Remnant movement 9781614517467


196 39 2MB

English Pages 322 pages : [320] Year 2015

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

Remnant movement
 9781614517467

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Günther Grewendorf (Ed.) Remnant Movement

Studies in Generative Grammar

Editors Norbert Corver Harry van der Hulst Roumyana Pancheva Founding editors Jan Koster Henk van Riemsdijk

Volume 123

Remnant Movement

Edited by Günther Grewendorf

ISBN 978-1-61451-746-7 e-ISBN (PDF) 978-1-61451-633-0 e-ISBN (EPUB) 978-1-5015-0050-3 ISSN 0167-4331 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. 6 2015 Walter de Gruyter, Inc., Berlin/Boston Typesetting: RoyalStandard, Hong Kong Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents I

Introduction Günther Grewendorf Problems of remnant movement

3

II Constraints on remnant movement Yuji Takano Minimality for Merge

35

Gereon Müller Remnant movement in local derivational grammar Chris Collins and Joachim Sabel A C/I-interface constraint on remnant movement

53

93

III The domain of remnant movement Jean-Yves Pollock and Cecilia Poletto Arguing for remnant movement in Romance

135

Carlo Cecchetto and Caterina Donati Please do not move that remnant too much!

179

Edwin Williams Head movement and remnant movement

201

IV Remnant movement and the interfaces Mamoru Saito Remnant movement, radical reconstruction, and binding relations Erich Groat Total Transfer, Dynamic Labeling, and Transfer Remnants Index

321

257

221

Günther Grewendorf

Problems of remnant movement 1 Introduction In this paper I would like to give a brief survey of some of the problems associated with remnant movement and the solutions that have been suggested in the literature (following the lines of Grewendorf 2003). I will start with a comparison of relevant facts in Japanese and German, since these facts have given rise to traditional accounts for the phenomenon of remnant movement.

2 Remnant movement and the Proper Binding Condition (Saito 1992) Japanese shows certain asymmetries in the scrambling of categories which contain the trace of a scrambled element. In the well-formed example (1a), the embedded CP3 “that Mary bought that book” has been adjoined to the AgrsP2 “Bill said . . .”, and the object sono hon-o of the scrambled clause CP3 has undergone long scrambling to the matrix AgrsP1. In this configuration, the trace of the long-scrambled object is bound by its antecedent. In the ungrammatical example (1b), the object of the embedded clause CP2 has undergone long scrambling to the matrix AgrsP1 (“Taroo said . . .”), followed by short scramling of CP2 itself. (1) a.

[ AgrsP1 sono hon-oj [ AgrsP1 John-ga [CP2 [ AgrsP2[CP3 tj’ Mary-ga tj that book-ACC John-NOM Mary-NOM katta to]i [AgrsP2 Bill-ga ti itta]] to] omotteiru]]. bought COMP Bill-NOM said COMP think (That bookj, John thinks that [that Mary bought tj]i Bill said ti.)

b. *[ AgrsP1[CP2 ti’ Hanako-ga ti yonda to]j [ AgrsP1 sono hon-oi Hanako-NOM read COMP that book-ACC [Taroo-ga tj itta]]]. Taroo-NOM said

(Saito 1994a)

4

Günther Grewendorf

In German, the counterparts of the Japanese examples are both ungrammatical. In other words, scrambling of an XP out of a (short-)scrambled category is impossible in German irrespective of whether or not the scrambled XP binds its trace. [tj zu füttern]i keiner ti versuchte (2) a. *dass [den Hund]j zweifellos that the dog-ACC undoubtedly to feed nobody tried ‘that undoubtedly nobody tried to feed the dog’ [den Hund]i keiner tj versuchte b. *dass [ti zu füttern]j zweifellos that to feed undoubtedly the dog-ACC nobody tried (Grewendorf/Sabel 1994) Saito (1992) accounts for the asymmetry observed in Japanese in terms of the Proper Binding Condition (PBC). Since scrambling can be freely undone at LF, PBC is assumed to not only apply at LF but also in the overt syntax. (3)

Proper Binding Condition Traces must be bound (at LF but also in the overt syntax).

There are several empirical problems with an account in terms of PBC. The first problem has to do with the contrast between remnant scrambling and remnant topicalization in German. As can be seen from example (4a), a category XP from which scrambling has taken place cannot be followed by scrambling of the remnant to a position outside of the c-command domain of XP. However, as shown by (4b), topicalization of the remnant to SpecCP is well-formed even though the trace of the scrambled XP is no longer bound. [den Hund]i keiner tj versuchte (4) a. *dass [ti zu füttern]j zweifellos that to feed undoubtedly the dog-ACC nobody tried ‘that nobody undoubtly tried to feed the dog’ tj versucht]]]. b. [ti zu füttern]j [C’ hat [ AgrsP [den Hund]i [ AgrsP keiner to feed has the dog-ACC nobody-NOM tried ‘Nobody has tried to feed the dog.’ (Grewendorf/Sabel 1994) Another problem has been pointed out by Takano (1995). It concerns the combination of A-movement and remnant wh-movement. Given the VP-internal Subject Hypothesis, predicate fronting in (5) illustrates remnant wh-movement with an unbound A-trace (5b), which should be ungrammatical if the PBC holds in overt syntax.

Problems of remnant movement

(5)

a.

How proud of Bill is John?

b.

[CP [Pred ti how proud of Bill]j is [IP Johni tj]]?

5

(Takano 1995:332)

The third problem has to do with subject raising constructions in Japanese. As pointed out by Hiraiwa (2003), in Japanese subject raising constructions, the constituent from which subject raising takes place (v*P) can undergo scrambling. (6) shows that A-movement of the embedded subject Taroo-ga can be followed by scrambling of the remnant v*P. The result is well-formed although PCB is violated in the overt syntax. (6) a.

b.

[TP Taroo-gai [ν*P ti Hanako-wo tataki]-sae si-ta]. Taroo-NOM Hanako-ACC hit even do-PST ‘Taroo even hit Hanako.’ [TP[ν*P ti Hanako-wo tataki]-saej [Taroo-gai tj si-ta]]. Hanako-ACC hit even Taroo-NOM do-PST (Hiraiwa 2003)

3 Remnant movement and the Condition of Unambiguous Domination (Müller 1996, 1998) Rather than relying on the PCB, Müller (1996, 1998) has suggested to account for the ungrammaticality of (1b) and (2b) in terms of what he calls the Condition of Unambiguous Domination. According to this condition, remnant XPs are not allowed to undergo a certain type of movement if the antecedent of the unbound trace has undergone the same type of movement: (7)

Unambiguous Domination In a structure . . .[ A. . .[B. . .]. . .]. . . , A may not undergo α-movement if B has undergone α-movement.

The various types of movement are taken to be scrambling, topicalization, whmovement, A-movement. The Condition of Unambiguous Domination provides Müller with a simple account for the contrast between remnant scrambling and remnant topicalization in German, i.e. the contrast between (4a) and (4b). Since scrambling in German is adjunction and thus constitutes an instance of A’movement (Grewendorf/Sabel 1999), the trace of the scrambled element in example (4b) is dominated by a category which has itself undergone scrambling. This configuration violates (7). As far as the well-formed example (4b) is

6

Günther Grewendorf

concerned, Müller points out that unlike the scrambled remnant in (4a), the remnant in (4b) has undergone topicalization so that in this case, the trace of the scrambled element is not dominated by an element which has undergone the same kind of movement. Condition (7) is therefore not violated. Condition (7) also accounts for the ungrammaticality of the German example (2a). Since scrambling in German is without exception adjunction, the trace of the scrambled element in (2a) is dominated by a category which has itself undergone scrambling. A problem arises with ungrammatical remnant scrambling in the Japanese example (1b). In this example, the direct object of the embedded CP has undergone long scrambling, which is adjunction, while the CP itself has undergone short scrambling, which according to standard analysis, is A-movement in Japanese. Consequently, two different kinds of movement are involved so that condition (7) would be satisfied, contrary to fact. Note that this problem cannot be avoided by claiming that both kinds of scrambling belong to the same movement type (adjunction), since in that case example (1a) would be predicted to be ungrammatical, again contrary to fact. A similar empirical problem arises from German control facts. As is wellknown, German permits long scrambling out of subject control infinitives embedded under certain control verbs: (8) weil Maria den Studenteni vergeblich [PRO ti zu küssen] versuchte since Mary-NOM the student-ACC in vain to kiss tried ‘since Mary tried in vain to kiss the student’ Since the adverbial vergeblich (‘in vain’) in (8) can only modify the matrix verb, the embedded object den Studenten (‘the student’) must have undergone long scrambling into the matrix clause. Furthermore, the control verb can be passivized with the infinitival object undergoing A-movement to the subject position, as in (9). (9) weil [PRO den Studenten zu küssen] von Maria vergeblich versucht wurde since the student-ACC to kiss by Mary in vain tried was ‘since to kiss the student was tried by Mary in vain’ If we now combine long scrambling out of the infinitival object with A-movement of the latter, the Condition of Unambiguous Domination would predict the result to be well-formed, since two different kinds of movement are involved. However, as can be seen from the examples in (10), the result is ungrammatical:

Problems of remnant movement

7

(10) a. *weil [PRO ti zu küssen]j den Studenteni von Maria tj since to kiss the student by Mary vergeblich versucht wurde in vain tried was b. *weil [PRO ti zu küssen]j von Maria den Studenteni since to kiss by Mary the student vergeblich versucht wurde in vain tried was

4 Remnant movement and the Proper Binding Condition as a constraint on Merge (Saito 2002) Saito (2002) offers a new account of the remnant movement problem by restating the PBC as a constraint on Merge: (11) The Proper Binding Condition as a constraint on Merge α is subject to Merge only if α is a complete constituent. He considers a constituent incomplete if it contains only part of a chain, e.g. a trace but not its antecedent. According to condition (11), the remnant category in (1b) is prevented from being merged. Condition (11) is thus taken to explain the ungrammaticality of scrambling a remnant constituent which contains an unbound trace. Although new and interesting, this approach to the remnant movement problem also faces a series of problems. The first problem again arises from the combination of scrambling and remnant topicalization in German, as repeated in (12): tj versucht]]]. (12) [ti zu füttern]j [C’ hat [ AgrsP [den Hund]i [ AgrsP keiner to feed has the dog-ACC nobody-NOM tried ‘Nobody has tried to feed the dog.’ There are two crucial assumptions that Saito (2002) makes in order to account for the well-formedness of (12). The first assumption is that the scrambling operation which creates the topicalized remnant in (12) behaves like NP-movement,

8

Günther Grewendorf

and the second assumption adopts Lasnik’s (1999) claim that NP-movement does not leave a trace. On the basis of these assumptions, the well-formedness of (12) may be derived. However, this account raises a new problem. If scrambling in German were NP-movement and did not leave a trace, condition (11) would predict remnant scrambling in (4a) (repeated here as (13)) to be likewise well-formed since there would be no trace in the scrambled infinitive. But, as shown by (13), this prediction is not borne out by the facts. [den Hund]i keiner tj versuchte (13) *dass [ti zu füttern]j zweifellos that to feed undoubtedly the dog-ACC nobody tried ‘that undoubtedly nobody tried to feed the dog’ A further empirical problem for Saito’s (2002) analysis arises from an observation made by Hiraiwa (2003). Hiraiwa shows that topicalization of a remnant which contains an NP-trace is ungrammatical in Japanese in cases where the NP-trace results from a Raising-to-Object operation, as in (14b). If NP-movement did not leave a trace, the topicalized constituent in (14b) should not be incomplete and the example should be well-formed on the basis of condition (11). (14) a.

[ν*P Hanako-woi kokorokara [TPTaroo-ga Taroo-NOM Hanako-ACC sincerely to] omot]-ta]. [CP ti baka-da foolish-PRES COMP think-PST ‘Taroo sincerely considered Hanako to be a fool.’

to] (-wa)j [TPTaroo-ga b. *[CP ti baka-da foolish-PRES COMP TOP Taroo-NOM [ν*P Hanako-woi (kokorokara) tj omot]-ta]. Hanako-ACC sincerely think-PST Finally, there is a conceptual problem with Saito’s (2002) account of the remnant movement problem. Restating the Proper Binding Condition as a condition on Merge, seems merely to recast its descriptive nature; furthermore, Saito’s condition adds extra computation that constrains the operation Merge without casting much light on why such computation is needed or how it might be derived.

Problems of remnant movement

9

5 Remnant movement and the Phase Impenetrability Condition (Hiraiwa 2003) Pursuing the aim of explanatory adequacy, Hiraiwa (2003) has tried to derive the PBC from more general principles of language. The crucial generalization that Hiraiwa derives from his investigation of proper binding effects in Japanese is as follows: remnant categories cannot be moved if the antecedent of the unbound trace has undergone movement to the periphery of CP or v*P. This generalization covers two cases. Remnants that arise from scrambling or topicalization to the periphery of CP are prevented from movement, and the same holds for remnants that arise from scrambling to the periphery of v*P. As far as the latter case is concerned, there is an interesting contrast in Japanese raising constructions. While Raising-to-Object blocks movement of the remnant (as already shown in (14b)), this is not the case with Raising-to-Subject; compare the examples in (14) with (15) and notice the contrast between (14b) and (15b): (15) a.

[CP ti baka-da to] [TPTaroo-gai minna-ni Taroo-NOM everyone-DAT foolish-PRES COMP omow-arete-iru]. think-PASS-PRES ‘Taroo is considered to be stupid by everyone.’

b.

to] (-wa)j [TPTaroo-gai minna-ni [CP ti baka-da foolish-PRES COMP-TOP Taroo-NOM everyone-DAT tj omow-arete-iru]. think-PASS-PRES

(Hiraiwa 2003)

On the basis of these observations, Hiraiwa arrives at the generalization in (16): (16) Hiraiwa’s generalization (Hiraiwa 2003) Movement of β to the periphery of CP/v*P blocks subsequent movement of the remnant α containing the trace/copy of β, but A-movement of β to the periphery of TP does not. This generalization leads Hiraiwa to account for the remnant movement problem in terms of Chomsky’s theory of phases (Chomsky 2001, 2008). Given that CP and vP are phases in Chomsky’s theory, illicit remnant movement can be attributed to the so-called Phase Impenetrability Condition. According to (a specific version

10

Günther Grewendorf

of) this condition, a phase is closed off, if the edge of a phase head is extended by some XP. The theoretical basis of this condition is the theory of Cyclic SpellOut: completion of a phase by extending the phase head’s edge triggers the operation Transfer of the phase head’s complement to the external systems. The consequence of Transfer is that the complement of the phase head is no longer accessible to further syntactic operations. So if an XP moves to the edge of a phase, the remnant in the domain of the phase head is prevented from undergoing movement. Although Hiraiwa’s analysis of the remnant movement problem has explanatory force and is based on recent theoretical developments in syntactic theory, it is not without problems. The first problem concerns the target position of scrambling. There is strong evidence (Saito 1992, 1994a, Grewendorf/Sabel 1999, among others) that long scrambling in Japanese is adjunction to IP/AgrsP, but IP/AgrsP, let alone its counterpart TP in a minimalist phrase structure, is not a phase. In the ungrammatical Japanese example (1b), long scrambling of the embedded object is followed by short scrambling of the embedded clause itself. Should Hiraiwa intend to apply his account to examples such as (1b), he would either be forced to make non-standard assumptions about the target position of long scrambling or to modify the theory of phases. A further problem arises from the contrast between remnant scrambling and remnant topicalization in German. It has been argued on the basis of rather strong evidence (Müller/Sternefeld 1993, Müller 1996, Grewendorf/Sabel 1999, among others) that scrambling in German is generally adjunction to IP/AgrsP. Since the landing site of German scrambling is thus not a phase edge either, a remnant which contains a scrambling trace should be able to undergo further movement. This implies that remnant scrambling should be well-formed, contrary to fact. To assume that adjunction to IP/AgrsP is in fact movement to the edge of a phase would not help. A solution along these lines would lead to the wrong prediction that remnant topicalization in German is ungrammatical. A third problem arises in the case of remnant topicalization out of a whisland. As is well-known, wh-movement out of wh-islands is ungrammatical in German: wissen [wer ti betrogen hat]? (17) *Weni wollte Hans whom wanted Hans-NOM to-know who cheated has However, as observed by Fanselow (1987), topicalization out of a wh-clause is marginally possible in German, resulting only in a weak subjacency violation in cases of long object movement:

Problems of remnant movement

11

(18) a. ??[Den Professor zu betrügen]i weiß ich nicht the professor-ACC to cheat know I not [CP warum sie ti versucht hat]. why she tried has ‘I don’t know why she tried to cheat the professor.’ weiß ich nicht b. ??[Den teuren Computer]i the expensive computer-ACC know I not sie ti geschenkt hat]. [CP wem to-whom she given has ‘I don’t know to whom she gave the expensive computer.’ A problem for Hiraiwa’s analysis arises from the fact that remnant topicalization out of a wh-island is no less acceptable than the examples in (18) (see also Müller 1996:384): sie tj (19) a. ??[ti zu betrügen]j weiß ich nicht [CP weni to cheat know I not who-ACC she versucht hat]. tried has ‘I don’t know who she tried to cheat.’ b. ??[NP Ein Buch ti]j weiß ich wohl [CP worüberi ich tj a book know I well about-what I schreiben sollte]. write should ‘I know well about what I should write a book.’ Since CP is a phase by standard assumptions of phase theory, wh-movement to SpecCP should be accompanied by Transfer of the phase head’s complement. Hiraiwa’s analysis would then wrongly predict that the remnant created by whmovement cannot undergo any kind of movement.

6 Remnant movement and the hierarchy of movement types (Grewendorf 2003) 6.1 The generalization From now on I will call the movement which creates the remnant remnant creating movement. The new generalization about a constraint on remnant movement that I defend in Grewendorf (2003) is in the spirit of Müller’s generalization and

12

Günther Grewendorf

Williams’ (2003) Representation Theory but extends their insights by referring to a particular hierarchy of movement types already familiar to a certain extent from the problem of improper movement. If we compare the pros and cons of the analyses presented so far, a crucial observation about remnant movement emerges. Remnant movement appears to be not only prohibited if remnant creating movement and remnant movement is of the same type but also in cases where specific combinations of different movement types are involved. Grewendorf (2003) argues that the relevant constraint has to do with a certain hierarchy of movement types and licenses remnant movement only in cases where the latter is of a higher type than remnant creating movement. The generalization on improper remnant movement is stated as in (20): (20) Improper remnant movement Remnant movement is prohibited unless it is of a higher type than remnant creating movement. It is obvious that (20) is insufficient unless the hierarchy of movement types that (20) refers to can be established on independent grounds. This goal can be achieved if we relate such a hierarchy to the well-known fact that particular types of movement may not feed other types of movement. In other words, we can try to derive such a hierarchy from basic properties of improper movement. For example, we can argue that A’-movement is located higher in the hierarchy than A-movement by referring to the well-known fact that A-movement of an XP may be followed by A’-movement of XP while A’-movement can never feed A-movement. As far as A’-movement is concerned, we can distinguish between A’-movement which is operator movement (as wh-movement in English), and A’-movement which is non-operator movement (as topicalization in German), basing this distinction on the familiar diagnostic (Lasnik/Stowell 1991) that only the former triggers Weak Crossover Effects, while it shares with the latter the typical A’properties such as reconstruction effects, locality effects, the licensing of parasitic gaps and the inability to induce anaphoric binding. Comparing operator and non-operator A’-movement with respect to the hierarchy of movement, we can find evidence that non-operator A’-movement can feed operator A’movement but not the other way round. Evidence for the former claim can be derived from Rizzi’s (1997) constraints on movement operations within the extended left clausal periphery as well as from the observation made in Grewendorf (2005) that short wh-movement in German is non-operator movement, while long wh-movement is operator movement. The claim that operator A’-movement

Problems of remnant movement

13

cannot feed non-operator A’-movement already follows from the fact that a whelement that has undergone movement to the specifier of an interrogative head, is frozen in place (see Lasnik/Saito 1992, Chomsky 1995, among others). Considering other types of movement such as scrambling in the sense of adjunction movement, we can refer to Grewendorf/Sabel (1994, 1999), where evidence is presented for a Constraint on Adjunction Movement according to which scrambling cannot feed any other type of movement on principled grounds. The fact that operator A’-movement cannot be followed by adjunction again follows from the criterial nature of operator movement. Empirical considerations imply that non-operator A’-movement cannot be followed by adjunction either. As shown by Müller/Sternefeld (1993), if movement to SpecCP could be followed by scrambling, ungrammatical long-distance scrambling in German could be derived. It is well-known that A-movement can feed adjunction movement. In languages such as Hindi and Persian, the passivized subject of an embedded clause can undergo long scrambling to the matrix clause, and in languages like German, which have overt object shift, the object can be scrambled into a presubject position. The fact that adjunction movement cannot feed A-movement not only follows from Grewendorf/Sabel’s Constraint on Adjunction Movement but can also be seen from the illicitness of superraising. Finally, Müller (1996) has observed that remnant scrambling in German improves greatly if the antecedent of the unbound trace is a pronominal clitic. If we analyze clitic movement (or movement of “weak pronouns” in general) as an instance of head movement (as independently argued by Cardinaletti/ Roberts (2002)), we can conclude from this observation that head movement is located lower in the hierarchy of movement types than adjunction movement. Summarizing our considerations on the various types of movement we arrive at the hierarchy in (21), according to which operator A’-movement is ranked the highest and A-movement the lowest type (ignoring head movement, which has not yet been discussed in detail): (21) Hierarchy of movement types (see also Grohmann 2003, Williams 2002) – A’-movement as operator movement (‘focus movement’) – A’-movement as non-operator movement (‘topicalization’) – Adjunction movement (‘scrambling’) – A-movement – (Head movement) If the hierarchy in (21) is in fact correct, it provides us with the means to subject generalization (20) to closer empirical scrutiny.

14

Günther Grewendorf

6.2 The explanatory force It is obvious that generalization (20) accounts for all the cases that are explained by Müller’s generalization (7). For example, it rules out the combination of remnant creating scrambling and remnant scrambling as in the familiar example (2b), repeated here as (22): i. Remnant scrambling and remnant creating scrambling: [den Hund]i keiner tj versuchte (22) *dass [ti zu füttern]j zweifellos that to feed undoubtedly the dog-ACC nobody tried ‘that undoubtedly nobody tried to feed the dog’ Notice that there is a contrast between remnant scrambling and remnant creating object raising which provides further evidence for (20). Object raising is an instance of A-movement such that (20) predicts that remnant scrambling should be possible in this case. (23) shows that this prediction is borne out:1 (23) ?dass [ti zu füttern]j zweifellos keiner [den Hund]i tj versuchte that to feed undoubtedly nobody the dog-ACC tried ‘that undoubtedly nobody tried to feed the dog’ ii. Remnant topicalization and remnant creating scrambling: Furthermore, (20) correctly predicts the well-formedness of remnant topicalization, as represented by (24):2 (24)

[ti zu füttern versucht]j [C’ hat [ AgrsP [den Hund]i [ AgrsP keiner tj]]]. to feed tried has the dog-ACC nobody-NOM ‘Nobody tried to feed the dog.’

1 The fact that (23) is still not perfect may be due to independent factors. For example, it may have to do with the presence of a trace within an adjoined category. 2 Recall that topicalization in German is understood as movement to SpecFinP triggered by the EPP. Although it may turn out that differences between movement to SpecFinP and movement to SpecTopP require a more fine-grained distinction with respect to the hierarchical status of non-operator A’-movement, I will ignore such differences for present purposes and subsume either type of movement under the general term “A’-movement as non-operator movement”.

Problems of remnant movement

15

In (22), remnant creating movement and remnant movement are of the same type while in (24) remnant movement, which is topicalization, is of a higher type than remnant creating movement, which is scrambling.3 iii. Illicit remnant scrambling in Japanese (the contrast in (1)) Generalization (20) also explains the ungrammaticality of the Japanese example (1b), repeated here as (25): (25) *[ AgrsP [CP ti’ Hanako-ga ti yonda to]j [ AgrsP sono Hanako-NOM read COMP that [Taroo-ga tj itta]]]. hon-oi book-ACC Taroo-NOM said It has been argued (Grewendorf/Sabel 1999, among others) that long scrambling in Japanese is adjunction movement while short scrambling is A-movement. The ungrammaticality of (25) then follows straightforwardly from the fact that remnant creating movement of the object sono hon-o is long scrambling and remnant movement of the CP is short scrambling. Remnant movement in (25) is thus of a lower type than remnant creating movement. One may object that the scrambled CP in (25) has also undergone adjunction movement on the grounds that the long-scrambled sono hon-o is adjoined to AgrsP and the CP has been scrambled to a position in front of sono hon-o. But notice that even then can (25) be ruled out by generalization (20) since in that case, remnant movement as well as remnant creating movement would be instances of the same type of movement. iv. Remnant wh-movement and remnant creating A-movement If we refrain from making the assumption (Lasnik 1999, Saito 2003) that NPmovement does not leave a trace, generalization (20) also captures English examples such as (5), repeated here as (26), where remnant creating movement is an instance of raising (A-movement) and remnant movement is wh-movement (operator A’-movement):

3 Note that the trace in the topicalized remnant cannot be an A-trace left behind by moving the object to the embedded Spec AgroP. Since the remnant contains the matrix verb, the topicalized category must include the AgroP of the embedded infinitive and consequently contains a scrambling trace.

16

Günther Grewendorf

(26) a. b.

How proud of Bill is John? [CP[Pred ti how proud of Bill]j is [IP Johni tj]]?

v. Remnant A-movement and remnant creating scrambling The German examples in (27) (see (10a) and (10b) above) proved to be problematic for Müller’s account, since they are ungrammatical despite the fact that remnant creating movement and remnant movement are of a different type: (27) a. *weil [PRO ti zu küssen]j den Studenteni von Maria since to kiss the student by Mary vergeblich versucht wurde in vain tried was b. *weil [PRO ti zu küssen]j von Maria den Studenteni since to kiss by Mary the student vergeblich versucht wurde in vain tried was Unlike Müller’s account, the generalization (20) correctly predicts the ungrammaticality of (27). The examples in (27) show remnant creating scrambling of the embedded object followed by remnant A-movement of the infinitival clause. Since the latter is of a lower type than the former, the ungrammaticality of these examples is in line with generalization (20). It is interesting and at first sight unexpected that the ungrammaticality persists if remnant A-movement of the infinitival clause is followed by remnant topicalization, as in (28): (28) *[PRO ti zu küssen]j wurde tj den Studenteni tj von Maria to kiss was the student by Mary vergeblich versucht. in vain tried Since the combination of remnant topicalization and remnant creating scrambling should be grammatical, as illustrated by (24), the ungrammaticality of (28) must be attributed to the illicit intermediate configuration (present in (27)) with remnant A-movement of the infinitive and remnant creating scrambling of the embedded object.

Problems of remnant movement

17

vi. Remnant scrambling and remnant creating A-movement In contrast to (27), where remnant creating scrambling precedes remnant Amovement, the inverse situation with remnant creating A-movement and remnant scrambling is correctly predicted to be grammatical. In the well-formed example (29), the infinitival object has been scrambled and the embedded object has undergone long passivization, an option that exists in German with certain subject control verbs such as versuchen (‘try’): von Maria tj (29) weil [PRO ti zu küssen]j der Studenti since to kiss the student-NOM by Mary versucht wurde tried was lit.: since the student was tried by Mary to kiss There is an interesting contrast in grammaticality between examples such as (29), where remnant creating A-movement precedes remnant scrambling, and examples such as (30), where the remnant is created by scrambling followed by remnant scrambling. Notice the difference in Case between the extracted NPs in (29) and (30): tj versucht hat (30) *weil [PRO ti zu küssen]j den Studenteni Maria since to kiss the student-ACC Mary-NOM tried has ‘since Mary has tried to kiss the student’ As we have already seen, the ungrammaticality of (30) is correctly predicted by generalization (20). vii. Remnant topicalization and remnant creating A-movement While remnant topicalization in (28) is ungrammatical because of the intermediate step of remnant A-movement, remnant topicalization turns out to be well-formed when it is combined with remnant creating A-movement. This is also predicted by generalization (20): (31)

von Maria tj versucht. [PRO ti zu küssen]j wurde der Studenti to kiss was the student-NOM by Mary tried

Note that Italian topicalization behaves like topicalization in German in the relevant respect: the combination of remnant topicalization and remnant creating A-movement is also well-formed in Italian:

18 (32)

Günther Grewendorf

[Tradito ti da sua moglie]j credo che Marioi non sia mai stato tj. betrayed by his wife I-believe that Mario not has ever been ‘I believe that Mario has never been betrayed by his wife.’ (Longobardi 1985:186)

viii. Remnant scrambling and remnant creating wh-movement Let us finally consider the combination of remnant scrambling and remnant creating wh-movement. Since scrambling is of a lower type in the movement hierarchy than wh-movement, generalization (20) predicts this combination to be ungrammatical. In German, long scrambling is only possible out of certain control infinitives but since there are no infinitival wh-questions with the infinitival complementizer zu (‘to’), the relevant case cannot be tested with respect to German. Unlike German, Japanese provides us with the relevant configuration if we recall Kuwabara’s (1998) analysis of long wh-scrambling into a [+Q] clause. According to Kuwabara, the scope-fixing effect observed with long-distance whscrambling should be analyzed as scrambling followed by overt movement of a phonetically empty operator. (33)

[CP WH [IP wh-phrasei [IP [CP [IP . . . ti . . .] Q] . . . ]][+wh]] C overt wh-movement

However, this analysis raises the question of why in examples such as (34), whmovement cannot take place prior to the application of wh-scrambling, resulting in the configuration represented in (35) (see Kuwabara 1998:7): (34)

Taroo-wa [Hanako-ga ti katta ka] siritagatteiru no? Nani-oi what-ACC Taroo-TOP Hanako-NOM bought Q want-to-know Q ‘What does Taroo want to know whether Hanako bought?’

(35)

[CP [IP [ti wh-phrase]j [IP . . . [CP WHi [IP . . . tj . . . ] Q]. . .]]. . .Q]

If (35) were in fact a permissible derivation, Kuwabara would wrongly predict that (34) allows an embedded reading of the wh-phrase. The problem Kuwabara faces with the analysis in (33) can be solved on the basis of generalization (20). In the problematic derivation (35), remnant creating movement is wh-movement

Problems of remnant movement

19

and it is followed by remnant scrambling. (35) thus constitutes a case where remnant movement is of a lower type than remnant creating movement. Such a derivation is excluded by generalization (20).4

6.3 Problems with generalization (20) Generalization (20) seems not to be without problems either. The first problem concerns the combination of remnant topicalization and remnant creating whmovement. i. Remnant topicalization and remnant creating wh-movement Although generalization (20) correctly rules out the derivation in (35), the German examples in (19), repeated here as (36), seem to raise a serious problem: (36) a. ??[ti zu betrügen]j weiß ich nicht [CP weni sie tj to cheat know I not who-ACC she versucht hat]. tried has ‘I don’t know who she tried to cheat.’ b. ??[NP Ein Buch ti]j weiß ich wohl [CP worüberi ich tj a book know I well about-what I schreiben sollte]. write should ‘I know well about what I should write a book.’ In these examples, the movement operation that creates the remnant is whmovement and the remnant undergoes topicalization, the latter extraction being an exceptional option with wh-islands. For such a configuration, generalization (20) predicts a degree of ungrammaticality which exceeds the ungrammaticality associated with a subjacency violation. 4 Note that this argument is only valid if Takahashi’s generalization on the scope-fixing effect in (34) is correct. However, according to other Japanese speakers the lower scope construal is also permissible in principle and there is only a strong preference for the matrix scope reading (see Sabel 2002). If this observation is on the right track, embedded scope of the wh-element in (34) would be a consequence of reconstruction followed by short wh-movement, and the preference for the wide scope reading could be explained as a processing effect along the lines of Sabel (2002).

20

Günther Grewendorf

In order to solve this problem, we have to take into account some specific properties of the movement operations involved in (36). We will then see that the examples in (36) no longer conflict with generalization (20). The properties that are relevant in the present case concern the status of overt short wh-movement in German. As already pointed out above, short whmovement in German, unlike long wh-movement, displays the typical properties of non-operator A’-movement and lacks the properties of operator movement. This can be seen from the fact that short wh-movement in German does not trigger Weak Crossover Effects, as illustrated by the examples in (37): (37) a.

b.

hat seini Vater einen Mercedes geschenkt? Wemi who-DAT has his Vater a Mercedes given ‘To whom has his father given a Mercedes?’ hat seini Trainer heftig kritisiert? Welchen Spieleri which player-ACC has his coach-NOM strongly criticized

We can therefore conclude that short wh-movement in German is movement to SpecFinP rather than to SpecFocP (as argued by Rizzi 1997 for the case of Italian). But this conclusion would be insufficient to bring (36) in line with generalization (20), since (36) would still be derived by two instances of the same movement type: remnant creating movement as well as remnant movement would be instances of non-operator A’-movement. In order to solve this problem we have to consider specific properties of topicalization out of whislands. It can be shown that in this particular case, topicalization is in fact focus movement rather than non-operator A’-movement. Evidence for this assumption is provided by the fact that unlike “ordinary” topicalization (in the sense of movement to SpecFinP), topicalization out of a wh-island triggers a Weak Crossover Effect: weiß seini Trainer nicht [wer ti kritisiert hat]. (38) *Den Spieleri the player-ACC knows his coach-NOM not who criticized has ‘His coach does not know who has criticized the player.’ If this argument is on the right track, then the examples in (36) turn out to represent a combination of different movement types. Remnant movement would in fact constitute an instance of operator A’-movement, while remnant

Problems of remnant movement

21

creating wh-movement would be an instance of non-operator A’-movement. (36) is then no longer at variance with generalization (20).5 Confirmation for this analysis may be derived from an interesting prediction. As already pointed out above, long wh-movement in German shows operator properties, which can be seen from the fact that unlike short wh-movement, it triggers a Weak Crossover Effect, see the contrast between (39a) and (39b). (39) a.

hat seini Trainer heftig kritisiert? Welchen Spieleri which player-ACC has his coach-NOM strongly criticized

b. *Welchen Spieleri hat seini Trainer which player-ACC has his coach-NOM verkaufen ti]? angenommen [CP ti werde der Verein assumed will the club-NOM sell ‘Which player has his coach assumed will the club sell?’ The prediction then is that remnant topicalization out of a wh-island should be ungrammatical when remnant creating wh-movement is long wh-movement. This prediction is in fact borne out: Peter glaubt (40) *[ti zu überreden]j weiß ich nicht [weni to persuade know I not who-ACC Peter thinks [ti dass Maria tj versucht hat]]. that Maria tried has ‘I don’t know who Peter thinks that Mary has tried to persuade.’ Given that wh-movement in English is always operator movement, generalization (20) further predicts that the combination of remnant topicalization and remnant creating wh-movement is ungrammatical in English. This prediction is also borne out by the facts, as can be seen from (41):

5 Note also that the ungrammaticality of the German example (i) is not at variance with generalization (20): (i) *[Welches Buch ti]j fragst du dich [CP [über wen]i du tj lesen sollst]? which book ask you REFL about whom you read should ‘Which book do you wonder about whom to read?’ (Müller 1996:374) Although the claim that short wh-movement is non-operator A’-movement and long whmovement is operator movement would predict the remnant movement in (i) to be well-formed, (i) is nevertheless ungrammatical due to the fact that wh-extraction in German strictly obeys the wh-island constraint.

22

Günther Grewendorf

(41) a.

Give a book to Mary, I wonder whether Peter did.

b. *Give twhat to Mary, I wonder what Peter did.

(Ausin 1998)

ii. Remnant topicalization and remnant creating Raising-to-Object (vs. remnant creating Raising-to-Subject) Another problem is raised by Hiraiwa’s (2003) interesting observation that remnant topicalization is well-formed in Japanese when the remnant is created by Raising-to-Subject movement, while remnant topicalization is ungrammatical when the remnant is created by Raising-to-Object. The relevant contrast is represented by the examples (14b) and (15b), repeated as (42) and (43): (42)

to] (-wa)j [TPTaroo-ga [ν*P Hanako-woi *[CP ti baka-da foolish-PRES COMP TOP Taroo-NOM Hanako-ACC (kokorokara) tj omot]-ta]. sincerely think-PST ‘Taroo sincerely considered Hanako to be a fool.’

(43)

to] (-wa)j [TPTaroo-gai minna-ni tj [CP ti baka-da foolish-PRES COMP TOP Taroo-NOM everyone-DAT omow-arete-iru]. think-PASS-PRES ‘Taroo is considered to be stupid by everyone.’

So far we have not been dealing with the hierarchical properties of “true topicalization” in the sense of movement that is triggered by a topic feature. Without going into the details of such an investigation, let us just assume that topicalization in Japanese is adjunction movement, since topicalized elements can be preceded by long-scrambled elements, as can be seen from example (34). If this assumption is on the right track, then the well-formedness of (43) is expected on the basis of generalization (20): remnant adjunction movement is of a higher type than A-movement. The problem then is (42). Without having a well-founded solution I will tackle this problem by assuming that despite appearances, the alleged Raising-to-Object operation in (42) does not constitute a case of Amovement but is in fact an instance of scrambling.6 If this assumption can be 6 One may object that as shown by Saito (1994b), long scrambling out of a finite clause obligatorily targets IP. However, if (42) represents an instance of Exceptional Case Marking it may be that this kind of long object shift in Japanese is different from scrambling out of finite clauses as well as scrambling out of control infinitives, which is usually taken to be A-movement (see Grewendorf/Sabel 1999). Confirming familiar ideas about Japanese scrambling, this sort of movement may have A- as well as A’-properties and thus also count as adjunction movement.

Problems of remnant movement

23

supported by further evidence, the movement which creates the remnant in (42) would be of the same type as remnant movement, namely adjunction movement. The ungrammaticality of (42) would then no longer conflict with generalization (20). iii. Accounting for the hierarchy of movement types We have seen that remnant movement is not only subject to a constraint which prohibits the combination of similar movement types but is also constrained by a stronger restriction which is related to a hierarchy of movement types and requires that remnant movement be of a higher type than the movement that creates the remnant. This is the crucial claim expressed by generalization (20). The important question that arises at this point is how this generalization might be theoretically implemented in a feature-based approach to syntactic derivations. Although an answer to this question is beyond the limits of this paper, I will briefly outline how this problem might be tackled. The hierarchy of movement types might be expressed in terms of a hierarchy of features that trigger the various kinds of movement. If such a hierarchy could be established on independent grounds, generalization (20) might be theoretically implemented within Chomsky’s Probe-Goal system as a constraint along the lines of Rizzi’s Relativized Minimality. The basic idea is as follows. Let us assume that a Goal G1 (the later remnant) bears an uninterpretable feature F1 and contains another Goal G2 which bears an uninterpretable feature F2. Either Goal needs to be subject to an operation Agree in order to get rid of its uninterpretable feature. Let us further assume that the required Agree operations involve movement for independent reasons. In terms of the theory of phases we could say that G1 and G2 have to leave the domain of the next higher Probe P, which as a Probe acts as a phasal head and requires uninterpretable elements to leave its domain before the operation Transfer takes place. On the basis of a hierarchy of features we could then state a constraint on Agree along the following lines. We could claim that a Probe P2 for Goal G2 with feature F2 cannot access G2 if an intervening “container” G1 is present which bears a feature F1 of the same type as F2 or of a lower type than F2. In other words, in such a situation G2 cannot undergo movement to the edge of P2. The basic idea of this suggestion thus consists of two components: a hierarchy of features and a minimality constraint that is relativized with respect to such a hierarchy:

24

Günther Grewendorf

(45) Constraint on remnant movement a.

hierarchical order of features

b.

A probe P2 cannot access a goal G2 with feature F2 when an intervening “container” G1 is present which bears a feature F1 of the same type as F2 or of a lower type than F2.

Constraint (45) can schematically be represented as in (46) (46) P2 [G1 (F1) [G2 (F2) . . .] ] (ok with F2 = Case and F1 = Op, not ok with F2 = Σ (scrambling) and F1 = Case) The question of how this idea could be worked out in detail is left for another paper.

7 Improper movement and improper remnant movement: UCOOL and GenPIM (Abels 2006) Abels (2006) makes further interesting suggestions as to how remnant movement is constrained. The general aim that he pursues is to show that improper movement, extraction out of moved phrases, and remnant movement are subject to the same constraint Generalized prohibition against improper movement (GenPIM). GenPIM is based on the Universal constraint on operational ordering (UCOOL) given in (47): (47) Universal constraint on operational ordering (UCOOL) Θ