmain Vmain object]] [VPaux Vaux tpredPmain]]]]
174
I>ORBJORG HROARSDOTTIR c.
PredPaux moves to SpecPredPfin :
[PredPfin [PredPaux [PredPmain [VPmain Vmain object]] [VPaux Vaux tPredPmain]] [VPfin Vfin tpredPaux]]
d.
Object moves long:
object [PredPfin [PredPaux [PredPmain [VPmain Vmain tobject]] [VPaux Vaux tPredPmain]] [VPfin Vfin tPredPaux]] e.
VPfin moves to F:
[VPfin Vfin tPredPaux] object [PredPfin [PredPaux [PredPmain [VPmain Vmain tobject]] [VPaux Vaux tPredPmain]] [tvpfin]]
Given the remnant VP-preposing hypothesis, all the OV word order patterns shown above would disappear as soon as the possibility of the extraction of VPs out of VPs (PredPs into Spec of PredPs) was lost, as already illustrated in ( 1 8) above. In other words, the remnant finite VP will always necessarily contain all the non-finite verbs in addition to the finite verb (in the appropriate order), and the preposing of this remnant VP will always mask the object movement, deriving va word order only. A crucial factor in support of this analysis is related to the behavior of verbal particles in the Germanic va and OV languages. There are two interest ing facts to note in this respect. First, the ordering with respect to the verb is [particle-verb] in the OV languages, while it is [verb-particle] in the va lan guages. Second, while the particle can precede the DP object in (most) va lan guages, it invariably follows all complements in the Germanic OV languages. We will propose that certain occurrences of verbal particles in the Germanic languages cannot be derived by incorporation in terms of head movement, but must involve some XP-movement instead (see also Taraldsen, this volume). Assuming a uniform and universal SHe order of constituents (in the spirit of Kayne 1 994), it not only becomes necessary for the direct object to follow the verb in base word order (regardless of whether it is a surface OV or va lan guage), it also becomes necessary for the verbal particle to occur in a postverbal position. Assuming the particle to constitute a small clause together with the DP object, the base word order for both the OV and the va languages must be along the lines shown in (27).
(27)
[vp verb [s c DP [particle]] ]
a.
taka hattinn af
(Icelandic)
take hatthe off b.
skakeZ die Zig turn
af
the light off
(Afrikaans)
PARAMETER CHANGE IN ICELANDIC
175
According to our remnant VP-preposing hypothesis, the direct object must obligatorily and universally raise out of the VP into SpecAgrOP. The second type of movement, then, distinguishes the two languages, and OV and VO lan guages in general, namely the remnant VP raises to SpecFP above the extracted direct object. As a result, the verb is situated to the left of its complement (even in embedded non-verb-second clauses). The particle can at this point be situated inside the VP, as a result of it raising together with the verb within the remnant VP, acquiring its position to the left of the object. As no such movement applies in OV languages, according to our hypothesis, the particle cannot raise across the object in OV languages. The third step in the derivation, then, raises the finite verb to the verb-second position. This movement applies (both in main and subordinate clauses) in Icelandic and (in main clauses) in the Germanic verb second languages, but never in English. We claim that the verb movement to the verb-second position is a head-movement, rather than VP-raising. If the verb movement to the verb-second position is a head-movement, it follows directly that only the finite verb, and not the particle, can raise higher than the negation. This correctly excludes the particle from preceding the DP object in Dutch and other OV languages, since they lack the remnant VP-preposing.
5.
Summary
In this paper, various ways to account for the loss of OV word order patterns in the history of Icelandic were discussed, all in line with the uniform VO base hypothesis. First, by means of loss of overt leftward movements of objects, leading to covert movements only in Modem Icelandic. On that hypothesis, the parameter change in question would have to do with the loss of the relevant strong N-features in the functional domain; the N-features of both AgrOP and PredP. Hence, the word order change, the decline of OV word order, was described as a reflex of a parameter change, eliminating the relevant strong N-features in AgrOP and PredP, within the VP, leading to the unavailability of overt movement of objects (DPs, PPs and small clause predicates) to the left of the main verb, which in response has lead to the unavailability of scrambling (movement of objects to SpecAgrOP and SpecPredP outside the VP). It was argued that although this approach might yield positive results in the analysis of the Old(er) Icelandic VP, it does not easily handle the diachronic aspect. Second, it was claimed that the word order variation in the Old(er) Icelandic VP might be accounted for in terms of interacting leftward movements of both the object(s) and the non-finite verb(s) . On this view, the cross-linguistic
176
pORBJORG HROARSDOTTIR
superficial word order differences depend on the number of positions/functions that attract the movement of both the non-finite verb and its complements. Thus, the objects still have to move to the left in overt syntax in Modem Icelandic. The hypothesis that the main verb has begun to move to a higher position, to the left of the moved objects, however, also has to face high empirical hurdles. Therefore, the uniform VO base hypothesis was revised in the direction of a theory of remnant VP-preposing, which has properties in common with Kayne's (1998) analysis of negative/focus constructions. In short, we assumed three main transformations: a) obligatory leftward movement of the direct object out of the VP (to SpecAgrOP in the functional domain) at all stages in the history of Ice landic, in addition to b) optional extraction of the embedded VP from the matrix VP in Old(er) Icelandic, followed by c) obligatory preposing of the remnant VP, containing (at least) the finite auxiliary verb, in Old and Modem Icelandic. The only major drawback we can see with this proposal is that it needs to explain why the third leftward movement process, the extraction of VPs out of VPs, was lost - a question we do not have an answer to. 3 However, any other analyses seem to have to face this same problem, and at least we have pinpointed the diachronic aspect (the language change in question) to only one parameter change - the loss of a singe movement, which is a lot better than having to explain three unrelated but simultaneous changes. Of course, this proposal also has to assume a certain optionality with respect to the derivation of OV word order patterns in Old(er) Icelandic, especially with regard to long versus short object movement, but that seems to be a side effect that every possible frame work has to face to some extent. At least the generalized Kaynean approach does not do any worse in this respect than the other possibilities that we have studied, or that have been put forward in recent literature. In sum, the uniform VO base hypothesis, together with the tools of a generalized Kaynean (1998) proposal, can account for all the various attested pure and mixed word order patterns in Old(er) Icelandic, beside providing the diachronic account; explaining the language change in question in terms of a single parameter change; the loss of [Vmain Vaux] or in more general terms, the ' loss of extracting VPs out of VPs (PredPs into Spec of PredPs), leading to the superficial loss of all the other attested OV word order patterns. In the end, the relative base order of the verb and the object plays no role in the analysis, in particular since the object always moves. This gives the following comparative Germanic correlation: 4 -
177
PARAMETER CHANGE IN ICELANDIC Table 3. Comparative Germanic Correlation
Icelandic, English, The mainland Scandinavian languages Old(er) Icelandic German Dutch
VP-preposing
VP-out-of-VP
yes
no
yes no no
yes yes no
Notes 1.
Furthermore, if only arguments of F can be 'bare' VPs, while arguments of the verb must be an extended VP (an S), then it is possible to assume that only bare VPs raise to SpecPredP, while a full extended V-projection cannot be licensed there. Hence, bare VPs and extended VPs do not move into the same position. This correctly excludes the ungrammatical pattern [Vfin [Vmain Object] Vaux] .
2.
We use the label 'IC' for 'Incoherent Complement' i n the following examples. HinterhOlzl (1997), among others, assumes that there are two types of full sentential infinitival comple ments: incoherent/opaque CPs and coherent/transparent CPs. While the coherent infinitives are transparent for several types of extraction processes/restructuring effects, the incoherent infinitives block long distance scrambling/restructuring.
3.
Maybe there is no explanation in structural linguistic terms, but only in sociolinguistic terms: a certain type of word order became unfashionable (Holmberg, p.c.).
4.
In addition, there is a head movement (of the finite verb) to IIC in some of these languages:
Modern Icelandic English Mainland Scandinavian Old(er) Icelandic German Dutch
Main clauses
Subordinate clauses
yes no yes yes yes yes
yes no no yes no no
References Chomsky, Noam. 1993 . "A minimalist program for linguistic theory". In Kenneth Hale and S amuel Jay Keyser (eds), The View from Building 20: Essays in linguistics in honor of Sylvain Bromberger. Cambridge, MA: MIT Press. (Also included in Chomsky 1995-0riginally distributed as MIT Occasional Papers in Linguistics 1, 1992) Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
178
pORBJORG HROARSD OTTIR
Hinterholzl, Roland. 1997. A XP-movement account of restructuring. Ms., University of Southern California. Hr6arsd6ttir, porbjorg. 1995. Setningafneoilegar breytingar a 19. old. I>r6un ]xiggja m:ilbreytinga. Master's thesis, University of Iceland, Reykjavik. (Published in 1998 as M:ilfrreoiranns6knir 10, M:ilvfsindastofnun Hask61a Islands, Reykjavik) Hr6arsd6ttir, porbjorg. 1996. "The decline of OV word order in the Icelandic VP; a diachronic study". Working Papers in Scandinavian Syntax 57: 92-14 1 . Hr6arsd6ttir, porbjorg. 1 998. "Interacting movements in the history of Icelandic". Paper presented at the Fifth Diachronic Generative Syntax Conference (DIGS 5), Universi ty of York, 30 May-l June, 1 998. (To appear in Pintzuk, Susan, George Tsoulas, and Anthony Warner (eds), DIGS 5 volume. Diachronic Syntax: Models and Mechanisms. Oxford: OUP) Hr6arsd6ttir, porbjorg. 1999a. VP-preposing in Icelandic. Paper presented at 14th Comparative Germanic Syntax Workshop, Lund, Sweden, January 8-9, 1999. Hr6arsd6ttir, porbjorg. 1999b. Verb Phrase Syntax in the History of Icelandic. Ph.D. dissertation, University of Troms¢. Indrioason, Porsteinn. 1987. Skyrsla urn oroaroo f sagnlio. Ms., University of Iceland, Reykjavik. Johnson, Kyle. 1996. In search of the English middle field. Ms., University of Massachusetts, Amherst. Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, MA : MIT Press. Kayne, Richard S. 1998. "Overt vs. covert movement". Syntax 1: 128-1 9 1 . Koopman, Hilda and Anna Szabo1csi. 1997. The Hungarian verbal complex: Incorporation as XP-movement. Ms., UCLA. Koster, Jan. 1994. "Predicate incorporation and the word order of Dutch". In Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi, and Raffaella Zanuttini (eds), Paths to Universal Grammar: Studies in honor ofRichard S. Kayne. Washington DC: Georgetown University Press. 255-276. Lattewitz, Karen. 1996. Movement of verbal complements. Ms. University of Groningen. Roberts, Ian G. 1997. "Directionality and word order change in the history of English". In van Kemenade, Ans and Nigel Vincent (eds), Parameters of Morphosyntactic Change. Cambridge: CUP. Rognvaldsson, Ein1rur. 1994/1995. "Breytileg oroaroo f sagnlio". fslenskt mal og almenn malfrceOi 16-17: 27-66. Rognvaldsson, Ein1rur. 1996. "Word order variation in the VP in Old Icelandic". Working Papers in Scandinavian Syntax 58: 55-86. Sigurosson, Halld6r Armann. 1988. "From OV to VO: Evidence from Old Icelandic". Working Papers in Scandinavian Syntax 34.
Wurff, Wim van der. 1997. "Deriving object-verb order in late Middle English". Journal of Linguistics 33: 485-509.
PARAMETER CHANGE IN ICELANDIC
179
Zwart, Cornelius Jan-Wouter. 1993 . Dutch Syntax. A minimalist approach. Ph.D. dissertation, University of Groningen. [Distributed as vol. 10 of Groningen Disserta tions in Linguistics] .
Zwart, Cornelius Jan-Wouter. 1997. Morphosyntax of Verb Movement. A minimalist approach to the Syntax of Dutch. Dordrecht: Kluwer.
1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1
The
.......... li..........
"" Periphery
SOV Languages
English and Dutch Sjef Barbiers Leiden University
Abstract This paper discusses the distribution of CP, DP, PP and Small Clause comple ments in the right periphery in English and Dutch from a modified anti symmetric perspective. English and Dutch are argued to be SOY underlyingly: all constituents except propositional CPs originate to the left of position V in S OY. English differs from Dutch in that it has overt short V movement. Arguments in English and Dutch basically have the same distribution. DP arguments and SC complements do not occur to the right of the underlying V position in SOY. This follows from the VP Intraposition analysis (Barbiers 1995) of extraposition according to which the trigger of VP Intraposition is to establish a predication relation. Since DP arguments and SC complements are saturated, VP cannot become their subject. The optional occurrence of PP complements in post-V position follows from the VP Intraposition analysis if PP complements are predicates of a V projection, not arguments. Most importantly, new evidence is provided to show that CP and DP complements have a different syntactic distribution because they bear different semantic relations to the verb. Whereas DP complements are subjects of internal predication, factive CPs are predicates of (extended) VP, while propositional CPs are the only real complements.
1.
Introduction
The starting point of this paper is the following hypothesis:
(1)
Syntactic structure i s generated from bottom t o top b y uniform leftward application of merge and move, starting out with a verb.
1 82
SJEF BARBIERS
This hypothesis entails that rightward movement and right-adjunction are not available, as in Kayne 1 994 and Haider 1997, and that languages are SOY underlyingly. I assume that the different surface position of V in English and Dutch is the result of short V-movement applying overtly in English (cf. Larson 1988; Johnson 1 99 1 ; Costa 1 996; Haider 1997; among others). The idea that a bottom-up derivation starts out with the verb is implicit or explicit in many existing syntactic theories, the background being that it is mostly V that deter mines which other constituents must or can occur in the sentence. From the perspective of the hypothesis in (l), this paper seeks to explain the following generalizations about the position X in the linear order S Vs 0 V X in English and Dutch (V indicates the position of the verb before short verb movement; Vs indicates the position of the verb after short verb movement). 1
(2)
(i)
DP arguments and Small Clause complements cannot surface in X. (ii) PP complements may surface in X. 2 (iii) Propositional CP complements must surface in X. (iv) In English, PP adjuncts and factive CP complements must surface in X, while in Dutch they may surface there.
Given ( l ), there are two possibilities for a constituent yP to end up in position X: (3)
yP is a verbal complement of the matrix V, i.e., a VP, IP or CPo Matrix V is merged to the left of YP. (ii) yP is generated to the left of V; an (extended) projection of V moves leftward, crossing YP (henceforth: VP Intraposition), yielding the linear order VP yp.3 (i)
_
Thus, X in the order S V s 0 V X always indicates a base position. However, V indicates a base position only in the case of verbal complements (3i) ; otherwise, it is a derived position, the landing site of VP Intraposition (3ii). In view of (3), two questions must now be answered to explain the generalizations in (2):
(4)
(i)
Why do certain constituents (e.g, PP complements and PP adjuncts) trigger/allow VP Intraposition, while others (e.g., DP arguments and SC complements) do not? (ii) Why is it impossible for CP complements to surface in 0 in the order SOY, either by base generation or by movement?
THE RIGHT PERIPHERY IN SOY LANGUAGES 2.
183
The data
Abstracting away from the position of V, English and Dutch show the same unmarked linear and hierarchical organization in double object constructions (5a-b), prepositional dative constructions (5c-d) , transitive Small Clause constructions (5e-f), DPgoaCPPtheme constructions (5g-h) and double PP construc tions (5i-j). This is shown for linear order in (5). It can be demonstrated with existing diagnostics for hierarchy (Barss and Lasnik 1 986), that in all of these constructions the first complement asymmetrically c-commands the second one. (5)
a. b. c. d. e. f. g. h. i. j.
dat Jan
Marie het boek
that John gave
Mary the book
dat Jan
het boek aan Marie
that John gave
the book to Mary
dat Jan
de doos op de tafel
that John put
the box on the table
dat Jan
Marie van boeken
gaf gaf zette voorzag
that John provided Mary with books
dat Jan
met Marie over Piet
sprak
that John talked
with Mary about Pete
___
Reordering possibilities are the same in both languages. Reordering in the prepositional dative construction (6c-d) and the small clause construction (6e-f) requires that the DP complement be interpreted as new in the discourse, giving rise to "heaviness" effects : the DP must have focus, be heavy and cannot be pronominal. (Changes in) extraction possibilities are the same in both languages. For example, whereas extraction from the about PP in (5i-j) is possible, it is impossible from the about PP in (6i-j). Also, the binding properties of the two PPs in the double PP construction ((5i-j) and (6i-j)) are the same in the two languages (cf. Barbiers, to appear). (6)
a. b. c. d. e. f. g. h. 1.
j.
*dat Jan
het boek Marie
*that John gave
the book Mary
gaf
dat Jan
aan Marie DAT boekl*het gaf
that John gave
to Mary THAT bookl*it
dat Jan
op de tafel de DOOSl*het zette
that John put
on the table the BOXl*it
*dat Jan
van boeken Marie
---
voorzag
*that John provided with books Mary
dat Jan
over Piet met Marie
that John talked
about Pete with Mary
sprak
1 84
SJEP BARBIERS
The null hypothesis is that the complements in (5) are in their base position both in English and in Dutch. There is no evidence that any of the constituents has moved before reaching the surface positions. The well-formed reorderings in (6) may be the result of scrambling or base-generation: both possibilities can explain why the first complement in the sentences in (6) behaves like an adjunct. Given the similarities between English and. Dutch in the middle field illustrated in (5) and (6), which can be supplemented with data from the distribution of adverbials, it seems to be safe to assume that the difference between English and Dutch in this domain is that English has overt short verb movement. This is not a very costly assumption: cross-linguistic parametrization of V positions is empirically well-motivated (cf. Pollock 1 989 and subsequent work) . If a movement operation that must be assumed for one language must be assumed for all other languages, we may say that short verb movement takes place in Dutch as well, but that the verb is spelled out in its base position, while it is spelled out in its landing site in English. The SOY hypothesis also makes it possible to show that English and Dutch have the same restrictions on the end field (position X in S Vs 0 V X). The distribution of CP complements will be discussed in Section 4. DP arguments and Small Clause-complements cannot occur in X:
(7) S a. b. c. d. e. f. g. h.
*
Ed
zal
Ed
will
*
*
*
0
V
de deur rood
verven
paint
the door red
de deur rood
Zal
*
*
Vs
Will
Ed
zal
Ed
will
Ed
zal
Ed
will
paint
the door red
paint
red
rood de deur paint
the door
verven
x Ed Ed
verven
de deur the door
verven
rood on Monday red
PP complements and PP adjuncts may occur in X; in English, PP adjuncts cannot occur in the middle field (between S and V) :
THE RIGHT PERIPHERY IN SOY LANGUAGES
1 85
(8) a. b. c. d.
Ed zal
het boek aan Marie
E d will read
the book to Mary
voorlezen
Ed zal
het boek
Ed will read
the book
tomorrow to Mary
S
Vs
V
e. Ed zal op zondag Ed will *on Sunday switch g. Ed zal h. Ed will switch
f.
3.
X
V
Vs
S
voorlezen
aan Marie X
de lichten uitdoen *on Sunday the lights off
de lichten uitdoen
op zondag
the lights off
on Sunday
Explanation of the distribution of DPs, PPs and Small Clause complements
The distribution of DPs, PPs and Small Clause Complements in the end field X (the data in (7) and (8» can be explained if we assume that DPs and PPs must be generated to the left of position V and can only surface in X as a result of VP Intraposition. On this view, the linear position V in (7) and (8) is a derived position. Consider first VP Intraposition with PP adjuncts. The linear effect of VP Intraposition (9a-b) corresponds to the structural analysis in ( 1 0a-b) .4,5 (9)
( 1 0)
a. b.
Ik zal
[vp de lichten uitdoen] [pp op zondag] I will switch [yp the lights o ff -switeh] [pp o n Sunday] [vp de lichten uittleen] . [vp the lights off switch]. __
VP
VP
�
W
on Sunday
------
W
-----PP switch off the lights � �
� � op zondag
�
VP
de lichten uitdoen
�
VPi
de lichten uitdoen
switch off the lights
op zondag
on Sunday
I assume that the trigger of this movement is semantic: only by moving into SpecPP can the VP become the subject of PP, such that PP is interpreted as
1 86
SJEF BARBIERS
predicating of VP. 6,7 If this is correct, then we can immediately explain why SC complements cannot "extrapose". The crucial property of a small clause comple ment is that it has its own subject, the internal DP argument. Under the assump tion that a PP can only have one subject (whichs need not be stipulated but follows from the theory proposed in Barbiers 1 995), VP Intraposition is blocked by the presence of the subject of the small clause ( 1 1 a-b). (11)
a.
Jan zal
__
fxp lOp de doos] IPp op de tafel]] &,rp zetten] . I t
>k
b.
John will pu�
fxp IDp the box] IPp on the table] [yp ...M ... ] . I t
>1
orsson, and especially porbjorg Hroarsdottir for indispens able discussion.
Notes 1.
In Norwegian, as in English, noun phrases may precede a verbal particle: throw it out; thus a noun phrase may undergo OS in the presence of a particle. In Swedish, however, noun phrases must follow particles, and cannot be moved to the left of them by OS. Thus the example illustrating this is taken from Swedish. Cf. Holmberg 1986 for the OS facts, Taraldsen 1 983, this volume, Svenonius 1994, Haider 1997 for the facts about particles.
2.
Henry 1995 makes the extremely interesting observation that Belfast English has OS (of pronouns) in exactly in those contexts in which there is verb movement (imperatives). The fact that verb movement is restricted to imperative constructions means that evidence for the learner of an OS rule cannot be terribly abundant; OS will only be observed in imperatives with pronominal objects and suitable adverbials. The fact that Belfast English observes HG then suggests that HG is the result of some innate constraints, or at least some which are very easy to learn.
3.
Some speakers get main clause phenomena in certain embedded clauses; those speakers will accept (5b), but the same speakers will accept the order V-Adv in conditional clauses with hvis 'if'.
4.
I follow Diesing in assuming that the leftward movement of the object in (52b), across the negation marker, is not a case of QM, given that QM does not cross adverbial elements cf. (29) above. If in fact (52b) involves QM, then whether OS induces scope fixing is still an open question.
References Beghelli, Filippo, and Stowell, Tim. 1 997. "Distributivity and negation: The syntax of each and every." In Anna Szabo1csi (ed.), Ways oJScope Taking. Dordrecht: Kluwer. 7 1-107. Burzio, Luigi. 1 986. Italian Syntax: A government-binding approach. Dordrecht: ReideL Chomsky, Noam. 1 99 1 . "Some notes on economy of derivation and representation." In Robert Freidin (ed.), Principles and Parameters in Comparative Grammar. Cam bridge, MA: MIT Press. 417-454. Chomsky, Noam. 1 995. The Minimalist Program. Cambridge, MA: MIT Press. Christensen, Kirsti Koch. 1 99 1 . Quantifier-movement and its syntactic reflexes. Ms. University of Bergen. Cinque, Guglielmo. 1 990. Types ojA ' -dependencies. Cambridge, MA: MIT Press. Diesing, Molly. 1 992. Indefinites. Cambridge, MA: MIT Press.
290
PETER SVENONIUS
Diesing, Molly. 1996. "Semantic variables and object shift." In Hoskuldur Thniinsson, Samuel David Epstein, and Steve Peter (eds), Studies in Comparative Germanic Syntax, vol. II. Dordrecht: Kluwer. 66-84. Diesing, Molly. 1 997. "Yiddish VP order and the typology of object movement in Germanic." Natural Language and Linguistic Theory 15: 369-427. Farkas, Donka. 1992. "On the semantics of subjunctive complements." In Paul Hirsch buhler and Konrad Koerner (eds), Romance Languages and Modern Linguistic Theory. Amsterdam and Philadelphia: John Benjamins [Current Issues in Linguistic Theory 91]. 69-104. Fodor, Jerry, and Sag, Ivan. 1982. "Referential and quantificational indefinites." Linguis tics and Philosophy 5: 355-398. Fox, Danny. 1 999. "Reconstruction, binding theory and the interpretation of chains." Linguistic Inquiry 30: 157-1 96. Fukui, Naoki. 1 993. "Parameters and optionality." Linguistic Inquiry 24: 399-420. Giannakidou, Anastasia. 1 998. The Landscape of Polarity Items. Ph.D. dissertation, Rijksuniversiteit Groningen. Groningen: Grodil. Haegeman, Liliane. 1 995. The Syntax ofNegation. Cambridge: CUP. Haegeman, Liliane, and Zanuttini, Raffaella. 1 99 1 . "Negative heads and the Neg criteri on." The Linguistic Review 8: 233-25 1 . Haider, Hubert. 1 997. "Precedence among predicates." The Journal of Comparative Germanic Linguistics 1 : 3-4 1 . Henry, Alison. 1 995. Belfast English and Standard English. Oxford: OUP. Holmberg, Anders. 1 986. Word order and syntactic features. Ph.D. dissertation, University of Stockholm. Holmberg, Anders. 1 997. "The true nature of Holmberg's Generalization." In Kiyomi Kusumoto (ed.), Proceedings of the North East Linguistic Society 27. Amherst: GLSA. 203-217. Holmberg, Anders. 1 999. "Remarks on Holmberg's generalization." Studia Linguistica 53: 1-39. Holmberg, Anders and Platzack, Christer 1 995. The Role of Inflection in Scandinavian Syntax. Oxford: OUP. Josefsson, GunlOg. 1992. "Object shift and weak pronominals in Swedish." Working Papers in Scandinavian Syntax 49: 59-94. J6nsson, J6hannes Gisli. 1 996. Clausal Architecture and Case in Icelandic. Ph.D. dissertation, University of Massachusetts at Amherst. Kuroda, S.-Y. 1 972. "The categorical and the thetic judgment." Foundations ofLanguage 9: 153-185. Ladusaw, William. 1 992. "Expressing negation." In Chris Barker and David Dowty (eds), SALT II: Proceedings of the conference on Semantics and Linguistic Theory 2. The Ohio State University. 237-259. Nilsen, 0ystein. 1 997. "Adverbs and A-shift." Working Papers in Scandinavian Syntax 59: 1-3 1 .
QUANTIFIER MOVEMENT IN ICELANDIC
29 1
Obenauer, Hans-Georg. 1998. Negative movement in French. Handout CNRS, Paris; paper presented at Going Romance, Utrecht University, December 1 998. Partee, Barbara. 1 995. "Quantificational structures and compositionality." In Emmon Bach, Eloise Jelinek, Angelika Kratzer, and Barbara H. Partee (eds.), Quantification in Natural Languages, vol. II. Dordrecht: Kluwer. 541-602. Pesetsky, David. 1989. Language-particular processes and the Earliness Principle. Ms., MIT. Reinhart, Tanya. 1 997. "Quantifier scope: How labor is divided between QR and choice functions." Linguistics & Philosophy 20: 335-397. Rizzi, Luigi. 1 990. Relativized Minimality. Cambridge, MA: MIT Press. Ross, John Robert. 1 967. Gapping and the order of constituents. Ms. MIT; presented at the 10th International Congress of Linguistics. Published 1970 in Manfred Bierwisch and Karl Erich Heidolph (eds.), Progress in Linguistics. The Hague: Mouton. 247-259. R6gnvaldsson, Eirikur. 1987. "On word order in Icelandic." In Proceedings of the Seventh Biennial Conference of Teachers of Scandinavian Studies in Great Britain and Northern Ireland. London: University College London. 33-49.
Steele, Susan. 1 978. "Word order variation: A typological study." In Joseph Greenberg (ed.), Universals of Human Language, vol. 4: Syntax. Stanford: Stanford University Press. 585-624. Stowell, Tim. 1 99 1 . "Small clause restructuring." In Robert Freidin (ed.), Principles and Parameters in Comparative Grammar. Cambridge, MA: MIT Press. 1 82-218. Svenonius, Peter. 1 994. Dependent Nexus: Subordinate Predication Structures in English and the Scandinavian Languages. Ph.D. dissertation, UCSc. Svenonius, Peter. 1 996a. "The optionality of particle shift." Working Papers in Scandina vian Syntax 57: 47-75 . Svenonius, Peter. 1 996b. "Predication and functional heads." In Jose Camacho, Lina Choueiri, and Maki Watanabe (eds), The Proceedings of the Fourteenth West Coast Conference on Formal Linguistics. Stanford: CSLI. 493-508. Svenonius, Peter. 1 998. "Clefts in Scandinavian: An investigation." In Artemis Alexiadou, Nanna Fuhrhop, Paul Law, and Ursula Kleinhenz (eds), ZAS Papers in Linguistics 10. Berlin: Zentrum fur Allgemeine Sprachwissenschaft, Sprachtypologie und Universalienforschung. 1 63-190. Szabo1csi, Anna. 1997. "Strategies for scope taking." In Anna Szabo1csi (ed.), Ways of Scope Taking. Dordrecht: Kluwer. 109-154. Taraldsen, Knut Tarald. 1 983. Parametric Variation in Phrase Structure: A case study. Ph.D. dissertation, University of Troms¢. Taraldsen, Knut Tarald. 1 99 1 . "A directionality parameter for subject-object linking." In Robert Freidin (ed.), Principles and Parameters in Comparative Grammar. Cam bridge, MA: MIT Press. 219-268.
292
PETER SVENONIUS
Thnlinsson, Hoskuldur. 1 993. "On the structure of infinitival complements." In Hoskuldur Thrrunsson, Samuel D. Epstein, and Susumu Kuno (eds), Harvard Working Papers in Linguistics, vol. 3. Cambridge, Ma.: Harvard University. 1 8 1-213. Vikner, Sten. 1 995. Verb Movement and Expletive Subjects in the Germanic languages. Oxford: OUP. Webelhuth, Gert. 1 992. Principles and Parameters of Syntactic Saturation. Oxford: OUP. Williams, Edwin. 1 983. "Against small clauses." Linguistic Inquiry 14: 287-308. Zwart, Jan Wouter. 1 993. Dutch Syntax: A minimalist approach. Ph.D. dissertation, University of Groningen. Groningen: Grodil.
Licensing Movement the West
Languages
Roland Hinterh6lzl
Abstract This paper investigates the distribution of syntactic elements in the Mittelfeld and Nachfeld from the point of view of the universal base hypothesis. Two approaches will be compared: Kayne's ( 1994) proposal in terms of licensing leftward movement that may strand syntactic elements in the VP and Haider's (1993, 1 995) account in terms of directionality of licensing. Providing evidence for licensing movement out of the VP, we will argue in favor of the licensing movement and stranding approach. First, we argue that a proper analysis of the syntax of the infinitival marker implies that not only the arguments of the verb but also VP-intemal predicates, including verb-particles, small clause predicates and directional PPs move out of the VP into specific licensing positions in the middle field. Then we show how a particular implementation of the mechanism of feature checking in terms of copy and delete (partial deletion) explains how particular elements can be stranded by licensing movement. Finally, we discuss some of the consequences of this approach for the analysis of verb-preposing (cf. Kayne 1 998) in English.
1.
Introduction
The structure of the German clause has been traditionally described in terms of the three fields in which the clause is partitioned by the complementizer and the verb in its base position, as illustrated in ( I a). Syntactic elements preceding the finite verb in CO occupy the Voifeld (pre-field) or are said to have been topicalized. Elements following the complementizer but preceding the verb in its base position occupy the Mittelfeld (middle field), while elements following the verb in its base
ROLAND HINTERHOLZL
294
posItIOn occupy the Nachfeld (post-field). Within the traditional OV-based approach, it was assumed that elements occurring in the middle field either occupy their base position in the VP or have been scrambled out of the VP to a desig nated position within IP or left-adjoined to IP, while elements in the Nachfeld were assumed to have been extraposed, that is, moved out from their base position preceding the verb to a position that is right-adjoined to either VP or IP. In this paper, we are concerned with the issue of how the distribution of syntactic elements in the Mittelfeld and Nachfeld is to be described in a YO-based approach. More specifically, we want to investigate what factors determine which elements appear in the Mittelfeld and which elements can or must stay in the Nachfeld. Two proposals have been made to account for the distribution of syntactic elements in the Mittelfeld and Nachfeld within the general approach that assumes unidirectionality of branching. Kayne ( 1 994) proposed that elements occupying the Mittelfeld have undergone licensing movement out of the VP to specific functional positions to the left of the verb, while elements occupying the Nachfeld are stranded by this type of licensing movement in the VP, as is illustrated in ( 1b). Haider ( 1 993, 1 995) proposed that elements occupying the Mittelfeld are directly licensed in OV languages, while elements occupying the Nachfeld are indirectly licensed. (1)
a. b. c.
Vorfeld C Topicalization (Kayne 94) (Haider 95)
Mittelfeld V middle field licensing movement direct licensing
Nachfeld Extraposition stranding indirect licensing
Assuming that arguments of the verb are base-generated solely in Specifier positions, Haider (this volume) argues that the order OV is more basic than the order VO and that all arguments of the verb are licensed within the VP in both OV and VO languages with the differences following from the directionality of licensing. Providing evidence for licensing movement out of the VP, we will argue in favor of the licensing movement and stranding approach as outlined in ( 1b) . In Section 2, we argue with the help of the infinitival marker in IPP constructions that not only the arguments of the verb but also VP-internal predicates, including verb-particles, small clause predicates and directional PPs, move out of the VP into specific licensing positions in the middle field. We will thus adopt the approach outlined in ( 1 b) and explore in Section 3 the ways and means by which an element ends up in the Nachfeld. In Section 4, we will discuss some of the consequences of the licensing movement and stranding approach for English. This approach implies that arguments and verb-particles in English, like in
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
295
continental West Gennanic, move into the middle field. The English verb or verb phrase moves in front of its arguments, most likely into T, making the assump tion necessary that negation and frequency adverbs do not occur between the verb and T but higher up in the clause structure.
2.
The syntax of IPP constructions and the structure of the lower middle field in West Germanic
In this section, we investigate the structure of the lower middle field in West Gennanic. The arguments for the particular structure that we propose are in part based on an analysis of the syntax of IPP constructions in West Flemish, Afrikaans and Gennan. We will argue that a YO-based approach (Kayne 1 994) is superior to the traditional OV-based approach in explaining the syntax of IPP complements. We will show that the distribution of verbal elements in the West Gennanic clause cannot be properly accounted for in tenns of rightward head movement of the verb and/or extraposition of verb-projections, and present empirical evidence for leftward V-movement in embedded clauses in West Gennanic. Furthennore, we will demonstrate that not only DP arguments of the verb, but also verb-particles, VP-internal predicates and CP complements have to move out of the VP to be licensed in specific positions in the lower middle field. Since Haider's proposal ( 1 995, this volume) assumes, like the traditional SOY approach, that these elements are licensed in the VP, these facts provide evidence for Kayne's antisymmetry hypothesis and against Haider' s alternative asymmetric approach. In the following discussion of the data, the traditional SOY approach is compared to the antisymmetric SVO approach. 2.1
The verb-final Pattern: aV and va accounts
Besides Haider's proposal, two approaches have been proposed to account for the verb-final embedded pattern in the West Gennanic SOY languages, the tradition al SOY approach and the anti symmetric SVO approach. The traditional approach allows for parametric variation in the head-complement order. For the West Gennanic SOY languages it is assumed that the functional and lexical projections in the IP domain are head-final. In this approach, the embedded verb-final position in (2a) is compatible either with the lack of V-movement (2b) or with rightward V-movement to a functional head within IP (2c). (2)
a.
wei! Hans gestern
das Buch las
since Hans yesterday the book read.pAST
296
ROLAND HINTERHOLZL b. c.
[cp weil [cp weil
[IP Hans gestem [vp das Buch las]]] [IP Hans gestem [vp das Buch ta lasi]]
A slight complication for the traditional approach is posed by the position that the infinitival marker occupies in infinitival clauses, as is illustrated in (3a). The sequence zu lesen (to read) looks very much like a head-initial right-branching structure, in which the infinitival marker heading a functional position takes the infinitival VP as its complement to the right. To accommodate this order with the particular assumptions about the word order in the IP domain within the standard approach, it is either assumed that the infinitival marker in the West Germanic SOY languages is not an independent head but rather a verbal affix (Haider 1 993), as illustrated in (3b), or that the infinitival verb undergoes rightward head movement to (right-)adjoin to the head-final infinitival marker in the IP domain, as illustrated in (3c). (3)
a.
ohne
b. c.
without yesterday the book to read.INF [cp ohne [IP PRO gestem [vp das Buch [y zu-Iesen]]]] [cp ohne [IP PRO gestem [vp das Buch tJ [I zu+lesenJ]]
gestem
das Buch zu lesen
In the antisymmetry approach (Kayne 1 994; Zwart 1 993), it is assumed that all structure is head-initial. In this approach, the embedded verb-final order is derived by some sort of feature driven leftward movement of VP-intemal material. This is illustrated for arguments of the verb in (4a) and for VP-intemal predicates in (4b) . While the movement of DPs into AgrPs is in accordance with recent minimalist assumptions about Case-licensing of arguments, the movement of APs, PPs and other VP-intemal constituents required within the antisymmetry approach appears stipulative in the absence of any empirical evidence. (4)
a.
[wei! Hans gestem
b.
[weil Hans gestem
das Buchi [vp las
since Hans yesterday the book
read.p AST
til ]
kranki [vp war ti]]
since Hans yesterday sick
was
The crucial empirical evidence that supports the assumptions necessary within the anti symmetry approach and at the same time disqualifies assumptions necessary within the traditional approach comes from the position of the infinitival marker in non-finite IPP constructions in West Flemish (Sa), taken from Haegeman ( 1 995), and Afrikaans (5b), taken from Donaldson ( 1 993).
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES (5)
a.
297
mee Valere te [willen [dienen boek kuopen]] een with Valere to want that book buy have 'with Valere having wanted to buy that book'
b.
Die banke moes
oop gewees het, om dit gister
the bank should open been
have
te [kan
it yesterday to can
betaal] het. buy have 'The bank should have been open to have been able to buy it yesterday. ' Since material, given in square brackets in (5), can intervene between the infinitival marker and the corresponding infinitival verb, it follows that the infinitival marker in the West Germanic SOY languages cannot be analyzed as a verbal affix, but, like in English, has to be analyzed as occupying a functional position within the IP domain. That this functional position is not head-final and that the sequence te+ V cannot be accounted for by assuming rightward head movement of the verb within the traditional approach, also follows directly from (5a). In (5a), the constituent that has been moved to the right of the infinitival marker cannot possibly be a head since it contains the DP that book. We thus assume that the infinitival marker occupies a functional head to the left of VP, which Haegeman ( 1 995) coined F l . In Section 2.3, we will argue that it can be shown with the help of the infinitival marker that VP-intemal constitu ents, including APs, verb-particles and PPs move out of the VP. But first we want to explain what IPP constructions are. 2.2 The syntax ofIPP constructions The infinitival clauses in (5) involve IPP complements. The Infinitivus Pro Participio (IPP) Effect occurs when a restructuring verb, e.g. a modal verb, selecting an infinitival complement (the dependent infinitive) is used in a perfect tense. In this case, the selecting verb does not show up in its participial form but is realized as a bare infinitive (the IPP infinitive). This is illustrated for Dutch in (6) and German in (7), where the infinitival forms willen/wollen replace the participial forms gewildlgewollt of the modal verb want. (6)
a.
*dat Elsje hem een Brief heeft gewild that Elsje him a
b.
letter has
schrijven
wanted(pp) write dat Elsje hem een Brief heeft willen schrijven that Elsje him a letter has want(INF) write 'that Elsje has wanted to write him a letter'
298
ROLAND HINTERHOLZL (7)
a.
*dass Else ihm einen Brief schreiben gewollt
hat
that Else him a letter write wanted(pp) has 'that Else has wanted to write him a letter' b. *dass Else ihm einen Brief schreiben wollen hat that Else him a letter write want(INF) has dass Else ihm einen Brief hat schreiben wollen c. that Else him a letter has write want(INF) As the contrast between (6b) and (7b) shows, in German, it is not sufficient to simply replace the participle with a bare infinitive, as in Dutch. In order to yield a grammatical sentence in German, the auxiliary has to invert with the cluster comprised of the dependent infinitive and the IPP infinitive (7c). We will give an account of inversion in German in this context when we talk about the internal syntax of IPP complements. The interesting issue that the IPP effect raises is the question of whether IPP infinitives are real infinitives or hidden participles of some sort. Most notoriously, Jakob Grimm ( 1 969/1 898: 1 95) put forth the hypothesis that the IPP infinitive is a prefixless participle. We will adopt the hidden participle account for the following reasons. Based on the distribution of participles, infinitivals and IPP complements in West Flemish and Afrikaans, we will argue in 2.2. 1 that IPP complements behave like participles and unlike infinitives. Secondly, the hidden participle account allows us to assume that for the purpose of checking the subcategorization of the auxiliary (which selects for a participial phrase) and for the purpose of temporal interpretation, the IPP infinitive counts as a participle. 2.2. 1 The external syntax of IPP complements Let us look at the distribution of infinitives, IPP complements and participles in West Flemish first. The following data are taken from Haegeman ( 1 995). In West Flemish, infinitives always follow their selecting verb, while participles always precede the selecting auxiliary. This is illustrated in (8) and (9), respectively. (8)
a.
da Valere dienen boek wilt that Valere that
kuopen
book wants buy
b. *da Valere dienen boek kuopen wilt that Valere that book buy wants 'that Valere wants to buy that book' (9)
a.
da Valere dienen boek gekocht eet that Valere that book bought has 'that Valere has bought that boek'
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
299
b. *da Valere dienen hoek eet gekocht that Valere that book has bought In contrast, IPP complements can both precede and follow the selecting finite verb. In the standard analysis, we may assume that IPP complements can optionally be extraposed (cf. ( l Oa-b)). However, the IPP complement must follow the selecting verb if the negative particle en is to be spelled out on the finite verb. This is illustrated by the contrast in ( l Oc-d). ( l 0)
a.
da Jan ee [willen
[Marie nen hoek geven]]
that Jan has want(IPp) Marie a b.
da Jan [willen
book give
[Marie nen hoek geven]] eet
that Jan want(IPp) Marie a book give 'that Jan has wanted to give Marie a book' c.
da Jan nooit en-eet willen
has
an Valere nen hoek geven
that Jan never en has want(IPp) to Valere a book give 'that Jan has never wanted to give a book to Valere' d.
da Jan nooit willen
an Valere nen hoek geven (*!!!JJ -eet book give en has
that Jan never want(IPp) to Valere a
On the other hand, the IPP complement must precede the selecting verb if the auxiliary is non-finite. In this case the IPP complement must occur between the infinitival marker and the infinitive as we have seen in (5a). Before we provide an analysis of the distribution of IPP complements within the anti symmetry approach, let us see how we would account for these data in the traditional SOY approach. Bringing the facts illustrated in (8)-( 1 0) into a single picture, it is hard to see what rule might govern extraposition in West Flemish. Remember that the regularities are the following. An IPP complement may or may not be extraposed if the selecting verb is finite. However, it must be extraposed if the selecting finite verb incorporates the negative particle en. On the other hand, it may not be extraposed at all when the selecting verb is non-finite. Finally, an infinitival complement must always be extraposed independently of the finiteness of the selecting verb. To capture these regularities in a single rule that follows from other properties of West Flemish or from general principles of grammar seems almost impossible. Hence we will try to give a better explanation of these regularities within the antisymmetry approach. In the antisymmetry approach all complements, that is, infinitives, IPP complements and participles in our case, start out to the right of the selecting verb. All we have to say about infinitival complements is that they stay in their base position, as illustrated in ( l l a).
300
ROLAND HINTERHOLZL (U) a. b. c. d. e.
F1 (verb)
Spec
te
IPP IPP IPP IPP
Auxfin en-Aux
F2 (verb) Aux Auxfin
tauxfin taux
Complement infinitive
tIPp tIPp tIPp tIPp
Remember, as shown in (9) above, that participles as opposed to infinitives have to precede the selecting auxiliary and that the infinitive in an IPP complement stands for a participle. Thus, we would like to propose that IPP complements are Participle Phrases (PartPs) and that they move like participles into a Specifier, say SpecF2P, of the selecting auxiliary to check the auxiliary's subcategorization. Now we would like to adopt the ingenious account by Haegeman ( 1 995), who proposes that the complex pattern illustrated in (10) follows from the availability of leftward head-movement of the selecting verb. If we assume that non-finite verbs as opposed to finite ones cannot move to F 1 , then we derive the fact that an IPP complement may not be extraposed if the selecting verb is non-finite, as noted above (Ub). All we have to say to derive the fact that an IPP complement may or may not be extraposed if the selecting verb is finite (and does not incorporate the negative marker en) is to assume that finite verbs may optionally move to F1 ( 1 1 c-d). Still the question arises why an IPP complement can be extraposed while a simple participle, as is shown in (9b), cannot. The answer is that a participle moves even higher than F l . ( 1 2) shows that a participle cannot intervene between the infinitival marker and the auxiliary but must always precede the infinitival marker. ( 1 2)
a.
mee Valere dienen boek gewild te een with Valere that
book wanted to have
(vu zenen verjoardag) for his birthday 'Valere having wanted that book for his birthday' b. *mee Valere dienen boek te gewild een (vu zenen verjoardag) c. ? *mee Valere dienen boek t' een gewild (vu zenen verjoardag) Finally, if we assume, following Haegeman's ( 1 995) proposal that a) the negative particle en is licensed in F 1 and b) that it has to incorporate into the verb, then we derive the fact that an IPP complement has to be extraposed if the finite verb incorporates en l ( U e). In ( 1 3), we summarize the distribution of participles, infinitives and IPP complements with respect to the selecting verb and infinitival marker te.
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES ( 1 3)
301
participle te IPP verb infinitival complement
To minimize the difference between participles and IPP complements (hidden participles) we propose that participles are not moved in one swoop from their base position to the right of the selecting verb to their surface position but that they, like IPP complements, first undergo XP-movement to SpecF2P of the selecting auxiliary (to check its subcategorisation) and subsequently undergo XO-movement to Fl . The latter (additional) movement of participles must be triggered by the participial morphology that IPP infinitives lack. That participles indeed undergo this complex two-step movement in West Flemish and the other West Germanic languages is supported by the behavior of participles in Afrikaans. In this language, the intermediate step of the complex movement of participles, absent in all other West Germanic languages, is evidenced. As is illustrated in ( 14), the participle is spelled out between the infinitival marker and the infinitival verb. ( 14) also provides the ultimate confirmation for our hypothesis that IPP infinitives are hidden participles: in Afrikaans, participles and IPP infinitives have exactly the same distribution (cf. (14) and (5b)). (14)
af � geskagel het. you ought the light off to turned have 'You should have turned the light off. '
Jy behoort die lig
2.2.2 The internal syntax of IPP complements In this section, we provide an account of the IPP effect which we left unex plained so far. We will also explain why participles and IPP infinitives pattern exactly alike in Afrikaans but have a slightly different distribution in the other West Germanic languages, as illustrated above for the case of West Flemish. The IPP effect occurs in restructuring contexts. One important feature of restructuring is the formation of verbal complexes (cf. Haider 1 993; Rutten 1 99 1 ) . In Hinterh61z1 ( 1 999), we argue that in restructuring contexts, due to a defective complementizer, the dependent infinitive moves into SpecF2 of the selecting verb to check its subcategorisation. Following Bech ( 1 955), we assume that a verb selects for the status of its non-finite complement. That is, it deter mines whether the dependent nonfinite verb is a participle, a bare infinitive or a to-infinitive. Thus we will henceforth refer to F2P as Status Phrase (StatP). To explain the IPP effect, we will make use of the particular structure of participle phrases in West Germanic. We note that the languages and dialects in which the participle is formed without the participial prefix ge, namely Frisian and Low German, do not display an IPP effect. In the following we will show
ROLAND HINTERHOLZL
302
how the IPP effect can be reduced to a structural incompatibility between the prefix and the infinitive dependent on the restructuring verb. In the West Germanic languages displaying the IPP effect, the participle is formed by affixation of the prefix ge and the suffix tid. We follow Halle & Marantz ( 1 993) in assuming that inflected forms are (partially) derived in the syntax. More specifically, we propose that the participial prefix ge is inserted in SpecStatPIF2P of the participial phrase. The verb in the participial phrase will then first move to F2, to check its prefix, and then up to Fl to adjoin to its suffix. The prefix will then left-adjoin to the complex of verb and suffix to form the participle before Spell-out. This is illustrated in ( 1 5). (15)
[PIP -t [F2P [ge] F2 [vp V]]]
Gr LLJ
If the verb in the participle phrase is a restructuring verb, then the dependent infinitive will at some point in the derivation (before Spell-out in German, but after Spell-out in Dutch, West Flemish and Afrikaans) move into SpecF2P. It follows that a verb in participial form and a bare infinitive selected by such a verb rule each other out. In this case the participial prefix is blocked by the dependent infinitive, that is to say, it cannot be inserted. We argue that the blocking of the prefix leads to a violation of a morphological constraint at MF, namely, of the requirement that a participle consist of a suffix and a prefix. This violation is avoided in that a) no phonological material is inserted in Fl (that is, the suffix is dropped) and b) the verb remains in F2 and is spelled out with the default morphology of a bare infinitive. This is illustrated in ( 1 6). ( 1 6)
[F IP 0 [F2P [dependent infinitive] IPP infinitivei [vp tiJJ]
The morpheme in Fl in ( 1 6), though not containing any phonological feature (represented as "0" in ( 1 6)), arguably contains the formal feature [+participle] and a semantic feature [+ Past] (or the condition that event time precede reference time in a Reichenbachian system). We propose that the semantic feature of the participle has to move to Fl of the auxiliary (which we now identify as an Aspect-head) to be linked with the matrix tense and argue that it is movement of this feature (after the formal feature of the participle has been checked by XP-movement of the Participle Phrase into StatP of the auxiliary) that pied-pipes a phonetically realized participle but remains invisible in the case of an IPP infinitive. Returning to obligatory inversion with the IPP complement of the auxiliary in German (cf. (7b-c)), we propose that the movement of the semantic feature of
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
303
the participle is in a way made visible by the concomitant movement of the auxiliary, which, like auxiliaries in West Flemish, moves up to F 1 , presumably, in order to help license the empty morpheme, indicated as "0" in ( 17). We assume that the empty morpheme is licensed before Spell-out in German by attaching to the higher head which for this reason needs to be lexicalized by movement of the auxiliary. (17)
daft Else ihm einen Brief [PIP hati [F2P [PIP 0 [[schreiben] wollen]] ti [vP til]] '"
that Else him a
letter
has
/
write
want.!pp
That movement of the auxiliary in ( 17) is obligatory is probably due to the 2 strength of the participial feature. Thus, ( 1 7) provides an example for a case of leftward V-movement in German. Why then do IPP infinitives and participles behave alike in Afrikaans while they differ in their distribution in the other West Germanic languages? Note that verbs in Afrikaans have lost all their endings. In particular, participles, while retaining the ge-prefix, have lost their dlt-suffix. It stands to reason that partici ples in Afrikaans, like IPP infinitives in general, contain an empty morpheme in F 1 , movement of which will fail to pied-pipe the participle in F2P below. Thus, we have assimilated our account of the IPP effect in West Germanic to an independent fact in one of its member languages, namely Afrikaans. In both cases, we find the morphology of the participle to be defective. In the case of IPP infinitives, this is caused by the blocking of the prefixal part of the participi al circumfix in the syntax. In the case of the participle in Afrikaans, this is due to the general loss of verbal endings in this language. We also have now an explanation for why it is that verbal elements in West Germanic do not normally appear between the infinitival marker and the infinitival verb, that is, in German and Dutch. The explanation is that only nonfinite verbs with defective morphology will remain there, while all others will just move through SpecF2P on their way to F1 of the selecting verb. To conclude, the behavior of participles in Afrikaans provides strong, independent evidence for our account of IPP infinitives in West Germanic. 2.3 The structure of the lower middle field in West Germanic In this section, we investigate the structure of the lower middle field in West Germanic from the perspective of the antisymmetric SVO approach. The most important results of this investigation are the following generalisations: a)
304
ROLAND HINTERHOLZL
Nominal complements of the verb always have to leave the VP before Spell-out (independently of whether they are definite or indefinite) and are licensed in functional projections above the position of manner adverbs. b) Also small clauses, idioms and directional PPs have to move out of the VP and are licensed in a position below manner adverbs. c) Full sentential complements are not licensed in their base position within VP either. They undergo short movement and are licensed in a functional projection directly below the position to which the verb in embedded clauses moves. That arguments move out of the VP obligatorily can already be shown with the help of manner adverbs. Adverbs like sorgfiiltig ( 'carefully' ) , genau ( 'pre cisely' , 'exactly'), gut ( 'well'), schlecht (badly) and so forth show that both definite and indefinite NPs have to leave the VP (cf. Haiden 1 995; Brugger & Poletto 1 993; Hinterholzl 1 999). Since manner adverbs are usually analyzed as being adjoined to VP (we will later show that they actually occupy a higher position in the tree), an NP preceding a manner adverb must have moved out of the VP. ( 1 8) and ( 1 9) show that in sentences with neutral intonation, the direct object precedes the manner adverb. a.
( 1 8)
weil Hans das Buchlein Buch sorgfiiltig gelesen hat since Hans the bookJa book carefully read has ' since Hans has read the/a book carefully' b. ??weil Hans sorgfiiltig das Buchlein Buch gelesen hat! since Hans carefully the bookJa book read has
( 1 9)
a.
weil Hans den Planleinen Plan genau ausfohrte
since Hans the plan/a plan exactly executed ' since Hans executed the/a plan exactly' b. ??weil Hans genau den Planleinen Plan ausfohrte since Hans exactly the plan/a plan executed We assume that sentences with neutral intonation correspond to the unmarked or basic word order, from which sentences with marked word order (often accompa nied with a special intonation pattern) are derived by additional movement. These considerations are important since in a scrambling language like German alternative orders are often equally good but can - due to the assumptions above - be treated as marked or secondary. The negative marker nicht ( 'not') obligatorily precedes manner adverbs as is shown in (20). (20)
a.
weil der Hans das Buch nicht sorgfiiltig gelesen hat since the Hans the book not carefully read has ' since Hans has not read the book carefully'
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
305
b. ??wei! der Hans das Buch sorgfaltig nicht gelesen hat since the Hans the book carefully not read has All definite nominal arguments precede the negative marker (2 1). If an indefinite NP precedes the negative marker it is interpreted as specific (22a). If an indefi nite 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 (this holds for definite and indefinite NPs), 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 (22b). (2 1 )
a.
wei! der Hans das Buch nicht gelesen hat
since the Hans the book not read has ' since Hans did not read the book' b. ??wei! der Hans nicht das Buch gelesen hat4 since the Hans not the book read has (22)
a.
b.
wei! Hans ein Buch nicht gelesen hat since Hans a book not read has 'there is a book that Hans did not read' (only specific interpre tation) wei! Hans nicht ein Buch gelesen hat (sondern zwei; sondern since Hans not a book read has but two but
ein Journal) a j ournal 'Hans did not read ONE book, he read TWO books; Hans did not read a B OOK, 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 be simply adjoined to an XP which it narrowly or exclusively negates (23 a-b). (23)
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 (24a). In this case, the
ROLAND HINTERHOLZL
306
negative marker is non-overt or fused5 (24b) . Thus, we cannot determine in (24) whether a nonspecific indefinite NP has to move across the negative marker. (24)
a.
wei! Hans kein Buch gelesen hat since Hans no 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
no book
not
read
has
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 B avarian dialects, the phenomenon of negative concord which allows for the Spell-out of the negative marker even in the presence of negative constituents. As (25) shows, the negative existential NP kein Buch ( 'no book' ) has to precede the negative marker net ('not' ) . In the following, we will assume that a negative existential NP with a kein-determiner occupies SpecNegP. (25)
a.
woei da Hans ka Buach net glesn hot since the Hans no book not read has 'since it's not the case that John read a book' b. *woei da Hans net ka Buach glesn hot since the Hans not no 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, i.e., checks the negative marker in SpecNegP. This is shown in (26). (26)
a. b.
c.
wei! ein Mann einer Frau
eine Blume schenkte
since a man a woman.DAT a flower gave weil kein Mann einer Frau eine Blume schenkte since no man a woman.DAT a flower gave ' since it is not the case that some man gave some flower to some woman' wei! ein Mann keiner Frau eine Blume schenkte since a man no woman.DAT a flower gave 'since a (certain) man did not give some flower to some woman'
(26a) is a sentence that contains three indefinites. If all three DPs are interpreted nonspecific ally, the negation of (26a) must be (26b). In (26c), where the indirect object is spelled out with a kein-determiner, the higher subject must be interpreted
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
307
as specific, if we exclude focus-affected and quantificational readings. (26b) in conjunction with (25) and (26c) shows - if we analyze the kein-phrase as occupying SpecNegP - that nonspecific indefinites may follow negation. Also quantified NPs may follow the negative marker without giving rise to a constituent negation-interpretation (27a-b). In (27a-b), the negative marker can act as sentence negation. As (27c) 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 quanti fied NPs need not move across negation. (27)
a.
b.
c.
weil der Hans nicht viele Bucher liest since the Hans not many books reads ' since it is not the case that Hans reads many books' wei! der Hans nicht jede Frau anbetet since the Hans not every woman adores ' since it is not the case that Hans adores every woman' Der HANS hat nicht viele Bucher gelesen, der PETER hat the Hans has not many books read the Peter has viele gelesen. many read 'it was not Hans but Peter who read many books.'
To summarize the above discussion, we conclude that all nominal arguments have to leave the VP before Spell-out. It seems that nominal arguments in German have to undergo a type of scrambling that moves them across the position of manner adverbs (short scrambling). This type of scrambling does not affect the interpretation of the moved NP. On the other hand, movement across the negative marker seems to be motivated by semantic features. Specific NPs obligatorily move across the negative marker while nonspecific indefinite NPs, unless they move into SpecNegP to check sentential negation, remain below the negative marker. This is reminiscent of the well-known facts, reported and discussed by Diesing ( 1 992) and Kratzer ( 1 9 89), that indefinite NPs in German differ in their interpretation depending on whether they follow or precede sentential adverbs like oft (often), as illustrated in (28). (28)
a.
wei! Hans ein Buch Qft gelesen hat since Hans a book often read has 'since Hans often read a certain book' (only specific interpreta tion)
308
ROLAND HINTERHOLZL b.
weil Hans qft
ein Buch gelesen hat since Hans often a book read has ' since Hans often read some book or other' (only nonspecific interpretation)
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 SpecNegP only if they receive a contrastive interpretation as illustrated in (22b) 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. (29a) 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 (29b). (29)
a.
HANS hat nicht viele der Hans has not
viele davon
Bucher gelesen, PETER hat
many of.the books read
Peter
has
gelesen.
many thereof read b.
weil der Hans viele Bucher nicht gelesen hat since the Hans many books not read has ' since for many of the books it holds that Hans did not read them' (only specific interpretation)
So the movement of NPs across manner adverbs (short scrambling) has to be distinguished from the type of scrambling that moves them across the negative marker and other scope-bearing sentential adverbs. While the latter kind of movement is apparently triggered by a semantic feature, namely specificity, the former kind seems to occur for reasons of purely formal licensing. For purposes of reference, we will call the latter kind of movement long scrambling. We like to propose that the Case-licensing positions in German occur between manner adverbs and sentential negation. So short scrambling will be motivated by Case-licensing. Any further movement of an NP (long scrambling) must be motivated with semantic properties of that NP. We classify movement across sentential negation as long scrambling. To account for the fact that DPs that follow sentence negation receive a contrastive interpretation, we will take up a proposal by Richard Kayne (p.c.) and assume that there is a Focus phrase just below negation into which contrastively focussed elements move (cf. also Brody 1 990). The resultant structure is illustrated in (30).
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES (30)
309
[CP [TP T [Specifics [oft [Neg [Focus [AgrNom [AgrDat [AgrAcc [Manner [V]]]]]]]]]]]
At this point, we will not provide a formal account of the regularities discussed above but simply restate them as descriptive generalisations. 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 SpecNegP. 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 clear to me how to properly express these regularities in a system of feature checking. One possibility is to assume that something like (3 1 ) holds. (3 1)
Once Case is checked, only the feature of a DP with the closest licensing head is checked overtly
Given (3 1), a DP with a specificity and a contrastive focus feature will move overtly only to the closest licenser, that is in this case, into SpecFocusP. In a VO-based approach, we have to assume that the non-verbal predicates in (32) have been moved leftward from a position to the right of the verb. We assume that these nonverbal predicates, together with the "direct object" of the verb, form a Small Clause which in tum forms the complement of the verb as is illustrated in (33). (32)
a. b. c.
(33)
a. b. c.
wei! Hans das Haus gelb
farbte
since Hans the house yellow painted wei! Hans ihn einen Idioten nannte since Hans him an idiot called weil Hans Maria intelligent findet since Hans Maria intelligent finds ' since Hans considers Maria intelligent'
wei! Hans wei! Hans wei! Hans
[vp !arbte Esc das Haus gelb]] [vp nannte Esc ihn einen Idioten]] [vp findet Esc Maria intelligent]]
Following Zwart ( 1 993) and Koster ( 1 994), we assume that these small clauses undergo XP movement to check some feature of the verb and are moved into the Specifier of a Predicate Phrase (PredP) that dominates the VP. (34) shows that this Predicate Phrase occupies a position below the position of manner adverbs. (34a) also shows that the subject of the small clause den Zaun is moved out of the domain of the manner adverb. Presumably it moves, like the nominal
310
ROLAND IDNTERHOLZL
arguments of verbs, to its Case-licensing position below the negative marker, and from there, it moves like other specifics (if not contrastively focused) to a licensing position above the negative marker, as is indicated in (34c). (34)
a.
wei! Hans den Zaun sorgfiiltig gelb
angestrichen hat
since Hans the fence carefully yellow up.painted
has b. ??weil Hans den Zaun gelb sorgfiiltig angestrichen hat since Hans the fence yellow carefully up.painted has c. wei! Hans den Zaunj nicht ti [ti gelb] angestrichen hat since Hans the fence not yellow up.painted has That VP-internal predicates (including small clauses, idioms and directional PPs) in fact occupy a position in the middle field, namely PredP, rather than remain within a right-headed VP, can be shown with the help of the infinitival marker. The data in (35) show that an adjectival small clause predicate cannot remain within the VP or for that matter, incorporate into the verb, as is often assumed (cf. Neeleman 1 994), but has to move out of the VP to a licensing position above the infinitival marker. (35) shows that PredP dominates F 1 P in German (cf. (35a-b)) and in West Flemish (cf. (35c-d)) . The contrast in (35c-d) is even more significant since we have seen in (5a) above that the infinitival marker in West Flemish can in principle be separated from the verb. We have argued above that Small Clauses are licensed in PredP. (35) shows that the head of a Small Clause predicate obligatorily precedes the infinitival marker occupying F 1 , hence PredP must dominate F 1 P. (35)
a.
ohne
die Tilr griin zu fiirben
without the door green to paint 'without painting the door green' b.
*ohne
die Tilr zu griln fiirben
without the door to green paint c.
K goan proberen van die deure groen te verwen.
1 go
try of the door green to paint 'I will try to paint the door green. ' (Haegeman 1 995;(23))
d. *K goan proberen van die deure te groen verwen. 1 go try of the door to green paint The test with sentential negation indicates that idiomatic expressions and directional PPs are licensed in PredP too. Since these elements can only occur between the negative marker and the infinitival marker, even if they have a definite reference as the PPs in (36c-d) do, they must occupy PredP in (36) .
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES (36)
a.
urn
es ihr nicht zur
311
Verfiigung zu stellen
in. order it her not to.AGR availability to put 'in order to not make it available for her' b. *urn es ihr zur Verfiigung nicht zu stellen in. order it her to.AGR availability not to put urn die Milch nicht in den Kithlschrank zu stellen c. in.order the milk not in the refrigerator to put 'in order not to put the milk into the fridge' d. *urn die Milch in den Kithlschrank nicht zu stellen in.order the milk into the refrigerator not to put So far we have arrived at the following picture of the lower middle field in German. Nominal arguments (including prepositional arguments) are Case licensed in functional positions above the position of manner adverbs. The negative marker dominates Manner adverbs. Small clauses, idiomatic expressions and directional PPs are licensed in PredP directly below the position of VP adverbs. This Predicate Phrase dominates F l , a position which itself immediately dominates VP (37). (37)
[DPs [Neg [DPs [VP-adverbs [PredO [FI [yp]] ]
Let u s now look at the behavior o f sentential complements (CPs) . CPs in German appear invariably to the right of the verb selecting them. Given that it is standardly assumed that the German verb in embedded clauses does not move, the CPs in (38a-b) seem to be in their base position, as is indicated by the analysis of (3 8b) in (3 8c). Although there is clear evidence that the German embedded verb does not move to Tense - as we have seen above it cannot even move across negation or even VP adverbs - it is still possible that it undergoes some type of short local movement that moves it across a sentential complement that has itself moved into the Specifier of a functional projection for purposes of licensing, as is illustrated in (38d) for the sentence in (38b). (3 8)
a.
weil Hans der Maria
b.
ohne
c. d.
without the Maria.DAT to say that Peter sick is ' without telling Maria that Peter is sick' ohne der Maria [F I zu [yp sagen dajJ Peter krank ist]] ohne der Maria [F I zu [F2 sagen [F3 CP [vp tv tcp]]]]
nicht sagte, dajJ Peter krank ist since Hans the Maria.DAT not told that Peter sick is 'since Hans did not tell Maria that Peter is sick'
der Maria
zu sagen, dajJ Peter krank ist
ROLAND HINTERHOLZL
3 12
From the licensing movement of adj ectives (cf. (35) above) an argument can be constructed that CP complements cannot remain within the VP. (39a) shows the only possible order between adj ective and CP complement in an infinitival clause. The adjectival phrase undergoes licensing movement into PredP above the infinitival marker. As (39b) shows, pied-piping of the CP complement leads to ungrammaticality. It follows then that the CP complement has to move out of the VP before the adjectival phrase moves to PredP, in order to derive (39a) from the underlying structure in (39c). (39)
a.
ohne froh zu sein, dajJ der Hans nicht kam without happy to be that the Hans not came 'without being happy that Hans did not come'
b. *ohne rJroh, daft der Hans nicht kam] zu sein without happy that the Hans not came to be [cP ohne . . . [F IP ZU [yp sein [AdjP froh [CP]]]]] c. In the standard theory, cases like (39) have been accounted for by the operation of extraposition that right-adjoins CP complements to the local VP or !P. Since, following Kayne ( 1 994), we want to refrain from positing right-adjunction, we propose that the CP complements are licensed in SpecF3P directly above VP. This entails that the verb in German moves up (at least) to F2, a functional projection dominating the licensing projection for CPs. We thus arrive at the following structure of the German sentence, a structure that we assume also holds for Dutch and West Flemish. In (40), L-NPs stands for "long-scrambled NPs", S-NPs for "short-scrambled NPs" and S-Advs for "sentential adverbs" like "often". For the sake of simplicity, we assume that the domain of L-NPs corresponds to the traditional TP. (40)
3.
[L-NPs [S-Advs [Neg [S-NPs [VP-Advs [Predo [F I P zu [F2P V [F3P CP [yp]]]]]]]]]]
Of what remains in the VP: Stranding as Copy and Delete
In the previous section, we have argued that the arguments of the verb, including sentential complements, and VP-internal predicates all move out of the VP to specific licensing positions in the middle field. In this section, we want to investigate which elements can remain in the VP. The elements that can remain in the VP are essentially CP-adjuncts and PP-adjuncts. As (4 1 ) shows, CP-adjuncts to DPs can follow the verb that has
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
3 13
moved out of the VP into F2 in the middle field. In the absence of rightward movement, the simplest analysis of the position of the relative clause in (41 a) and of the so-called Noun-complement clause in (4 1b), is one in which it is assumed that these clauses remain in the VP stranded by the licensing movement of the DP they modify. -
(4 1 )
a.
Hans hat die Frau
e inge laden, die ich ihm empfohlen
Hans has the woman invited
who I
him recommended
habe. have b.
Hans hat die Behauptung zuritckgewiesen, daft die Erde rund Hans has the claim
refuted
that the earth round
ist. is As (42) shows, also PP-adjuncts to DPs (42a) or the verb (42b) can appear in a position following the verb in F2. Cases like (42a), which we will call PP-out-of NP, have traditionally been analyzed as involving extraction of the PP out of the NP it modifies plus subsequent right-adjunction of the PP to either VP or IP. Cases like (42b) , which we will call PP-extraposition, have traditionally been analyzed as involving right-adjunction of the PP to either VP or IP. This is illustrated in (42c) . (42)
a.
Hans hat ein Buch (itber Chomsky) gekauft (itber Chomsky)
b.
Hans hat (in Wien) ein Buch gekauft (in Wien)
c.
Hans has in Vienna a book bought in Vienna [cp Hans hat [IP PP [vp [NP ein Buch PP] gekauft] hat]] I I
Hans has a
book about Chomsky bought about Chomsky
I
In the following section, we want to investigate in which manner PPs modifying DPs can be stranded in positions following the verb. 6 We will argue against analyses that assume rightward movement. Furthermore, we will argue that PP-out-of-NP cases should not be accounted for in terms of extraction of the PP and subse quent remnant movement of the constituent containing the verb and the NP. 3 . 1 PP-out-of-NP The data and arguments presented in this section are drawn from Brugger & HinterhOlzl ( 1 998). Brugger & Hinterholzl ( 1 998) argue that cases of PP-out-of-
ROLAND HINTERHOLZL
314
NP should not be accounted for in terms of extraction of the PP, be it rightward extraction or leftward extraction plus subsequent remnant movement. Here we will briefly sketch the most important arguments. One important property of cases of PP-out-of-NP is the fact that the PP following the verb can receive nuclear stress and give rise to a wide focus reading, such that (43) is a perfect answer to an out-of-the-blue question "What happened?" (43)
Ich glaube daj3 Hans ein Buch gekauft hat aber CHOMSKY. I
believe that Hans a
book bought has about Chomsky
The fact that the PP in (43) can receive nuclear stress (with the DP ein Buch receiving secondary stress) immediately militates against the assumption that the PP has been right-adj oined to VP or IP, after having been extracted out of the DP ein Buch. If the PP were right-adjoined to VP or IP, nuclear stress should fall on the DP ein Buch, according to standard assumptions about the relation between phrase structure and stress assignment (cf. Cinque 1 993 ; Zubizarreta 1 998). The fact that the PP in (43) can receive nuclear stress and give rise to a wide focus reading can also be taken as an argument against an analysis in which it is assumed that the PP is extracted leftward followed by remnant movement of the constituent ein Buch gekauft hat. In German, elements · that undergo leftward movement out of their licensing positions are either deaccented (long scrambling) or if stressed give only rise to a narrow focus (contrastive focus) reading. Secondly, if sentences like (43) were derived via extraction of the PP out of the DP, then it remains unclear why cases of PP-out-of-NP in German neither violate the Specified Subject Condition nor the Specificity Condition. (44a) shows that a specified subject blocks extraction of DP-intemal material. (44b) shows that cases of PP-out-of-NP are unaffected by a specified subject. (44)
a. *Who did you read Peter's book of? b. Hans hat Peters Buch gelesen aber Chomsky. Hans has Peter's book read about Chomsky
Muller ( 1 995) takes the contrasts between wh-extraction out of a DP and the extraction of a PP out of a DP, exemplified above for the Specified Subject condition, as an argument in favor of a rightward movement analysis of these cases and of extraposition in general. He surmises that leftward subextraction out of a DP must proceed through SpecDP, and will thus be affected by a specified subject, while rightward subextraction out of a DP uses right-adjunction to the DP and is thus unhampered by a specified subject, as is illustrated in (45).
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES (45)
315
that [IP [Hans [yp [DP [ a book til ti] bought]] PPJ
The problem with Muller's solution to the extraction paradox, illustrated in (44), is that the resultant structure of rightward movement makes the wrong predic tions about quantifier binding. In cases where more than one PP has been "extrapo sed" , an analysis in terms of right-adjunction predicts for the sequence XP PP I PP2 either that PP2 c-commands PP 1 if adjunction is to different maximal projections or that PP1 and PP2 c-command each other if adjunction is to the same maximal projection. As the pronominal binding facts in (46) indicates an analysis in terms of right-adjunction is untenable. The contrast in (46a-b) requires that PP 1 c-com mands PP2. For cases where only one PP has been extraposed, Muller's account of extraposition fails to explain why pronominal binding out of an extraposed PP is impossible, as is illustrated in (46c). (46)
a.
weil Hans eine Rede since Hans a
gehalten hat [in jeder Stadt] [uber
speech given
has in every city
about
ihre UW-Probleme] its environmental.problems b. ??weil Hans eine Rede gehalten hat [uber ihre Umweltprobleme]
[in jeder St.] ?? c. weil Hans ihren Burgermeister getroffen hat in jeder Stadt since Hans its
mayor
met
has in every city
The binding facts in (46) as well as the prosody and focus properties in (43) follow straightforwardly if we assume that the PPs in (46) and (43) are simply stranded in the VP. Kayne ( 1 994) proposes that a relative clause is stranded in the VP by sub extraction of the DP modified by the relative clause for reasons of case licensing. Since in the structure in (47), D and NP do not form a constituent excluding the PP that could be sub extracted for reasons of Case-licensing, leaving behind the PP within VP, and since subextraction of PPs is excluded also, as we have argued above, we need to look for an alternative means of stranding PPs that modify DPs. (47)
[DP D [NP N [pp P [DP]]]] a book about Chomsky
3 .2 Extraction as partial deletion In the Minimalist Program (Chomsky 1 993, 1 995) movement is viewed as the result of the application of two basic operations, copy and delete. The standard
ROLAND HINTERHOLZL
316
application of copy-and-delete has it that one copy, normally the material that corresponds to the trace in the traditional treatment of movement, is entirely deleted, as illustrated in (48a). In (48a) movement targets the constituent C consisting of the lexical items X and Y, in which process both X and Y are deleted in their pre-movement position. This is the standard case of total deletion of one copy . yielding the traditional pattern of the antecedent-trace relationship of movement. (48)
a. b.
[c X Y] . . . . . . [c X-¥] [c X ¥] . . . . . . [c X YJ
However, total deletion of one copy is not a logical necessity. We can imagine cases of partial deletion. In (48b), movement again targets constituent C consist ing of the lexical items X and Y, but this time X has been affected by forward deletion while Y has been affected by backward deletion. 7 Given the operations of forward and backward deletion, the structure in (49a) can be derived from (49b), where two full sentences have been coordinat ed, by forward deletion of John and backward deletion of the newspaper. (49)
a. b.
John bought and read the newspaper John bought the BewsJ)aper and JeIm read the newspaper
Since in Minimalism, the operation of move alpha is deconstructed into indepen dent copy and deletion processes, Wilder ( 1 995) proposes to investigate whether the operations of FWD and BWD are available in ordinary antecedent-trace relationships and formulates the condition of Chain-Internal Selective Deletion (CISD) given in (50) as a means of constraining partial deletion. The deletion operations illustrated in (48) are two possible outcomes of applying CISD in an antecedent-trace configuration. (50)
Phonological deletion can remove part of the antecedent and the complementary part of the trace (Wilder 1 995; (57))
In the next section, we will argue that the deletion processes operative in antecedent-trace configurations may delete parts of constituents but have different properties from the deletion processes that Wilder ( 1 994) argues occur in contexts of ellipsis. Deletion in ellipsis contexts is deletion of phonological features only, while deletion within antecedent-trace configurations is essentially guided by the deletion of formal features. Only in the cases where deletion of formal features is not at stake, that is in cases of deletion of material that is pied piped by movement operations, does deletion in antecedent-trace configurations obey the same restrictions that are relevant for deletion in cases of ellipsis. In the
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
3 17
following, we will continue to use FWD and BWD as descriptive tenns to refer to deletion operations as illustrated in (48b). 3 .2. 1 Deletion andfeature checking Given (50), a sentence like (5 1 ) could be derived simply by selective deletion operations, that is, without employing any extraction operation. As we have argued in 2.3 above, the direct object in Gennan moves overtly into SpecAgrAccP to check its Case, leaving behind a copy in its base position following the verb. This is illustrated in (52). Under these assumptions (5 1 ) can be derived by forward deleting ein Buch and backward deleting uber Chomsky, as is illustrated in (53a). However, (50) would allow us to derive also (53b-d) which are ungrammatical. (5 1 )
wei! Hans ein Buch liest uber Chomsky since Hans a
(52) (53)
book reads about Chomsky
wei! Hans [AgrAcc [ein Buch] [P2P liest [yp [ein Buch]]]] a. wei! Hans [ein Buch libel' Chemsky] liest [ein Buch Chomsky] b. **weil Hans rein Buch uber Chemsky] liest [ein Buch Chomsky] c. **wei! Hans [ein Buch libel' Chemsky] liest [em- Buch Chomsky] d. *weil Hans [ein Buch libel' Chemsky] liest [ein Buch Chomsky]
uber libel'
uber uber
It is clear from the examples in (53) that CISD is too unconstrained as it stands. Intuitively speaking, what is going on in cases like (53) is the following. The DP ein Buch moves out of its VP-internal base position into AgrAccP in order to check its case. This operation pied-pipes the DP-internal PP. Given this view, (53a) can be derived from the stipulations in (54) (cf. HinterhOlzl 1 997) . (54)
Free Deletion of Pied-piped Material (FDPM) Material that is moved to check a feature is subject to forward a. deletion b. Material that is pied-piped by such movement is subject to optional backward deletion
The FDPM, if correct, should follow from more basic principles in the grammar. We will later argue that clause a) of the FDPM follows from the elementary mechanics of the operation of feature checking in a derivation and that clause b)
ROLAND HINTERHOLZL
318
can be derived from the LCA. Before doing that, we would like to point out that the FDPM raises an interesting question concerning the operation of feature checking. With respect to feature checking, the question arises what the scope of the checked feature is. For example, in order to derive (53a) from (54), we have to assume that "material that is moved to check the Case-feature" comprises the determiner and the head noun but not the rest of the DP (the preposition and the DP within its scope). The traditional assumption has been that Case is a property of the head noun of an NP and that the determiner agrees with its head noun in Case (and possibly other features) . Within the DP-based approach we have to assume that Case is minimally a property of the determiner. In an LGB-framework, we would assume that when a DP is moved to check Case the scope of this operation comprises the head of the DP and all heads governed by it. In the Minimalist Framework, we have to assume that when a D P moves to check Case, Case is checked on the determiner only. Additional movement of the head noun to the determiner is necessary to check Case (and possibly some other features) 8 on the head noun as well. The question then arises how to derive (54a) from more basic principles. The question is relevant since in fact, in Chomsky's ( 1 993, 1 995) execution of the feature checking operation, it has to be stipulated which copy is to be spelled out. There it is assumed that if an element X has to check a feature f and X is copied and merged with the target category containing feature f, feature f is checked in all copies of X. In such a system, it is indeed necessary to stipulate that it is always the highest copy that is spelled out. Nunes ( 1 995) argued that a particular execution of the feature checking mechanism gives us for free the effect that it is always the highest copy that is spelled out. This follows simply from the assumption that only the feature of the copy is checked that merges with the target category containing the relevant feature. This is illustrated in (55), where the DP ein Buch is moved into SpecAgrAccP to check its case. In both (55 a-b) , the Case feature is only checked off in the copy in SpecAgrAccP but left unchecked in the copy in the VP-internal position. The derivation then converges only if, as in (55a), the lower copy is deleted, since deletion of the lower copy also effects deletion of the offending unchecked Case-feature. If, on the other hand, as in (55b), the higher copy is deleted, the copy with the offending unchecked feature remains and causes the derivation to crash. (55)
a. b.
weil Hans [AgrOP [DP ein Buch d [vp kaufte [DP ein Buc.� J *weil Hans [AgrOP [DP ein Buck J [vp kaufte [DP ein Buch d
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
319
So it i s the feature checking mechanism itself that decides which copy i s spelled out and which copy is deleted. This immediately predicts clause b) of (54), namely that the computational system is silent with respect to pied-piped material. Copies of pied-piped material are identical in their feature content and it only follows from the LCA, namely from the fact that two or more identical copies cannot be ordered with respect to third elements in the clause, that only one of the identical copies can be spelled out but it is left to the liberty of the speaker to decide which one. Now let us go back to (53) and see whether we can derive from the specific interpretation of (54) why (53a) is grammatical and (53b-d) are not. Let us start with the easy case (53d) . (53d) is ungrammatical for the very same reason (55b) is ungrammatical: the copy with the offending Case feature remains and causes the derivation to crash. In order to converge, at least the determiner containing the offending Case feature must be spelled out in the higher copy. This is the case of (53c). Why is (53c) still ungrammatical? (53c) is ungrammatical since the head noun still has a Case-feature to check, which it can only do in the higher copy, since in the lower copy the determiner with the relevant feature has been deleted. 9 This leaves us with the choice between (53a-b). Let us look at the Spell-out of the preposition. The preposition being pied-piped by the operation of Case checking of the containing DP is free to be spelled out in the higher or in the lower copy. Note, however, that if the preposition is spelled out in the higher copy its complement must be spelled out in the higher copy as well. And if the preposition is spelled out in the lower copy then its complement must as well. This is so because the preposition, so we assume, checks the Case of its DP argument. If the preposition were to be spelled out in the higher copy and its argument in the lower copy, as in (53b), Case-checking would be impossible due to the deletion of the Case feature of the preposition in the lower copy. In this case, the offensive Case-feature of the DP argument of the preposition remains unchecked and causes the derivation to crash. That it is actually Case-checking that is relevant for distinguishing between the grammatical (53a) and the ungrammatical (53b) can be shown with the following interesting contrast. In German, there are essentially three ways of expressing the possession relation between two DPs. a) The postnominal Possessor can be marked with genitive Case as in (56a). b) The postnominal Possessor, typically a name, is marked with the genitive s-morpheme, as in (56b). Or c) The possessive preposition von is inserted between the head noun and the postnominal Possessor, as in (56c). It is interesting to note that only option c) allows for "extraposition" of the possessor DP, as is shown in (57).
320
ROLAND HINTERHOLZL (56)
a.
wei! Hans das Buch der Maria
gelesen hat
since Hans the book [the Maria] .GEN read ' since Hans has read Maria's book' b.
wei! Hans das Buch Marias gelesen hat since Hans the book Maria's read
c.
has
has
wei! Hans das Buch von der Maria
gelesen hat
since Hans the book of [the Maria] .GEN read (57)
a.
has
*weil Hans das Buch gelesen hat der Maria
since b. *wei! since c. weil since
Hans the book read
has [the Maria] .GEN
Hans das Buch gelesen hat Marias Hans the book read
has Maria's
Hans das Buch gelesen hat von der Maria Hans the book read
has of [the Maria] .GEN
The difference between (57a-b) on the one hand and (57c) on the other hand follows if we assume that in (56a-b)/(57a-b), the Case of the possessor is checked in an Agreement position by the head noun Buch in the case at hand, while in (56c/57c) the inserted preposition von checks the Case of the post nominal possessor. 3 .2.2 Prosodic constraints on partial deletion Partial deletion creates discontinuous relations between (parts of) constituents that seem to be subject to prosodic constraints. In (58a), the PP von Chomsky must be construed with the direct object ein Buch, yielding the reading the student is reading a book by Chomsky. The extraposed PP in (58a) cannot be construed with the subject DP yielding the reading the student of Chomsky 's is reading a book. However, if, as in (58b) the object is scrambled across the subject, the extraposed PP must be construed with the subject and cannot be construed with the direct object, allowing only the reading a student of Chomsky 's is reading the book. (58)
a.
wei! die Studentin ein Buch liest von Chomsky
b.
wei! das Buch eine Studentin liest von Chomsky
since the student since the book a
a
book reads of Chomsky
student
reads of Chomsky
It is clear that the contrast in (58) cannot be easily reduced to a syntactic distinc tion. If one only looks at the contrast in (58a), one may conclude that "extra position" is possible from objects but not from subjects. However, (58b) immediately shows that the restriction at hand cannot be reduced to an instance of the famous subject-object asymmetry.
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
321
Rather what is at stake in (58) is a notion of prosodic distance between the elements of a discontinuous relation. The backward deleted PP in (58) is construed with the DP that is contained within a certain restricted domain of the preceding clause. From (58) it may be concluded that the backward deleted PP is simply construed with the closest DP. That this is not the case is shown by the contrast in (59). In both (59a) and (59b), the direct object is the closest "antecedent" for interpreting the backward deleted PP, but (59b) is considerably worse than (59a). The difference between (59a-b) is that in (59a), the antecedent is contained in the immediately preceding phonological phrase, whereas in (59b), another phonological phrase - constituted by the adjunct - intervenes between the backward deleted PP and its "antecedent". This is illustrated in (60) where constituents belonging to the same phonological phrase are given in round brackets. These observations were first made by Truckenbrodt ( 1 995). The phonological constraint at work in (59) can thus be stated as given in (6 1 ) (adopted from Truckenbrodt 1 995). (59)
a.
wei! der Peter nach Zanger UberZegung das Buch gekauft since the Peter after long
hat I aber Chomsky has
(60)
(6 1 )
consideration the book bought
about Chomsky
b. ??wei! der Peter das Buch nach Zanger UberZegung gekauft hat I aber Chomsky (wei! der Peter) (nach Zanger UberZegung) (das Buch) (gekauft a. hat aber Chomsky) (wei! der Peter) (das Buch) (nach Zanger UberZegung) (gekauft b. hat aber Chomsky) The antecedent o f a backward deleted phrase must b e contained in the immediately preceding phonological phrase
While we agree with most of Truckenbrodt's observations, we do not agree with the conclusion he draws from them, namely, that extraposition is the result of a PF-movement rule (cf. also Rochemont 1 978 ; Chomsky 1 986). Such an approach is untenable in the light of evidence that extraposition has syntactic consequenc es. For instance, as pointed out already by Gueron ( 1 980), extraposition can affect the licensing of (N)Pls, as is illustrated in (62). (62)
a. * [The names of � of these composers] weren't called out b. The names weren't called out [of � of these composers]
322
ROLAND HINTERHOLZL
In our account, "extraposition" is the result of the partial deletion of phonological features, lO guided by checking of the formal features, in the copies created by syntactic movement. Since Spell-out, the deletion or insertion of phonological features, is the last syntactic operation, its result is visible both at PF and at LF. Thus it is to be expected that partial deletion, being an operation of Spell-out, is subject both to prosodic constraints, as sketched above, as well as to interpretive constraints that, as Gueron ( 1 9 80) and Gueron & May ( 19 84) have argued, apply at LF. Syntactic movement is usually not subject to prosodic locality conditions but obeys constraints like the Specified Subject condition, while "extraposition" is subject to the former type of conditions but violates the latter type of conditions. Since extraposition does not display the typical properties of syntactic movement operations and since its effects are visible at LF, we conclude that extraposition must be treated as the result of the Spell-out operation of partial deletion. To summarize, we have provided empirical evidence showing that VP-inter nal elements undergo licensing movement into the middle field and sketched a general mechanism for stranding material in the VP, in terms of partial deletion.
4.
VP-preposing in English
In this section, we will outline some consequences of the account we have argued for in the previous sections for the proper analysis of English. More specifically, we will address the question of whether verb-preposing in English (Kayne 1998; cf. also Hr6arsd6ttir, this volume, for arguments for VP-preposing in Icelandic) made necessary by this account, as we will argue, is to be charac terized as head-movement or XP-movement. Haider (this volume) points out that the stranding approach to Extraposition leads to impossible word orders. If the relative clause were simply stranded in the base position of the DPs that have undergone passive movement, then the sentences in (67) should be grammatical. (67)
a. b. c. d. e.
*A man was awarded [that noone knew] the prize. (DP) *A book was put [that no one knew] on the shelf. (PP) *The barn was painted [that Mary liked so much] red. (AP) * A man came [who noone knew] in. (particle) [came [[a man who . . . ] in]]
A tentative solution to this problem would be to assume that arguments (67a), directional PPs (67b), small clause predicates (67c) and particles (cf. (67d) as
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
323
derived from (67e)), like in West Germanic, move into their licensing positions in the middle field. However, this immediately creates a new problem concerning the position of the verb in English. The English verb too must move into the middle field. Where does the English verb move to? Movement of the English verb must occur for a very general reason, since it occurs in every clause. The most plausible assumption is thus that the English verb (phrase) moves to the local Tense-head. Assuming verb (phrase) movement to Tense has, of course, far-reaching consequences, of which we can only sketch the most important ones in the remainder of this paper. The first problem with the assumption that the English verb moves to Tense concerns the position of Negation. It is standardly assumed that negation, which precedes the main verb in English, occurs between Tense and the verb. Given that we now have to assume that the English verb moves to Tense, negation must occur higher in the tree, presumably between Tense and AgrS . The second problem concerns auxiliaries. If auxiliaries are base-generated in functional projections outside of the VP, including Tense, then we get the wrong word order whether we assume that the verb undergoes head- or XP movement to Tense, as is illustrated in (68). (68)
a. [TP [T' has [AspP been [vp seen]]]] b. * [TP [T' seeI\ has [AspP been [vp ti]]]] V-movement c. * [TP [vp seen] [T' has [AspP been tvp]]] VP-movement
The right word order can only be derived if we assume that auxiliaries are part of the VP. Assuming that auxiliaries head their own VP-shells, verb-movement must affect all VP-shells containing either auxiliaries or the main verb. Thus verb movement in English must be XP movement given these assumptions. The third problem concerns the infinitival marker, which is standardly assumed to be base-generated in Tense in English. Since VP-movement into SpecTP would derive the wrong word order, we would have to assume that the infinitival marker is base-generated somewhere else. One possibility is that the infinitival marker, like in the West Germanic languages, occupies F l . VP-preposing in English would then be akin to movement of F1P that occurs in restructuring contexts in West Germanic (cf. HinterhOlzl 1 999). Another possibility is to assume that the infinitival marker is a prepositional complementizer. If the whole VP undergoes verb-movement in English, then we expect that PPs and clauses that have been stranded in the VP precede the other arguments of the verb, contrary to fact, as is illustrated in (69).
324
ROLAND HINTERHOLZL (69)
a. *10hn [yp bought about Chomsky] a book b. *10hn [yp bought that Peter liked] the book
However, cases like (69) can be excluded if we adopt partial deletion as a mechanism of stranding, since they violate an important constraint on backward deletion (BWD). Wilder ( 1 995) shows with independent evidence from deletion within coordinate structures that a BWD-site must precede its antecedent. To simplify things, this constraint has the effect that BWD-material can only surface in the rightmost copy. In (69), backward deleted material about Chomsky and that Peter liked surfaces in the leftmost copy.
Notes 1.
Haegeman (1 995) also argues that if non-finite verbs cannot move to Fl, as we have assumed, then one can derive the fact that en can only be spelled out on a finite verb.
2.
Note that in German, even non-finite auxiliaries have to invert with IPP complements (i), while the opposite holds in West Flemish (ii). It follows that in West Flemish, due to the inability of non-finite verbs to move to Fl, the empty morpheme may be licensed at LF. Else wird ihm einen Brief haben schreiben wollen (i) a. Else will him a letter have write want(IPp) *Else wird ihm einen Brief schreiben wollen haben b. Else will him a letter write want(IPp) have 'Else will have wanted to write him a letter' (ii) a. dan-ze kosten willen dienen boek kuopen een that-they could want(IPp) that boek buy have b. *dan-ze kosten een willen dienen boek kuopen that-they could have want(IPp) that boek buy 'that they could have wanted to buy that book'
3.
Often, as in (1 8b) the order manner adverb < nominal argument yields a perfect sentence. This is always then the case when the manner adverb is eligible for an alternative interpretation. So, for instance, (1 8b) is perfect under the interpretation "it was careful of Hans to read the book", where the adverb is interpreted as sentential rather than as a VP-adverb. Also (19b) is perfect under the interpretation "Hans executed exactly one/this plan", where the adverb is construed as modifying the DP.
4.
(2lb) is perfect if the negative marker is interpreted as constituent negation (see (22) and (23) below).
5.
The determiner kein has been analyzed as created by fusing a determiner with existential force with negation (cf. Kratzer 1 989).
6.
PP-extraposition has properties that differ considerably from the properties of cases of PP-out of-NP. These issues are discussed in detail in Brugger & Hinterh6lz1 (1998).
7.
The terms forward deletion (FWD) and backward deletion (BWD) are loaned from the literature on coordination (cf. Wilder 1994). The operation of FWD targets left-peripheral material in the first conjunct and deletes this material under identity in all conjuncts following. The operation
MOVEMENT AND STRANDING IN GERMANIC OV LANGUAGES
325
of BWD, in contrast, targets right-peripheral material in the last conjunct and deletes this material under identity in all preceding conjuncts. 8.
In the German DP, determiner, noun and adjectives agree in Case, Number and Gender.
9.
It is thus important to distinguish between feature checking and feature deletion. Feature deletion removes a feature from the computation. Feature checking, so to say, only removes the offensive character of a feature for the rest of the computation (cf. Nunes 1995). Secondly, feature deletion is total, that is to say, it deletes all formal and phonological features. To allow for reconstruction, we have to assume that semantic features are not affected by this deletion operation.
10.
Whether Spell-out is to be described as deletion or insertion of phonological features is immaterial for our purposes here. However, since, in our account of the IPP effect, we treated Spell-out as the insertion of phonological features into a syntactic tree, we have to assume that phonological features are inserted into the one copy in which all strong formal features have been checked.
References Bech, Gunnar. 1 955/1983. Studien Uber das Deutsche Verbum Infinitum. Tlibingen: Niemeyer. Brody, Michael. 1 990. "Some remarks on the focus field in Hungarian." UeL Working Papers in Linguistics 2, University College London. Brugger, Gerhard and Hinterh61zl, Roland 1998. PP-extraposition: Cases of complex inversion and cases of partial deletion. Ms., USC & UCLA. Brugger, Gerhard and Poletto, Cecelia. 1 993 . Negation in Bavarian. Ms. , University of Venice. Chomsky, Noam. 1 986. Barriers. Cambridge, MA: MIT Press. Chomsky, Noam. 1 993 . "A minimalist program for linguistic theory." In Ken Hale and Jay Keyser (eds.), The View from Building 20. Cambridge, MA: MIT Press. 1-52. Chomsky, Noam. 1 995. "Bare phrase structure." In Gert Webelhuth (ed.), Government and Binding Theory and the Minimalist Program. Cambridge, MA: Blackwell. 383-439. Cinque, Guglielmo. 1 993. "A null theory of phrase and compound stress." Linguistic Inquiry 24: 239-298. Diesing, Molly. 1 992. Indefinites. Cambridge, MA: MIT Press [Linguistic Inquiry Monograph 20]. Donaldson, Bruce C. 1 993. A Grammar of Afrikaans. Berlin & New York: Mouton de Gruyter. Grimm, Jakob. 1 898. Deutsche Grammatik. 1 969 edition, Gottingen: Glitersloh. Gueron, Jacqueline. 1 980. "On the syntax and semantics of PP extraposition." Linguistic Inquiry 1 1 : 637-678. Gueron, Jacqueline and Robert May. 1 984. "Extraposition and Logical Form." Linguistic Inquiry 15: 1-3 1 .
326
ROLAND HINTERHOLZL
Haegeman, Liliane. 1 995. Verb projection raising and finite V movement in West Flemish. Ms., University of Geneva. Haiden, Martin. 1 995. On the Semantic Evaluation ofNP-Syntax. Diplomarbeit, University of Vienna. Haider, Hubert. 1 993. Deutsche Syntax, generativ. Vorstudien zur Theone einer projektiven Grammatik. Tubingen: Narr. Haider, Hubert. 1 995. "Downright down to the right." In Uli Lutz & Jtirgen Pafel (eds), On Extraction and Extraposition in German. Amsterdam and Philadelphia: John Benjamins [Linguistics Today 1 1 : 245-27 1]. Halle, Morris and Marantz, Alec. 1 993. "Distributed morphology and the pieces of inflection." In Ken Hale & Jay Keyser (eds), The Viewfrom Building 20. Cambridge, MA: MIT Press. 1 1 1-176. Hinterholzl, Roland. 1 997. CP-licensing, stranded relatives and the nesting requirement on multiple extraposed clauses. Ms., University of Southern California. Hinterholzl, Roland. 1 999. Restructuring Infinitives and the Theory of Complementation. Ph.D. dissertation, University of Southern California. Kayne, Richard S. 1 994. The Antisymmetry of Syntax. Cambridge, Ma. : MIT Press [Linguistic Inquiry Monograph 25] . Kayne, Richard S. 1 998. "Overt versus covert movement." Syntax 1 : 1 28-19 1 . Koster, Jan, 1 994. "Predication incorporation and the word order of Dutch." In Cinque et al. Kratzer, Angelika. 1 989. "Stage-level and individual-level predicates." In Gregory N. Carlson & Francis Jeffry Pelletier (eds), The Generic Book. Chicago. 125-175. Muller, Gereon. 1 995. "On extraposition and successive cyclicity." In UIi Lutz & Jtirgen Pafel (eds), On Extraction and Extraposition in German. Amsterdam and Philadel phia: John Benjamins. [Linguistics Today 1 1 : 2 13-243]. Neeleman, Ad. 1 994. Complex Predicates. Ph.D. dissertation, University of UtrechtiOTS. Nunes, Jairo. 1 995. The Copy Theory of Movement and Linearization of Chains in the Minimalist Program. Ph.D. dissertation, University of Maryland. Rochemont, Michael. 1 978. A Theory of Stylistic Rules in English. NY: Garland. Rutten, Jean, 1 99 1 . Infinitival Complements and Auxiliaries. Ph.D. dissertation., Universi ty of Amsterdam. Truckenbrodt, Hubert. 1 995. Phonological Phrases. Ph.D. dissertation, MIT. Wilder, Christopher. 1 994. "Coordination, ATP and ellipsis." Groninger Arbeiten zur Germanistischen Linguistik 37: 291-329. Wilder, Christopher. 1 995. "Rightward movement as leftward deletion." In Uli Lutz and Jtirgen Pafel (eds), On Extraction and Extraposition in German. Amsterdam and Philadelphia: John Benjamins. [Linguistics Today 1 1 : 273-309] . Zubizarreta, Maria Luisa. 1 998. Prosody, Focus and Word Order. Cambridge, MA: MIT Press [Linguistic Inquiry Monograph 33] Zwart, C. Jan-Wouter. 1 993. Dutch Syntax. Ph.D. dissertation, University of Groningen.
Two Types
Languages
Matthew Pearson
Abstract In this paper I argue that VO languages do not form a homogeneous class, as distinct from OV languages. Rather, VO languages fall into two separate types, direct and inverse, with respect to the order of objects and adverbs within the predicate phrase. In direct VO languages, the order of these elements is the same as in OV languages, while in inverse VO languages, they occur in the mirror image of the OV order. I argue that the differences between direct and inverse languages reduce to a difference in how the lexical verb V raises up to substitute into the heads of higher VP shells. In direct languages, this is accomplished by means of successive head movement, while in inverse languages, head movement is unavailable and successive XP-movement is employed instead. In addition to explaining the word order symmetries between the two language types, this analysis also accounts for a number of other facts, including some unexpected differences between English (a direct VO language) and Malagasy (an inverse VO language) with respect to binding and weak crossover effects in double object constructions.
1.
Introduction
In discussions of word order variation, it is generally assumed that the most basic typological split is between languages with O V order, such as Turkish, and those with VO order, such as English. Traditionally, OV and VO languages have been taken to have distinct base structures, reflecting different settings of a fundamen tal 'headedness' parameter (perhaps supplemented by additional parameters relating to directionality of Case- and 6-assignment, as in Travis 1 9 89). More recently, a number of researchers have suggested that OV and VO languages have the same base structure, but differ in terms of movement. For
MATTHEW PEARSON
328
example, Kayne ( 1 994) has suggested that all languages are undedyingly VO, and that OV languages are derived by leftward movement of the object to a specifier position above the highest position of the verb. Some support for this view has been provided by Zwart ( 1 993), Carstens ( 1 997), Hinterh61zl ( 1 997), Koopman and Szabolcsi ( 1 998), Nakajima ( 1 999), and others, who draw on data from a variety of OV languages, including Dutch, German, Japanese, and the West African language Ijo. Taking a different approach, Fukui and Takano ( 1 998) and Haider ( 1 994, this volume), have argued that all languages are undedyingly OV, and that VO languages are derived by leftward movement of the verb over the object. However, these movement-based approaches to word order resemble directionality-based approaches insofar as OV and VO are treated as fundamental word order types. Among languages of a given type, it is commonly assumed that cross-linguistic differences are relatively superficial. VO languages, for example, are generally taken to differ from each other primarily in the position of the verb at spell-out (low in English, somewhat higher in French, higher still in Irish). In this paper I argue for a different view, based on data from languages not considered by the researchers mentioned above. In particular, I claim that VO languages do not form a homogeneous class, but rather fall into two separate types. With respect to a variety of word order phenomena to be considered below, VO languages of the first type, including familiar languages like English and French, exhibit the same word orders as those found in OV languages. Languages of the second VO type, among them the Austronesian language Malagasy, exhibit word orders which are the mirror image of their counterparts in OV languages. In light of this, I will argue that VO languages such as English and French should be grouped together with OV languages like Turkish to form a single word order class, which I will call direct languages. VO languages of the Malagasy type, on the other hand, constitute a separate class which I refer to as inverse languages. This alternative typology is shown schematically in ( 1 ) : (1)
�
direct
�
OV
VO
inverse
I
VO
The structure of this paper is as follows: In Section 2, I present data illustrating the word order differences between direct and inverse languages. In Section 3, I argue against a directionality style analysis, whereby inverse languages are taken to be the literal mirror image of OV languages, and in favor of an
TWO TYPES OF VO LANGUAGES
329
' antisymmetric' analysis, according to which direct and inverse languages differ in terms of the movements involved in constructing a multi-layered VP-what could be thought of as the 'extended projection' of the lexical verb, in a sense (somewhat different from Grimshaw 1 99 1 ) to be made precise below. In Section 4, I apply this movement-based analysis to the constructions discussed in Section 2. Finally, in Section 5, I summarize my analysis.
2.
The data
The phenomena to be considered here involve the order of elements within what could loosely be called the predicate phrase (roughly equivalent to Chomsky' s vP). A s I show in 2. 1-2.3, direct and inverse languages differ with respect to the order of objects in double object constructions, the relative order of certain classes of adverbs, and the (apparent) direction of movement in object shift constructions. 2.1 Double object constructions As Haider ( 1 994) observes, va languages such as English pattern with OV languages such as Dutch with respect to the order of objects in double object 1 constructions. In the unmarked case, the indirect object precedes the direct object in both types of languages, as shown in (2): (2)
a. b.
John will give Mary the book. (dat) lan Mane het boek zal geven. that Jan Marie the book will give ' (that) Jan will give Marie the book. '
(English) (Dutch)
In va languages of the Malagasy type, however, double object constructions exhibit the mirror image of this order: As shown in (3a), the direct object precedes the indirect object in this language. (3b) and (3c) give examples from Palauan (from Polinsky 1 997, who cites Georgopoulos 1 99 1 ) and Quiavini Zapotec, showing that this pattern is not unique to Malagasy: (3)
a.
Nanolotra ny dite ny vahiny ny zazavavy.
b.
PAsT.offer DET tea DET guest DET girl 'The girl offered the guests the tea. ' Ng-mils-terir a buu ' a rngalek a Sabino. 3sG-give-3PL DET betel DET children DET Sabino 'Sabino gave the children some betelnut. '
(Malagasy)
(Palauan)
MATTHEW PEARSON
330 c.
Bdeidy
Gyeeihlly liiebr studiann. PERF.give Miguel book student 'Miguel gave the student the book. '
(Q. Zapotec)
The contrast in (2)-(3) is also attested with 'derived' double object predicates viz. dative and benefactive applicative constructions. (4a) below illustrates a typical applicative construction from Indonesian, a VO language of the English type. Here the indirect (benefactive) object precedes the direct object. This order is replicated in applicative constructions in the OV language Dutch, as shown in 2 (4b). However in Tzotzil, a VO language of the Malagasy type, the indirect (dative) object follows the direct object, as shown in (5) (Haviland 1 9 8 1 ) : (4)
a.
Ali membuka-kan bapak pintu.
b.
Ali open-APPL father door 'Ali opened the door for his father. ' (dat) Jan zijn vrienden uitnodigingen ver-stuurde. that Jan his friends invitations APPL-sent ' (that) Jan sent his friends invitations. '
tak'in Ii Petul e. INcPL- 1 s.ERG-give-APPL money DET Peter PART 'I gave Peter the money. '
(5)
Ch-k-ak' -be
(Indonesian)
(Dutch)
(Tzotzil)
2.2 Adverb order Using data from various languages, Cinque ( 1 999) argues that adverbs can be grouped into a number of classes, each associated with a different functional projection, and that the order in which these projections occur in the tree follows a universal hierarchy. In VO languages such as French, this hierarchy is directly reflected in the surface left-to-right order of adverbs. For example, frequency adverbs such as ' always' precede manner adverbs such as 'well' , over which they take scope, as in (6a). The same order is found in the OV language Dutch, as shown in (6b): (6)
a.
b.
Jean lave
toujours bien ses vetements. Jean washes always well his clothes 'Jean always washes his clothes well.' (dat) Jan altijd zijn kleren goed wast. that Jan always his clothes well washes ' (that) Jan always washes his clothes well. '
(French)
(Dutch)
In Malagasy, however, the relative order of adverbs is reversed, such that
TWO TYPES OF VO LANGUAGES
331
adverbs with wider scope appear t o the right o f adverbs with narrower scope. As shown in (7), the frequency adverb 'always' follows the manner adverb 'well' (Rackowski 1 998; Pearson 1998b; Rackowski and Travis 1 999). Comparing (7) with (6b) above shows that here, as with double object constructions, the Malagasy order is the mirror image of the Dutch. (7)
Manasa tsara ny lambany Joana i
Ketaka.
(Malagasy)
wash well DET clothes.3 always DET Ketaka ' Ketaka always washes his clothes well. ' 2 . 3 Object shift VO languages of the Malagasy sort also mirror OV languages in terms of the (apparent) direction of object shift. Consider the following sentences from Turkish: Here we see that the non-specific direct object in (8a) appears left adjacent to the verb, following the adverb 'immediately' , whereas the specific direct object, marked with morphological accusative case, appears in a displaced position to the left of the adverb (Murat Kural, p.c.; examples based on En9 1 99 1 ): (8)
a.
b.
Ali hemen
bir piyano kiralamak istiyor. Ali immediately one piano rent.INF wants 'Ali wants to rent a piano immediately. ' Ali bir piyanoyu hemen kiralamak istiyor. Ali one piano.Acc immediately rent.INF wants 'Ali wants to rent a (specific) piano immediately.'
(Turkish)
In Malagasy, we find the opposite order: In (9a), the non-specific direct object occurs right-adjacent to the verb, preceding the adverb 'quickly' , while the specific direct object in (9b), marked with the determiner ny, appears in a scram bled position following the adverb (Polinsky 1 994; Pearson 1 998b): (9)
a.
Nijinja vary haingana ny mpamboly.
b.
PAST.cut rice quickly DET farmer 'The farmer harvested rice quickly. ' Nijinja haingana ny vary ny mpamboly. PAST .cut quickly DET rice DET farmer 'The farmer harvested the rice quickly. '
(Malagasy)
VO languages of the English-French type either lack object shift altogether, or show leftward shift of the kind found in OV languages (e.g. object shift in Icelandic; see Bobaljik & Jonas 1 996 and references therein).
332
MATTHEW PEARSON
2.4 Direct and inverse order languages Examining the above data, we see that VO languages can be separated into two groups with respect to the order of objects and predicate-internal adverbs. VO languages of the first group, such as English and French, exhibit the same orders as one finds in OV languages like Dutch and Turkish. In this group, the surface linear order of objects and adverbs appears to correlate directly with their hierarchical relations : In double object constructions, where the indirect object c-commands the direct object (Barss and Lasnik 1 9 86 ; Larson 1 98 8), it also precedes the direct object. Similarly, frequency adverbs like 'always' precede manner adverbs like 'well' , over which they take scope, while shifted objects both precede and c-command their traces. To reflect this correlation between c-command and left-to-right order, I group VO languages of the English-French type together with OV languages like Turkish, and refer to them as direct order predicate languages (or direct languages, for short) . VO languages of the second group, such as Malagasy, exhibit orders which are the mirror image of what one finds in direct languages. Here the surface left to-right order of objects and adverbs appears to be an inverse mapping of their c-command relations. I will thus refer to languages of the Malagasy type as inverse order predicate (or simply inverse) languages. Examples of direct and inverse languages are listed in ( 10), while the table in (i) summarizes the data from 2. 1-2.3 in schematic form. ( 1 0)
Direct: Inverse:
(i)
Double obj. constructions
English, French, Indonesian, Dutch, Turkish, . . . Malagasy, Tzotzil, Q. Zapotec, Palauan, . . .
direct OV
VO
VO
10 DO V
V IO DO
V DO IO
V Adv2 Advl
V Advl Adv2
Adv order (Adv2 » Advl) Adv2 Advl V
inverse
Object shift
3.
Symmetry versus anti-symmetry
3.1
The symmetry hypothesis
What sort of analysis should we give for inverse languages? One obvious approach would be to claim that such languages have a base structure which is
333
TWO TYPES OF VO LANGUAGES
literally the mirror image of the structure found in OV languages. If we adopt the traditional assumption that phrase markers in OV languages have the order specifier-complement-head, as in ( l l a), then we might suppose that phrase markers in inverse languages have the opposite order, head-complement specifier, as in ( l Ib). I will refer to this as the symmetry hypothesis. ( l 1)
a.
b.
CP
�
Spec
C'
�
IP
�
Spec
I'
�
VP
�
Spec
C
I
V'
�
V
CP
�
Spec
C'
�
C
IP
�
Spec
I'
�
I
VP
�
V'
�
Spec
V
Though the symmetry hypothesis appears to offer a straightforward explanation of the word order symmetries discussed above, there are reasons to believe that it is on the wrong track. For example, notice that in ( l Ib) the specifier of IP occurs to the right of VP. Given this, we predict that the subject should follow the direct object in all inverse languages. However, while object-subject order is found in some inverse languages, such as Tzotzil, there are others where the opposite order is required. Quiavini Zapotec, for example, is strictly VSO, as 3 illustrated in ( 1 2): ( 1 2)
Ytaa 'z
Gyeeihlly Lieeb.
(Q. Zapotec)
IRR.beat Miguel Felipe 'Miguel will beat Felipe. ' * 'Felipe will beat Miguel. ' Furthermore, since the specifier of CP occurs to the right of IP in ( l l b), we would expect wh-movement in inverse languages to target the right periphery of the clause. However, in every inverse language that I am familiar with, wh-phrases move to the left periphery, just as in many direct VO languages like English. Examples from Tzotzil and Quiavini Zapotec are given in ( 1 3):
334
MATTHEW PEARSON ( 1 3)
a.
b.
Buch'u i-s-mil Ii Petul e ? who INCPL-3sG.ERG-kill DET Peter PART 'Who killed Peter?' (or 'Who did Peter kill?') Xi rralloh lluuw gauw Gyeeihlly ? what think you FUT.eat Miguel 'What do you think Miguel will eat?'
(Tzotzil)
(Q. Zapotec)
These facts suggest that the specifier of CP and (in at least some inverse languages) the specifier of IP are generated to the left of their heads rather than the right, in which case we must modify the mirror-image structure in ( l Ib), restricting the occurrence of right-specifiers to projections below IP, as in ( 14). Not only does this modification obscure the apparent symmetry between inverse languages and OV languages, but it forces us to ask why inverse ordering should be confined to constituents within VP, a question to which no ready answer suggests itself. (14)
CP
�
Spec
C'
�
C
IP
�
Spec
I'
�
I
VP
�
V'
�
Spec
V
However, even if we accept the modified structure in ( 14), we still encounter empirical problems with the symmetry hypothesis, related to the syntax of double object constructions in languages like Malagasy. Larson ( 1 988) argues that the indirect object in a double object construction occupies a VP-internal specifier position, from which it asymmetrically c-commands the direct object, as schema tized in ( 1 5) (position of the verb irrelevant) :
TWO TYPES OF VO LANGUAGES ( 1 5)
335
VP
�
10
V'
� O D
V
Larson develops this structure in part to explain certain asymmetries between the two objects first noted by B arss and Lasnik ( 1986). For example, as shown in ( 1 6), a quantified indirect object may bind a pronoun within the direct object, but not vice versa. Assuming that a quantifier must c-command a pronoun in order to bind it, these data suggest that the indirect object asymmetrically c-commands the direct object. ( 1 6)
a. John gave every workerj hisj paycheck. b. *John gave itsj owner every paychec�.
Furthermore, Barss and Lasnik: show that A'-movement of a direct object over an indirect object containing a pronoun yields a weak crossover effect, as in (17). Since weak crossover effects only result when the constituent containing the pronoun c-commands the trace of the A'-operator, ( 1 7) provides further evidence that the indirect object c-commands the direct object in double object constructions.4 ( 1 7)
*Which childj did you show hisj mother tj?
If VPs in inverse languages were the mirror image of VPs in OV languages, then double object constructions in a language like Malagasy would have the structure in ( 1 8), where the indirect object asymmetrically c-commands the direct object to its left: ( 1 8)
VP
�
V'
�O
V
10
D
However, when we apply B arss and Lasnik's c-command tests to double object constructions in Malagasy, the results are somewhat surprising. When it comes to pronoun binding, Malagasy behaves just like English, modulo the difference in linear order: A quantified indirect object may bind a pronoun (here, the third person enclitic -ny) within the direct object that precedes it ( 1 9a), but not vice versa ( 1 9b). This is just what we would expect given the structure in ( 1 8).
MATTHEW PEARSON
336 ( 1 9) a.
Nanaseho ny sarinYi
ny ankizy rehetrai Rasoa. (Malagazy)
PAST .show DET picture.3 DET child every Rasoa 'Rasoa showed every chil� his/heri picture. ' b.
*Nanaseho ny ankizy rehetrai ny
reninYi
Rasoa.
PAST.show DET child every DET mother.3 Rasoa 'Rasoa showed his/heri mother every childi.' However, when we look at weak crossover, we get apparently contradictory results. To see this, we must first have a look at some properties of Malagasy phrase structure. As Keenan ( 1 994) and Pearson ( 1 998a, in preparation) show, clauses in Malagasy have the bipartite structure in (20a): Each clause is com posed of two maj or constituents, a verb-initial predicate phrase, and a predicate external constituent which I will loosely call the topic. Frequently the subject raises to the topic position, as shown in (20b) , resulting in the VOS order for which Malagasy is well known (d. footnote 3). However, other arguments and adjuncts may also freely topicalize. Raising of the object, for example, results in the VSO order in (20c): (20) a. b. c.
[PredP V S [PredP V ti [PredP V S
o o ti
(PP)] (PP)] (PP)]
There is evidence from anaphor binding to suggest that the subject asymmetrical ly c-commands the direct object when both are within the predicate phrase. Consider the sentence in (2 1 ) (adapted from Keenan 1 993), where a benefactive argument has been topicalized: As this example shows, a predicate-internal subject may antecede a predicate-internal reflexive object: (2 1 ) [Namonoan ' ny vehivavYi tenaJ ny zanany. PAST. kill DET woman self DET child.3 'The woman killed herself for her child. '
(Malagasy)
In Pearson (in preparation), I argue that topicalization in Malagasy involves movement to an A'-position (contra Guilfoyle et al. 1 992). Evidence for this comes from the fact that when a quantified direct object is topicalized over a subject containing a pronoun, as in (22), a weak crossover effect results. 5
,
(22) * [Novangian ny reninYi ti tamin 'ny zoma] ny ankizy rehetrai. PAST.visit DET mother.3 PAST.on-DET Friday DET child every 'Every chil�, hisi mother visited ti on Friday. ' Turning to double object constructions, we might predict that if the indirect
TWO TYPES OF VO LANGUAGES
337
object c-commands the direct object, as in ( 1 8), then topicalizing a quantified direct object over an indirect object containing a pronoun should yield a weak crossover effect. However, this is not what we find. As (23) shows, no weak crossover effect results : A topicalized direct object may freely bind a pronoun within the indirect object. Thus it seems as though the indirect object asymmetri cally c-commands the direct object for purposes of pronoun binding (cf. (19a-b)), but not for purposes of weak crossover. However, the symmetry hypothesis, as exemplified by the structure in ( 1 8), fails to account for this apparent contradiction. (23)
[Nasehon-dRasoa ti ny reninYi] ny ankizy rehetrai. PAsT.show-Rasoa DET mother.3 DET child every 'Every childi, Rasoa showed hisi mother ti.'
In short, we see that the symmetry hypothesis is inadequate, insofar as it fails to explain why the symmetry between direct and inverse languages is confined to elements within the VP, and fails to predict the differences between English and Malagasy double object constructions with respect to B arss and Lasnik's pronoun binding and weak crossover tests. I now present an alternative to the symmetry hypothesis which does account for these facts. 3 .2 Word order and v- V complex formation Based on various typological considerations - e.g., the cross-linguistic absence of wh-movement to the right periphery - Kayne ( 1994) concludes that all languages order specifiers to the left of heads, a result which is predicted to follow from his linear correspondence axiom (LeA) . Here I will argue for an analysis of inverse languages which is consistent with this position: Rather than appealing to some directionality parameter whereby specifiers may be ordered before or after heads, I will instead assume that phrase markers in all languages conform to a specifier-head-complement order, and account for the symmetry effects discussed above in terms of movement. The basic analysis is as follows: Recall that the kinds of mirror-image effects we find in inverse languages appear to be confined to elements which are arguably within the VP domain of the clause, such as direct and indirect objects, manner adverbs, and so on. When it comes to elements which are arguably higher in the tree, such as subjects and wh-operators, no mirror image effects are observed. Here I will argue that this is non-accidental - that the crucial difference between direct and inverse languages has to do with how verb phrases are put together.
MATTHEW PEARSON
338
Following work of Travis ( 1 99 1) , Sportiche ( 1999), and others, I assume that verb phrases have the basic architecture shown in (24). Rather than consist ing of a single projection, a verb phrase involves two or more VP ' shells ' (Larson 1 98 8), the lowest o f which i s the lexical projection V P (in which the verb is generated), and the highest of which is vP, headed by the phonologically null light verb v, in the specifier of which the verb's external argument is introduced (Chomsky 1 995). Intervening between these shells is a set of func tional projections, notated FP in (24). These projections are associated with various functions: Some introduce adjuncts and secondary predicates into the structure, others participate in feature-checking (e.g. Case licensing) and trigger agreement on the verb, others provide landing sites for 'short movement' scrambling, and so on. (24)
vP
�
v
FP
�
F
VP
�
V
Following Chomsky ( 1 995), I assume that syntactic objects are constructed in 'bottom-up' fashion through successive applications of Merge and Move. Now suppose that, as the structure in (24) is built up, the head of VP must combine with the head of vP (along with the heads of any VP shells intervening between them) to form a single complex predicate head. In order to do this, the lexical verb must raise and substitute into the head of vP, thereby licensing the light verb by providing it with lexical material. I will refer to this process of V-to-v raising and substitution as v-V complex formation. Suppose further that, in order for v-V complex formation to take place, any functional proj ections intervening between the verb and the v head must in some sense be 'invisible' for purposes of substitution. I formulate this idea as a constraint on substitution, given in (25iii): (25)
A head X may substitute for a head Y iff: (i) X and Y are lexical heads with non-conflicting categorial features, (ii) Y is not associated with any phonological features, and (iii) for all heads Z, if Z c-commands X but does not c-command Y, then the projections of Z must be categorially non-distinct from X and Y.
TWO TYPES OF VO LANGUAGES
339
How does a functional projection become categorially non-distinct from the verb? Here I will assume that certain functional projections can acquire a lexical feature from the verb, in effect becoming VPs 'by inheritance' .6 This inheritance takes place just in case a constituent containing the verb's lexical feature raises into the checking domain of the functional projection at some point in the derivation. This idea is formulated in (26a-b): (26)
a. b.
A functional projection FP is categorially non-distinct from a lexical head L iff it has inherited a lexical feature from L. A functional projection FP inherits a feature