The Nature of Scrambling and Its Resulting Chains: Operator or Mediator of Various Constructions

Jun Abe

March, 2015 [Slightly Revised in August, 2016]

E-mail: [email protected]

Contents 1

Introduction

2

Scrambling as Operator Movement

19

2.1

The Nature of Scrambling and Focus Movement

20

2.2

Scrambling of Wh-Phrases and QPs in Japanese

34

2.3

An Apparent Paradox in Long-Distance Scrambling of Wh-Phrases

38

2.3.1

Superiority Effects

55

2.3.2

Additional Wh-Effects

61

3

1

2.4

Scrambling of QPs as Focus Movement

66

2.5

Clause-Internal Scrambling of Wh-Phrases and Phrases Carrying FPs

87

2.6

Multiple Scrambling

100

Scrambling as A-Movement: A Residue of NIC and Super-Raising 131 3.1

Proposal: Scrambling as Mediator 3.1.1 3.1.2

Scrambling as an Instantiation of Overt Operator Movement: Abe (2012)

136

An Extension to A-Movement

141

3.2

Minimality

153

3.3

The Correlation of Pronunciation and LF Interpretation in A-Chains

166

3.3.1

Cases of Subject-to-Object Raising

171

3.3.2

Cases of Nominative-Genitive Conversion

176

Cases of A-Movement Involving More Than One θ-Role

184

3.4.1

“Backward Binding” in the Case of Subject-to-Subject Raising

191

3.4.2

“Backward Binding” in the Case of Subject-to-Object Raising

200

3.4

4

136

Appendix: Is Overt Raising Possible with the NGC Construction?

210

Scrambling for Relativization: String-Vacuity and Island Effects

215

4.1

Inoue’s (1976) Generalization

217

4.2

Proposal

229 i

4.3

4.2.1

String-Vacuous Movement and Island Sensitivity

230

4.2.2

Sakai’s (1994) approach

238

4.2.3

Some Consequences

244

Why Movement in Japanese Relativization?

5 Scrambling for Right Dislocation: Bi-Clausal Analysis

6

254

267

5.1

Bi-Clausal Analysis of Japanese Right Dislocation

269

5.2

The Status of the Gap in Japanese Right Dislocation

280

5.3

Evidence from Quantifier Scope Interaction

286

5.4

Proper Binding Condition Violations

297

5.5

Condition C Effects

303

5.6

The VP Scrambling Approach to Multiple Right Dislocation

Conclusions

308

322

References

328

ii

1

Introduction

This manuscript aims to elucidate the nature of scrambling with Japanese as a target of research. It has been well-established since Saito (1985) that the property of free word order attested in many languages, including Japanese, is best captured by scrambling, an instance of Move. Thus, Japanese allows various word orders as long as a verb occupies the final position of a given sentence, as shown below: (1) a.

John-ga

Mary-ni Bill-o

syookaisita.

John-Nom Mary-Dat Bill-Acc introduced ‘John introduced Bill to Mary.’ b.

Mary-ni John-ga Bill-o syookaisita.

c.

Bill-o John-ga Mary-ni syookaisita.

d.

John-ga Bill-o Mary-ni syookaisita.

e.

Mary-ni Bill-o John-ga syookaisita.

f.

Bill-o Mary-ni John-ga syookaisita.

There is no significant meaning difference detectable among these sentences. The standard assumption is that (1a) reflects the underlying word order (cf. Hoji 1985) and the other instances are derived from it by applying scrambling. Thus, (1b), for instance, is derived from (1a) by scrambling the indirect object Mary-ni to the top of the sentence, as shown below: (2) Mary-ni1 John-ga t1 Bill-o syookaisita

1

Scrambling is claimed to be different from a typical instance of A’-movement such as topicalization, illustrated below, in that it does not make the scrambled phrase bear a special semantic function. (3) a.

John met Mary.

b. Mary, John met. It is obvious that in such a topicalized case, the movement operation involved induces semantic effects; in (3b), Mary functions as the topic of the sentence, unlike that in (2). Such movement is usually called operator movement. Scrambling has given rise to much controversy about how it is accommodated into a general theory of movement. There are at least two properties peculiar to scrambling that constitute the focus of controversy: (i) optionality and (ii) semantic vacuity. Under the checking theory of movement, proposed by Chomsky (1995) and advocated since in the Minimalist Program, it is supposed that each application of movement has some driving force, expressed with a morphological feature that needs to be checked. Thus, it is claimed in the original version of checking theory that in a passive construction, for instance, the object has a Case feature to check and since the passive morpheme deprives the verb of its Case checking ability, it needs to move to the nearest Spec of TP to check its Case with T. Or in an overt wh-movement construction, a wh-phrase has a wh-feature to check and hence needs to move to a Spec-CP to check this feature with a C introducing an interrogative clause. Under this system of movement, it is further supposed that movement does not take place unless it has some driving force; this is called the Last Resort Principle. Given this mechanism of movement, scrambling is 2

peculiar in that it does not seem to be the kind of movement that is triggered by a morphological feature to be checked by moving its bearer to the Spec of some designated category, but rather to be the one that applies “whenever you like” in order to change word order. Accordingly, it is also taken to be free from the Last Resort Principle. This property has been called optionality of scrambling (see Fukui (1993), among others). The second property of scrambling, namely semantic vacuity, is most typically illustrated by the fact that in a free word order language such as Japanese, changing the word order of the arguments of a verb keeps the meaning of the sentence unchanged, as shown in (1) above. This property is also peculiar to scrambling, as is clear when it is compared with operator movement, illustrated in (3). The above observations have led to the temptation that scrambling is not an instance of transformation but rather a stylistic rule that applies outside the syntactic component (see Chomsky (1965) and Ross (1967), among others). Despite the conceptual plausibility of this conjecture, there has been accumulating empirical evidence for the contrary claim that scrambling operates in the syntactic component, as witnessed by the fact that this operation affects the binding possibilities of anaphors and pronouns and the scope interaction among quantifier phrases (henceforth, QPs), etc. Hence there is good reason to assume, at least on empirical grounds, that scrambling is an instance of Move that operates in the syntactic component. The optionality issue regarding scrambling was very often tackled in such a way that scrambling has in fact a driving feature that needs to be checked, just like other instances of movement, as witnessed by such a work as Bošković and Takahashi (1998). 3

But the way to approach this issue has been changed since Chomsky (2004) expressed the view that Move (or Internal Merge according to his terminology) is “an operation that is freely available,” and “accordingly, displacement is not an ‘imperfection’ of language; its absence would be an imperfection.” (p. 8) Once we admit that Move applies freely in principle, it is not unnatural to assume a movement operation like scrambling that is free from the Last Resort Principle and to make it serve for a functional utility such as changing the word order among arguments of a verb. This is directly relevant for the consideration of the second property of scrambling, namely semantic vacuity. It is not unreasonable to claim that the main functional utility of scrambling is to change word order and hence that this operation has an effect only on the PF interface and no effect on the LF side. However, this brings us back to the question whether scrambling is actually a syntactic operation rather than simply a stylistic rule, given the tenet expressed by Chomsky (2008) according to which “there is a basic asymmetry in the contribution to ‘language design’ of the two interface systems: the primary contribution to the structure of FL [= faculty of language] may be optimization of mapping to the C-I interface.” (p. 136) Under this tenet, it will be conceptually odd if we assume an operation that has no relevance whatsoever for the C-I interface, affecting solely the PF interface. This manuscript aims to demonstrate that scrambling in fact may have semantic effects mainly in either one of the following two ways: (i) a chain produced by scrambling is reinterpreted as an operator-variable chain or (ii) it serves as a mediator of various constructions. In Chapter 2 we make a detailed discussion on the first case, 4

arguing that the reinterpretation for an operator-variable chain occurs whenever it is possible and that when it is prohibited, the chain produced by scrambling is regarded as a semantically vacuous chain, hence collapsed into a single-membered one or incorporated into a larger chain. The claim that a chain produced by scrambling is reinterpreted as an operator-variable chain gains initial support from cases of what Saito (1992) calls long-distance scrambling such as the following: (4) Mary-o1

Bill-ga

[John-ga t1 hihansita to]

Mary-Acc Bill-Nom John-Nom

itta.

criticized Comp said

‘Mary1, Bill said that John had criticized t1.’ In this case, the scrambled phrase Mary-o normally receives focus interpretation, which thus indicates that long-distance scrambling gives rise to a focus chain. This is further supported by Takahashi’s (1993) observation that long-distance scrambling of a wh-phrase behaves like regular wh-movement. Let us consider the following examples provided by Takahashi (1993): (5) a.

John-wa [Mary-ga

nani-o

tabeta ka] siri-tagatteiru no?

John-Top Mary-Nom what-Acc ate

Q

want-to-know Q

‘Does John want to know what Mary ate?’ or ‘What does John want to know whether Mary ate?’ b.

Nani1-o

John-wa [Mary-ga t1 tabeta ka]

sirit-agatteiru no?

what-Acc

John-Top Mary-Nom ate

want-to-know Q

Q

‘What does John want to know whether Mary ate?’ 5

(Takahashi 1993:657)

In each of these examples, there are two Q-markers; one is ka in the embedded clause and the other is no in the matrix clause. Takahashi observes that (5a) is ambiguous depending upon whether the in-situ wh-phrase nani ‘what’ takes scope over the embedded clause or the matrix clause. When nani takes embedded scope, the whole sentence is interpreted as a yes-no question, and when it takes matrix scope, the embedded ka is interpreted as ‘whether’. Takahashi makes an interesting observation with (5b), where nani-o ‘what-Acc’ is scrambled out of the embedded interrogative clause: in this sentence, nani can only take matrix scope. If long-distance scrambling of a wh-phrase were simply taken to produce a semantically vacuous chain, then nani in (5b) should be interpreted in its original position, hence giving rise to ambiguity with respect to whether it takes scope over the embedded or matrix clause, just like in (5a). The fact that nani can only take matrix scope in (5b) then indicates that long-distance scrambling of a wh-phrase can establish an operator-variable chain. This claim is further supported by Takahashi’s (1993) observation that long-distance scrambling of a wh-phrase crossing another shows superiority effects. We cannot claim, however, that long-distance scrambling of a wh-phrase always establishes an operator-variable chain, as evidenced by well-known examples such as the following, provided by Saito (1989), to show that scrambling may produce a semantically vacuous chain: (6) ?Dono which

hon-o1

[Mary-ga [John-ga t1

tosyokan-kara kari-dasita

book-Acc Mary-Nom John-Nom library-from

6

checked-out

ka] Q

siri-tagatteiru] (koto) want-to-know fact ‘Lit. Which book, Mary wants to know Q John checked out from the library.’ (Saito 1989:192) In this case, the chain produced by scrambling dono hon-o ‘which book-Acc’ long-distance needs to be undone, so that this wh-phrase is properly located within the scope of the embedded Q-marker ka. Thus, we need to claim that scrambling is reinterpreted as an operator-variable chain whenever possible. In Chapter 2, we also consider how long-distance scrambling of QPs affects scope interaction. Oka (1989) and Tada (1993) observe that long-distance scrambling does not change scope order, as illustrated below: (7) a.

Dareka-ga

[sensei-ga

daremo-ni

kisusita to]

someone-Nom teacher-Nom everyone-Dat kissed

sinziteiru.

Comp believe

‘Someone believes that the teacher kissed everyone.’ b.

Daremo-ni1

dareka-ga

[sensei-ga t1

everyone-Dat someone-Nom teacher-Nom

kisusita to] kissed

sinziteiru.

Comp believe

‘Everyone, someone believes that the teacher kissed.’ (7b) is derived from (7a) by applying long-distance scrambling to the embedded object QP daremo-ni ‘everyone-Dat’. Although this QP crosses dareka ‘someone’ in the matrix subject position, this does not make the sentence ambiguous regarding the scope order of these two QPs. Based upon Saito’s (1989) claim that long-distance scrambling can be semantically vacuous movement, Oka (1989) makes the stronger claim that 7

long-distance scrambling produces an unlicensed chain according to Full Interpretation and hence must be undone at LF. This accounts for why long-distance scrambling does not extend the scope domain of the scrambled phrase to its landing site, as witnessed in (7). I argue against this claim by demonstrating that the pattern of scope facts observed in (7) is also observed with long-distance operator movement and hence that it has nothing to do with the nature of scrambling as semantically vacuous movement. I argue that the clause-boundedness condition on Quantifier Raising (henceforth, QR) is responsible for limiting the scope domain of a QP scrambled long-distance to the embedded clause, so that in (7b) daremo ‘everyone’ cannot scope over dareka ‘someone’. With this account, we can maintain our claim that scrambling is reinterpreted as an operator-variable chain whenever possible. Things are different with what Saito (1992) calls clause-internal scrambling, which does not appear to establish an operator-variable chain. Consider a typical case of such scrambling: (8) Mary-o1

John-ga t1 hihansita.

Mary-Acc John-Nom

criticized

‘Mary1, John criticized t1.’ As noted in the beginning of this chapter, scrambling of Mary-o in this case does not cause a significant change in meaning, unlike the long-distance scrambling case (4). I propose, following Abe (1993), that in such a case, a phrase scrambled clause-internally can be adjoined to so local a position where a [Focus] feature will not be licensed that the resulting chain can be semantically vacuous. I demonstrate that this explains why 8

clause-internal scrambling of a wh-phrase crossing another does not show superiority effects. Alternatively, I argue that a chain produced by clause-internal scrambling may end up being part of a larger chain, as is the case when an inherently focus phrase undergoes clause-internal scrambling (the following example is cited from Kuroda (1971)): (9) S.S.-o-sae1

John-dake-ga t1

S.S.-Acc-even John-only-Nom

yonda. read

‘Even S.S., only John read.’ I argue that in this case, the focused phrase S.S.-o-sae ‘S.S.-Acc-even’ undergoes covert movement from the position to which it is scrambled to license its own [Focus] feature, so that it takes scope over the other focused phrase John-dake-ga ‘John-only-Nom’. In the final section of this chapter, we analyze multiple scrambling in Japanese, illustrated below: (10) Mary-ni1 Mary-Dat

Bill-o2

John-ga t1 t2 syookaisita.

Bill-Acc John-Nom

introduced

‘Lit. To Mary, Bill, John introduced.’ I argue for Koizumi’s (1995, 2000) VP scrambling approach to multiple scrambling, according to which after V raises out of its maximal projection, the remnant VP undergoes scrambling. I argue that literal application of multiple scrambling is severely restricted under our hypothesis that scrambling produces a focus chain whenever possible with the further assumption that a focus chain cannot cross another. I demonstrate that this VP scrambling approach properly deals with the relevant data 9

concerning scope interaction of QPs. In Chapters 3 through 5, I discuss those cases in which chains produced by scrambling serve as “mediators” of various constructions. Abe (2012) argues that no overt operator movement exists in Japanese and that what appears to be an instance of such movement in fact involves scrambling. Assuming that a [Focus] feature is responsible for triggering overt operator movement, Abe hypothesizes the following: (11) [Focus] is not attracted in Japanese. Here “attract” is used in the sense of Chomsky (1995) and (11) means that there is no functional category in Japanese that attracts a phrase carrying a [Focus] feature. Assuming that this feature is assigned during a derivation, Abe claims that Japanese instead exploits scrambling to move a phrase that is going to be assigned a [Focus] feature after the scrambling takes place. Abe suggests that property (11) has to do with the defectiveness of functional categories in the sense of Fukui (1986) and that this may ultimately be attributed to the lack of agreement in this language. In this manuscript, I argue that (11) should be generalized so as to include A-movement cases, thus hypothesizing the following: (12) No phrase is attracted in Japanese. It follows from (12) that what appears to be an instance of A-movement in Japanese also involves scrambling. In Chapter 3, I discuss three cases of A-movement in Japanese, illustrated below, to give support to hypothesis (12):

10

(13) Nominative-genitive conversion: a.

[John-ga

sukina] hon

John-Nom like

book

‘the book John likes.’ b. [John-no sukina] John-Gen like

hon book

(14) Subject-to-subject raising: John-ga1

saikin [t1 motto

benkyoosuru yoo-ni]

natta.

John-Nom

recently more

study

became

Comp

‘Recently, John has come to study harder.’ (15) Subject-to-object raising: a.

John-ga

[Bill-ga baka

da

to]

John-Nom Bill-Nom stupid be

omotteiru.

Comp think

‘John thinks that Bill is stupid.’ b.

John-ga [Bill-o

baka da

to]

John-Nom Bill-Acc stupid be Comp

omotteiru. think

These constructions can be claimed to share one crucial property with respect to the A-movement involved: it takes place across a nominative subject position without inducing a violation of a minimality condition of the sort that derives the effects of Nominative Island Condition (henceforth, NIC), proposed by Chomsky (1980). (13) illustrates the phenomenon of so-called nominative-genitive conversion (henceforth, NGC) in which the subject of an adnominal clause, which is normally marked with 11

nominative Case, can be marked with genitive Case, as shown in (13b). Given that genitive Case is licensed in Spec-DP, it is natural to assume that a genitive subject such as John-no in (13b) undergoes movement to Spec-DP, as shown below: (16) [DP John-no [TP [vP John-no sukina]] hon]

This movement takes place from the subject position of a finite clause across the embedded Spec-TP, in which John could be assigned nominative Case. (14) illustrates a case that Uchibori (2000, 2001) regards as a case of subject-to-subject raising (henceforth, SSR) in which movement takes place from the subject position of an embedded finite clause. Thus, this construction also instantiates a case where A-movement skips a nominative subject position. Finally, (15b) illustrates a case of subject-to-object raising (henceforth, SOR), according to Kuno (1976), in which the embedded subject Bill-o is assigned accusative Case by the higher verb. Under this assumption, (15b) will involve such movement as shown below: (17) John-ga [VP Bill-o [CP [TP [vP Bill-o baka da]] to] omotteiru]

In this case as well, the movement takes place from the subject position of a finite clause across the embedded Spec-TP in which Bill could be assigned nominative Case. I argue that these apparent violations of the NIC are explained by proposing a mechanism in which the A-chains involved in the three Japanese constructions are produced by scrambling. On the other hand, these A-chains are subject to a condition like the Specified Subject Condition in the sense of Chomsky (1976). I argue that this fact is 12

explained by a crossing constraint that applies to the resulting A-chains produced by scrambling, which prohibits a phrase carrying a feature X from crossing another also carrying X. Further examining the nature of the A-chains produced by scrambling with respect to pronunciation and LF interpretation, I argue for the Agree-less approach, advocated recently by Hornstein (2009), in which the effect of overt vs. covert movement is captured by which copy of a produced chain is pronounced. Following Abe (2016), I demonstrate that in the A-chains produced by scrambling in NGC and SOR constructions in Japanese, any member can be the target for pronunciation in principle. This is actually instantiated with the SOR cases, as illustrated below: (18) John-ga [mada John-Nom still

Mary-o

kodomo da to]

Mary-Acc child

omotta.

be Comp thought

‘John thought that Mary was still a child.’ (19) Yamada-wa Tanaka-o Yamada-Top Tanaka-Acc

orokanimo [tensai da

to]

stupidly

Comp thought

genius be

omotteita.

‘Yamada stupidly thought that Tanaka was a genius.’ (18) illustrates a case where the bottom copy of the resulting A-chain is pronounced and (19) a case where the top copy is pronounced. This optionality of pronunciation, however, does not hold for the NGC cases, in which only the bottom copy is pronounceable. I follow Abe (2016) in his claim that this is due to the fact that the A-chains involved in NGC cross adnominal clause islands and that in order to avoid this violation, the bottom copies of the A-chains must be pronounced, on the assumption 13

made by Abe and Hornstein (2012) that only “overt” movement, namely movement with the top copy of the resulting chain pronounced, is subject to island conditions. I further argue, following Abe (2016), that pronunciation and LF interpretation correlate with each other in the A-chains involved in NGC and SOR constructions; that is, the member of a given A-chain that is pronounced must be the one that serves for LF interpretation (I refer to this correlation condition as P-L Match Condition, following Abe (2016)). There is one exception to this condition: only those cases that involve string-vacuous movement of genitive subject in the NGC construction can violate the P-L Match Condition; that is, the top copy of an A-chain can be active for LF interpretation even though the bottom copy is pronounced. I argue, following Abe (2016), that this discrepancy between pronunciation and LF interpretation is properly captured by reformulating Abe and Hornstein’s (2012) relevant condition on chain production in such a way that when a given instance of movement is string-vacuous, a PF adjustment rule dictates that the instruction of pronouncing the top copy of the resulting chain be canceled and instead forces the bottom copy to be pronounced. We also discuss cases of A-chains produced by scrambling that involve movement into θ-position. Abe (2014) argues, along the lines of Hornstein’s (1999, 2001) movement theory of control, that when a null subject takes an argument in the next clause up as its antecedent, the anaphoric relation involved is best captured as an instance of A-movement. Thus, the following example has the derivation given in (21): (20) John1-ga [e1 atama-ga John-Nom

head-Nom

warui

to]

bad

Comp think 14

omotteiru/itteiru. /say

‘John1 thinks/says that e1 is stupid.’ (21)

[TP John-ga [vP John-ga [VP [CP [TP John-ga [vP John-ga [VP atama-ga warui]]] to] …

In this derivation, the second step of movement is movement into θ-position, so that the whole A-chain involves two members carrying different θ-roles. Given that in A-chains, any member can be the target for pronunciation in principle, it is expected that the lower copies of the resulting A-chains in (21) are pronounceable. If this is in fact possible, it will instantiate a case similar to so-called backward control in the sense of Polinsky and Potsdam (2002) (I refer to such a case as an instance of “backward binding.”). I demonstrate that cases of backward binding are in fact attested in such SSR cases as (20) as well as in the SOR construction. This will give strong support to the Agree-less approach to these constructions. In Chapter 4, I discuss Japanese relativization. It has been standardly assumed since Kuno (1973) that Japanese relativization does not involve movement, at least in any significant way. A main reason comes from the fact that it does not show island sensitivity, as illustrated below: (22) a.

[[e1 kawaigatteita] was-fond-of

inu-ga

sinde-simatta]

kodomo1

dog-Nom dying-ended-up

child

‘Lit. the child1 who [the dog [e1 was fond of] died]’ b. [[e1 kiteiru]

yoohuku-ga

is-wearing suit-Nom

yogoreteiru]

sinsi1

is-dirty

gentleman

‘Lit. the gentleman1 who [the suit [that e1 is wearing] is dirty]’ 15

(Kuno 1973:239) (23) a.

[[e1 sinda node]

minna-ga

kanasinda]

hito1

died because everyone-Nom was-distressed person ‘Lit. the person1 who, because e1 died, everyone was distressed’ b. [[pro e1 hara ippai belly-full

tabeta-ra] geri-o

site-simatta]

okasi1

ate-when diarrhea-Acc doing-ended-up-with cookies

‘Lit. cookies1 which, when (we) had stuffed ourselves with e1, (we) ended up with diarrhea’

(ibid.:237-8)

(22) illustrates cases of relative clause island violations and (23) cases of adjunct island violations. Based upon the observation that those languages that allow island violations in relativization also allow “pronoun drop,” Perlmutter (1972) claims that relativization in these languages involves deletion of an ordinary pronoun rather than movement of a relative pronoun, like in English. In this chapter, I argue, along the lines of Ishii (1991), that contrary to the standard view on Japanese relativization, this construction involves movement; more specifically, it involves an overt operator scrambled to a designated position and the overt operator is then deleted in the PF component a la Chomsky (1977). It is observed that island conditions are violable in Japanese relativization only when the gap that induces such a violation is located at the left edge of the whole relative clause, as is the case with those examples in (22) and (23). I argue that this is explained under the overt operator scrambling approach by assuming Abe and Hornstein’s (2012) theory of movement, according to which string-vacuous movement is immune to island conditions. Given 16

that scrambling takes place leftward, it follows that the overt operator scrambling involved in relativization is string-vacuous when it takes place from the left edge position of the whole relative clause, hence immune to island conditions. Finally, in Chapter 5, I discuss so-called right dislocation (henceforth, RD) constructions in Japanese, which appears to involve rightward movement, as illustrated below: (24) a.

John-wa Mary-o

hihansita yo.

John-Top Mary-Acc

criticized

‘John criticized Mary.’ b. John-wa e1 hihansita yo, Mary-o1. Haraguchi (1973) claims that (24b) is an instance of Japanese RD in which Mary-o is moved rightward from the underlying order given in (24a). This proposal, however, is not compatible with what has been claimed about the relationship between the directionality of movement and the head parameter. Fukui (1993) claims that the value of the head parameter should be preserved in derived structures in such a way that an adjunction operation should create a structure that is consistent with the value of the head parameter in a given language. Under this theory, in a head-final language such as Japanese, rightward adjunction should be prohibited. Alternatively, Kayne (1994), Takano (1996, 1998), and Fukui and Takano (1998) propose what is called asymmetrical structure thesis, according to which hierarchical structure unambiguously determines the order of terminal symbols, thereby denying the existence of symmetrical structures. According to Kayne, hierarchy and precedence correspond roughly in such a 17

way that what precedes is structurally higher than what follows. Under this thesis, rightward movement is excluded altogether. In this chapter, I argue, following Tanaka (2001), that Japanese RD involves a bi-clausal structure in which the second clause functions as an afterthought for the first clause, filling the gap of the latter with semantic content. According to this analysis, a postverbal phrase such as Mary-o in (24b) undergoes leftward scrambling in the second clause, with the other materials deleted under identity with the corresponding ones in the first clause, hence serving as a filler of the gap of the first clause. An obvious advantage of this bi-clausal analysis is that the Japanese RD construction is not an exception to the generalization that Japanese is a strictly head-final language, and that this analysis is compatible with both the restricted theory of adjunction and the asymmetrical structure thesis, mentioned above. I will demonstrate in this chapter that the bi-clausal analysis of the Japanese RD construction is also empirically well-motivated. In particular, this analysis properly captures the island sensitivity of this construction since it involves leftward scrambling. More importantly, it nicely captures the fact that this island sensitivity persists even if the gap of the first clause is replaced by an overt resumptive pronoun.

18

2

Scrambling as Operator Movement

This chapter aims to demonstrate that scrambling produces an operator-variable chain whenever possible; otherwise, it is regarded as a semantically vacuous chain, hence collapsed into a singled-membered one or incorporated into a larger chain. I argue that when wh-phrases and QPs undergo scrambling, the produced chains are reinterpreted as focus chains, unless this is somehow prohibited, by assigning [Focus] features to the heads of these chains, following the mechanism proposed by Abe (2012). In so doing, I address an issue concerning the possibility of the “radical reconstruction” of long-distance scrambling of wh-phrases with which apparently contradictory claims have been made: whereas Saito (1989) claims that long-distance scrambling can be undone, Takahashi (1993) shows that some instances of long-distance scrambling of wh-phrases behave like operator movement. I also argue for the hypothesis that the chains produced by scrambling QPs are identified as focus chains, by demonstrating that QPs moved long-distance interact in their scope with other QPs in the predicted ways given this hypothesis and other independently motivated assumptions. Finally, I argue,

following

Abe

(1993),

that

clause-internal

scrambling

can

produce

non-operator-variable chains that are ultimately incorporated into larger chains and that the availability of this option is the main source of the different behaviors of long-distance and clause-internal scrambling.

19

2.1

The Nature of Scrambling and Focus Movement

It has been standardly claimed that the main role of scrambling is to change the word order of a given sentence. In order to capture this property, let us first assume that in a chain created by Move, features can be scattered among its members. In this sense, Move is not the operation of Copy and Merge, but rather is characterized as movement of relevant features without leaving any copy or trace of them. Given that lexical items (LIs) consist of phonological features [PF], semantic features [SF] and formal features [FF], we can, for instance, have the following chains: (1) Given an LI named K, a.

({PF(K), FF(K)}, {SF(K)})

b. ({FF(K)}, {PF(K), SF(K)}) c.

({PF(K), SF(K)}, {FF(K)})

d. ({SF(K)}, {PF(K), FF(K)}) e.

({PF(K)}, {FF(K), SF(K)})

In this system, overt vs. covert movement is distinguished in terms of which member of a given chain carries its [PF]. Thus, (1a) corresponds to a chain created by overt movement to check the [FF] of K whereas (1b) corresponds to a chain created by covert movement. Let us call the movement triggered by [FF] agreement-driven movement. There is another type of movement, one triggered by [SF], called scope-driven movement. Among such movement operations are QR, topicalization and heavy NP shift.1 (1c) is a chain created by an overt application of scope-driven movement while 1

See Abe (forthcoming) for how these types of scope-driven movement are analyzed

20

(1d) is a covert counterpart of scope-driven movement. I argue that (1e) is a chain created by scrambling in which only [PF] is carried along by an application of Move. This directly captures the nature of scrambling: it does not have any driving force in Chomsky’s (1995) sense, hence its application being optional, and further it does not give rise to any semantic effects. It follows then that its main role is simply to change word order. Although the above characterization of scrambling is simple and straightforward, hence desirable, it seems to require some complication. True, scrambling does not induce any semantic effects in its essentials, as shown in (1.1) above, but it is also the case that it gives rise to “side effects” in some cases. Thus, consider the following examples, taken from Saito (1992): (2) a.?*[Otagai1-no

sensei]-ga

karera1-o hihansita (koto)

each other-Gen teacher-Nom they-Acc criticized fact ‘Each other1’s teachers criticized them1.’ b. ?Karera1-o they-Acc

[otagai1-no

sensei]-ga t1 hihansita (koto)

each other-Gen teacher-Nom criticized fact

‘Them1, each other1’s teachers criticized.’

(Saito 1992:74-75)

(2a) is a violation of Condition A, since the anaphor otagai ‘each other’ is not bound by its antecedent karera ‘they’. The grammaticality of (2b) then suggests that karera-o, which is fronted by scrambling, can serve as the antecedent of the anaphor otagai. If scrambling were an instance of movement in which only [PF] is carried along, as under the minimalist framework.

21

claimed above, then it will be expected that (2b) would behave the same as (2a) with respect to the licensing of otagai, contrary to fact. This, I take to indicate that scrambling allows some features other than [PF] to be carried along as a free ride. Thus, in (2b), the [SF] of karera-o is carried along as a free ride by application of scrambling to this object, and hence can serve as the antecedent of otagai. Note that in this case, the produced chain is nothing different from the one in (1c). The only difference between scrambling and scope-driven movement in this case comes down to what is the trigger of producing such a chain. While in the cases of scope-driven movement, [SF] serves as its driving force, in the cases of scrambling, neither [SF] nor [FF] serves as a driving force but they simply can be free riders. Hence, scrambling in principle allows these features to be left behind as an alternative option, unlike the cases of scope-driven movement. From the fact that scrambling does not have any driving force, it follows that it is free from the Last Resort Principle. This guarantees that it can be applied “whenever you like,” but it does not mean that the way it applies is also free. Hoji (1985) notes some Japanese data that suggest that a string-vacuous application of scrambling must be prohibited. Let us consider the following example: (3) [John-ga

kiratteiru]

John-Nom hate

sensei-o

kenasita (koto)

teacher-Acc

criticized fact

‘[e] criticized the teacher John hates.’ This sentence cannot have the reading where John is coreferential to the empty subject. If scrambling could apply string-vacuously, then (3) could have (4b) besides (4a) as its

22

resulting structure: (4) a. *[TP pro1 [vP [John1-ga kiratteiru] sensee-o kenasi]ta] b.

[TP [[John1-ga kiratteiru] sensee-o]2 [TP pro1 [vP t2 kenasi]ta]]

In (4a), pro cannot be coreferential to John, since it c-commands this R-expression, hence violating Condition C of the binding theory. In (4b), in contrast, such coreference should be allowed, since pro does not c-command John. In fact, if pro is replaced by an overt pronoun, the sentence is acceptable, as shown below: (5) [John1-ga John-Nom

kiratteiru] sensei-o hate

kare1-ga

teacher-Acc he-Nom

kenasita (koto) criticized fact

‘The teacher John1 hates, he1 criticized.’ Thus, representation (4b) should be excluded as illegitimate for (3). The intuition behind the ban on string-vacuous application of scrambling is clear: since its main function is to change word order, its application is illegitimate if the resulting structure gives rise to no change of word order. It is suggested in Abe (1993) that this condition follows from the Fewest Steps Condition. It is proposed there that the derivations that should be compared in terms of this condition are those which have the same numeration and the same output for the PF component, where “the same output” is taken as the same sequence of strings with respect to phonetic content. On this assumption, the derivation that leads to (4b) is excluded by the Fewest Steps Condition, since there is another derivation that produces the same PF output, namely, the one that leads to (4a), which involves no application of scrambling. So far we have considered scrambling as movement that carries along only [PF],

23

apart from those cases in which other features are also carried along as a free ride, as illustrated in (2). Hence, this operation is naturally constrained by an economy condition to the effect that it cannot apply if it has no effect on PF output, i.e., no effect on word order. Putting such a free ride case as (2) aside for the moment, I now move on to discuss the second nature of scrambling, i.e., its characterization as movement that produces a semantically significant chain. I argue for the following hypothesis: (6) Scrambling is reinterpreted as operator movement whenever possible. I assume that the operator movement relevant for (6) amounts to the scope-driven movement that is applied in order to license a [Focus] feature. In Abe (forthcoming), it is claimed that among such cases are topicalization, as illustrated in (7a), and heavy NP shift, as illustrated in (7b): (7) a. b.

Mary, John met. John met yesterday [the woman you had been talking about].

Given that no morphological manifestation of agreement seems to be involved in these phenomena, it is not unreasonable to assume that these cases involve scope-driven movement rather than agreement-driven movement. Abe (forthcoming) argues that such movement involves satisfying the feature [Focus] and that this feature is satisfied by moving a phrase carrying it to a “peripheral” position. Abe also argues that topicalization and focalization such as heavy NP shift involve satisfaction of the same feature in CHL, and that the different functions they bear at the LF interface are read off from positions that those phrases carrying [Focus] occupy; in English, for instance, those occupying “left-peripheral” positions are interpreted as topics whereas those

24

occupying “right-peripheral” positions are interpreted as foci. Notice that in such cases as (7a, b), it is not obvious what is the locus of the trigger of the movement involved. Based upon this observation, Abe (2012) proposes that a relevant triggering feature is not carried by a particular head but rather assigned to a whole category that undergoes movement. Abe then puts forward the following hypothesis: (8) The feature [Focus] can be assigned to a syntactic object during the derivation. Since such a language as English does not allow the option of scrambling, the Last Resort Principle dictates that overt operator movement involve the assignment of [Focus] to a phrase before it undergoes movement. Abe (2012) then reasons that since scrambling is free from the Last Resort Principle, such a language as Japanese that does allow scrambling should have the possibility that [Focus] is assigned to a phrase after it undergoes scrambling. Given this possibility, we can restate our hypothesis (6) as something like the following: (9) A chain produced by scrambling is assigned [Focus] whenever possible. Before examining the validity of this hypothesis in the next section, let us discuss how the feature [Focus] is licensed in Japanese. Let us first note that while such cases as (7a, b) involve overt scope-driven movement, Japanese usually exploits particles to mark topic and focus phrases, as shown below: (10) a.

Mary1-wa John-ga pro1 atta. Mary-Top John-Nom

saw

‘Mary, John met.’ 25

b. John-ga

Mary-wa atta.

John-Nom Mary-Foc saw ‘John loves MARY.’ Since Kuno (1973) demonstrated that topicalization such as (10a) is not constrained by the Subjacency Condition, it has been widely assumed that the topic in Japanese is base-generated in a sentence-initial position. The wa-phrase in (10b), which is located in its original position, is named by Kuno (1973) contrastive -wa, which conveys the meaning of “not others but X” or “at least X.” That this case involves the same morphological form -wa as the topic case (10a) indicates that there is some common property shared by topicalization and focalization; under the present assumptions, it is natural to claim that the shared property is that -wa carries a [Focus] feature. Given the above assumptions, this feature must be licensed in a peripheral position. In such a case of topicalization as (10a), the [Focus] feature borne by Mary-wa is immediately licensed when this phrase is introduced into a sentence-initial position. The more interesting case is focalization. Contrary to a case of focalization such as heavy NP shift in English, this case does not involve overt movement of the focused phrase. It is natural to claim that this is because such a phrase can be identified as focused by overt manifestation of a focus particle (henceforth, FP) attached to it.2

2

In fact, the wa-phrase in (10a) is ambiguously interpreted as topic or contrastive.

See Hoji (1985) for the claim that the contrastive wa-phrase put in a sentence-initial position is derived by movement, possibly scrambling, from such an underlying structure as (10b). See Abe (2012) for relevant discussion.

26

A question immediately arises as to how a phrase carrying an FP is licensed when it appears overtly in its original θ-position. There is evidence that such a phrase undergoes covert scope-driven movement. Aoyagi (1994) observes that Japanese FPs such as -sae ‘even’ and -mo ‘also’, when attached to NPs, show ambiguity with respect to its scope, as illustrated below:3 (11) John-wa [Mary-ga John-Top Mary-Nom

sinabita ringo-mo/-sae

tabeta to]

omotteiru.

wilted

ate

think

apple-also/-even

Comp

‘John thinks that Mary ate also/even a wilted apple.’ i. [embedded scope] John thinks that Mary ate A WILTED APPLE in addition to some other things. ii. [matrix scope] Even for A WILTED APPLE, John has an idea that Mary ate it (in addition to some other idea about some other things).

(Aoyagi 1994:31)

This difference in scope can be naturally captured by assuming that those phrases bearing FPs undergo covert scope-driven movement. Suppose that such phrases carry [Focus] features, like those phrases in English that undergo topicalization and 3

The contrastive -wa is also regarded naturally as an FP. However, it is hard to judge

the scope ambiguity observed in (11) with -wa. I guess that this is why Aoyagi does not include -wa to make relevant examples. The scope ambiguity observed in (11) is also observed by Taglicht (1984) with English only and even. Further, there is one caveat to be noted with (11). Aoyagi (1994) notes that although this example involves a Caseless FP phrase, the latter can be replaced by the Case marked FP phrase sinabita ringo-o-mo/-sae-o and in this case, it is hard to obtain the matrix scope interpretation. Sano (2001) also argues that some FPs such as -made ‘as far as’ do not allow the matrix scope interpretation. I must leave such ramifications aside.

27

focalization, and that these features are licensed by moving their carriers to “peripheral” positions.4 Then, sentence (11) can have the following two LF representations: (12) a.

John-wa [TP sinabita ringo-mo/-sae1 [TP Mary-ga t1 tabeta]] to omotteiru [Focus]

b. [TP sinabita ringo-mo/-sae1 [TP John-wa [CP [TP Mary-ga t1 tabeta] to] [Focus] omotteiru]] (12a) represents the embedded scope of the FP phrase and (12b) its matrix scope. It is expected under this proposal that FP phrases may not extend their scope across islands. Aoyagi (1994) claims that this is in fact borne out by such an example as the following: (13) Mary-ga [e1 gakubusei-zidai-ni Mary-Nom

Barriers-mo/-sae

yonda] hito1-ni

undergraduate-time-at Barriers-also/-even read

atta.

person-Dat met

‘Mary met a person who also/even read Barriers when he/she was in undergraduate school.’

(Aoyagi 1994:32)

This sentence only allows the embedded scope reading as indicated in the English translation, and does not have the matrix scope reading. This is straightforwardly accounted for by assuming that FP phrases carry [Focus] features and hence undergo 4

I will not address here the question why movement for licensing a [Focus] feature is

overt in English topicalization and focalization whereas that for licensing a [Focus] feature of a phrase bearing a FP is covert. I simply follow Abe (forthcoming) in assuming the following: (i)

[Focus] carries along [PF] for being licensed unless the phrase bearing it has an FP.

28

covert movement. Let us now discuss how FP phrases interact in scope. Consider first the following example, taken from Kuroda (1971): (14) John-dake-ga

S.S.-o-sae

John-only-Nom S.S.-Acc-even

yonda. read

‘Only John read even Syntactic Structures.’ Kuroda observes that this sentence allows only the reading in which John-dake-ga ‘only John’ takes scope over S.S.-o-sae ‘even S.S.’ and crucially it does not have the reading of the opposite scope order; this is called “rigidity” effect. Thus, (14) can be properly uttered, for instance, in the situation given in (15a), but not in the situation given in (15b): (15) a.

The students of a linguistics department were expected to read some of Chomsky’s books and most of them refrained from reading S.S. because it was too hard for them. It then turned out that it was only John who read even S.S.

b.

The students of a linguistics department were expected to read some important books of Chomsky’s and S.S. was the most important one, but everyone except John was so lazy that they read only one or two. It then turned out that even S.S. was read only by John.

In order to capture such a rigidity effect of scope, in which a structurally higher phrase necessarily takes scope over a lower one in surface structure, let us first assume that a [Focus] feature can be licensed in situ in a sentence that involves more than one 29

occurrence of a phrase carrying a [Focus] feature, so that we modify the licensing condition on [Focus] into the following:5 (16) [Focus] is licensed (i) in a peripheral position or (ii) in situ when it is bound by another occurrence of [Focus]. We can now account for the rigidity effect in question by a minimality condition on movement such as the following:6 (17) Crossing Constraint on Features (CCF) A feature [X]1 cannot cross another feature [X]2 unless it is carried along as a free ride. Here “crossing” is defined based upon c-command: α crosses β iff movement of α creates a structure in which α c-commands β and β (asymmetrically) c-commands the trace of α created by the movement in question. Ignoring the unless-clause in (17) for the moment, let us consider how this minimality condition captures such a rigidity effect as observed in (14). Let us consider the following derivation of this sentence: (18) a.

[TP John-dake-ga S.S.-o-sae yonda] [Focus]

5

[Focus]

Here I follow Abe (forthcoming) in assuming that peripheral positions are either

specifier or adjoined positions of functional categories. See Section 2.5 for related discussion. 6

See Abe (forthcoming) for a more precise formulation of the relevant minimality

condition in terms of minimal Search under the probe-goal system assumed in Chomsky (2008).

30

b. *[TP S.S.-o-sae1 [TP John-dake-ga t1 yonda]] [Focus]

[Focus]

(18b) is derived from (18a) by adjoining S.S.-o-sae ‘even S.S.’ covertly to TP to satisfy its [Focus]. Given the standard assumption that the c-command domain of an element corresponds to its scope domain, this representation would represent the reading in which S.S.-o-sae takes scope over John-dake-ga ‘only John’. However, the covert movement in question violates the CCF, since it crosses the subject, which also carries [Focus]. This accounts for why sentence (14) cannot have the reading in question. How about the reading of the opposite scope order? I assume that (18a) is the LF representation expressing this reading. According to the licensing condition on [Focus] given in (17), the [Focus] feature of John-dake-ga is licensed in a peripheral position and that of S.S.-o-sae is licensed in situ since it is bound by John-dake-ga. We can motivate the need of a minimality condition such as the CCF independently with such data involving FP phrases. Let us consider the following example taken from Sano (2001): (19) a.

Taro-wa [Aiko-ga

Jiro-no

okane-sae

nusunda] to

Taro-Top Aiko-Nom Jiro-Gen money-even stole

syutyoosita.

Comp claimed

‘Taro claimed that Aiko had stolen even Jiro’s money.’ b.

Taro-wa [Aiko-wa

Jiro-no

okane-sae

Taro-Top Aiko-Top Jiro-Gen money-even

nusunda] to stole

syutyoosita.

Comp claimed (Sano 2001:20)

(19a) has not only the embedded scope reading of Jiro-no okane-sae ‘even Jiro’s money’ 31

but also its matrix scope reading, exactly like (11). As for (19b), where the embedded subject is changed into Aiko-wa ‘Aiko-Foc’, which involves contrastive -wa, Sano claims, as I think correctly, that it lacks the matrix reading of Jiro-no okane-sae. Given that a phrase with contrastive -wa carries [Focus], this fact follows straightforwardly from the CCF: Jiro-no okane-sae cannot cross Aiko-wa to move to a peripheral position covertly in the matrix clause to license its [Focus] feature, since the latter phrase also has the same feature.7 Keeping the above assumptions in mind, let us now consider what readings obtain when the object S.S.-o-sae is overtly preposed to the top of the sentence in (14): (20) S.S.-o-sae

John-dake-ga

S.S.-Acc-even John-only-Nom

yonda. read

There are, in principle, two possible derivations for this sentence: (21) a.

[TP S.S.-o-sae1 [TP John-dake-ga t1 yonda]] [Focus]

b.

[Focus]

[TP S.S.-o-sae1 [TP John-dake-ga t1 [Focus]

yonda]]

[Focus]

While in (21b), S.S.-o-sae undergoes scrambling, carrying along only its [PF], it carries along its [Focus] as well in (21a). The latter case appears to violate the CCF since the

7

Sano (2001) also attributes the unacceptability of the matrix reading of Jiro-no

okane-sae in (19b) to an intervention effect, though it remains to be seen whether the whole theory he assumes is compatible with that assumed here. See Sano (2001) for his original analysis of FP phrases and their licensing.

32

object S.S.-o-sae crosses the subject John-dake-ga when it moves to the top of the sentence and both phrases carry [Focus]. Here the unless-clause stated in (17) is relevant since the [Focus] feature of the scrambled phrase may be taken to be carried along as a free ride, hence free from a violation of the CCF.8 Note that in these representations, all occurrences of [Focus] are licensed according to the licensing condition on [Focus] given in (16), and that these two representations express different readings with respect to scope: while (21a) represents the reading where S.S.-o-sae takes scope over John-dake-ga, (21b) represents the reading of the opposite scope. Considering the possible interpretations of (20), Kuroda (1971) finds difficulty getting any clear interpretation of this sentence, but as far as I can determine, if this sentence has any interpretation at all, it allows only the reading in which S.S.-o-sae takes scope over John-dake-ga and crucially does not have the reading of the opposite scope, namely, the one that is the only available reading for (14). Thus, (20) is true in the situation given in (15b), but not in (15a). This indicates that when a focused phrase undergoes scrambling, it must be licensed in its scrambled position, heading a focus chain. Thus, this lends initial support to our hypothesis that scrambling is reinterpreted as operator movement whenever possible, as stated in (6). Now the cases we have been considering so far involve inherently focused phrases bearing FPs. In the next section, we deal with cases where scrambled phrases are not inherently assigned [Focus] and give support to the version of our hypothesis stated in (9), reproduced below: (22) A chain produced by scrambling is assigned [Focus] whenever possible. 8

See Section 2.5 for exactly how the free ride exemption from the CCF works.

33

2.2

Scrambling of Wh-Phrases and QPs in Japanese

I argue in what follows that scrambling of wh-phrases and QPs in Japanese produces focus chains unless the derivations crash, following our hypothesis just stated above. It is well known that QPs and wh-phrases in some languages behave as if inherently focused, hence undergoing overt movement to focus positions, as witnessed by Hungarian QPs and Serbo-Croatian wh-phrases (see Stjepanović (1995) and Bošković (1997) for the latter). Given these facts, it is not unreasonable to claim that QPs and wh-phrases in general have some close relationship to focus features that might be ascribed to semantic reasons and that they differ among languages as to whether they are required to have an overt manifestation of this relationship. Though Japanese is not required to manifest it in the form of overt movement, I would like to argue that when QPs and wh-phrases undergo scrambling, their property of “being closely related to focus” reveals itself; that is, the movement in question produces a focus chain.9 Let us first consider the derivations of the following wh-question sentences: 9

Given that topicalization and focalization are not compatible with each other, the fact

that QPs and wh-phrases in Japanese are not used with the topic marker -wa, as shown below, will indicate that they function like focused phrases. (i) a.

Dare-ga/*-wa

kita no?

who-Nom/-Top came Q ‘Who came?’ b.

Daremo-ga/*-wa

kita.

everyone-Nom/-Top came ‘Everyone came.’

34

(23) a.

John-ga

Mary-ni nani-o

agetanodesu ka?

John-Nom Mary-Dat what-Acc gave

Q

‘What did John give to Mary?’ b. Nani-o1

John-ga

Mary-ni t1 agetanodesu ka?

what-Acc John-Nom Mary-Dat

gave

Q

(23b) is derived from (23a) by scrambling the wh-phrase nani-o ‘what-Acc’ to the top of the sentence. Let us first assume, following Abe (1997), that a wh-phrase has a [WH] feature and that this feature can be licensed in the following way: (24) [WH] is licensed by means of being bound by a [+WH] Comp. Given this, (23a) has the following representation: (25) [CP [TP John-ga Mary-ni nani-o agetanodesu] ka] [WH]

[+WH]

In this representation, the [WH] feature of nani-o is licensed by the [+WH] Comp ka, since the latter c-commands the former. Let us now consider (23b), to which we could assign the following three representations under the present assumptions:10 (26) a. *[CP [TP nani-o1 John-ga Mary-ni t1 agetanodesu] ka] [WH]

[+WH]

b. *[CP [TP nani-o1 John-ga Mary-ni t1 agetanodesu] ka] [WH]

10

[+WH]

Here it is tacitly assumed that the features [Focus] and [WH] cannot be separated,

probably because of the “close relationship” they bear. See the next section for the claim that these two features are in fact merged into one.

35

c.

[CP [TP nani-o1 John-ga Mary-ni t1 agetanodesu] ka] [WH][Focus]

[+WH]

In (26a), nani-o has undergone “genuine” scrambling, namely movement in which only the [PF] of nani-o is carried along. This representation has nothing different from (25) in the respects relevant for LF licensing and interpretation. (26b) also involves scrambling of nani-o, but in this case, [WH] is carried along as a free ride by this movement. In (26c), [Focus] is assigned to nani-o after this phrase has undergone scrambling, carrying along its [WH] feature. Given our hypothesis in (22), the derivations that lead to (26a) and (26b) should be excluded as illegitimate since (26c) is a possible derivation for (23b). Hence, the chain produced by scrambling nani-o is interpreted as a focus chain. The above mechanism works in much the same way for scrambling QPs. Let us consider the following examples: (27) a.

John-ga

Mary-ni daremo-o

syookaisita.

John-Nom Mary-Dat everyone-Acc introduced ‘John introduced everyone to Mary.’ b. Daremo-o1

John-ga

Mary-ni t1 syookaisita.

everyone-Acc John-Nom Mary-Dat

introduced

Here again, (27b) is derived from (27a) by scrambling the QP daremo-o ‘everyone-Acc’ to the top of the sentence. Let us suppose, following Abe (1993), that a QP carries a [Scope] feature and the latter is licensed in a position where the QP can take scope over a clause, typically in a TP-adjoined position. (27a) will then have the following LF 36

representation:11 (28) [TP 1 [TP John-ga Mary-ni daremo-o1 syookaisita]] [Scope] In this representation, daremo-o ‘everyone-Acc’ adjoins covertly to TP to license its [Scope] feature. Let us now consider the possible derivations of (27b); again there are three possibilities: (29) a. *[TP daremo-o1 [TP John-ga Mary-ni

t1

syookaisita]]

[Scope] b. *[TP daremo-o1 [TP John-ga Mary-ni t1 syookaisita]] [Scope] c.

[TP daremo-o1 [TP John-ga Mary-ni t1 syookaisita]] [Scope][Focus]

In (29a), daremo-o has undergone genuine scrambling, in which only its [PF] is carried along. In (29b), [Scope] is carried along as a free ride by this movement. In (29c), [Focus] is assigned to daremo-o after this phrase has undergone movement. Our hypothesis in (22) dictates that the derivations indicated in (29a) and (29b) be excluded as illegitimate since there is an alternative derivation, namely the one indicated in (29c), which produces a focus chain. I argue in what follows that the present mechanism of 11

In what follows, we represent a chain produced by covert movement in the way

standardly assumed in the copy theory of movement with the top copy enclosed with angled brackets. On the other hand, we follow a more standard way of representing a chain produced by overt movement, using a trace, for just ease of presentation, unless otherwise indicated.

37

licensing [WH] and [Scope] features, augmented with the hypothesis in (22), is in fact empirically well-supported in dealing with overt movement of wh-phrases and QPs.

2.3

An Apparent Paradox in Long-Distance Scrambling of Wh-Phrases

Saito (1989) argues that scrambling differs from other types of A’-movement in that it does not establish a semantically significant operator-variable chain but rather that it is a semantically vacuous movement. In order to capture this property, he proposes that scrambling can be freely undone in the LF component, unlike topicalization. To illustrate this point, let us first consider the following sentences: (30) a.??Who said that [the man that bought what]1, John knows whether Mary likes t1? b. *Mary thinks that [the man that bought what]1, John knows who likes t1. (Saito 1989:188) In each sentence, the man that bought what is topicalized in the embedded clause, and this movement violates the Wh-Island Condition. This is reflected in the slight degradation of (30a). Notice, however, that (30b) is much worse than (30a). Saito attributes this contrast to the Proper Binding Condition (henceforth, PBC), which requires that traces be bound, under the assumption that wh-phrases in situ move in the LF component. Under the present assumption in which wh-phrases in situ can be licensed without recourse to covert movement, we can attribute the contrast in question to the licensing condition on [WH] given in (24), repeated below: (31) [WH] is licensed by means of being bound by a [+WH] Comp. 38

The [WH] feature borne by what in (30a) is licensed according to this condition, since it is bound by the matrix [+WH] Comp. In contrast, the [WH] feature borne by what in (30b) remains unlicensed, since it is not bound by any [+WH] Comp; hence the total ungrammaticality of this sentence. Keeping this in mind, let us consider the following Japanese sentence, reproduced from (1.6): (32) ?Dono

hon -o1 [Mary-ga

[John-ga t1 tosyokan-kara karidasita

ka]

which book-Acc Mary-Nom John-Nom library-from checked-out Q siri-tagatteiru] (koto) want-to-know fact ‘Lit. Which book, Mary wants to know Q John checked out from the library.’ In this sentence, the wh-phrase dono hon-o ‘which book-Acc’ is scrambled out of the embedded question clause. If this phrase remains at the top of the sentence, then its [WH] feature will not be licensed according to condition (31), since it is not bound by ka, a marker of [+WH] Comp, in the embedded clause. Despite this, (32) is acceptable. To account for this fact, Saito proposes that scrambling can be freely undone in the LF component due to its nature as semantically vacuous movement. Under this proposal, the wh-phrase dono hon-o, which has undergone long-distance scrambling, can be moved back to its original position, in which its [WH] feature can be licensed by the embedded [+WH] Comp. On the other hand, Takahashi (1993) shows that certain cases of long-distance scrambling of wh-phrases exhibit the property of syntactic wh-movement in Japanese. 39

The relevant examples are reproduced from (1.5): (33) a.

John-wa [Mary-ga

nani-o

tabeta ka] siri-tagatteiru

John-Top Mary-Nom what-Acc ate

Q

want-to-know

no? Q

‘Does John want to know what Mary ate?’ or ‘What does John want to know whether Mary ate?’ b.

Nani1-o

John-wa [Mary-ga t1 tabeta ka]

what-Acc John-Top Mary-Nom ate

Q

siri-tagatteiru no? want-to-know Q

‘What does John want to know whether Mary ate?’ In each of these examples, there are two Q-markers; one is ka in the embedded clause and the other is no in the matrix clause. Takahashi observes that (33a) is ambiguous depending upon whether the wh-phrase in situ nani ‘what’ takes scope over the embedded clause or the matrix clause.12 When nani takes embedded scope, the whole sentence is interpreted as a yes-no question, and when it takes matrix scope, the embedded ka is interpreted as ‘whether’. Takahashi makes an interesting observation with (33b), where nani-o is scrambled out of the embedded question clause. He observes that in this sentence, nani can only take matrix scope. This appears to illustrate a case where LF undoing is prohibited. What is, then, the crucial difference between a case such as (32) that allows LF undoing and a case such as (33b) that prohibits it? Takahashi gives the following descriptive generalization:

12

Though I agree with Nishigauchi (1990) that the matrix scope reading is very hard to

get. I will briefly discuss this matter toward the end of this section. See also Abe (forthcoming) for relevant discussion.

40

(34) Movement of a wh-phrase to the initial position of a clause headed by a [+WH] Comp counts as wh-movement in Japanese. Here, the hidden assumption is that if a given instance of movement counts as operator movement such as wh-movement and topicalization, it cannot be undone in the LF component, as shown in (30b). According to this generalization, the movement of nani-o in (33b) counts as an instance of wh-movement, since the matrix clause is headed by a [+WH] Comp, and hence it cannot be undone. Thus, (33b) has only the matrix scope reading of nani. On the other hand, the movement of dono hon-o in (32) does not count as an instance of wh-movement, since the matrix clause is headed by a [-WH] Comp, and hence it can be undone. Thus, this wh-phrase properly takes embedded scope. Takahashi (1993) gives an account for the generalization (34). First, he assumes that A-scrambling is movement to Spec-IP, which amounts to Spec-TP under the present assumptions, whereas A’-scrambling is movement to Spec-CP. He follows Mahajan (1990), Tada (1993) and Saito (1992) in that clause-internal scrambling can be A- or A’-movement whereas long-distance scrambling is uniformly A’-movement. It then follows that long-distance scrambling is always movement to Spec-CP. Second, Takahashi assumes that the wh-feature of a wh-phrase must be checked off under the Spec/head configuration with a [+WH] Comp. Finally, he assumes, following Epstein (1992), the economy principle proposed by Chomsky (1991), which prohibits unnecessary steps. Keeping these assumptions in mind, let us see how the relevant contrast shown in (32) and (33b) is accounted for. These sentences should have the 41

following schematic structures: (35) a.

[CP dono hon-o1 [C’ [TP ... [CP ... t1 ...+Q]]] -Q]

b. [CP nani-o1 [C’ [TP ... [CP ... t1 ...+Q]]] +Q] Both dono hon-o and nani-o have undergone long-distance scrambling, and hence occupy the matrix Spec-CP according to the above assumptions. In (35b), the wh-feature of nani-o is checked off in this position, so that undoing is an unnecessary step. Hence it is prohibited according to the economy principle, and as a result, nani can only take matrix scope. In (35a), by contrast, the wh-feature of dono hon cannot be checked off in this position, so that undoing is a necessary step to satisfy this morphological feature. Hence, dono hon can take embedded scope. Thus, Takahashi convincingly shows that the LF undoing operation is not a free option but is subject to the economy principle prohibiting unnecessary steps. Although his conclusion seems to be correct, his analysis faces a couple of technical problems. First and most importantly, it is a sheer stipulation that long-distance scrambling is movement to Spec-CP. As far as I can see, there is no independent evidence for this stipulation; thus, there is no motivation other than to account for the data that we have been considering. Second, he assumes that the wh-feature of a wh-pharse is checked off under the Spec/head configuration. This implies that wh-phrases in situ must also move to Spec-CP in the LF component, as claimed by Huang (1982). But it has been standardly assumed under the Minimalist Program since Chomsky (1993) that wh-phrases in situ do not have to move at LF and are licensed in situ in a way that does not involve movement; by way of binding, 42

according to Abe (1993) and Tsai (1994). Thus, it is worthwhile to pursue a possibility of modifying Takahashi’s (1993) approach to the LF undoing operation without assuming covert LF wh-movement. Recall that we have been assuming the licensing condition on [WH] features by means of binding, as repeated below: (36) [WH] is licensed by means of being bound by a [+WH] Comp. I will now show how to derive the facts given in (32) and (33b) under our assumptions concerning scrambling introduced in the preceding sections. The explanations to be offered are crucially based upon Takahashi’s (1993) insights on the last resort nature of the LF undoing effects, while getting over the technical problems pointed out in the preceding paragraph. I follow Abe (1997) in assuming the following: (37) The licensing condition on [WH] in (36) is a condition for convergence. With this in mind, let us first consider the possible derivations for (33b): (38) a. *[CP [TP nani1-o John-wa [CP [TP Mary-ga t1 ([WH]) b. [CP [TP nani1-o

tabeta] ka] siritagatteiru] no]

([WH])

[+WH]

[+WH]

John-wa [CP [TP Mary-ga t1 tabeta] ka] siritagatteiru] no]

[WH][Focus]

[+WH]

[+WH]

(38a) represents the derivation in which nani-o has undergone scrambling and the [WH] feature of nani-o is either left behind or carried along as a free ride; the parentheses put in the two occurrences of [WH] represent this optionality. In this representation, the [WH] feature of nani-o is properly licensed by either the embedded [+WH] Comp or the matrix [+WH] Comp by means of binding when left behind in its original position, 43

while it is licensed by only the matrix [+WH] Comp when carried along by the movement involved. In (38b), on the other hand, [Focus] is assigned to nani-o after this phrase has undergone scrambling. In this case, the [WH] feature of nani-o is licensed only by the matrix [+WH] Comp, since the resulting chain is now considered as a focus chain, hence resisting LF undoing. Given our hypothesis in (22), repeated below, (39) A chain produced by scrambling is assigned [Focus] whenever possible. the derivation that leads to (38a) is excluded as illegitimate since there is an alternative legitimate derivation, namely the one indicated in (38b). This correctly explains the fact that in (33b), nani-o can take only matrix scope. Let us now consider the possible derivations for (32): (40) a.

[CP [TP dono hon-o1 Mary-ga [CP [TP John-ga t1 [WH]

tosyokan-kara karidasita] ka] [+WH]

siritagatteiru]] b. *[CP [TP dono hon-o1 Mary-ga [CP [TP John-ga t1 tosyokan-kara karidasita] ka] [WH][Focus]

[+WH]

siritagatteiru]] Notice that the derivation given in (40b) does not converge since the [WH] feature of dono hon-o is not licensed in this position. Hence, there is no alternative derivation other than (40a) that will lead to convergence. In this representation, the [WH] feature of dono hon-o is properly licensed by the embedded [+WH] Comp when left behind by the movement involved. This correctly captures the undoing effect of dono hon-o in (32). Note that under the present system, undoing is not a real operation but rather is 44

captured as a reflection of genuine instances of scrambling in which no features relevant for LF interpretation are carried along by this operation and hence the scrambled phrases are interpreted in situ at the LF interface. Notice that the present system has enabled us to correctly capture the impossibility of “undoing” in such cases as (33b) and at the same time to get rid of the stipulation made by Takahashi (1993) that long-distance scrambling is a movement to Spec-CP. On top of this conceptual desirability, there seems to be a couple of empirical advantages of our approach over Takahashi’s. Let us first consider the schematic structure given below: (41) [CP ... [CP wh1 [TP ... [CP ... t1 ...+Q]...] -Q] ...+Q] Here, the wh-phrase is scrambled out of the question clause into the intermediate clause headed by [-WH] Comp, and the matrix clause is headed by [+WH] Comp. According to Takahashi’s analysis, the wh-phrase should be moved to the intermediate Spec-CP, and since it cannot be checked off in this position, it is allowed to move to either the matrix Spec-CP or the most embedded Spec-CP. Hence, his theory predicts that a relevant sentence has both the matrix scope and the most embedded scope readings. In contrast, according to my analysis, the scrambled wh-phrase should carry [WH] and [Focus] features in accordance with condition (39) and the [WH] feature should be licensed by the matrix [+WH] Comp by means of binding. Hence, the present theory predicts that the relevant sentence only has the matrix scope reading. In fact, noting the relevance of a sentence of the form given in (41), Takahashi (1993) provides the following data: 45

(42) a.

Kimi-wa [John-ga [Mary-ga you-Top

nani-o

tabeta ka] sitteiru

John-Nom Mary-Nom what-Acc ate

Q

know

to] Comp

omotteiru no? think

Q

‘Do you think that John knows what Mary ate?’ or ‘What do you think that John knows whether Mary ate?’ b.

Kimi-wa

[nani1-o John-ga [Mary-ga t1

tabeta ka] sitteiru

you-Top

what-Acc John-Nom Mary-Nom ate

Q know

to] Comp

omotteiru no? think

Q

‘Lit. Do you think what, John knows Q/whether Mary ate?’ c.

Nani1-o

kimi-wa [John-ga [Mary-ga t1 tabeta ka] sitteiru to]

what-Acc you-Top John-Nom Mary-Nom ate

Q

omotteiru

know Comp think

no? Q ‘What do you think that John knows whether Mary ate?’ (Takahashi 1993:660) In (42a), nani-o stays in its original position, and it can take scope over either the most embedded clause or the matrix clause, though the latter reading is only marginally accepted for some native speakers (cf. fn. 12). In (42c), nani-o is scrambled long-distance to the top of the sentence, and it only takes matrix scope, just as in (33b). The crucial case is (42b), where nani-o is scrambled into the top of the intermediate 46

clause headed by a [-WH] Comp. Takahashi reports that this sentence is exactly what he predicts, that is, nani takes not only matrix scope but also the most embedded scope. On the other hand, one reviewer of Takahashi’s article reports that he/she gets only the matrix scope reading of nani with (42b), which I agree on. Thus, we probably cannot decide which prediction is the right one, solely based upon a sentence like (42b). A second case to test the two approaches is discussed in Abe (1997) in a similar context. This is concerned with the following schematic structure: (43) [CP ... [CP wh1 [TP ... [CP ... t1 ...-Q]...] +Q] ...+Q] In this case, the wh-phrase is scrambled long-distance to the intermediate clause headed by a [+WH] Comp and the matrix clause is also headed by [+WH] Comp. According to Takahashi’s analysis, long-distance scrambling is to Spec-CP and hence the wh-feature of the wh-phrase in question should be checked with the intermediate [+WH] Comp. Hence his theory predicts that a relevant sentence has only the reading where the wh-phrase takes the intermediate scope. On the other hand, according to the present theory, the wh-phrase scrambled long-distance can be licensed by either the intermediate [+WH] Comp or the matrix [+WH] Comp. Hence the present theory predicts that a relevant sentence can have either the intermediate scope or the matrix scope reading. It seems that the relevant data favor the present theory rather than Takahashi’s: (44) a.

Kimi-wa [John-ga [Mary-ga you-Top

nani-o

tabeta to]

John-Nom Mary-Nom what-Acc ate

47

omotteiru ka] sitteiru

Comp think

Q

know

no? Q ‘Do you know what John thinks that Mary ate?’ or ‘What do you know whether John thinks that Mary ate?’ b.

Kimi-wa

[nani-o1 John-ga

[Mary-ga t1 tabeta to]

you-Top

what-Acc John-Nom Mary-Nom ate

Comp

omotteiru ka] think

Q

sitteiru no? know Q ‘Do you know what John thinks that Mary ate?’ or ‘What do you know whether John thinks that Mary ate?’ In (44a), the wh-phrase nani-o stays in its underlying position and this sentence can have both the intermediate scope and the matrix scope readings, though again it is hard to get the matrix scope reading. In (44b), nani-o is moved long-distance to the intermediate clause headed by a [+WH] Comp. It is fairly clear that this sentence allows the matrix scope reading of nani-o. Thus, this lends strong support to the present theory. We have seen that our “focus movement” analysis of long-distance scrambling of a wh-phrase

is

both

conceptually

and

empirically

preferable

to

Takahashi’s

“wh-movement” analysis, which stipulates that such movement is always to Spec-CP. Nonetheless, there is evidence that appears to support Takahashi’s position. This comes from the observation that long-distance movement of a wh-phrase in Japanese exhibits the same restrictions on pied-piping that are observed in typical cases of wh-movement in English. It has been well known that there are some restrictions on what size of 48

phrases that include a wh-head can be pied-piped in wh-movement, as shown below: (45) a.

Which picture of Mary did John see?

b. Whose picture of Mary did John see? c. *A picture of who did John see? d. * A picture that who took did John see? Given the assumption that which and whose in (45a, b) occupy Spec-DP, we can draw the following generalization with respect to pied-piping of wh-phrases: (46) A DP can be pied-piped by wh-movement if its Spec is occupied by a wh-phrase. Interestingly, the availability of pied-piping of wh-phrases in English correlates with the unavailability of undoing of a phrase that includes a wh-phrase in long-distance scrambling in Japanese. Let us consider the following examples: (47) a.

John-wa [Mary-ga

dare-no

hon-o

yonda ka] siri-tagatteiru no?

John-Top Mary-Nom who-Gen book-Acc read

Q

want-to-know Q

‘Does John want to know whose book Mary read?’ or ‘Whose book does John want to know if Mary read?’ b.

Dare-no hon-o1

John-wa [Mary-ga t1 yonda ka]

who-Gen book-Acc John-Top Mary-Nom read

Q

siri-tagatteiru no? want-to-know Q

‘Whose book does John want to know if Mary read?’ (48) a.

John-wa [Mary1-ga [pro1 nani-o John-Top Mary-Nom

yonda to]

what-Acc read

itta ka] siri-tagatteiru no?

Comp said Q want-to-know Q

‘Does John want to know what Mary said that she had read?’ or ‘What does John want to know if Mary said that she had read?’ 49

b.

[pro1 nani-o

yonda to]2

what-Acc read

John-wa [Mary1-ga t2 itta ka] siri-tagatteiru no?

Comp John-Top Mary-Nom said Q want-to-know Q

‘Does John want to know what Mary said that she had read?’ or ‘What does John want to know if Mary said that she had read?’ (47b) is derived from (47a) by moving the wh-phrase dare-no hon-o ‘whose book-Acc’ across the embedded interrogative clause to the tope of the sentence, and it only allows the reading in which this wh-phrase takes scope over the matrix clause, just as in (33b). (48b) is derived from (48a) by moving the whole complement clause of itta ‘said’, including nani-o ‘what’, to the top of the sentence, which has much the same configuration as in (47b) in the relevant respects. Contrary to (47b), however, this sentence allows not only the matrix scope reading of nani ‘what’ but also its embedded scope reading. One may reasonably claim that this pattern of facts will give support to Takahashi’s approach, according to which movement of dare-no hon-o ‘whose book’ in (47b) is taken as an instance of wh-movement to the matrix Spec-CP and hence the wh-feature of dare ‘who’ is forced to be checked off with the matrix [+WH] Comp just in the same way as the wh-feature of whose is checked in (45b). In (48b), in contrast, movement of the whole complement clause cannot be taken as an instance of wh-movement, as is clear from the ungrammaticality of (45d) for instance, so that the wh-feature of nani ‘what’ is not forced to be checked off with the matrix [+WH] Comp. This will account for why the embedded scope reading is possible for nani in (48b). Although the above claim sounds straightforward in capturing the correlation of the possibility of pied-piping of wh-phrases in English and the unavailability of LF 50

undoing of scrambled wh-phrases in Japanese, it lacks the exact mechanism of pied-piping that would accounted for the whole paradigm of the above data. I argue that the mechanism involves the way a [WH] feature is percolated up and merged into a [Focus] feature assigned to a phrase including the [WH] feature, and that the Japanese facts in (47) and (48) are given an equally straightforward and yet more principled explanation under the present theory of long-distance scrambling of wh-phrases. In so doing, I propose the following: (49) A phrase that undergoes wh-movement satisfies the following requirements: (i) it carries or includes a [WH] feature. (ii) it is assigned a [Focus] feature. (iii) the [WH] feature is merged into the [Focus] feature. As for the requirement in (49iii), I assume that the merging in question takes place through a percolation mechanism. Thus, a [WH] feature is merged with a [Focus] feature if it percolates up to a phrase to which the [Focus] feature is assigned. I assume the following convention about percolation: (50) Features can percolate up to the next higher projection from either a head or a Spec. With these assumptions in mind, let us consider the derivation of (45b). This sentence has the following representation before wh-movement takes place: (51) [CP [TP John1 Past [vP t1 see [DP[Focus] whose picture of Mary]]]] [WH] In this representation, whose carries a [WH] feature and the DP phrase whose picture of 51

Mary is assigned a [Focus] feature. In this case, the [WH] feature is successfully merged with the [Focus] feature through percolation, since whose occupies the Spec of the DP in question and hence its [WH] feature can percolate up to that DP. Let us express the outcome of the merging of the two features as [WH-F(ocus)]. Then, we can claim that the [+WH] Comp targets the phrase that carries a [WH-F] feature in (51), so that whose picture of Mary is moved to its Spec. Let us now consider the derivation of (45c), which will have the following representation before wh-movement takes place: (52) *[CP [TP John1 Past [vP t1 see [DP[Focus] a picture of who]]]] [WH] In this case, the [WH] feature of who cannot be merged with the [Focus] feature assigned to the DP a picture of who, since who occupies a position inside the complement position of that DP and hence its [WH] feature cannot reach the top of that DP. This accounts for the ungrammaticality of (45c). Let us now consider the Japanese examples in (47) and (48). Given the above discussion of the pied-piping mechanism, we can provide a rather straightforward definition of wh-phrases: (53) Wh-phrases are ones to which [WH] features percolate up to [Focus] phrases. Then, what we want to claim is that when a phrase that includes a [WH] feature is overtly moved, it must be taken as a wh-phrase in order to be licensed by a [+WH] Comp. We can express it as a licensing condition on [WH] in the following way: (54) A [+WH] Comp cannot license a [WH] feature that is included in a [Focus] phrase. 52

From this it follows that: (55) If a phrase that includes a [WH] feature is assigned a [Focus] feature, these features must be merged into a [WH-F] feature. With this in mind, let us first consider the possible derivations of (47b): (56) a. *[CP dare-no hon-o1 John-wa [CP Mary-ga t1 ([WH])

yonda ka] sirit-agatteiru no]

([WH])

b. [CP dare-no hon-o1 John-wa [CP Mary-ga t1 [WH-F]

[+WH]

[+WH]

yonda ka] siri-tagatteiru no] [+WH]

[+WH]

(56a) represents the derivation in which dare-no hon-o ‘whose book-Acc’ undergoes scrambling, with its [WH] feature either left behind or carried along as a free ride. In (56b), dare-no hon-o is assigned a [Focus] feature after it undergoes scrambling. Notice that in this case, the [WH] feature of dare is successfully percolated up to the phrase dare-no hon-o and merged with its [Focus] feature since dare occupies the Spec of that DP. Hence, condition (39) excludes the derivation in (56a), since there is an alternative derivation, given in (56b), which involves assignment of [Focus] to the scrambled phrase. In this representation, the [WH-F] is properly licensed by the matrix [+WH] Comp but not by the embedded one. This explains the fact that (47b) only allows the matrix scope reading of dare ‘who’. Let us now consider the possible derivations of (48b): (57) a.

[CP [pro1 nani-o yonda to]2 John-wa [CP Mary1-ga t2 itta ka] siri-tagatteiru no] [WH]

53

b. [CP pro1 nani-o yonda to]2 John-wa [CP Mary1-ga t2 itta ka] siri-tagatteiru no] [WH] c. *[CP [CP[Focus] pro1 nani-o yonda to]2 John-wa [CP Mary1-ga t2 itta ka] [WH] siri-tagatteiru no] (57a) represents the derivation in which only [PF] is carried along by scrambling the complement clause of itta ‘said’,13 whereas (57b) represents the derivation in which the [WH] feature of nani ‘what’ is carried along as a free ride by applying scrambling to the same clause. (57c) represents the derivation in which [Focus] is assigned to the whole complement clause in question after the latter undergoes scrambling. In this case, the [WH] feature of nani cannot be merged with the [Focus] feature due to the percolation convention in (50), which results in a violation of condition (55). This makes condition (39) inapplicable to those derivations in (57). Hence, both (57a) and (57b) represent the legitimate derivations for (48b). In (57a), the [WH] feature is licensed by either the embedded [+WH] or the matrix Comp, and hence it represents not only the embedded scope reading but also the matrix scope reading for (48b). In (57b), on the other hand, the [WH] feature is licensed only by the matrix [+WH] Comp, and hence it represents

13

It is natural to assume that in this representation, the clause scrambled long-distance

is deleted at LF, since it has no semantic content. See Section 2.4 for relevant discussion. Notice that we have been using trace to indicate the position from which overt movement has taken place but that this is just a matter of exposition. In particular, the full content of the [SF] and [FF] of the clause moved to the top of the sentence in (57a) must remain in the position indicated by trace.

54

the matrix scope reading of (48b). This representation appears to be redundant for the purpose of representing the possible readings of the sentence in question, since (57a) alone represents both readings available for it. This will not be the case, however, taking into consideration the fact that we obtain the matrix scope reading of (48b) quite easily, compared with (48a). Recall that I have pointed out again and again that the matrix scope reading is hard to obtain in (48a) and other examples of the same configuration at least for some native speakers. We can reasonably attribute this fact to some minimality constraint that requires that the licensing of [WH] features by means of binding be done in as minimal a domain as possible in a representation such as (57a) as well as the one for (48a), given below:14 (58) [CP John-wa [CP Mary1-ga [CP pro1 nani-o yonda to] itta ka] siritagatteiru no] [WH]

[+WH]

[+WH]

The minimality constraint in question will dictate that the [WH] feature of nani in both (57a) and (58) be licensed by the embedded [+WH] Comp. This will account for the fact that the matrix scope reading is hard to obtain for (48a). The fact that the same reading is easy to obtain for (48b), then, indicates that this sentence has a possible representation other than (57a) which represents its matrix scope reading. The present theory provides the one that serves for this purpose, namely (57b).

2.3.1

Superiority Effects

Takahashi (1993) gives further argument for his claim that long-distance scrambling of 14

See Abe (1993, forthcoming) for relevant discussion on this minimality constraint.

55

wh-phrases behaves like wh-movement. This is concerned with superiority effects, as illustrated below: (59) a.

Who saw what?

b. *What did who see? These effects will be characterized as a descriptive generalization in the following way: (60) When more than one wh-phrase appears in a sentence, the structurally highest one must undergo wh-movement. The ungrammaticality of (59b) is, then, attributed to the fact that the lower wh-phrase has undergone wh-movement. With this in mind, let us consider the following Japanese examples, provided by Takahashi (1993): (61) a.

John-ga dare-ni

[Mary-ga

John-Nom who-Dat Mary-Nom

nani-o

tabeta to]

what-Acc ate

itta no?

Comp said Q

‘Who did John tell that Mary ate what?’ b.?? Nani-o1 John-ga

dare-ni [Mary-ga t1 tabeta to]

what-Acc John-Nom who-Dat Mary-Nom ate

itta no?

Comp said Q (Takahashi 1993:664)

(61b) is derived from (61a) by scrambling nani-o ‘what-Acc’ long-distance to the top of the sentence. Takahashi (1993) claims that if we take this movement as an instance of wh-movement, then the degradation of (61b) can be attributed to a superiority effect, since in this sentence the lower wh-phrase has undergone movement across the higher one dare-ni ‘who-Dat’. The present theory of long-distance scrambling of wh-phrases can provide a 56

principled explanation for this fact. First, the superiority effects under consideration will follow from the CCF stated in (17), repeated below: (62) Crossing Constraint on Features (CCF) A feature [X]1 cannot cross another feature [X]2 unless it is carried along as a free ride. Given that a wh-phrase has a [WH] feature, both sentences in (59) will have the following representation before wh-movement takes place: (63) [CP [TP who1 Past [vP t1 see what]]] [WH]

[WH]

Under the present mechanism of wh-movement, (59a) will be derived from (63) by assigning a [Focus] feature to who, merging [WH] and [Focus] into [WH-F], and then moving this wh-phrase to the matrix Spec-CP. Likewise, (59b) will be derived from (63) by assigning a [Focus] feature to what, merging [WH] and [Focus] into [WH-F] and then moving this wh-phrase to the matrix Spec-CP. Suppose that [WH] and [WH-F] are classified into the same wh-feature for the sake of the way the CCF in (62) is evaluated. Then, the wh-movement involved in deriving (59b) will violate the CCF, since the [WH-F] carried by what crosses the [WH] of who. Let us now consider (61b), a Japanese counterpart exhibiting a superiority effect according to Takahashi’s (1993) analysis. There are at least two possible derivations for this sentence: (64) a. *[CP nani-o1 [TP John-ga dare-ni [CP Mary-ga [WH]

t1 [WH]

57

tabeta to] itta] no]

b. [CP nani-o1 [TP John-ga dare-ni [CP Mary-ga t1 tabeta to] itta] no] [WH-F]

[WH]

In (64a), nani-o has undergone scrambling with its [PF] alone carried along. In (64b), a [Focus] feature is assigned to nani-o after the latter has undergone scrambling and then this feature is merged with [WH] into [WH-F]. This derivation induces a violation of the CCF for the same reason as (59b), whereas (64a) does not violate this condition. Condition (39), reproduced below, however, excludes the derivation given in (64a). (65) A chain produced by scrambling is assigned [Focus] whenever possible. As a result, there is no derivation that is free from a violation of either the condition on [Focus] assignment or the CCF; hence this accounts for the ungrammaticality of (61b). Note that we do not want to claim that a violation of the CCF allows us to take the option of not assigning [Focus] to the relevant scrambling chain. Note further that the CCF cannot be taken as a strictly derivational constraint since [Focus] is assigned to a resulting chain right after scramling takes place. For this reason, I characterize this constraint as one on chains: (66) Crossing Constraint on Features (CCF) Given a chain C = (α1, … αi, … αn), *αi [X]

…

β …

αi+1

[X]

unless feature [X] is carried along as a free ride. Here the unless-clause is interpreted as follows: Suppose that a chain is characterized as a Y-chain in terms of the feature carried by α1 such as a [Focus]-chain, [Scope]-chain 58

and [WH]-chain. Then feature [X] is taken as a feature carried along as a free ride just in case where X ≠ Y. With this new characterization of the CCF, we correctly capture the fact that a violation of the CCF does not affect the determination of whether a chain produced by scrambling is assigned [Focus]; here the qualification “whenever possible” in (65) is understand as seeking the convergence of a derivation. The claim that superiority effects are relevant for the degradation of such a Japanese sentence as (61b) is supported by the fact that when one wh-phrase is not in the c-command relation with the other in much the same configuration as that of (61b), these effects go away. Consider the following examples:15 (67) a.

John-ga dare-ni [Mary-ga [dare-ga

tutta

sakana-o] tabeta to]

John-Nom who-Dat Mary-Nom who-Nom angled fish-Acc ate

Comp

itta no? said Q ‘Who did John tell that Mary ate fish that who angled for?’ b. [Dare-ga tutta who-Nom angled

15

sakana-o]1 John-ga fish-Acc

dare-ni [Mary-ga t1 tabeta to]

John-Nom who-Dat Mary-Nom ate

Comp

It is well known that the c-command relation is necessary for one wh-phrase to block

the movement of the other in the superiority phenomena in English (example (i) is taken from Fiengo et al. 1988 and those in (ii) from Oka 1993a, b): (i) a. b.

What did people from where try to buy t? Who did pictures of who please t?

(Fiengo et al. 1988:83)

(ii) a. ?Whom did you persuade friends of t to buy what? b. ?What did you persuade friends of whom to buy t?

59

(Oka 1993b:257)

itta no? said Q (67b) is derived from (67a) by preposing the whole object of tabeta ‘ate’, which includes a wh-phrase dare-ga ‘who-Nom’. No significant change in acceptability is observed with this sentence, compared with (67a); the latter sentence is a somewhat awkward and complicated sentence and this property carries over to (67b). Under the present mechanism of movement, (67b) should have the following derivation: (68) [CP [dare-ga tutta sakana-o]1 [TP John-ga dare-ni [CP Mary-ga t1 tabeta to] itta] ([WH])

[WH]

([WH])

no]

Note that in this case, a [Focus] feature cannot be assigned to the scrambled object, since it would fail to merge with [WH], hence violating condition (55). The scrambling involved in (68) does not violate the CCF even if the [WH] feature is carried along, since in such movement, the [WH] feature of dare-ga does not cross that of dare-ni; recall that “crossing” was defined in the following way: α crosses β iff movement of α creates a structure in which α c-commands β and β (asymmetrically) c-commands the trace of α created by the movement in question. Thus we correctly capture the acceptability of (67b).16

16

Takahashi (1993) observes that superiority effects do not occur when clause-internal

scrambling of a wh-phrase is involved, unlike a case such as (61b). I will address this question in Section 2.5. Further, Takahashi claims that long-distance movement of a wh-phrase to the top of a clause headed by a [-WH] Comp does not yield a superiority effect, as shown below:

60

2.3.2

Additional Wh-Effects

There is another piece of evidence for the present mechanism of long-distance scrambling, which is concerned with additional wh-effects, discussed by Saito (1994). He observes that in Japanese an adjunct wh-phrase such as naze ‘why’ can be saved by a higher wh-argument when it appears in a typical island violation configuration, as shown below: (69) a. *John-wa John-Top

[sono hon-o

naze katta]

hito-o

sagasiteiru

that book-Acc why bought person-Acc

no?

is-looking-for Q

‘Lit. Q John is looking for the person [that bought that book why]?’ b.??John-wa John-Top

[nani-o

naze katta]

hito-o

sagasiteiru

what-Acc why bought person-Acc is-looking-for

no? Q

‘Lit. Q John is looking for the person [that bought what why]?’ (i)

John-ga [nani-o1

Bill-ga

dare-ni [Mary-ga t1 tabeta to]

John-Nom what-Acc Bill-Nom who-Dat Mary-Nom ate

itta to]

Comp said Comp

omotteiru no? think

Q

‘Lit. Q John thinks that what, Bill told who that Mary ate?’ (Takahashi 1993:668) Though Takahashi takes this sentence to be acceptable, unlike (61b), it does not seem to me that there is any significant difference in acceptability between the two sentences. Note that Takahashi’s theory will predict that a superiority effect does not occur in such a case since the movement involved is not an instance of wh-movement, whereas my theory will predict that such an effect should occur since the movement involved must be taken as an instance of focus movement.

61

(69a) is a typical instance of island effects, because the adjunct wh-phrase naze appears within a complex NP island. This is expected under the assumption made by Abe (1993) and Tsai (1994) that adjunct wh-phrases in situ must move covertly to Spec-CP to be licensed, unlike argument wh-phrases, which can be licensed by way of binding, as has been assumed here, hence not exhibiting island effects.17 In (69b), the object within the relative clause is changed to a wh-phrase, and the sentence shows dramatic improvement. To account for this contrast, Saito gives the following characterization for the way wh-adjuncts are licensed:18 (70) A wh-adjunct is licensed by means of either (a) or (b): a.

it moves to Spec-CP to agree with a [+WH] Comp.

b. it is adjoined to a higher wh-argument. According to this licensing, naze in (69b) can be licensed by way of adjoining to nani-o 17

Relevant examples are given below:

(i)

Kimi-wa [nani-o

katta]

you-Top what-Acc bought

hito-o

sagasiteiru

no?

person-Acc

be-looking-for Q

‘Lit. Q you are looking for the person [who bought what]?’ (ii)

Kimi-wa [Taroo-ga you-Top

dare-to

hanasita kara]

Taroo-Nom who-with talked

because

sittositeiru no? is-jealous Q

‘Lit. Q you are jealous because [Taroo talked with who]?’ In both sentences, the wh-phrases in situ nani ‘what’ and dare ‘who’ can be licensed by the matrix [+WH] Comp by means of binding, and hence the sentences are grammatical even though these wh-phrases appear within islands. 18

Saito (1994) crucially relies on LF wh-movement in his analysis of additional wh-

effects. Thus his particular analysis is not compatible with the present assumption that wh-arguments do not have to move. (70) is just a descriptive statement of his proposal. See Saito (1994) for details of his analysis.

62

‘what-Acc’ and it does not have to move out of the complex NP island; hence this sentence improves in its acceptability. Considering what are the appropriate syntactic configurations in which such saving effects take place, Sohn (1994) notes that one of the requirements on such effects is that a higher wh-argument that is intended to save an adjunct wh-phrase must be a clause-mate with this adjunct at S-Structure. He provides the following example: (71) *Nani-o1 John-wa [t1 naze katta] hito-o what-Acc John-Top

sagasiteiru

no?

why bought person-Acc is-looking-for Q

‘Lit. What1 is John looking for the person [that bought t1 why]?’ This sentence is derived from (69b) by preposing nani-o ‘what-Acc’, which would serve to save naze from inducing an island violation if it stayed in its original position. The total unacceptability of (71) suggests that the saving effect in question is not at work here. To account for why this is so, Sohn (1994) suggests the possibility that the long-distance scrambling involved in (71) is in fact an instance of wh-movement, as claimed by Takahashi (1993), and hence cannot be undone. Notice that if the scrambled wh-phrase nani-o must stay at the initial position of the sentence, it will not be able to save naze from an island violation; in such a case, naze must move out of the complex NP island. Sohn (1994), however, dismisses this possibility, noting that one of the problems with this approach is that “the same effect obtains even when we embed [71] inside another clause.” (p. 333) Although he does not provide any relevant example, it is clear what kind of sentence he has in mind:

63

(72) *Mary-wa [nani-o1

John-ga [t1 naze katta]

Mary-Top what-Acc John-Nom why bought

hito-o

sagasiteiru

to]

person-Acc is-looking-for Comp

omotteiru no? think

Q

‘Lit. Q Mary thinks that what1, John is looking for the person [that bought t1 why]?’ Here, nani-o is scrambled long-distance to the intermediate clause headed by a [-WH] Comp. This sentence is just as bad as (71). According to Takahashi’s (1993) theory, however, it would be predicted that (72) should be grammatical, since the long-distance scrambling of nani-o in this case is not an instance of wh-movement and hence can be undone at LF, so that nani-o should be able to save naze from an island violation. Under the present theory of movement, in contrast, the ungrammaticality of both (71) and (72) follows straightforwardly. According to this theory, there will be at least two possible derivations for them, as schematically shown below: (73) a. * ... nani-o1 ... [Relative ...

t1 [WH]

b.

naze ... ] ... [WH]

... nani-o1 ... [Relative ... t1 naze ... ] ... [WH-F]

[WH]

Condition (65) excludes the derivation in (73a) that involves a genuine instance of scrambling, since there is an alternative derivation in (73b) that involves assignment of [Focus] to the scrambled wh-phrase. Suppose that the licensing condition given in (70b) that induces the saving effects in question is more strictly stated as follows: 64

(74) The [WH] feature of a wh-adjunct is licensed if this adjunct is adjoined to a higher wh-argument bearing [WH]. Then, naze cannot be saved from an island violation in the configuration given in (73b), since even if naze is adjoined to the trace of nani-o, its [WH] feature will not be licensed by means of (74). This correctly explains why the saving effects of naze are not observed in either (71) or (72).19 19

Sohn (1994) argues that some cases in which the saving effects in question are not

detected can be straightforwardly explained by Oka’s (1989) claim that unlicensed chains created by long-distance scrambling must be undone. He takes the relevant cases to be those in which wh-adjuncts in situ are not clause-mates with higher wh-arguments, as shown by Saito (1994) with the following examples: (i) a. *Kimi-wa [naze you-Top why hito-o

Mary-ni [John-ga

nani-o

katta

tte]

itta]

Mary-Dat John-Nom what-Acc bought Comp said

sagasiteiru

no?

person-Acc is-looking-for Q ‘Lit. Q you are looking for the person [that told Mary [that John bought what] why]?’ b. *Kimi-wa [nani-o1

naze Mary-ni [John-ga t1 katta

tte]

itta]

you-Top what-Acc why Mary-Dat John-Nom bought Comp said hito-o

sagasiteiru

no?

person-Acc is-looking-for Q (ia) is an island violation, since naze is within a complex NP island. In (ib), nani-o is scrambled long-distance just in front of naze. According to the licensing condition in (74), naze should be licensed by means of adjoining to the scrambled wh-argument nani-o, and yet this sentence is unacceptable. Sohn claims that this is straightforwardly explained under the assumption that unlicensed chains created by long-distance scrambling must be undone. Thus, since nani-o in (ib) has undergone long-distance scrambling, it must be undone, so that it cannot serve to save naze. Notice that we

65

2.4

Scrambling of QPs as Focus Movement

In this section, I argue that scrambling of QPs gives rise to a focus chain in Japanese. We are largely concerned with scope interaction among QPs to make this point. For this reason, let us first outline the mechanism of the licensing of QPs and their interpretation proposed by Abe (forthcoming) and slightly modified here in such a way as to be compatible with the present assumptions. There is a well-known fact about scope interaction among QPs according to which the structural height among QPs directly reflects the scope relation among them; this is named the rigidity condition. Japanese is one of the well-attested languages that observe this condition, as shown below: (75) Dareka-ga

daremo-o

hihansita.

someone-Nom everyone-Acc criticized ‘Someone criticized everyone.’

cannot adopt this account since it is not compatible with the present theory of movement, according to which long-distance scrambling must keep the chain it has created as a focus chain whenever possible. Notice further that the ungrammaticality of (ib) is at least partially explained as a violation of the CCF that induces a superiority effect, since in this case, nani-o, which carries a [WH-F] feature, crosses naze, which also carries a [WH] feature. Given Saito’s (1994) claim that (ib) is much worse than those cases that simply induce superiority effects in Japanese, however, this account may not be sufficient. It might be claimed that licensing a wh-adjunct by way of (74) is prohibited if the licensing wh-argument bears [Focus], but more examination is necessary to justify this claim.

66

This sentence reflects the basic word order and it observes the rigidity condition; that is, the structurally higher QP must take scope over the lower one. On the other hand, when the surface order of a sentence is a derived word order, as in the following, (76) Daremo-o1

dareka-ga

t1

hihansita.

everyone-Acc someone-Nom

criticized

then either QP can take scope over the other. This shows that a QP that undergoes scrambling, crossing another QP, makes the scope order ambiguous. In order to capture these facts, let us first assume the following interpretive rules: (77) a.

The scope domain of a [Scope] is its c-command domain.

b. If [Scope]1 asymmetrically c-commands [Scope]2, then [Scope]1 >[Scope]2. c.

If [Scope]1 and [Scope]2 c-command each other, then [Scope]1 > <[Scope]2.

(77a) indicates how the absolute scope of a QP carrying [Scope] is determined, and (77b, c) indicate how the relative scope among QPs is determined. We have assumed in Section 2.2 that [Scope] is satisfied by moving its carrier to a scope-taking position, typically either a TP-adjoined or Spec-TP position. Further, in Abe (forthcoming), the following qualification is added: (78) [Scope] in situ can be satisfied by means of being bound by another instance of [Scope] that satisfies itself by movement. It is intended that in this way of satisfaction of [Scope], the [Scope] in situ takes the same scope domain as that which licenses it. Note that in such a case, the [Scope] in situ is interpreted as under the scope of the other c-commanding [Scope], according to interpretive rule (77b). This way of licensing may be characterized more generally; 67

recall that we have assumed the same sort of licensing with [Focus] features, as stated in (16), and with [WH] features, as stated in (36). Given these assumptions, (75) can have the following LF representation: (79) [TP dareka-ga1 [vP t1 daremo-o hihansita]] [Scope]

[Scope]

In this representation, the [Scope] feature of dareka is satisfied in the Spec-TP position and it also licenses the [Scope] feature of daremo, according to (78). This represents the reading in which the subject QP takes scope over the object QP. In order to capture the rigidity effects of scope interaction, it is necessary to exclude such an LF representation as the following, in which daremo-o is adjoined covertly to TP to represent the reading in which the object QP takes scope over the subject QP: (80) *[TP [TP dareka-ga2 [vP t2 daremo-o hihansita]]] [Scope]

[Scope]

This is straightforwardly ruled out by the CCF, repeated below: (81) Crossing Constraint on Features (CCF) Given a chain C = (α1, … αi, … αn), *αi [X]

…

β …

αi+1

[X]

unless feature [X] is carried along as a free ride. The unless-clause in (81) is relevant to account for the scope ambiguity of (76). Let us now consider possible derivations for this sentence, taking into consideration the condition on scrambling we have been assuming: 68

(82) A chain produced by scrambling is assigned [Focus] whenever possible. Under the present assumptions, (76) can have the following representations: (83) a. *[TP daremo-o1 [TP dareka-ga2 [vP t2 ([Scope]) b. [TP

daremo-o1

[Focus][Scope] c.

[Scope]

hihansita]]]

([Scope])

[TP dareka-ga2 [vP

t2

t1

hihansita]]]

[Scope]

[TP daremo-o1 [TP dareka-ga2 [vP [Focus]

t1

t2

[Scope]

t1

hihansita]]]

[Scope]

(83a) represents the derivation in which daremo-o ‘everyone-Acc’ undergoes scrambling with the [Scope] feature either left behind or carried along as a free ride. This derivation, however, is excluded by condition (82), since there are alternative derivations indicated in (83b, c) that involve assignment of [Focus] to the scrambled phrase. In (83b), [Scope] is taken to be carried along as a free ride since the chain produced by scrambling is now regarded as a focus chain. Hence, even though it crosses that of dareka when its carrier is adjoined to TP, it does not violate the CCF. This represents the reading in which the object QP takes scope over the subject QP. In (83c), the [Scope] feature of daremo is left behind by scrambling and is thus licensed by that of dareka by way of binding according to (78). This represents the reading in which the subject QP takes scope over the object QP. As a result, (76) is scopally ambiguous with (83b) and (83c) for its legitimate LF representations of the two readings. When we consider how long-distance scrambling of QPs affects scope interaction, we find interesting facts that, at first sight, appear to discourage the focus movement 69

analysis of QP scrambling. Oka (1989) and Tada (1993) observe that long-distance scrambling does not change scope order, as illustrated below: (84) a.

Dareka-ga

[sensei-ga

daremo-ni

someone-Nom teacher-Nom everyone-Dat

kisusita to] kissed

sinziteiru.

Comp believe

‘Someone believes that the teacher kissed everyone.’ b. Daremo-ni1

dareka-ga

[sensei-ga t1

kisusita to]

everyone-Dat someone-Nom teacher-Nom kissed

sinziteiru.

Comp believe

‘Everyone, someone believes that the teacher kissed.’ (84b) is derived from (84a) by applying long-distance scrambling to the embedded object daremo-ni ‘everyone-Dat’. Although this QP crosses dareka ‘someone’ in the matrix subject position, this does not make the sentence ambiguous regarding the scope order of these two QPs, unlike such cases of clause-internal scrambling as (76). Oka (1989) and Tada (1993) attribute this fact to the nature of long-distance scrambling as semantically vacuous movement, originally claimed by Saito (1989). Specifically, Oka (1989) claims that long-distance scrambling produces an unlicensed chain according to Full Interpretation and hence must be undone at LF. This accounts for why long-distance scrambling does not extend the scope domain of the scrambled phrase to its landing site. There is a good reason, however, to believe that this account may not be true, since the same pattern of facts also seems to be observed in English, a language supposedly lacking scrambling operations. Reinhart (1976) notes that clause-internal topicalization alters scope order, as shown below: 70

(85) a.

Someone criticized many students.

b. Many students, someone criticized. According to Reinhart’s observation, many students cannot take scope over someone in (85a), but when topicalized, it can take scope over someone, as shown in (85b). Though there is a discrepancy among English native speakers as to whether (85a) is ambiguous, it is interesting to note that the judgments reported by Reinhart show the same pattern as those judgments on Japanese quantifier interaction that involves clause-internal scrambling of a QP, as illustrated in (75) and (76).20 This is predicted under the present analysis. The sentences in (85) will have the following representations: (86) [TP someone1 [vP t1 criticized many students]] [Scope] (87) a.

[TP many students2 [TP someone1 [vP t1 criticized t2]]] [Focus][Scope]

20

[Scope]

[Scope]

There are some speakers that have difficulty getting the wide scope of a topicalized

DP, so that they cannot get the reading in which many students takes scope over someone in (85b). I do not have anything to say about why such a restriction holds, but this difficulty might be gotten over in a case where PP topicalization is involved. Reinhart (1967) provides the following data: (i) a.

Someone found scratches in all of Ben’s pictures.

b.

In all of Ben’s pictures, someone found scratches.

(Reinhart 1967:73)

According to Reinhart, (ia) has only the reading in which someone takes scope over all of Ben’s pictures, while (1b) is ambiguous, depending upon the scope order of these two QPs.

71

b. [TP many students2 [TP someone1 [vP t1 criticized t2]]] [Focus]

[Scope]

[Scope]

In (86), the [Scope] feature of someone is satisfied in the Spec-TP and that of many students is satisfied, according to (78), by means of being bound by that of someone. In (87a), the movement of many students to license its [Focus] feature can carry along its [Scope] feature without violating the CCF, since this feature is carried along as a free ride. This represents the reading where many students takes scope over someone, according to interpretive rule (77b). In (87b), the [Scope] feature of many students is left behind by the movement in question, so that this representation expresses the same distribution of [Scope] features as that given in (86), hence representing the reading in which someone takes scope over many students. Interestingly, when a QP undergoes long-distance topicalization, the topicalized QP cannot take scope over another QP in the same clausal domain, just like Japanese cases involving long-distance scrambling of a QP, as shown below: (88) a.

Someone said that John criticized many students.

b. Many students, someone said that John criticized. It is very hard to get the reading where many students takes scope over someone in (88b), though this reading may be easier to get than in (88a), a situation similar to that observed in the Japanese counterparts. We can provide the same pattern of facts with Japanese cleft constructions; consider the following examples: (89) [Dareka-ga

[sensei-ga t1 kisusita to]

someone-Nom teacher-Nom kissed

sinziteiru no]-wa

Comp believe Comp-Top 72

daremo-ni1/

ooku-no

seito-ni1

da.

everyone-Dat/ many-Gen student-Dat be ‘It is everyone/many students that someone believes that the teacher kissed.’ This cleft sentence does not have the reading in which daremo ‘everyone’ or ooku-no seito ‘many students’ takes scope over dareka ‘someone’. If Hoji (1987) is right in claiming that Japanese cleft constructions such as (89) involve null operator movement, then we must seek an explanation for the above data without recourse to the device of LF undoing. Under our hypothesis that long-distance scrambling of a QP produces a focus chain, it is very likely that such an explanation can be extended to accommodate the pattern of scope interaction that involves long-distance scrambling of a QP, as illustrated in (84b). Abe (forthcoming) argues that the above facts are explained by the clause-boundedness condition on QR, which forces a QP that undergoes long-distance movement to leave its [Scope] feature in the embedded TP-adjoined position on its way. Given this assumption, (84b) will have the following LF representation: (90) [TP daremo-ni1 [TP dareka-ga2 [vP t2 [CP [TP [Focus]

[Scope]

t’1

[TP sensei-ga t1

[Scope]

kisusita]] to] sinziteiru]]] In this representation, daremo-ni, when scrambled long-distance, is adjoined to the embedded TP to satisfy its [Scope] feature in a way that does not violate the clause-boundedness condition. This represents the reading where dareka ‘someone’ takes scope over daremo ‘everyone’. Exactly the same explanation holds for the fact 73

that in (88b) and (89), the QP moved long-distance cannot take scope over the other QP in the matrix subject position. This explanation with recourse to the clause-boundedness condition on QR is supported by the fact that when a QP that undergoes long-distance movement interacts with a QP that occupies the embedded subject position, the sentence becomes ambiguous with respect to the relative scope of the two QPs. Consider the following examples: (91) Many students, John said that someone criticized. (92) Daremo1-ni sensei-ga

[dareka-ga t1

kisusita

everyone-Dat teacher-Nom someone-Nom kissed

to]

sinziteiru.

Comp believe

‘Everyone, the teacher believes that someone kissed.’ (93) [Sensei-ga [dareka-ga t1

kisusita to]

sinziteiru no]-wa

teacher-Nom someone-Nom kissed

Comp believe Comp-Top

daremo-ni1/

da.

ooku-no

seeto-ni1

everyone-Dat/ many-Gen student-Dat be ‘It is everyone/many students that the teacher believes that someone kissed.’ These sentences have the reading where the QP moved long-distance takes scope over the QP in the embedded subject position. To take sentence (92) for illustration, this reading is obtained from the following LF representation: (94) [TP daremo-ni1 [TP sensei-ga [CP [TP [Focus]

t’1 [Scope]

kisusi]ta]] to] sinziteiru]] 74

[TP dareka-ga2 [vP t2 t1 [Scope]

In this representation, the [Scope] feature of daremo ‘everyone’ is licensed in the embedded TP-adjoined position, following the clause-boundedness condition. In this case, this QP c-commands the other QP dareka ‘someone’. Hence, this representation properly represents the reading in which daremo takes scope over dareka. Exactly the same explanation applies to the sentences in (91) and (93) concerning how to derive the reading in which the QP moved long-distance takes scope over the QP in the embedded subject position. To the extent that we can provide a unified explanation for the scope interaction facts that involve not only long-distance operator movement of QPs but also long-distance scrambling of QPs, this will lend strong support to the present hypothesis that scrambling of QPs gives rise to focus chains. We can provide a further argument for this hypothesis. Let us consider the following examples, taken from Abe (1993), which differ from those in (84) only in that dareka ‘someone’ and daremo ‘everyone’ are swapped: (95) a.

Daremo-ga

[sensei-ga

dareka-ni

kisusita to]

sinziteiru.

everyone-Nom teacher-Nom someone-Dat kissed Comp believe ‘Everyone believes that the teacher kissed someone.’ b. Dareka-ni1 someone-Dat

daremo-ga

[sensei-ga t1

kisusita to]

everyone-Nom teacher-Nom kissed

sinziteiru.

Comp believe

‘Someone, everyone believes that the teacher kissed.’ While daremo can take scope over dareka in (95a), (95b) loses this reading and only has the reading where dareka takes scope over daremo. Let us further note that when daremo is put in the embedded subject position in a sentence comparable to (95b), it can 75

take scope over dareka, as shown below: (96) Dareka-ni1

sensei-ga

[daremo-ga t1

someone-Dat teacher-Nom

kisusita to]

sinziteiru.

everyone-Nom kissed Comp believe

‘Someone, the teacher believes that everyone kissed.’ These data undermine Oka’s (1989) claim that chains produced by long-distance scrambling are unlicensed and hence must be undone at LF; if this were the case, then it would be wrongly predicted that in both (96b) and (96), daremo could take scope over dareka. Note further that the same pattern of facts obtains with cleft constructions: (97) a.

[Daremo-ga

[sensei-ga t1

kisusita to]

everyone-Nom teacher-Nom kissed seito-ni1

da.

student-Dat

be

sinziteiru no]-wa

dareka

Comp believe Comp-Top some

‘It is some student that everyone believes that the teacher kissed.’ b. [Sensei-ga [daremo-ga t1

kisusita

teacher-Nom everyone-Nom kissed seito-ni1

to]

sinziteiru no]-wa

dareka

Comp believe Comp-Top some

da.

student-Dat be ‘It is some student that the teacher believes that everyone kissed.’ While it is very hard to get the reading in which daremo takes scope over dareka in (97a), this reading is easily available to (97b). To the extent that Hoji (1987) is right in claiming that Japanese cleft constructions involve null operator movement, it is natural to conjecture that the difference between (95b) and (96) with respect to the availability 76

of the wide scope reading of daremo has to do with the fact that the long-distance scrambling involved exhibits a relevant property of operator movement, so that we can provide a unified account for these cases and cases involving cleft constructions such as those given in (97) under the present assumptions. This conjecture gains support from a similar pattern of English facts observed by Sloan (1991) regarding the scope interaction of wh-phrases and QPs. Consider the following example: (98) Who1 does everyone think you saw t1? This sentence is unambiguous, lacking the reading on which everyone takes scope over who, namely, the distributed question reading in which you are asked about each individual who that person thinks you saw. This contrasts sharply with the following sentence: (99) Who1 do you think everyone saw t1 at the rally? Contrary to (98), this sentence allows the distributed question reading. Let us suppose, following Chomsky (1964) and Kuroda (1965), among others, that interrogative phrases are composed of WH and indefinites; thus, who and what are composed of WH + someone and WH + something, respectively. This amounts to saying, under the present assumptions, that wh-phrases have the features [WH] and [Scope].21 Assuming this, the 21

Given that indefinites allow not only existential but also specific readings, Abe

(1993) assumes that wh-phrases can optionaly carry [Scope] features, so that if they do not, they are interpreted as “specific”, meaning that they are not under the scope of a QP. The same optionality also applies to indefinites themselves, so that if they do not carry [Scope] features, they are interpreted as specific.

77

distributed question reading of (98) could be represented as follows: (100) *[CP who1 does [TP everyone2 [vP t2 think [CP [TP t’1 [WH-F]

[Scope]

[TP you saw t1]]]]]]

[Scope]

Here, who stops by the embedded TP-adjoined position on its way to the matrix Spec-CP to license its [Scope] feature so as not to violate the clause-boundedness condition on QR. Note that this representation has much the same structure in the relevant respects as that for the unavailable reading of (95b) if long-distance scrambling of dareka-ni is taken as focus movement, as shown below: (101) *[TP dareka-ni1 [TP daremo-ga2 [vP t2 [CP [TP t’1 [TP sensei-ga t1 kisusita]] [Focus]

[Scope]

[Scope]

to] sinziteiru]]] Let us now consider the representation for the distributed question reading of (99). Under the present assumptions, this reading will be represented as follows: (102) [CP who1 do [TP you think [CP [TP t’1 [TP everyone2 [vP t2 saw t1 at the [WH-F]

[Scope]

[Scope]

rally]]]]]] In this representation, who stops by the embedded TP-adjoined position to license its [Scope] feature. Let us now consider how the relative scope of the two occurrences of [Scope] is determined in this representation. Let us assume that the definition of c-command relevant for the interpretive rules given in (77) is something like the following: (103) a.

α c-commands β iff every category that dominates α contains β. 78

b.

α contains β iff some segment of α dominates β.

Given this definition, everyone and t’1 in (102) c-command each other; in particular, everyone c-commands t’1 since the category most immediately dominating everyone is the embedded TP and this category contains t’1. Hence, this representation represents the reading in which everyone takes scope over who as well as that of the opposite scope order. Likewise, the representation for the reading of (96) where either daremo or dareka takes scope over the other can have much the same structure in the relevant respects as (102), as shown below: (104) [TP dareka-ni1 [TP sensei-ga [CP [TP t’1 [Focus]

[TP daremo-ga2 [vP t2 t1

[Scope]

[Scope]

kisusi]ta]] to] sinziteiru]] Given the above definition of c-command, daremo-ga and t’1 in (104) c-command each other, and hence this representation represents the ambiguous scope order with respect to dareka and daremo. The question that remains to be answered is what makes the representations in (100) and (101) illegitimate and those in (102) and (104) legitimate. Abe (1993) proposes that the representations given in (100) and (101) are ruled out by what he calls an economy condition on dependency: (105) The Economy Condition on Dependency II (ECDII) If αi is a member of the chain (α1, ... αi, ...), then αi+1 must be the closest possible dependent for αi. “Possible dependent” and “closeness” are defined as follows: (106) α is a possible dependent for β if γ is dependent upon β, and α and γ have the 79

same feature. (107) α is closer to γ than β if α asymmetrically c-commands β. With these assumptions in mind, let us first consider (102) and (104). In these representations, everyone and daremo-ga are possible dependents for who and dareka-ni, respectively, since they share the feature [Scope] with those members adjoined to the embedded TP which are the real dependents of who and dareka-ni. Since these QPs are not closer to who and dareka-ni than those in the embedded TP-adjoined positions according to (107), the latter count as the closest possible dependents for who and dareka-ni, hence satisfying the ECDII. Let us now consider (100) and (101). In these representations, everyone and daremo-ga, which are possible dependents for who and dareka-ni, respectively, are closer to them than their real dependents in the embedded TP-adjoined positions, since they asymmetrically c-command these real dependents. Thus, those in the embedded TP-adjoined positions are not the closest possible dependents for who and dareka-ni, thereby violating the ECDII. Therefore, in sentences (95b) and (98), who and dareka-ni are not allowed to have the feature [Scope], with the consequence that these sentences have only the reading where no scope interaction exists. Notice that it is crucial in the above account of the availability of the wide scope reading of daremo for (95b) and (96) to assume that dareka-ni, scrambled long-distance, serves as an operator leading to a chain, so that the ECDII can properly constrain the dependency relation between the top member of dareka-ni and the next member marking the [Scope] feature. Therefore, if the account provided above is on the right track, then it will lend strong support to the present hypothesis that scrambling of QPs 80

gives rise to focus chains. Recall that this hypothesis is derived from condition (82), reproduced below: (108) A chain produced by scrambling is assigned [Focus] whenever possible. Given this, we predict the following: if we change such a sentence as (95b) into one in which the QP scrambled long-distance is required to undergo “LF undoing” to make the derivation converge, then the sentence should allow the wide scope reading of daremo, because in that case, the QP in question can be interpreted as if it stayed in its original position. This prediction is borne out; consider the following examples: (109) a.

Hutari-no onna-ni1 two-Gen to]

sannin-no otoko-ga

woman-Dat three-Gen man-Nom

[sensei-ga t1 kisusita teacher-Nom kissed

sinziteiru.

Comp believe ‘Two women, three men believe that the teacher kissed.’ b. Dono hutari-no onna-ni1

sannin-no otoko-ga [sensei-ga t1

which two-Gen woman-Dat three-Gen man-Nom teacher-Nom kisusita ka] siri-tagatteiru. kissed

Q

want-to-know

‘Lit. Which two women, three men want to know Q the teacher kissed.’ c.

Dono hutari-no which two-Gen

onna-ni1

sannin-no

otoko-ga [sensei-ga t1

woman-Dat three-Gen

man-Nom teacher-Nom

kisusita ka] siri-tagatteiru no? kissed

Q

want-to-know Q 81

‘Which two women do three men want to know if the teacher kissed?’ (109a) has exactly the same configuration as (95b) with respect to the scope interaction of two QPs in which one QP occupies the matrix subject position and the other is scrambled long-distance to the top of the sentence. Just like (95b), this sentence lacks the reading in which sannin-no otoko ‘three men’ takes scope over hutari-no onna ‘two women’, as predicted. (109b) has much the same configuration as (109a) except that the QP scrambled long-distance in the former bears the wh-determiner dono ‘which’ and needs to take scope in the embedded interrogative clause. Unlike (109a), this sentence has the reading in which sannin-no otoko takes scope over hutari-no onna. This is because the [WH] feature carried by dono hutari-no onna-ni ‘which two women-Dat’ must be licensed by the embedded [+WH] Comp, so that this wh-phrase cannot bear a [Focus] feature, as shown below: (110) *[TP dono hutari-no onna-ni1 [TP sannin-no otoko-ga2 [vP t2 [CP [TP [WH-F]

[Scope]

t’1 [Scope]

[TP sensei-ga t1 kisusita]] ka] siri-tagatteiru]]] [+WH] This representation violates condition (36), according to which [WH] is licensed by means of being bound by a [+WH] Comp, since the feature [WH], which is located in the uppermost occurrence of dono hutari-no onna-ni ‘which two women-Dat’ is not bound by the embedded [+WH] Comp marked by ka. It follows, then, that in the legitimate derivation of (109b), the wh-phrase dono hutari-no onna-ni must undergo scrambling with its [WH] feature left behind, as indicated in the following 82

representation: (111) [TP dono hutari-no onna-ni1 [TP sannin-no otoko-ga2 [vP t2 [CP [TP [Scope]

t’1 [Scope]

[TP sensei-ga t1 kisusita]] ka] siritagatteiru]]] [WH]

[+WH]

In this representation, the feature [WH], which is left behind in the original position of dono hutari-no onna-ni is bound by the [+WH] Comp marked by ka, hence satisfying condition (36). Since the uppermost member of the chain of dono hutari-no onna-ni carries only its [PF], it does not contribute to any semantic interpretation and hence it is natural to assume that it is deleted at LF. Thus, dono hutari-no onna-ni forms a two-membered chain at LF, one marking its quantificational scope in the embedded TP-adjoined position and the other indicating its original θ-position. This chain satisfies the ECDII, since no relevant phrase intervenes between these two members of the chain in question. Thus, the representation in (111) is correctly ruled legitimate, representing the reading of 3>2. (109c) contrasts with (109b) only in that the whole sentence is changed into an interrogative sentence, which thus amounts to a case of Takahashi’s (1993) that involves “wh-movement” according to his analysis. This sentence, unlike (109b), does not have the reading in which sannin-no otoko ‘three men’ takes scope over dono hutari-no onna ‘which two women’. This is exactly what we predict, since in this case, long-distance scrambling of dono hutari-no onna-ni can produce a focus chain with a [Focus] feature assigned to this scrambled phrase, so that its [WH] feature, after merged 83

with this [Focus] feature, can be licensed by the matrix [+WH] Comp, as shown below: (112) *[CP [TP dono hutari-no onna-ni1 [TP sannin-no otoko-ga2 [vP t2 [CP [TP [WH-F]

[Scope]

t’1

[Scope]

[TP sensei-ga t1 kisusita]] ka] siritagatteiru]]] no] [+WH]

[+WH]

In this representation, the chain produced by the long-distance scrambling of dono hutari-no onna-ni violates the ECDII for exactly the same reason as that produced by the long-distance movement of who and dareka-ni in (100) and (101). As a result, dono hutari-no onna-ni in (109c) is not allowed to have the feature [Scope], which results in producing only the reading where no scope interaction exists. Recall that in the discussion of pied-piping of wh-phrases, we have defined wh-phrases as in (53), reproduced below: (113) Wh-phrases are ones to which [WH] features percolate up to [Focus] phrases. We have also assumed the percolation mechanism given in (50), reproduced below: (114) Features can percolate up to the next higher projection from either a head or a Spec. Likewise, we can regard QPs as ones to which [Scope] features percolate up. Let us suppose that [Scope] features percolate up in accordance with (114), just like [WH] features. Keeping this in mind, consider the following sentences: (115) a.

Dareka-no

musume-ni sensei-ga

kisusita.

someone-Gen daughter-Dat teacher-Nom kissed ‘Someone’s daughter, the teacher kissed.’ 84

b. [Dareka-ga

Mary-o

sukida to]

someone-Nom Mary-Acc like

sensee-ga

itta.

Comp teacher-Nom said

‘That someone likes Mary, the teacher said.’ In (115a), dareka-no musume-ni ‘someone’s daughter-Dat’ can be regarded as a QP since the [Scope] feature assigned to dareka percolates up to the whole DP according to the percolation mechanism in (114). On the other hand, the whole complement clause scrambled to the top of the sentence in (115b) cannot be regarded as a QP since the [Scope] feature assigned to dareka does not percolate up to the whole clause. Given this, it is predicted that such a phrase as dareka-no musume-ni ‘someone’s daughter-Dat’ behaves like dareka-ni ‘someone-Dat’ in (95b) and (96) with respect to how it interacts in scope with another QP. This prediction is in fact borne out, as illustrated below: (116) a.

Dareka-no

musume-ni1 daremo-ga

[sensei-ga t1 kisusita to]

someone-Gen daughter-Dat everyone-Nom teacher-Nom kissed Comp sinziteiru. believe ‘Someone’s daughter, everyone believes that the teacher kissed.’ b. Dareka-no

musume-ni1

sensei-ga

[daremo-ga t1

kisusita

someone-Gen daughter-Dat teacher-Nom everyone-Nom kissed

to] Comp

sinziteiru. believe ‘Someone’s daughter, the teacher believes that everyone kissed.’ Just like the contrast observed between (95b) and (96) with respect to the scope 85

interaction of dareka ‘someone’ and daremo ‘everyone’, (116a) does not have the reading in which daremo takes scope over dareka, whereas (116b) does have this reading. Given that dareka-no musume-ni ‘someone’s daughter-Dat’ is identified as a QP, hence carrying a [Scope] feature, the LF representations of the wide scope reading of daremo for (116a) and (116b) correspond to (101) and (104), respectively, and hence the facts just noted follow in exactly the same way as with (95b) and (96). Let us now consider cases where a clause including a QP undergoes long-distance scrambling: (117) a.

[Dareka-ga

Mary-o

sukida to]1

someone-Nom Mary-Acc like to]

daremo-ga

[sensei-ga t1 itta

Comp everyone-Nom teacher-Nom said

sinziteiru.

Comp believe ‘That someone likes Mary, everyone believes that the teacher said.’ b. [Dareka-ga

Mary-o

sukida to]1

someone-Nom Mary-Acc like to]

sensei-ga

[daremo-ga t1

itta

Comp teacher-Nom everyone-Nom said

sinziteiru.

Comp believe ‘That someone likes Mary, the teacher believes that everyone said.’ Unlike in (116), both sentences in (117) have the reading in which daremo takes scope over dareka. This is because in these cases, the complement clause scrambled long-distance does not count as a QP and hence the ECDII is irrelevant for these cases. Under the present assumptions, the sentences in (117) will have the following LF 86

representation:22 (118) a.

[TP [dareka-ga Mary-o sukida to]1 [TP daremo-ga2 [vP t2 [Scope]

[Focus]

[CP [TP sensee-ga 1 itta] to] sinziteiru]]] [Scope] b. [TP [dareka-ga Mary-o sukida to]1 [TP sensei-ga [CP [TP daremo-ga2 [Focus]

[Scope]

[vP t2 1 it]ta] to] sinziteiru]] [Scope] In these representations, the [Scope] features of dareka fail to percolate up to the whole embedded clauses, hence taking scope on their own within these clauses. Suppose that these features are left behind when the whole complement clauses are scrambled, as indicated in (118). Then, these representations represent the reading where daremo takes scope over dareka, since the former asymmetrically c-commands the latter. This explains why this reading is available to both sentences in (117).

2.5

Clause-Internal Scrambling of Wh-Phrases and Phrases Carrying FPs

22

In these representations, we use a notation more accurate than those exploiting traces.

Following the copy theory of movement (cf. Chomsky 1993), we put the copies of the scrambled clauses in their original positions in order to make the account in the text more explicit.

87

In Sections 2.3 and 2.4, we have dealt mainly with cases of long-distance scrambling of wh-phrases and QPs to make the point that long-distance scrambling establishes a focus chain whenever possible. One might then expect that the same claim can be made with respect to cases of clause-internal scrambling of wh-phrases and QPs. This expectation does not seem to be fulfilled, however. Let us consider superiority cases, discussed in Section 2.3.1, where we find such cases when long-distance scrambling of wh-phrases is involved: (119) ??Nani-o1

John-ga

dare-ni [Mary-ga t1 tabeta to]

what-Acc John-Nom who-Dat Mary-Nom ate

itta no?

Comp said Q

‘What did John tell who that Mary ate?’ In this sentence, the wh-phrase nani-o ‘what-Acc’ is scrambled long-distance across the other wh-phrase dare-ni ‘who-Dat’ in the matrix indirect object position. I have argued that condition (108), reproduced below, forces this sentence to have only the derivation indicated in (121), in which long-distance scrambling of nani-o has produced a focus chain: (120) A chain produced by scrambling is assigned [Focus] whenever possible. (121) *[CP nani-o1 John-ga dare-ni [CP Mary-ga t1 tabeta to] itta no] [WH-F]

[WH]

The resulting focus chain, however, violates the CCF, reproduced below, since the [WH] feature of nani-o crosses that of dare-ni: (122) Crossing Constraint on Features (CCF) Given a chain C = (α1, … αi, … αn), 88

*αi [X]

…

β …

αi+1

[X]

unless feature [X] is carried along as a free ride. Thus, the ungrammaticality of sentence (119) follows. Takahashi (1993) observes, however, that superiority effects are not detected in those cases involving clause-internal scrambling of wh-phrases, as illustrated below: (123) a.

Dare-ga

nani-o

tabeta no?

who-Nom what-Acc ate

Q

‘Who ate what?’ b. Nani-o1 dare-ga t1 tabeta no? what-Acc who-Nom ate

Q

‘Lit. What did who eat?’ If nani-o had produced a focus chain in (123b), just as in (121), then the resulting chain should violate the CCF since the [WH] feature of nani-o crosses that of dare-ga, but this sentence is totally acceptable. I propose, following Abe (1993), that chains produced by clause-internal scrambling may be too short to license [Focus] features, hence not allowed to be assigned these features. Assuming that subject need not move up to Spec-TP in Japanese, Abe (1993) claims that a phrase that undergoes clause-internal scrambling can be adjoined to a V-projection. Adopting the standard V-v structure, we can now claim that the scrambled phrase nani-o ‘what-Acc’ in (123b) is adjoined to vP, as shown below: 89

(124) [CP [TP [vP nani-o1 [vP dare-ga ([WH])

[WH]

t1 tabe]] ta] no] ([WH])

[+WH]

Suppose that like [Scope] features, [Focus] features can be licensed in T-projections. Then, it is natural to claim that in (124), a [Focus] feature cannot be assigned to the scrambled phrase nani-o since it is not licensed in this position and hence that this case is exempt from condition (120). The above proposal is independently supported by considering a simple case of clause-internal scrambling such as the following: (125) Mary-o1

John-ga t1

Mary-Acc John-Nom

hihansita. criticized.’

‘Lit. Mary, John criticized.’ Under the present assumptions, Mary-o can be adjoined to vP, as shown below: (126) [TP [vP Mary1-o [vP John-ga t1 hihansi]]ta] Here, the chain produced by scrambling Mary-o cannot produce a focus chain since a [Focus] feature cannot be licensed in the position occupied by the scrambled phrase. Hence, this chain ends up being interpreted as a semantically vacuous chain. This accords well with the intuition that (125) has no significant difference in meaning from its canonical SOV order counterpart. On the other hand, things are different with long-distance scrambling. Let us consider a typical case of such scrambling: (127) Mary-o1 Mary-Acc

Bill-ga

[John-ga t1 hihansita to]

Bill-Nom John-Nom

itta.

criticized Comp said

Given the discussions in Sections 2.3 and 2.4, we need to claim that in this case, Mary-o 90

is identified as heading a focus chain. Following Abe (1993), it can be claimed that since Mary-o moves over the embedded TP, the resulting chain is eligible for a focus chain, no matter which position Mary-o has landed in the matrix clause. Alternatively, under

the

assumption

that

long-distance

scrambling

must

take

place

successive-cyclically, hence always going through the embedded Spec-CP, it can be claimed that the final landing site of Mary-o must be A’-position; otherwise it would give rise to so-called improper movement. In that case, Mary-o must be adjoined to the matrix TP or moved to the matrix Spec-CP. Either way, Mary-o is eligible to head a focus chain and hence must be according to (120). Again, this accords well with the intuition that (127) differs from (125) in that the scrambled phrase Mary-o receives focus interpretation in the normal situation. Going back to (124), we can now claim that clause-internal scrambling of nani-o in this case can be taken as a genuine instance of scrambling, carrying along only its [PF] and leaving its [WH] behind. In that case, the chain produced by scrambling nani-o does not violate the CCF.23 This explains the acceptability of (123b). We cannot assert that clause-internal scrambling is different from long-distance scrambling in that it never gives rise to a focus chain, since I have argued in Section 2.1 that Kuroda’s (1971) case of scrambling an FP phrase, reproduced below, indicates that 23

The resulting chain is finally collapsed into a one-membered chain since the top

member does not have any [SF]. When the [WH] feature is carried along by the scrambling in question in (124), on the other hand, it will violate the CCF since the resulting chain is regarded as a [WH]-chain with no additional feature other than [PF] residing in its top member.

91

clause-internal scrambling behaves like focus movement: (128) S.S.-o-sae

John-dake-ga

yonda.

S.S.-Acc-even John-only-Nom read ‘Even S.S., only John read.’ This sentence has only the reading where S.S.-o-sae takes scope over John-dake-ga. This fact follows immediately if we assume that clause-internal scrambling of the FP phrase S.S.-o-sae induces a focus chain, as indicated in the following representation: (129) [TP S.S.-o-sae1 [TP John-dake-ga t1 yonda]] [Focus]

[Focus]

Note, however, that this representation appears to violate the CCF since the object S.S.-o-sae crosses the subject John-dake-ga. Note further that the apparent exemption from the CCF is limited to cases of clause-internal scrambling of such FP phrases; long-distance scrambling of them leads to ungrammaticality, as illustrated below: (130) a.?*Mary-ni-sae1

John-dake-ga [sono sensei-ga t1 kisusita to]

Mary-Dat-even John-only-Nom that teacher-Nom kissed

omotteiru.

Comp think

‘Even Mary, only John thinks that that teacher kissed.’ b. Mary-sae1 Mary-even

John-dake-ga [sono sensei-ga e1 kisusita to] John-only-Nom that teacher-Nom kissed

omotteiru.

Comp think

In (130b), the FP phrase Mary-sae ‘even Mary’ lacks a Case-particle and hence is reasonably claimed to be base-generated in its surface position. This sentence clearly has the reading where this FP phrase takes scope over the other FP phrase John-dake-ga ‘only John-Nom’. Thus, it is interpreted in such a way that Mary is the least likely 92

person among the women for whom only John thinks that the teacher kissed them. (130a), on the other hand, does not have any clear reading with respect to the scope order between the two FP-phrases: it does not have the reading available to (130b), namely the reading of Mary-sae taking scope over John-dake, nor does it have the reading of the opposite scope order, namely, the one available to the non-scrambling version of (130a): (131) John-dake-ga

[sono sensei-ga

John-only-Nom that

Mary-ni-sae

kisusita to]

teacher-Nom Mary-Dat-even kissed

omotteiru.

Comp think

‘Only John thinks that that teacher kissed even Mary.’ Suppose that Mary-sae ‘even Mary’ in (130b) is based-generated in the Spec of Top(ic) position. Then, this sentence will have the following representation: (132) [TopP Mary-sae1 [TP John-dake-ga [sono sensee-ga pro1 kisusita to] omotteiru]] [Focus]

[Focus]

Here, no movement is involved in deriving this representation and hence the CCF has no relevance for this derivation. (132) correctly represents the reading where Mary-sae takes scope over John-dake. As for (130a), there are at least two possible derivations for this sentence, as indicated below:24 (133) a. *[TP Mary-ni-sae1 [TP John-dake-ga [sono sensee-ga t1 kisusita to] omotteiru]] [Focus]

24

[Focus]

Notice that we are now dealing with those cases involving FP phrases, which

inherently carry [Focus] features. Thus, (133b) is in principle a possible derivation where scrambling of the FP phrase Mary-ni-sae has left behind its [Focus] feature.

93

b. *[TP Mary-ni-sae1 [TP John-dake-ga [sono sensee-ga [Focus]

t1

kisusita to]

[Focus]

omotteiru]] (133a) indicates the derivation that involves focus movement of Mary-ni-sae, and represents the reading where the moved FP-phrase takes scope over John-dake-ga. According to condition (6), reproduced below, (134) Scrambling is reinterpreted as operator movement whenever possible. this derivation should be chosen over that indicated in (133b), but it violates the CCF, hence ruled out. (133b), in contrast, indicates the derivation that involves scrambling of Mary-ni-sae, leaving its [Focus] feature behind, hence representing the reading where John-dake takes scope over Mary-ni-sae. Although the scrambling in question does not violate the CCF, it does violate condition (134). Hence, (130a) does not have the reading represented in (133b), either. We are now left with the question why (129) does not lead to the ungrammaticality of sentence (128). Under the present assumptions, S.S.-o-sae can be adjoined to vP rather than TP, as shown below: (135) [TP [vP S.S.-o-sae1 [vP John-dake-ga t1 yon]]da] [Focus]

[Focus]

Here the relevant question is whether this derivation violates the CCF. Recall that we have assumed that [Focus] features are licensed in T-projections. Under this assumption, the [Focus] feature carried by S.S.-o-sae in (135) cannot be licensed in this scrambled position. Hence, in this case, it is natural to regard the scrambling of this FP phrase as a 94

genuine instance of scrambling in which the [Focus] feature is simply carried along. In order to license this feature, S.S.-o-sae needs to undergo covert movement to adjoin to TP, as shown below: (136) [TP < S.S.-o-sae1> [TP [vP S.S.-o-sae1 [vP John-dake-ga t1 yon]]da]] [Focus]

[Focus]

This representation does not violate the CCF, since the second member of the chain headed by S.S.-o-sae, namely the one adjoined to vP, does not carry a [Focus] feature and hence the intervenention of a phrase carrying this feature between this member and the next member does not lead to a violation of this constraint (cf. (122)). Thus (136) correctly represents the reading available to (128). Given that the scrambling of S.S.-o-sae in (135) is regarded as a genuine instance of scrambling, one might wonder what goes wrong if the [Focus] feature is left behind in this application of scrambling, thereby leading to the following derivation: (137) a.

[TP [vP S.S.-o-sae1 [vP John-dake-ga t1 [Focus]

yon]]da]

[Focus]

b. [TP [TP [vP S.S.-o-sae1 [vP John-dake-ga t1 [Focus]

yon]]da]]

[Focus]

In (137b), John-dake-ga has undergone covert movement to license its [Focus] feature. As a result, this representation represents the reading of John-dake taking scope over S.O.-o-sae, the one unavailable to (128). How is then the derivation indicated in (137) excluded? I suggest that a condition like (134) has direct relevance for the illegitimacy of this derivation. Note that in the derivation indicated in (136), the scrambled phrase 95

S.S.-o-sae constitutes a part of a focus chain unlike that in (137), though it does not function as the head of this chain. Let us now modify (134) into something like the following: (138) A chain produced by scrambling is reinterpreted as a focus chain or a part of it whenever possible. This condition correctly chooses the derivation in (136) over that in (137). What has been claimed about clause-internal scrambling carries over rather straightforwardly to what Tada (1993) calls short scrambling, the case of scrambling that takes place within VP. The latter behaves exactly like the former with respect to superiority effects and scope interaction of FP-phrases, as illustrated below: (139) a.

John-wa dare-ni

nani-o

ageta no?

John-Top who-Dat what-Acc gave Q ‘What did John give to who?’ b. John-wa nani-o1

dare-ni t1 ageta no?

John-Top what-Acc who-Dat gave Q (140) a.

John-wa Mary-dake-ni

S.S.-o-sae

ageta.

John-Top Mary-only-Dat S.S.-Acc-even gave ‘John gave even S.S. to only Mary.’ b. John-wa S.S.-o-sae1

Mary-dake-ni

John-Top S.S.-Acc-even Mary-only-Dat

t1 ageta. gave

If we assume, following Hoji (1985), that the underlying word order in Japanese is S-V-IO-DO, then (139b) and (140b) are derived from (139a) and (140a), respectively, 96

by scrambling the direct objects before the indirect objects. That (139b) is as good as (139a) indicates that short scrambling does not induce superiority effects. This follows immediately from the assumption that a phrase scrambled short-distance cannot be assigned a [Focus] feature due to the fact that this feature is not licensed in the scrambled position. Thus, the derivation of (139b) leads to the following representation: (141) [CP [TP John-wa1 [vP t1 [VP nani-o1 [VP dare-ni t1 tV]] tabe+v] ta] no] [WH] [WH] As represented here, the scrambling of nani-o is taken as a genuine instance of scrambling that leaves its [WH] feature behind and hence does not violate the CCF. Just like (128), (140b) has only the reading in which the structurally higher FP-phrase S.S.-o-sae takes scope over the lower phrase Mary-dake-ni. Under the present assumptions, the legitimate derivation for this sentence is as follows: (142) a.

[TP John-wa1 [vP t1 [VP S.S.-o-sae1 [VP Mary-dake-ni t1 tV]] age+v]ta] [Focus]

b.

[Focus]

[TP < S.S.-o-sae1> [TP John-wa1 [vP t1 [VP S.S.-o-sae1 [VP Mary-dake-ni t1 tV]] [Focus]

[Focus]

age+v]ta]] In (142a), the scrambling of S.S.-o-sae is taken as a genuine instance of scrambling with its [Focus] feature carried along as a free ride. Further, it must carry along this feature in this case, since this will lead to producing a focus chain of S.S.-o-sae in accordance with condition (138). In (142b), S.S.-o-sae has undergone covert movement to license its [Focus] feature. This representation does not violate the CCF since the second member 97

of the chain headed by S.S.-o-sae does not carry a [Focus] feature and hence the intervention of a phrase carrying this feature between this member and the next member does not induce a violation of this constraint. Thus (142b) correctly represents the reading of S.S.-o-sae taking scope over Mary-dake-ni. Finally,

there

is

evidence

that

long-distance

scrambling

may

involve

clause-internal scrambling on its way. Consider the following examples ((143) is taken from Takahashi (1993)): (143) Nani-o1 what-Acc

John-ga [dare-ga t1

tabeta to]

John-Nom who-Nom

ate

itta

Comp said

‘Lit. What did John say that who ate?’ (144) Mary-ni-sae1

sono sensei-ga

no? Q (Takahashi 1993:667)

[John-dake-ga t1 kisusita

Mary-Dat-even that teacher-Nom John-only-Nom kissed

to]

omotteiru.

Comp think

‘Even Mary, that teacher thinks that only John kissed.’ Takahashi (1993) notes that unlike (119), (143) does not display a superiority effect and attributes this fact to the availability of the derivation for (143) in which “the Wh-phrase can first undergo local A-scrambling over the embedded Wh-subject and then A’-move to the matrix clause, an option which is impossible in [119].” (p. 667) If we follow this claim, nani-o in (143) can first be adjoined to the embedded vP, as shown below: (145) [CP [TP [vP nani-o1 [vP dare-ga t1 tabe]]ta] to] [WH]

[WH]

Suppose that the CCF is applied phase by phase, with the assumption that CP is a phase, ignoring the phasehood of vP for the sake of discussion. Then, we must check whether 98

scrambling of nani-o in (145) violates the CCF at this stage of derivation. Notice that at this stage of derivation, the [WH] feature of nani-o has not been licensed since the [+WH] Comp ka is not introduced into the structure. We may interpret this as indicating that the chain produced by scrambling of nani-o at this stage has not been licensed as a [WH] chain yet and hence that it is produced by a genuine instance of scrambling with its [WH] feature carried along as a free ride. If so, the resulting chain is characterized as a scrambling chain and hence does not violate the CCF. Then, nani-o is further scrambled to the top of the sentence, where it is assigned a [Focus] feature in accordance with condition (120) and this feature is merged with its [WH] feature, yielding [WH-F], as shown below: (146) [CP [TP nani-o1 [TP John-ga2 [vP t2 [CP [TP [vP t’1 [vP dare-ga t1 tabe]]ta] to] it]ta]] no] [WH-F]

[WH]

[+WH]

In this way, (143) is successfully derived without inducing a violation of the CCF, unlike (119). Likewise, (144) can be derived in the following way: (147) a.

[CP [TP [vP Mary-ni-sae1 [vP John-dake-ga t1 kisusi]]ta] to] [Focus]

[Focus]

b. [TP Mary-ni-sae1 [TP sono sensei-ga2 [vP t2 [CP [TP [vP t’1 [vP John-dake-ga t1 [Focus]

[Focus]

kisusi]]ta] to] omotteiru]]] In (147a), Mary-ni-sae is adjoined to the embedded vP and since the [Focus] feature is not licensed in this adjoined position, this movement is taken as a genuine instance of 99

scrambling with its [Focus] feature carried along as a free ride, hence not violating the CCF. Next, Mary-ni-sae is further scrambled to the top of the sentence, as shown in (147b), where its [Focus] feature is licensed. Thus, (144) is successfully derived without violating the CCF, unlike (130a), and (147b) correctly represents the reading available to (144), namely, the one on which Mary is the least likely person among the women for which the teacher thinks that only John kissed them.

2.6

Multiple Scrambling

In this section, we discuss cases of multiple scrambling such as illustrated below to consider how they should be analyzed under the present assumptions: (148) Sensei-ni1

Mary-o2

John-ga t1 t2 syookaisita.

teacher-Dat Mary-Acc John-Nom

introduced

‘Lit. To the teacher, Mary, John introduced.’ A couple of questions arise with such a case of multiple scrambling: One has to do with the landing sites of the scrambled phrases. Recall that we have assumed in the previous section that clause-internal scrambling can be adjunction to vP. Given this assumption, one may wonder whether the two scrambled phrases in (148) can both be adjoined to vP or not. If we assume that no more than one phrase can be adjoined to a given phrase, then it follows that the higher scrambled phrase sensei-ni ‘teacher-Dat’ in (148) needs to be adjoined to TP, as shown below: (149) [TP sensei-ni1 [TP [vP Mary-o2 [vP John-ga t1 t2 syookaisi]]ta]] In this representation, sensei-ni must be assigned a [Focus] feature in accordance with 100

condition (120), reproduced below, since this feature can be satisfied in the TP-adjoined position: (150) A chain produced by scrambling is assigned [Focus] whenever possible. Another question with multiple scrambling is whether such a case is derived by multiple applications of scrambling or by a single application of scrambling of a larger phrase. Koizumi (1995, 2000) proposes the VP scrambling approach to such cases under which (148) can have the following structure: (151) [TP [vP [VP sensei-ni Mary-o tV]1 [vP John-ga t1 shookaisi+v]]ta] In this representation, the verb shookaisi ‘introduce’ is raised out of VP to adjoin to v (and possibly further up to T) and the remnant VP is adjoined to vP. Notice that in this case, the multiple phrases preposed by VP scrambling is not in a c-command relation with any phrase that follows them. In what follows, I argue (i) that multiple applications of scrambling is constrained by the ban on multiple adjunction, as illustrated in (149), so that the upper scrambled phrase is assigned a [Focus] feature in accordance with (150), and (ii) that VP scrambling is available in Japanese. Let us now consider cases of multiple scrambling of wh-phrases to see if they exhibit superiority effects. Clause-internal multiple scrambling shows no such effects, as illustrated below: (152) Dare-ni1 who-Dat

nani-o2

dare-ga t1 t2

what-Acc who-Nom

ageta no? gave Q

‘Lit. to who, what, who gave?’ If we take the two preposed phrases in (152) to have undergone separate applications of 101

scrambling, then this sentence should have the following output representation: (153) *[CP [TP dare-ni1 [TP [vP nani-o2 [vP dare-ga t1 [WH-F]

([WH])

[WH]

t2

age]]ta]]] no]

([WH])

[+WH]

Here, even though scrambling of nani-o to adjoin to vP does not violate the CCF when its [WH] feature is left behind, scrambling of dare-ni to adjoin to TP does violate this condition since the [WH] feature borne by the resulting focus chain crosses that borne by dare-ga. However, (152) can be derived by way of VP scrambling, as shown below: (154) [CP [TP [vP [VP dare-ni nani-o tV]1 [vP dare-ga t1 age+v]]ta] no] [WH]

[WH]

[WH]

This VP scrambling does not induce a violation of the CCF irrespective of whether it is assigned a [Focus] feature or not, since the two preposed wh-phrases do not c-command the wh-subject dare-ga in this structure. Hence, (152) is ruled grammatical with (154) as its output representation. It is then predicted that the multiple wh-phrases preposed by scrambling must be VP-mates, since they are immune to the CCF only when they are preposed by VP scrambling. This is borne out by the following pair of examples: (155) a.

Nani-o1 dare-ni2 John-ga [Mary-ga t2 t1 ageta to] what-Acc who-Dat John-Nom Mary-Nom

itta

gave Comp said

no? Q

‘Lit. what to who did John say that Mary gave?’ b.?*Nani-o1 dare-ni2 John-ga t2 what-Acc who-Dat John-Nom

[Mary-ga t1

moratta to]

Mary-Nom

got

‘Lit. what1 who2 did John tell t2 that Mary got t1?’ 102

itta no?

Comp said Q

In (155a) nani-o and dare-ni are both internal arguments of the embedded verb age ‘give’ and this sentence can be successfully derived in the following way:25 (156) a.

[CP [TP Mary-ga2 [vP t2 [VP nani-o1 [VP dare-ni

t1

tV]] age+v]ta to] …

[WH] [WH] b.

[CP [TP [VP[Focus] nani-o1 [VP dare-ni

t1

tV]]3 [TP John-ga4 [vP t4

[WH] [WH] [CP [TP Mary-ga2 [vP t2 t3 age+v]ta to] it]ta]] no] In (156a), nani-o has undergone short scrambling within the embedded VP, leaving its [WH] feature behind, and then the whole upper VP is scrambled to the top of the sentence, as shown in (156b). On the other hand, (155b) cannot be derived in a similar way via VP scrambling since the two preposed wh-phrases are not VP-mates. The only remaining option is to apply scrambling separately to the two wh-phrases, as shown below: (157) *[CP [TP nani-o1 [TP [vP dare-ni2 [vP John-ga t2 [CP Mary-ga t1 moratta to] it]]ta]] [WH-F]

([WH])

([WH])

no]

Since nani-o has undergone long-distance scrambling, it must be adjoined to the matrix TP (or a higher phrase) so that it is assigned a [Focus] feature. It then violates the CCF since the [WH] feature of the resulting focus chain crosses that of dare-ni. As a result, there is no legitimate way of deriving (155b), and this explains the ungrammaticality of

25

See the discussion toward the end of this section about how the [WH] features of the

two wh-objects in (156b) are merged with the [Focus] feature assigned to the scrambled VP.

103

this sentence. Further evidence for the present approach comes from multiple scrambling of QPs. Arguing for the VP scrambling approach, Abe (2012) provides the following examples as supporting evidence: (158) a.

Sannin-no otoko-ga

hutari-no onna-ni

dareka-o

syookaisita.

three-Gen man-Nom

two-Gen woman-Dat someone-Acc introduced

‘Three men introduced someone to two women.’ b. Hutari-no onna-ni1

sannin-no

two-Gen woman-Dat three-Gen c.

Hutari-no onna-ni1

otoko-ga t1

dareka-o

syookaisita.

man-Nom

someone-Acc introduced

sannin-no

otoko-o2 dareka-ga t1 t2 syookaisita.

two-Gen woman-Dat three-Gen

man-Acc someone-Nom introduced

‘Lit. To two women, three men, someone introduced.’

(Abe 2012:80)

Sentence (158a) reflects the basic word order according to Hoji (1985) and, as expected, it has the reading in which the subject QP sannin-no otoko ‘three men’ takes scope wider than both the indirect object QP hutari-no onna ‘two women’ and the direct object QP dareka ‘someone’ and between the latter two, hutari-no onna takes scope over dareka. Thus, this sentence can be uttered in the situation in which three men each did the act of introduction to different pairs of women and further each man introduced a different person to each woman. This reading will be represented as follows under the present assumptions: (159) [TP sannin-no otoko-ga1 [vP t1 hutari-no onna-ni dareka-o syookaisi]ta] [Scope]

[Scope] 104

[Scope]

In this representation, the [Scope] feature carried by sannin-no otoko ‘three men’ is licensed in the Spec of TP, whereas those carried by hutari-no onna ‘two women’ and dareka ‘someone’ are licensed by way of being bound by the [Scope] feature carried by sannin-no otoko; the relevant condition is reproduced below: (160) [Scope] in situ can be satisfied by means of being bound by another instance of [Scope] that satisfies itself by movement. According to the interpretive rules in (77), reproduced below: (161) a.

The scope domain of a [Scope] is its c-command domain.

b. If [Scope]1 asymmetrically c-commands [Scope]2, then [Scope]1 >[Scope]2. c.

If [Scope]1 and [Scope]2 c-command each other, then [Scope]1 > <[Scope]2.

(159) represents the reading of 3>2> ∃ . Now let us consider (158b), which is derived from (158a) by scrambling the indirect object QP hutari-no onna-ni ‘two women’ to the top of the sentence. This sentence has the reading derived from the scope order that reflects the hierarchical order of the indirect object, the subject and the direct object, namely the reading of 2>3> ∃ . Thus, this sentence can be uttered in the situation in which two women were each involved as recipients in the act of introduction by different groups of three men and further each man introduced a different person to each woman. This reading is represented as follows under the present assumptions: (162) [TP hutari-no onna-ni1 [TP sannin-no otoko-ga2 [vP t2 t1 dareka-o syookaisi]ta]] [Focus][Scope]

[Scope]

[Scope]

Here, hutari-no onna-ni is adjoined to TP, so that it is assigned a [Focus] feature. This movement does not violate the CCF with respect to [Scope] features, since it has 105

produced a focus chain and hence the [Scope] feature can be regarded as carried along as a free ride. According to the interpretive rule (161b), this representation correctly represents the reading in question. The interesting case is (158c), which involves multiple scrambling of indirect and direct object QPs. This sentence does not have the reading derived from the scope order that reflects the hierarchical order of the indirect object, the direct object and the subject, namely the reading of 2>3> ∃ . Thus, this sentence is infelicitous when uttered in the situation in which two women were each involved as recipients in the act of introduction of different groups of three men and further each man was introduced to each woman by a different person. This reading could be represented as either (163a) or (163b): (163) a. *[TP hutari-no onna-ni1 [vP sannin-no otoko-o2 [vP dareka-ga t1 t2 syookaisi]]ta] [Focus][Scope]

[Scope]

[Scope]

b. *[CP hutari-no onna-ni1 [CP [TP sannin-no otoko-o2 [TP dareka-ga3 [vP t3 t1 t2 [Focus][Scope]

[Focus][Scope]

[Scope]

syookaisi]ta]]]] In (163a), the indirect object QP hutari-no onna-ni ‘two women-Dat’ is adjoined to TP, hence licensing the assignment of a [Focus] feature, while the direct object QP sannin-no otoko-o ‘three men-Acc’ is adjoined to vP, hence prohibiting the assignment of this feature. In this case, the chain produced by scrambling of sannin-no otoko-o is regarded as a [Scope] chain, which is licensed by way of being bound by the indirect object QP hutari-no onna. Hence, this chain violates the CCF since the [Scope] feature 106

borne by this QP crosses that borne by the subject QP dareka. In (163b), the indirect and direct object QPs are adjoined to CP and TP, respectively, so that they are assigned [Focus] features. Though in this case there is no CCF violation with respect to [Scope] features since the relevant chains constitute focus chains, the focus chain of the indirect object QP induces a violation of the CCF since its [Focus] feature crosses that borne by the direct object QP. As a result, there is no legitimate derivation for representing the reading of 2>3> ∃ for (158c). If we apply VP scrambling to derive this sentence, we will have the following representation: (164) [TP [vP [VP hutari-no onna-ni sannin-no otoko-o tV]1 [vP dareka-ga t1 [Scope]

[Scope]

[Scope]

syookaisi+v]]ta] In this representation, hutari-no onna ‘two women’ asymmetrically c-commands sannin-no otoko ‘three men’ and no c-command relation holds between these QPs and dareka ‘someone’. Hence, this represents the reading where hutari-no onna takes scope over sannin-no otoko and no scope interaction exists between these two QPs and dareka. This reading amounts to that in which two women each acted as the recipients of introduction of different groups of three men by a particular person. Sentence (158c) in fact has this reading. Notice further that if the application of literal multiple scrambling were not constrained by a minimality condition like the CCF in the way presented in (163), then it would be incorrectly predicted that (158c) would have the reading where sannin-no 107

otoko ‘three men’ takes scope over hutari-no onna ‘two women’. To make this point, let us consider a simpler example such as the following: (165) Hutari-no two-Gen

onna-ni1

sannin-no

woman-Dat three-Gen

otoko-o2 John-ga t1 t2

syookaisita.

man-Acc John-Nom

introduced

‘Lit. to two women, three men, John introduced.’ If we applied scrambling separately to the two object QPs, the reading of 3>2 could be derived in either of the following two ways: (166) a. *[TP hutari-no onna-ni1 [vP sannin-no otoko-o2 [vP John-ga t1 [Focus]

[Scope]

t2 syookaisi]]ta]

[Scope]

b. *[CP hutari-no onna-ni1 [CP [TP sannin-no otoko-o2 [TP John-ga3 [vP t3 [Focus]

[Focus][Scope]

t1

t2

[Scope]

syookaisi]ta]]]] In both representations, when hutari-no onna-ni ‘two women-Dat’ is scrambled, it has left behind its [Scope] feature, and hence they represent the reading of 3>2. If these representations were allowed, then this reading should be available to (165). (166b) induces a violation of the CCF since the [Focus] feature borne by hutari-no onna-ni crosses that borne by sannin-no otoko-o. As for (166a), there is no violation of the CCF since the scrambling of sannin-no otoko-o in this case does not license its [Scope] feature in the vP-adjoined position and hence it is regarded as a genuine instance of scrambling with its [Scope] feature carried along as a free ride. However, (166a) violates the ECDII, which is reproduced below:

108

(167) The Economy Condition on Dependency II (ECDII) If αi is a member of the chain (α1, ... αi, ...), then αi+1 must be the closest possible dependent for αi. (168) α is a possible dependent for β if γ is dependent upon β, and α and γ have the same feature. In (166a), the scramble direct object QP sannin-no otoko ‘three men’ is regarded as a possible dependent for hutari-no onna-ni since the real dependent on the latter QP shares [Scope] features with the former. Since the direct object QP is closer to the top member of hutari-no onna-ni than its real dependent, it induces a violation of the ECDII.26 Note that (166b) also induces a violation of this condition. There are only two ways to derive (165) without inducing a violation of the CCF or ECDII: either to make the indirect object QP carry along its [Scope] feature when we apply literal multiple scrambling, as in (169a), or to derive (165) via VP scrambling, as in (169b): (169) a.

[TP hutari-no onna-ni1 [vP sannin-no otoko-o2 [vP John-ga t1 t2 syookaisi]]ta] [Focus][Scope]

b.

[TP [vP [VP hutari-no onna-ni sannin-no otoko-o tV]1 [vP John-ga t1 [Scope]

26

[Scope]

[Scope]

Strictly, the direct object QP sannin-no otoko-o needs to undergo further movement

to adjoin to TP, so that its [Scope] feature is properly licensed. Even with this movement, the final representation still violates the ECDII, since this direct object QP is closer to the head member of the indirect object QP hutari-no onna-ni than its second member.

109

syookaisi+v]]ta] In (169a), scrambling the indirect object QP hutari-no onna-ni ‘two women-Dat’ has produced a focus chain and hence its [Scope] feature can be regarded as carried along as a free ride, hence not violating the CCF. (169a, b) both represent the reading available to (165), namely that of 2>3. Yatsushiro (1996) and Nishida (1999) argue that such a sentence as (165) involves VP scrambling, so that it has only the reading where the indirect object QP takes scope over the direct object QP. Furthermore, Yatsushiro (1996) claims that literal application of multiple scrambling is prohibited since the option of VP scrambling involves fewer applications of scrambling. This is in consonance with Abe’s (1993) claim, discussed in Section 2.1, that string-vacuous scrambling is prohibited by the Fewest Steps Condition, but is not compatible with the present approach to multiple scrambling, which does not exclude the possibility of literal multiple scrambling altogether. In fact, Yatsushiro’s (1996) claim is countered by Sauerland and Elbourne (2002), who point out that it incorrectly predicts that in a sentence involving multiple scrambling of indirect and direct objects, even if only one of them is a QP, it cannot take scope over the subject QP. Consider the following example: (170) Sono onna-ni1 that

sannin-no otoko-o2

woman-Dat three-Gen man-Acc

dareka-ga t1 someone-Nom

t2 syookaisita. introduced

‘Lit. To that woman, three men, someone introduced.’ This sentence allows the reading where sannin-no otoko ‘three men’ takes scope over dareka ‘someone’. Notice that if VP scrambling were the only option for deriving an 110

appropriate representation for (170), as shown below, then we could not derive the reading in question. (171) [TP [vP [VP sono onna-ni sannin-no otoko-o tV]1 [vP dareka-ga t1 syookaisi+v]]ta] [Scope]

[Scope]

Since sannin-no otoko and dareka do not c-command each other in this representation, it represents only the reading where no scope interaction exists between these two QPs. This strongly suggests that unlike Yatsushiro (1996) (and Nishida (1999) and Koizumi (2000) for that matter, who also assume that VP scrambling is mandatory in such a case as (170), hence facing the same problem), we cannot entirely exclude the possibility of literal multiple scrambling. The present approach to multiple scrambling, which does not exclude this possibility, has no difficulty dealing with such a case as (170). This sentence is successfully derived in the following way: (172) a.

[TP sono onna-ni1 [vP sannin-no otoko-o2 [vP dareka-ga t1 t2 syookaisi]]ta] [Focus]

[Scope]

[Scope]

b. [TP [TP sono onna-ni1 [vP sannin-no otoko-o2 [Scope]

[Focus]

[vP dareka-ga t1 t2 syookaisi]]ta]] [Scope] At the stage of (172a), the QP sannin-no otoko-o ‘three men-Acc’ is adjoined to vP, and since the [Scope] feature is not licensed in this position, unlike in (163a) (where the [Scope] feature of the direct object QP is licensed by way of (160)), scrambling this QP is regarded as a genuine instance of scrambling with its [Scope] feature carried along as 111

a free ride. Next, this QP undergoes QR to satisfy its [Scope] feature, as shown in (172b), which thus does not violate the CCF with respect to [Scope] features. Thus, (170) illustrates a case where applying literal multiple scrambling leads to a successful derivation with the required scope order captured properly.27 Further support for the present approach to multiple scrambling comes from the behaviors of QPs with respect to reconstruction effects in multiple scrambling contexts. Compare the following examples: (173) a.

Dareka-o1

hutari-no onna-ga

sannin-no otoko-ni t1 syookaisita.

someone-Acc two-Gen woman-Nom three-Gen man-Dat

introduced

‘Someone, two women introduced to three men.’ b.

Dareka-o 1

hutari-no onna-ni2

someone-Acc two-Gen woman-Dat

sannin-no otoko-ga t2 t1 syookaisita. three-Gen man-Nom

introduced

‘Lit. Someone, to two women, three men introduced.’ (173a) has the reading where the scrambled object dareka-o ‘someone-Acc’ takes scope under the other two QPs, namely that of 2>3> ∃ , which indicates that this object can be “reconstructed” at LF. Under the present analysis, this sentence can have the following representation:

27

This argument may imply that Abe’s (1993) way of deriving the ban on

string-vacuous scrambling in terms of the Fewest Steps Condition is not on the right track because of its incorrect consequence that VP scrambling is always preferred to literal multiple scrambling. See Abe and Hornstein (2012) for relevant discussion. I am indebted to Mizuki Nishida for calling my attention to the issue concerning the obligatoriness of VP scrambling.

112

(174) [TP [vP dareka-o1 [vP hutari-no onna-ga sannin-no otoko-ni [Scope]

[Scope]

t1

syookaisi]]ta]

[Scope]

In this representation, dareka-o has undergone scrambling to adjoin to vP, leaving its [Scope] feature behind. (174) correctly represents the reading of 2>3> ∃ . Unlike this sentence, (173b) does not have the reading of 2>3> ∃ , which would be derived if dareka-o could be “reconstructed” at LF. Under the present analysis, this reading should be obtained from the following representation: (175) *[TP dareka-o2 [TP [vP hutari-no onna-ni1 [vP sannin-no otoko-ga [Focus]

[Scope]

[Scope]

t1

t2 [Scope]

syookaisi]]ta]] In this representation, there is no violation of the CCF, just like in (172a); since scrambling of hutari-no onna-ni ‘two women-Dat’ is regarded as a genuine instance of scrambling with its [Scope] feature carried along as a free ride, it does not violate the CCF. However, the focus chain headed by dareka-o ‘someone-Acc’ violates the ECDII: Even though hutari-no onna-ni undergoes QR and adjoins to TP to license its [Scope] feature, just like in (172b), this QP as well as the subject QP sannin-no otoko ‘three men’ intervenes in the focus chain of dareka-o. These QPs are regarded as possible dependents for dareka-o since the real dependent on the latter QP shares [Scope] features with the former. Since these QPs are closer to the top member of dareka than its real dependent, they induce a violation of the ECDII. The only way to make the derivation indicated in (175) legitimate is for dareka to carry along a [Scope] feature when it undergoes scrambling. In this case, the resulting representation represents the 113

reading of ∃ >2>3, which is in fact available to (173b). Alternatively, (173b) can be derived by way of VP scrambling, as shown below: (176) a.

[TP [vP [VP dareka-o2 hutari-no onna-ni t2 tV]1 [vP sannin-no otoko-ga t1 [Scope]

[Scope]

[Scope]

syookaisi+v]]ta] b. [TP [vP [VP dareka-o2 hutari-no onna-ni [Scope]

t2

tV]1 [vP sannin-no otoko-ga t1

[Scope]



[Scope]

syookaisi+v]]ta] c.

[TP [vP [VP dareka-o2 hutari-no onna-ni t2 tV]1 [vP sannin-no otoko-ga [Scope] syookaisi+v]]ta] [Scope]

[Scope]

d. [TP [vP [VP dareka-o2 hutari-no onna-ni t2 tV]1 [vP sannin-no otoko-ga [Scope]
t2

tV> syookaisi+v]]ta]

[Scope]

(176a) is derived by first scrambling the direct object QP dareka-o to the left of the indirect object QP hutari-no onna-ni, carrying along its [Scope], and then adjoining the whole VP, including both objects, to vP, again carrying along both [Scope] features. This represents the reading where dareka ‘someone’ takes scope over hutari-no onna ‘two women’ and no scope interaction exists between these QPs and the subject sannin-no otoko ‘three men’. (176b) is derived in much the same way as (176a) except 114

that when the direct object QP is scrambled within VP, it leaves its [Scope] behind. This represents the reading where hutari-no onna takes scope over dareka and no scope interaction exists between these QPs and the subject sannin-no otoko. In (176c), the scrambling of the direct object QP within VP carries its [Scope] feature along, but the VP scrambling leaves both [Scope] features behind. This represents the reading of 3> ∃ >2. (176d) differs from (175c) only in that the scrambling of the direct object QP leaves its [Scope] behind. This represents the reading of 3>2> ∃ . All these four readings are available to sentence (173b). Let us further consider the following examples: (177) a.

Dareka-ni1

hutari-no onna-ga t1

sannin-no otoko-o syookaisita.

someone-Dat two-Gen woman-Nom three-Gen man-Acc introduced ‘To someone, two women introduced three men.’ b. Dareka-ni1

hutari-no onna-o2

sannin-no

otoko-ga t1 t2 syookaisita.

someone-Dat two-Gen woman-Acc three-Gen man-Nom

introduced

‘Lit. To someone, two women, three men introduced.’ (177a) has the reading where the scrambled indirect object QP dareka ‘someone’ takes scope over the direct object QP sannin-no otoko ‘three men’ and under the subject QP hutari-no onna ‘two women’, which will be derived if the scrambled object is “reconstructed” into its original position at LF. Under the present analysis, this reading will be represented as follows: (178) [TP [vP dareka-ni1 [vP hutari-no onna-ga [Scope] 115

t1

sannin-no otoko-o syookaisi]]ta]

[Scope]

[Scope]

Here, the [Scope] feature of dareka-ni is left behind when it undergoes scrambling to adjoin to vP. The subject QP hutari-no onna-ga ‘two women-Nom’ must undergo QR to satisfy not only its own [Scope] feature but also those features borne by the other QPs by way of (160). From the hierarchical relations among the QPs, (178) represents the reading in question, namely that of 2> ∃ >3. (177b), in contrast, does not have the reading of 2>3> ∃ , which would be derived if dareka-ni could be “reconstructed” into its original position at LF. Under the present analysis, this reading would be represented as follows: (179) *[TP dareka-ni1 [TP [vP hutari-no onna-o2 [vP sannin-no otoko-ga [Focus]

[Scope]

[Scope]

t1

t2

[Scope]

syookaisi]]ta]]] However, the focus chain headed by dareka-ni violates the ECDII since both the direct object and subject QPs are regarded as possible dependents for the head of this chain and are closer to it than the real dependent on it. This correctly captures the fact that sentence (177b) does not have the reading of 2>3> ∃ . Again, (179) can lead to a legitimate derivation if dareka carries along a [Scope] feature when it undergoes scrambling and the resulting representation represents the reading of ∃ >2>3, which is available to (177b). On top of this, the legitimate derivations for this sentence can be derived by way of VP scrambling, as shown below: (180) a.

[TP [vP [VP dareka-ni hutari-no onna-o tV]1 [vP sannin-no otoko-ga t1 [Scope]

[Scope]

syookaisi+v]]ta] 116

[Scope]

b. [TP [vP [VP dareka-ni hutari-no onna-o tV]1 [vP sannin-no otoko-ga [Scope] syookaisi+v]]ta] [Scope]

[Scope]

In (180a), the [Scope] features of both objects are carried along by VP scrambling. This represents the reading where dareka ‘someone’ takes scope over hutari-no onna ‘two women’ and no scope interaction exists between these QPs and the subject QP sannin-no otoko ‘three men’. In (180b), in contrast, the [Scope] features of both objects are left behind when VP scrambling takes place. This represents the reading of 3> ∃ >2. As predicted, these two readings are available to (177b). Let us compare (173b) and (177b) with the following examples, in which the existential QP dareka is put in the second scrambled QP: (181) a.

Hutari-no onna-o1

dareka-ni2

sannin-no otoko-ga t2 t1

two-Gen woman-Acc someone-Dat three-Gen man-Nom

syookaisita. introduced

‘Lit. Two women to someone, three men introduced.’ b. Hutari-no onna-ni1

dareka-o2

sannin-no otoko-ga t1 t2

two-Gen woman-Dat someone-Acc three-Gen man-Nom

syookaisita. introduced

‘Lit. To someone, two women, three men introduced.’ Though delicate judgments may be required for the availability of the relevant readings for these sentences, there seems to be a significant contrast between (181a, b) on the one hand and (173b) and (177b) on the other with respect to the availability of the reading in which the existential QP dareka takes scope in its original position, namely the reading 117

of 2>3> ∃ . Under the present assumptions, (181a, b) can have the following representations: (182) a.

[TP hutari-no onna-o2 [TP [vP dareka-ni1 [vP sannin-no otoko-ga [Focus][Scope]

[Scope]

t1

t2

[Scope]

syookaisi]]ta]] b. [TP hutari-no onna-ni1 [TP [vP dareka-o2 [vP sannin-no otoko-ga [Focus][Scope]

[Scope]

t1

t2 [Scope]

syookaisi]]ta]] No violation of the CCF or the ECDII arises with these representations. Hence, the present theory predicts that the reading of 2>3> ∃ is available to (181a, b), which seems to be borne out. If so, these data give further support to the present claim that literal multiple scrambling is not a priori prohibited. Let us now consider those cases that involve long-distance multiple scrambling. It is predicted that unlike those cases of clause-internal multiple scrambling, these cases are never derived by multiple applications of scrambling but rather need to be derived by a single application of VP scrambling. Consider a typical example of such cases: (183) John-ni1 John-Dat

hon-o2

Mary-ga

[sensei-ga t1 t2

book-Acc Mary-Nom teacher-Nom

ageta to]

omotteiru.

gave Comp think

‘Lit. John a book, Mary thinks that the teacher gave.’ If we applied scrambling separately to John-ni and hon-o in this case, these phrases are assigned [Focus] features in their scrambled positions, hence inducing a violation of the CCF since the [Focus] feature borne by John-ni crosses that borne by hon-o. Thus, the 118

only way to derive (183) legitimately is to scramble the embedded VP in the following way: (184) [TP [VP[Focus] John-ni1 hon-o2 tV]1 [TP Mary-ga2 [vP t2 [CP [TP sensei-ga3 [vP t3 t1 age+v]ta] to] omotteiru]]] Koizumi (2000) provides one piece of evidence that demonstrates that VP scrambling is mandatory in deriving instances of long-distance multiple scrambling. This is concerned with the intonation patterns of those phrases that are moved long-distance by multiple scrambling. Koizumi (2000) observes that such an example as (183) is acceptable only “if it is parsed in such a way that the scrambled elements form an intonational phrase.” (p. 239) This indicates that the multiple phrases moved long-distance by scrambling must form a constituent, a VP according to Koizumi’s analysis. Interestingly, this requirement regarding intonation patterns is not imposed upon such a clause-internal scrambling case as (148), reproduced below: (185) Sensei-ni1

Mary-o2

John-ga t1 t2 syookaisita.

teacher-Dat Mary-Acc John-Nom

introduced

‘Lit. To the teacher, Mary, John introduced.’ The scrambled phrases in (185) may form an intonational phrase, so that a pause may be put after Mary-o, but it is also possible to utter the entire sentence without any significant pause, just as its corresponding sentence with the normal S-IO-DO-V order. This fact is immediately captured under the present approach to multiple scrambling since (185) can be assigned the following two representations: (186) a.

[TP [vP [VP sensei-ni Mary-o tV]1 [vP John-ga t1 syookaisi+v]]ta] 119

b. [TP sensei-ni1 [TP [vP Mary-o2 [vP John-ga t1 t2 syookaisi]]ta]] [Focus] (186a) is derived by applying VP scrambling and it will give rise to the intonation pattern in which sensei-ni and Mary-o form an intonational phrase. (186b), on the other hand, is derived by applying scrambling separately to these two objects and it is natural to claim that this representation gives rise to the intonation pattern in which no significant pause is put after these two objects. Thus, the above facts regarding intonation patterns give support to our prediction that VP scrambling is mandatory in the cases of long-distance multiple scrambling whereas it is not required in those of clause-internal multiple scrambling. So far we have not considered the question whether or not a chain produced by VP scrambling can be assigned a [Focus] feature. Recall that in the discussion of long-distance scrambling of wh-phrases in Section 2.3, we have assumed that a [WH] feature must be merged with a [Focus] feature; the relevant condition is reproduced below from (55): (187) If a phrase that includes a [WH] feature is assigned a [Focus] feature, these features must be merged into a [WH-F] feature. We have observed there that when a wh-phrase undergoes long-distance scrambling to the top of a clause headed by a [+WH] Comp, it must take scope in that clause, as illustrated in (47b), repeated below as (188a), and that when a clause that contains a wh-phrase undergoes long-distance scrambling, it can be “undone” and hence the scope of the wh-phrase contained in the scrambled clause is not fixed in that scrambled 120

position, as illustrated in (48b), repeated below as (188b). (188) a.

Dare-no hon-o1

John-wa [Mary-ga t1 yonda ka]

who-Gen book-Acc John-Top Mary-Nom read

Q

sirit-agatteiru no? want-to-know Q

‘Whose book does John want to know if Mary read?’ b.

[pro1 nani-o

yonda to]2

what-Acc read

John-wa [Mary1-ga t2 itta ka] siri-tagatteiru no?

Comp John-Top Mary-Nom said Q want-to-know Q

‘Does John want to know what Mary said that she had read?’ or ‘What does John want to know if Mary said that she had read?’ In (188a), dare-no hon-o ‘whose book-Acc’ is assigned a [Focus] feature after it undergoes scrambling. In this case, the [WH] feature of dare is percolated up to the whole phrase dare-no hon-o, according to the percolation mechanism assumed in (50), reproduced below: (189) Features can percolate up to the next higher projection from either a head or a Spec. Hence, the [WH] feature of dare is successfully merged with its [Focus] feature, satisfying condition (187), so that (188a) has the following representation: (190) [CP dare-no hon-o1 John-wa [CP Mary-ga t1 [WH-F]

yonda ka] siri-tagatteiru no] [+WH]

[+WH]

Here the wh-phrase dare-no hon must take matrix scope. In (188b), in contrast, the scrambled clause cannot be assigned a [Focus] feature since the [WH] feature of nani ‘what’ is unable to percolate up to the whole clause; if it were, this would violate condition (187). Hence, (188b) will have the following representation: 121

(191) [CP [pro1 nani-o yonda to]2 John-wa [CP Mary1-ga t2 itta ka] siri-tagatteiru no] ([WH])

([WH])

Here, the [WH] feature of nani can be carried along when the whole embedded clause is scrambled, thereby giving rise to its matrix scope reading or it can be left behind, thereby giving rise to its embedded scope reading. Keeping this much in mind, let us consider a case that involves long-distance multiple scrambling of wh-phrases such as follows: (192) Dare-ni1 who-Dat

nani-o2

John-wa [Mary-ga t1 t2 ageta ka] siri-tagatteiru no?

what-Acc John-Top Mary-Nom

gave Q want-to-know Q

‘Lit. Who what does John want to know whether Mary gave?’ (Takahashi 1993:670) Takahashi (1993) observes that in (192), the scope of the scrambled wh-phrases is fixed in the matrix clause. Since VP scrambling is mandatory in the case of long-distance multiple scrambling under the present analysis, this scope fact indicates that the scrambled VP is assigned a [Focus] feature and further that this feature should be able to be merged with the [WH] features borne by the two wh-phrases, as shown below: (193) [CP [TP [VP dare-ni nani-o tV]1 [TP John-wa2 [vP t2 [CP [TP Mary-ga3 [vP t3 t1 [WH-F] age+v]ta] ka] siri-tagatteiru]]] no] [+WH]

[+WH]

The question to be asked is how the [WH] features of the wh-phrases have managed to percolate up to the whole VP. I suggest that there are two relevant properties that make 122

this possible: (i) the scrambled VP is headless and (ii) VP is a lexical projection. Suppose, following Fukui and Speas (1986) and Fukui (1986), that specifier does not exist in lexical projections and further that in that case, the features of a phrase within any projection of a lexical head can percolate up in principle unless they do not conflict with those borne by its sister. With this assumption in mind, let us now consider the internal structure of the scrambled VP in (193), which I assume to be the following: (194)

VP

dare-ni

V’

[WH] nani-o

V

[WH] Since V has been moved out in this case, it is reasonable to assume that the [WH] feature of nani-o can percolate up to the dominating V’. Now that this V’ and its sister dare-ni share [WH] features, both of these features can percolate up to the whole VP. In this way, the two [WH] features can be successfully merged with the [Focus] feature assigned to the whole VP. Given this analysis, it is predicted that if one of the phrases preposed by VP scrambling is a non-wh-phrase, then the wh-feature of the other phrase is unable to percolate up to the whole VP, thereby preventing [Focus] feature assignment and that this should give rise to the scope ambiguity of the preposed wh-phrase. This prediction is in fact borne out by such an example as the following, provided by Watanabe (2000):

123

(195) Dare-ni1 ringo-o2

John-wa [Mary-ga t1 t2 ageta ka] siri-tagatteiru no?

who-Dat apple-Acc John-Top Mary-Nom

gave Q want-to-know Q

‘Who does John want to know whether Mary gave an apple to?’ ‘Does John want to know who Mary gave an apple to?’

(Watanabe 2000:289)

In this example, one of the two objects preposed to the top of the sentence is the non-wh-phrase ringo-o ‘apple-Acc’. Watanabe (2000) observes that this sentence has not only the reading where dare-ni ‘who-Dat’ takes matrix scope but also the one in which it takes embedded scope. Under the present analysis, this sentence has the following representation: (196) [CP [TP [VP dare-ni ringo-o tV]1 [TP John-wa2 [vP t2 [CP [TP Mary-ga3 [vP t3 ([WH])

t1

([WH])

age+v]ta] ka] siri-tagatteiru]]] no] [+WH]

[+WH]

Here, the [WH] feature of dare can be carried along when the embedded VP is scrambled, thereby giving rise to its matrix scope reading or it can be left behind, thereby giving rise to its embedded scope reading. Much the same argument can be made regarding long-distance multiple scrambling of QPs. I have argued in Section 2.4 that [Scope] features also percolate up, so that an expression such as dareka-no musume ‘someone’s daughter’ as a whole behaves like a QP. Consider (116a) again, reproduced below: (197) Dareka-no

musume-ni1 daremo-ga

[sensei-ga t1 kisusita to]

someone-Gen daughter-Dat everyone-Nom teacher-Nom kissed 124

Comp

sinziteiru. believe ‘Someone’s daughter, everyone believes that the teacher kissed.’ This sentence does not have the reading in which daremo takes scope over dareka. This follows from the ECDII, reproduced below, with the assumption that dareka-no musume-ni ‘someone’s daughter-Dat’ as a whole is identified as a QP, hence carrying a [Scope] feature: (198) The Economy Condition on Dependency II (ECDII) If αi is a member of the chain (α1, ... αi, ...), then αi+1 must be the closest possible dependent for αi. The LF representation of the wide scope reading of daremo for (197) is represented as follows: (199) *[TP dareka-no musume-ni1 [TP daremo-ga2 [vP t2 [CP [TP t’1 [TP sensei-ga t1 [Focus]

[Scope]

[Scope]

kisusita]] to] sinziteiru]]] Here, dareka-no musume-ni stops by the embedded TP-adjoined position on its way to the matrix TP-adjoined position to license its [Scope] feature without violating the clause-boundedness condition on QR. Then, the member marked as t’1 is not the closest possible dependent for its next higher member carrying a [Focus] feature, violating the ECDII, since daremo ‘everyone’ serves as a possible dependent for the latter member and is closer to it than the former member. Hence, the only reading available to (197) is the one that is derived from the representation in which dareka-no musume does not 125

carry a [Scope] feature, which amounts to its specific reading. On the other hand, when a clause including a QP undergoes long-distance scrambling, the reading unavailable to (197) becomes available, as shown in (117a), reproduced below: (200) [Dareka-ga

Mary-o

sukida to]1

someone-Nom Mary-Acc like to]

daremo-ga

Comp

[sensei-ga t1

itta

everyone-Nom teacher-Nom said

sinziteiru.

Comp believe ‘That someone likes Mary, everyone believes that the teacher said.’ Here daremo ‘everyone’ can take scope over dareka ‘someone’. This is because the clause scrambled long-distance as a whole cannot be identified as a QP and hence the [Scope] feature of dareka is allowed to be left behind when the clause in question is scrambled, as shown below: (201) [TP [dareka-ga Mary-o sukida to]1 [TP daremo-ga2 [vP t2 [Scope]

[Focus]

[CP [TP sensei-ga 1 itta] to] sinziteiru]]] [Scope] This representation represents the reading where daremo takes scope over dareka, since the former asymmetrically c-commands the latter. Let us now consider a case that involves long-distance multiple scrambling of QPs such as the following: (202) Dareka-ni1

nanika-o2

daremo-ga

[sensei-ga t1 t2

someone-Dat something-Acc everyone-Nom teacher-Nom 126

ageta to] gave Comp

sinziteiru. believe ‘Lit. Someone something, everyone believes that the teacher gave.’ This sentence does not have the reading where daremo ‘everyone’ takes scope over the two indefinites preposed to the top of the sentence. Under the present analysis, it will have the following representation: (203) *[CP [TP [VP dareka-ni nanika-o tV]1 [TP daremo-ga2 [vP t2 [CP [TP [Focus]

[Scope]

t’1 [TP sensei-ga3

[Scope]

[vP t3 t1 age+v]ta]] to] sinziteiru]]]] Given the above assumption about the way the features of phrases percolate up within a headless VP (see the discussion around (194)), it must be the case that the [Scope] features of the two indefinites dareka and nani both percolate up to the whole VP. Suppose that in that case, the two QPs are absorbed into a binary quantifier in the sense of Higginbotham and May (1981): (204) [some x: x a person] [some y: y a thing] -> [some x, y: x a person & y a thing] Then, the whole VP is identified as a binary QP and hence its [Scope] feature must be licensed in the embedded TP-adjoined position, just like in (199). This induces a violation of the ECDII, as seen above. The only reading available to (202) is the one that is derived from the specific readings of dareka and nanika. Again, it is predicted that if one of the indefinites preposed by VP scrambling in (202) is changed into a non-QP, then the [Scope] feature of the other phrase is unable to percolate up to the whole VP, hence being allowed to be left behind when VP 127

scrambling takes place, and hence that the reading unavailable to (202) should become available. This prediction is in fact borne out: (205) Dareka-ni1

sono

someone-Dat that

ringo-o2

daremo-ga

[sensei-ga t1 t2

apple-Acc everyone-Nom teacher-Nom

ageta to] gave Comp

sinziteiru. believe ‘Lit. Someone that apple, everyone believes that the teacher gave.’ This sentence has the reading where daremo ‘everyone’ takes scope over dareka ‘someone’. This is exactly what the present analysis predicts, since the whole VP scrambled long-distance cannot be identified as a QP and hence the [Scope] feature of dareka can be left behind when this VP is scrambled, just like in (201), so that this QP takes scope under daremo. To sum up this chapter, I have argued that scrambling is reinterpreted as focus movement whenever possible. This is most convincingly demonstrated by those cases that involve long-distance scrambling of wh-phrases and QPs. I have addressed an issue regarding the possibility of LF “undoing” of long-distance scrambling of wh-phrases with which apparently contradictory claims have been made: whereas Saito (1989) shows, on the basis of so-called radical reconstruction effects, that long-distance scrambling can be undone, Takahashi (1993) shows that some instances of long-distance scrambling of wh-phrases behave like operator movement and hence cannot be undone. I have provided a solution to this problem by claiming that long-distance scrambling of a wh-phrase produces a focus chain unless this choice leads 128

the derivation to crash. This analysis regards Takahashi’s (1993) case as unmarked and Saito’s case as the one in which the assignment of a [Focus] feature to a wh-phrase scrambled long-distance leads the derivation to crash and hence the alternative derivation without [Focus] feature assignment is chosen. Only in the latter case does long-distance scrambling produce a semantically vacuous chain. The hypothesis that wh-phrases scrambled long-distance give rise to focus chains has received further support from (i) Takahashi’s (1993) observation that long-distance scrambling of wh-phrases exhibits superiority effects and (ii) Sohn’s (1994) observation that a wh-phrase that has undergone long-distance scrambling cannot save a wh-adjunct from an island violation even when if it had stayed in its original position, it could have done so. I have also argued that those QPs scrambled long-distance interact with other QPs in the predicted ways under our hypothesis that they produce focus chains. In particular, I have refuted Oka’s (1989) claim that chains produced by long-distance scrambling are unlicensed and hence must be undone, by arguing that the fact that QPs scrambled long-distance cannot take scope in the scrambled positions is naturally explained by the clause-boundedness condition on QR. As for clause-internal scrambling, I have addressed the question why clause-internal scrambling of wh-phrases does not behave as if it produced a focus chain, in particular why superiority effects are not observed in clause-internal scrambling. I have argued, following Abe (1993), that clause-internal scrambling can be too short to license [Focus] feature assignment, hence not establishing a focus chain. In that case, the resulting chain is taken as semantically vacuous or ends up being part of a larger chain. Finally, I have shown good 129

consequences of our hypothesis that scrambling produces a focus chain whenever possible by examining the behaviors of multiple scrambling of wh-phrases and QP. I have argued for Koizumi’s (1995, 2000) VP scrambling approach to such cases and have reached the conclusion that VP scrambling is always involved in the case of long-distance multiple scrambling while literal multiple scrambling is available for limited cases of clause-internal multiple scrambling.

130

3

Scrambling as A-Movement: A Residue of NIC and Super-Raising

In this chapter, I argue that what appears to be an instance of A-movement in Japanese in fact involves scrambling. In so doing, I address the question why Japanese allows A-movement to take place across a nominative subject position without inducing a violation of a minimality condition of the sort that derives the effects of Nominative Island Condition (henceforth NIC), proposed by Chomsky (1980). There are at least three cases in Japanese that instantiate this property: (1) Nominative-genitive conversion: a.

[John-ga

sukina] hon

John-Nom like

book

‘the book John likes.’ b. [John-no sukina] John-Gen like

hon book

(2) Subject-to-subject raising: John-ga1

saikin [t1 motto

John-Nom recently more

benkyoosuru yoo-ni]

natta.

study

became

Comp

‘Recently, John has come to study harder.’ (3) Subject-to-object raising:

131

a.

John-ga

[Bill-ga baka

da

to]

John-Nom Bill-Nom stupid be

omotteiru.

Comp think

‘John thinks that Bill is stupid.’ b.

John-ga [Bill-o

baka da

to]

John-Nom Bill-Acc stupid be Comp

omotteiru. think

(1) illustrates the phenomenon of so-called nominative-genitive conversion (henceforth, NGC) in which the subject of the relative clause, which is normally marked with nominative Case, can be marked with genitive Case, as shown in (1b). Given the standard assumption that genitive Case is licensed in Spec-DP, it is natural to assume that a genitive subject such as John-no in (1b) undergoes movement to Spec-DP, as shown below: (4) [DP John-no [TP [vP John-no sukina]] hon]

Note that this movement takes place from the subject position of a finite clause across the embedded Spec-TP, in which John could be assigned nominative Case. Uchibori (2000, 2001) regards (2) as a case of subject-to-subject raising (henceforth, SSR) in which movement takes place from the subject position of a finite clause. Hence if Uchibori’s claim is right, this construction also instantiates a case where A-movement skips a nominative subject position. Finally, (3b) illustrates a case of subject-to-object raising (henceforth, SOR), according to Kuno (1976), in which the embedded subject Bill-o is assigned accusative Case by the higher verb. Under this assumption, (3b) will involve such movement as shown below: 132

(5) John-ga [VP Bill-o [CP [TP [vP Bill-o baka da]] to] omotteiru]

Note that in this case as well, the movement takes place from the subject position of a finite clause across the embedded Spec-TP in which Bill could be assigned nominative Case. These three cases of A-movement must have been taken as violations of the NIC under the framework in which subject is base-generated under S and the trace left by A-movement obeys the same locality conditions as such expressions as reflexives and reciprocals (cf. Chomsky 1976, 1980, 1981, among others). A typical instance of A-movment is illustrated by so-called subject raising: (6) John seems [TP t to be honest] Here the trace of John in the embedded Spec-TP (more traditionally, the subject position of S) is assumed by Chomsky (1980) to obey the Specified Subject Condition (henceforth, SSC), which prohibits A-movement from taking place across a subject, and the NIC, which prohibits A-movement from taking place from a nominative Case position.1 The A-movement of John in (6) satisfies these two conditions. Another instance of A-movement is illustrated by so-called raising to object, though this operation has been long denied until Lasnik and Saito (1991) revived Postal’s (1974) arguments for it:

1

The NIC was proposed by Chomsky (1980) as a replacement of Tensed-S Condition,

adopted by Chomsky (1976), which prohibits A-movement from taking place across a tensed clause.

133

(7) Mary believes John [TP t to be honest] Just like a subject raising case such as in (6), this raising to object also takes place from the embedded subject position of an infinitival clause, hence satisfying both the SSC and the NIC. On the other hand, the following example is ruled out as a NIC violation, since the trace of John occupies the nominative subject position: (8) *John seems (that) [TP t is honest] Likewise, the three Japanese constructions in question should also induce NIC violations. Interestingly, this fact correlates with the standard observation that Japanese does not show NIC effects regarding anaphors. Thus, compared with the English example (9), the Japanese example (10) demonstrates that the reciprocal otagai can violate the NIC. (9) *John and Mary thought that themselves/each other would leave. (10) John to Mary-ga

[otagai-ga

atama-ga ii

to]

omotteiru.

John and Mary-Nom each other-Nom head-Nom good Comp think ‘John and Mary think that each other is smart.’ Given this fact, at the time when it was assumed that A-trace is subject to the NIC as well as the SSC, it would have been predicted that A-trace can also violate the NIC in Japanese, and the three constructions in question seem to verify this prediction. Under the Minimalist Program, however, such a representational constrainst as the SSC and the NIC is abandoned and instead alternative constraints are formulated in terms of economy conditions on movement that produces an A-chain. Thus, such an NIC violation case as (8) is ruled out by the last resort principle on movement: given 134

that A-movement is triggered for a Case reason, John cannot move up from the embedded Spec-TP to the matrix one since its Case is already checked in the embedded Spec-TP.2 Or alternatively, if John skipped the embedded Spec-TP and moved directly to the matrix one, then the movement would violate not only the EPP requirement that Spec-TP be occupied by DP but also Chomsky and Lasnik’s (1993) Minimize chain links (henceforth, MCL), which requires that a moved element cannot skip a possible landing site. Given this minimalist approach to NIC violation cases such as (8), a question arises as to why the three Japanese constructions mentioned above are not subject to the relevant economy conditions. The violability of these conditions is most clearly witnessed by cases of the NGC and SOR, where Case alternation is possible: in (4), John-no moves to the Spec-DP, passing the Spec-TP within the relative clause, in which it could receive nominative Case and in (5), Bill-o moves to the higher Spec-VP (cf. Chomsky 2008), passing the embedded Spec-TP, in which it could receive nominative Case. We may regard this question as another case of the “residue of NIC.” The purpose of this chapter is two-fold. One is to provide a solution to our version of the residue of NIC by proposing a mechanism in which the A-chains involved in the three Japanese constructions are produced by scrambling. The other is to examine the properties of the A-chains produced by scrambling with respect to pronunciation and LF interpretation. With this demonstration, this chapter aims to give support to the kind of 2

See Chomsky (1995) and his later works for a more refined explanation in terms of

the operation of Attract or the probe-goal system.

135

approach, advocated recently by Hornstein (2009), in which the effect of overt vs. covert movement is captured by which copy of a produced chain is pronounced. This contrasts crucially with the Agree-based approach, proposed by Chomsky (2000), according to which long-distance dependency is captured by the operation of Agree rather than movement itself.

3.1

Proposal: Scrambling as Mediator

3.1.1

Scrambling as an Instantiation of Overt Operator Movement: Abe (2012)

Abe (2012) argues that what is standardly claimed to be an instance of overt operator movement in Japanese in fact involves scrambling. The argument goes as follows: First, Abe calls our attention to the observation, made in Section 2.1, that among those constructions that involve overt operator movement, there are cases in which the trigger of such movement will not be appropriately attributed to the inherent property of the head of a phrase to be moved. Cases in point are shown below: (11) a.

[John], Mary likes very much.

b. [The man you’re talking about], Mary likes very much. c.

John met yesterday [the woman you had been talking about].

(11a) and (11b) are instances of topicalization and (11c) an instance of heavy NP shift. In these cases, it does not seem possible to identify the locus of the trigger of the movement involved. Based upon this observation, Abe (2012) proposes that a relevant triggering feature is not carried by a particular head but rather assigned to a whole category that undergoes movement. Identifying the relevant feature as [Focus], Abe 136

then puts forward the following hypothesis: (12) The feature [Focus] can be assigned to a syntactic object during the derivation. Given the last resort principle on movement, assignment of [Focus] must take place before the movement of its bearer. However, Abe (2012) reasons that if we assume, following Fukui (1993) among others, that scrambling is immune to the last resort principle, a scrambling-language such as Japanese will have the possibility that [Focus] is assigned to a phrase after it undergoes trigger-free movement, i.e., scrambling. Thus, under this hypothesis, a Japanese cleft sentence such as (13) below can have the derivation given in (14), if we assume the deletion analysis of Japanese clefting, taken by Takano (2002), among others.3 (13) [John-ga John-Nom

hihansita no]-wa

Mary-o

da.

criticized NL-Top Mary-Acc be

‘It was Mary that John criticized.’ (14) a.

[Mary-o John-ga hihansita no]-wa Mary-o da (scrambling)

b. [Mary-o John-ga hihansita no]-wa Mary-o da [Focus] c.

([Focus] assignment)

[Mary-o John-ga hihansita no]-wa Mary-o da [Focus]

(deletion of Mary-o under identity)

Abe (2012) provides a couple of pieces of evidence for this derivation. One has to do 3

NL in the gloss stands for nominalizer. I will not be committed to the exact syntactic

status of the morpheme no in this manuscript.

137

with the fact that Japanese clefting shares with scrambling the property that it shows a very weak effect with respect to the wh-island though it shows a rather clear sensitivity to other islands. The following examples demonstrate that Japanese clefting is sensitive to the relative clause and adjunct clause islands: (15) a.?*[John-ga [Mary-ni t1 ageta] hito-o John-Nom Mary-Dat gave

kiratteiru no]-wa [sono hon-o]1

person-Acc hate

da.

NL-Top that book-Acc be

‘It is that book1 that John hates the person [who gave t1 to Mary].’ b.?*[John-ga [Mary-ga t1 John-Nom Mary-Nom [sono hito-o]1

syokuzi-ni sasotta node]

rakutansiteiru no]-wa

dinner-to invited because is-depressed NL-Top

da.

that person-Acc be ‘It is that person1 that John is depressed [because Mary invited t1 to dinner].’ (Abe 2012:67) Unlike these islands, the wh-island gives rise to only a weak effect with Japanese clefting, as shown below: (16) ?[John-ga [Bill-ga

dare-ni t1 ageta ka] siri-tagatteiru no]-wa [sono hon-o]1 da.

John-Nom Bill-Nom who-Dat gave Q want-to-know NL-Top that book-Acc be ‘It is that book1 that John wants to know who2 Bill gave t1 to t2.’

(ibid.:67)

A similar pattern of facts is observed by Saito (1985) with scrambling. Thus, (17a, b) demonstrate that scrambling is sensitive to the relative clause and adjunct clause islands, whereas (18) shows that scrambling shows a rather weak effect with the wh-island.

138

(17) a.?*[Sono hon-o]1 [John-ga

[Mary-ni t1 ageta] hito-o

that book-Acc John-Nom Mary-Dat gave

kiratteiru (koto)

person-Acc hate

fact

‘That book1, John hates the person [who gave t1 to Mary].’ b.?*[Sono hito-o]1

[John-ga

[Mary-ga t1 syokuzi-ni sasotta node]

that person-Acc John-Nom Mary-Nom dinner-to invited because rakutansiteiru (koto) is-depressed

fact

‘That person1, John is depressed [because Mary invited t1 to dinner].’ (18) ?[Sono hon-o]1 [John-ga [Bill-ga

dare-ni t1 ageta ka] siri-tagatteiru (koto)

that book-Acc John-Nom Bill-Nom who-Dat gave Q want-to-know fact ‘That book1, John wants to know who2 Bill gave t1 to t2.’

(ibid.:68)

According to Abe (1993), the fact that scrambling does not show wh-island effects is attributed to its nature as a “costless” movement in that it does not require any triggering feature. Abe follows Chomsky and Lasnik (1993) in assuming that wh-island effects are derived by MCL, which dictates that an application of movement cannot skip possible landing sites. Abe defines a possible landing site in terms of triggering features, as follows: (19) The possible landing sites for movement of α are possible checking positions of the feature(s) borne by α. Given that scrambling does not have any triggering feature, it follows that this operation

139

is insensitive to MCL, hence not exhibiting wh-island effects.4 The fact that scrambling does show other island effects, then, indicates that such effects have nothing to do with economy conditions that regulate movement triggered by feature-checking, but rather that such effects are derived in purely configurational terms such as the notion of barriers exploited by Chomsky (1986). Secondly, it has been observed in the literature that Japanese clefting allows more than one phrase to appear in the focus position, as illustrated below: (20) [John-ga John-Nom

ageta no]-wa [Mary-ni ringo-o] da. gave NL-Top Mary-Dat apple-Acc be

‘Lit. It was [an apple to Mary] that John gave.’ This is not a property widely observed among languages, as witnessed by the unacceptability of the English translation of (20). Abe (2012) attributes this peculiarity to the availability of scrambling to this langauge, which can be applied to more than one phrase, as shown below: (21) Mary-ni Mary-Dat

ringo-o

John-ga

ageta (koto)

apple-Acc John-Nom gave

fact

‘Lit. Mary an apple, John gave.’

4

An alternative minimality condition can be Minimal Link Condition (henceforth,

MLC), proposed by Chomsky (1995), which is formulated in terms of the operation of Attract. See Abe (forthcoming) for the claim that the movement operation in fact consists of two sub-operations, i.e., Search and Float, and that the former is subject to the MLC and the latter to MCL. Under this mechanism of movement, Abe (forthcoming) attributes wh-island effects to violations of the MLC.

140

Under the present assumptions, (21) will be derived in the following way:5 (22) a.

[Mary-ni ringo-o John-ga ageta no]-wa Mary-ni ringo-o da (multiple scrambling)

b.

[Mary-ni ringo-o John-ga ageta no]-wa Mary-ni ringo-o [Focus] [Focus]

da ([Focus] assignment)

c.

[Mary-ni ringo-o John-ga ageta no]-wa Mary-ni ringo-o [Focus] [Focus]

da (deletion of Mary-ni and ringo-o under identity)

In this way, the availability of multiple clefting can be attributed to that of multiple scrambling.

3.1.2

An Extension to A-Movement

Building upon the above mechanism for overt operator movement in Japanese, I propose that this mechanism be extended to cases of A-movement, so that scrambling may be involved in deriving the Japanese constructions mentioned above, i.e., NGC, SSR and SOR. Let us assume that the following holds for the Case checking system of

5

(21) can be derived by way of remnant VP scrambling. See Section 2.6 for this

possibility. The point will not be affected, however, as long as scrambling is involved in the derivation.

141

Japanese:6 (23) a.

The Case-feature can be assigned to a syntactic object during the derivation.

b. Case-feature checking may be conducted via the Spec/head relation. Given these assumptions, the three constructions in question can have the derivations given below:7 (24) Nominative-genitive conversion: a.

[DP John [TP [vP John sukina]] hon] (scrambling)

b. [DP John-no [TP [vP John-no sukina]] hon] (genitive Case-feature assigned and checked in Spec-DP)

6

Given that Japanese does not have any overt manifestation of agreement, it is not

unlikely that φ-feature checking simply does not apply to this language. Thus I concentrate on Case-feature checking in the text. The present mechanism of movement is very similar to Goto’s (2011), according to which Move can be applied to a phrase before that phrase is attracted, or probed in the more recent terminology, in some languages. Goto does not characterize the relevant movement as scrambling, however. 7

I assume that a Case-feature is assigned to a chain rather than a single member of that

chain. This may be implemented by assigning the same Case-feature to all the members of a chain, as indicated in the (b)-representations of (24)-(26) or by a sort of inheritance mechanism, adopted by Chomsky (1981), who characterizes the relevant condition on Case assignment as follows: (i) Case is assigned to an index [of an A-chain] and inherited by lexical [=pronounced] NP with this index.

(Chomsky 1981:268)

Such an assumption is necessary because we will be discussing in the next section which copy of a produced A-chain is pronounced.

142

(25) Subject-to-subject raising: a.

[TP John [CP [TP [vP John motto benkyoo-suru]] yoo-ni] natta] (scrambling)

b. [TP John-ga [CP [TP [vP John-ga motto benkyoo-suru]] yoo-ni] natta] (nominative Case-feature assigned and checked in the matrix Spec-TP) (26) Subject-to-object raising: a.

John-ga [VP Bill [CP [TP [vP Bill baka da]] to] omotteiru] (scrambling)

b.

John-ga [VP Bill-o [CP [TP [vP Bill-o baka da]] to] omotteiru] (accusative Case-feature assigned and checked in the matrix Spec-VP)

In each derivation, the target phrase undergoes scrambling, carrying no feature relevant for the movement. Hence even though it skips a position in which it could have its Case-feature checked, this will not give rise to a violation of MCL. So as long as it reaches the Spec position of a head appropriate for checking the Case assigned to it after scrambling, its Case requirement is satisfied in accordance with (23b). The only relevant conditions on scrambling that have been assumed so far are island conditions defined in terms of barriers in Chomsky’s (1986) sense. Interestingly, such a condition appears to be violated in the case of (24a), where John-no has moved out of the relative clause; I will address the question why this is possible directly. Notice that in order for this mechanism of movement to work properly, we need to say something about the status of the EPP in Japanese. If the EPP were at work for this language in the same way as for English, then all the derivations indicated in (24)-(26) would violate this 143

requirement, since the embedded Spec’s of TP are all empty. Let us then assume without any further argument that the EPP is not operative in Japanese, an assumption that may be innocuous, given that there is no strong evidence that the Spec of TP must be occupied in this language; insertion of an overt expletive would be a good indication of such a requirement, but no such element is attested in this language as far as I know. I have proposed above that scrambling is involved in producing A-chains in the three constructions in question, but we need to provide a more precise characterization of this operation. It has usually been taken for granted that scrambling is an instance of overt movement, but this assumption should not hold for those constructions, since some cases clearly show that the bottom copy of the resulting chain is pronounced: (27) [kinoo

John-no

kita]

yesterday John-Gen came

riyuu reason

‘the reason that John came yesterday’ (28) John-ga [mada John-Nom still

Mary-o

kodomo da to]

Mary-Acc child

omotta.

be Comp thought

‘John thought that Mary was still a child.’

(Hiraiwa 2001b:72)

(27) is a case in which the genitive-marked subject is preceded by a clausal modifier such as kinoo ‘yesterday’ and (28) is a case in which the accusative-marked subject is preceded by such a modifier as mada ‘still’ that is interpreted as modifying the embedded predicate.8 Under the present assumptions, these sentences will have the 8

Tanaka (2002) questions the acceptability of (28) and instead provides the following

example as totally unacceptable:

144

following representations: (29) a. b.

[DP [TP kinoo [vP John-no kita]] riyuu] John-ga [VP [CP [TP [vP mada Mary-o kodomo da]] to] omot]ta

In both representations, John and Mary have moved to the Spec-DP and the matrix Spec-VP, respectively, by scrambling and the bottom copies are pronounced. Thus, the movement operation that is involved in producing A-chains in the three constructions in question must be characterized as one that has no triggering feature and has optionality with respect to which copy is pronounced. This conclusion may be taken as indicating that “covert scrambling” is available in general. However, this does not seem to be in consonance with the idea that scrambling plays its main role on the PF side, i.e., to change word order and that this operation induces no semantic effect in itself, as Saito (1989) characterizes it as semantically vacuous movement. A typical example is reproduced below: (30) ?Dono

hon -o1

[Mary-ga [John-ga t1 tosyokan-kara karidasita

ka]

which book-Acc Mary-Nom John-Nom library-from checked-out Q

(i) *John-ga

[mada

John-Nom still

Mary-o

kodomo da

Mary-Acc child

be

‘John decided that Mary was still a child.’

to]

danteisita.

Comp decided (Tanaka 2002:647)

I am not sure how bad this sentence is, but it seems relatively clear that Hiraiwa’s example (28) is fairly acceptable; it seems to me that (i) is a little bit worse than (28). Although I may agree on Tanaka’s (2002) claim that “[28] does not conclusively support the optionality of RTO [= raising-to-object],” (p. 647), it is argued in the text that there is ample evidence that RTO is optional.

145

siri-tagatteiru] (koto) want-to-know fact ‘Lit. Which book, Mary wants to know Q John checked out from the library.’ In this sentence, the wh-phrase dono hon-o ‘which book-Acc’ is scrambled out of the embedded question clause. If this phrase remained at the top of the sentence, then its [WH] feature would not be licensed by the embedded [+WH] Comp. Hence, in this case, the resulting chain is reduced into a single-membered chain by deleting the top copy. Notice that the scrambling of dono-hon-o has a PF effect in the sense that it changes the word order of this sentence. Given an interface condition such as Chomsky’s (1995) to the effect that movement must have an effect on either the PF or LF interface, it will make sense to claim that a movement operation that has no LF effect must have a PF effect, as is the case with (30). However, such a condition also implies that a movement operation that has no PF effect is allowed as long as it gives rise to an LF effect. Notice that in such cases as (29a, b) the chains produced by scrambling have ended up with legitimate LF objects whose heads were involved in Case checking, hence no violation of the interface condition in question. Thus, we can reasonably conclude that “covert scrambling” is allowed as long as it has an effect on the LF interface. Under the Agree-less approach advocated by Hornstein (2009), Abe (2016) proposes the following with respect to which copy of an A-chain is pronounced: (31) In the case of an A-chain, any member can be the target for pronunciation. With this proposal, Abe (2016) captures the optionality of “raising to object” in ECM constructions in English, observed by Lasnik (1999). Thus, a sentence such as (32) will 146

have either (33a) or (33b), depending upon which copy of the produced A-chain is pronounced: (32) John believes Mary to be honest. (33) a.

John [vP v+believes [VP Mary tbelieves [TP to be honest]]]

b. John [vP v+believes [VP tbelieves [TP Mary to be honest]]] This optionality regarding pronunciation is more clearly shown with the make-out construction, first noted by Kayne (1985): (34) a.

John made Mary out to be a fool.

b. John made out Mary to be a fool.

(Lasnik 1999:201)

In this case, it is reasonable to claim that Mary is pronounced either in the higher Spec-VP, giving rise to (34a), or in the embedded Spec-TP, giving rise to (34b). Suppose, following Johnson (1991), that make+out is base-generated in V and only make undergoes head-movement to v, stranding the particle out in its original position, so that the schematic representations of (34a, b) are as follows: (35) a.

John [vP v+make [VP Mary tmake+out [TP to be a fool]]]

b. John [vP v+make [VP tmake+out [TP Mary to be a fool]]] Abe (2016) further claims that the fact that usually the top copy is pronounced in a typical A-chain such as one produced by subject raising, as shown below, is attributed to the EPP requirement. (36) John seems [TP to be honest]. Abe characterizes the EPP as follows: (37) a.

The EPP requires that Spec-TP must be occupied by an overt phrase. 147

b. Only the Spec of a tensed T is subject to the EPP. Given this characterization, the top copy of the John-chain must be pronounced in (36) in order to satisfy the EPP in the way stated in (37a). In the case of ECM, on the other hand, the EPP is off according to (37b) since no tensed T is involved in this construction, thus making “raising to object” optional, as indicated in (33) and (35). Given this mechanism of pronunciation, it is predicted that the A-chains involved in the three Japanese constructions under consideration are also allowed to have any member pronounced, since the EPP is inoperative in this language. This prediction is borne out with the SOR construction; we have not only a case such as (28), reproduced below, in which the bottom copy of the A-chain is pronounced, but also a case such as (39), in which the top copy is pronounced. (38) John-ga [mada John-Nom still

Mary-o

kodomo da to]

Mary-Acc child

omotta.

be Comp thought

‘John thought that Mary was still a child.’ (39) Yamada-wa Tanaka-o Yamada-Top Tanaka-Acc

orokanimo [tensai da

to]

stupidly

Comp thought

genius be

‘Yamada stupidly thought that Tanaka was a genius.’

omotteita.

(Kuno 1976:25)

In (39), the accusative-marked subject Tanaka-o precedes the matrix adverbial orokanimo ‘stupidly’, which thus indicates that the top copy of Tanaka-o is pronounced, as shown below:9 9

In the case of SSR, example (2) clearly illustrates a case in which the top copy of the

produced A-chain is pronounced. However, it is not an easy task to find a relevant

148

(40) Yamada-wa [VP Tanaka-o orokanimo [CP [TP [vP tensai da]] to] omotteita In the case of NGC, it has been generally claimed10 that “overt” movement of a genitive-marked subject is impossible, as shown most clearly by Terada (1987) with the following examples: (41) a.

[[utukusii] [John-no sukina] basyo] beautiful John-Gen like

place

‘the beautiful place that John likes’ b. *[John-no1 [utukusii] [t1 sukina] John-Gen beautiful

basyo]

like

place

(41b) would be derived by moving John to the above Spec-DP by scrambling, assigning genitive Case to it and then deleting the bottom copy, as indicated below: (42) *[DP John-no [NP utukusii [TP [vP John-no sukina]] basyo]]

example demonstrating that the bottom copy is pronounced. One might think that an example like the following is a case in point: (i)

[Nagai long

aida

John-ga

benkyoosuru yoo-ni]

period John-Nom study

Comp

natta. became

‘John has come to study for a long time.’ In such a case, however, it may be claimed that John-ga simply moves to the embedded Spec-TP to check its nominative Case and the matrix Spec-TP is left empty. For clearer cases, see Section 3.4.1, which discusses “backward binding” of this construction. 10

There is one notable exception: Ochi (2001). See Appendix for discussion on those

cases of NGC that are claimed to involve overt movement of genitive-marked subjects.

149

The unacceptability of (41b) then indicates that the top copy of the produced A-chain is not pronounced in this case. Recall that we have raised the question why scrambling of a subject that will be marked with genitive Case in this construction does not induce a violation of the relative clause island. Abe (2016) relates this violability of an island with the above fact that the top copy of the produced A-chain cannot be pronounced. Abe adopts Abe and Hornstein’s (2012) PF condition that regulates the determination of which copy of the resulting chain is pronounced: (43) The head of a chain produced by Move cannot be pronounced unless it has an effect on PF output. Abe and Hornstein motivate this PF condition by analyzing the right node raising (henceforth, RNR) construction in English, whose representative example is given below: (44) John likes

and Bill hates

, the linguistic professor teaching Ling 101.

They adopt Ross’s (1967) Across-the-Board (henceforth, ATB) movement analysis of this construction, according to which the shared element undergoes movement out of each conjunct in an ATB fashion and is right-adjoined to the whole coordinate structure, as shown below: (45) [Clause [Clause John likes

] and [Clause Bill hates

teaching Ling 101]] 150

] [the linguistic professor

Since Ross proposed this analysis, a major obstacle to it has been how to characterize the fact that RNR is immune to the bounding conditions, including island conditions, imposed upon rightward movement. Thus, it is free from the Right Roof Constraint (henceforth, RRC); compare (46a) with (46b):11 (46) a.*Max said [that he was going to return

to the library] yesterday each of the

books that he checked out last week.

(Sabbagh 2007:350)

b. Josh said that he thought that I should sell thought that she might want to buy

, and Jamie said that she

, each of the Rambaldi artifacts that

I have in my attic.

(ibid.:358)

Further, it has been well known since Wexler and Culicover (1980) that RNR is insensitive to islands in general, as shown below: (47) I know someone who wants to buy sell

, and you know someone who wants to

, a copy of this manuscript.

(48) Josh wonders who bought

(ibid.:352)

, and Bill will find out who sold

Fred. (49) Politicians win when they defend

, pictures of (ibid.:382)

, and lose when they attack

of a woman to an abortion.

, the right (ibid.:382)

These RNR examples have the following schematic structure:

11

Recently, Sabbagh (2007) supports an ATB movement analysis of RNR in English

by addressing the question why ATB movement shows insensitivity to locality conditions in RNR. Abe and Hornstein (2012) propose an alternative solution to this question. I have taken relevant RNR examples from Sabbagh (2007).

151

(50) [ … DP1] and [… DP2]

DP3

Here DP3 is an occurrence of the shared DP that has undergone ATB rightward movement and DP1 and DP2 are the occurrences in the sites from which it has undergone this movement. Thus, we have two chains whose head is shared: (DP3, DP1) and (DP3, DP2). Abe and Hornstein propose that while RNR involves overt ATB movement as a default case, hence DP3 being the target of pronunciation, the PF condition in (43) dictates that DP2 must be pronounced when the chain (DP3, DP2) involves string-vacuous movement. Thus, in all the RNR examples that have been presented so far, the shared DPs are pronounced in their original positions of the second conjuncts. Abe and Hornstein further propose the following: (51) Locality conditions such as the RRC and the island conditions apply only to “overt” movement. Given this proposal, the insensitivity of locality observed with the RNR examples above follows since in these cases, the shared DPs are pronounced in their original positions of the second conjuncts, hence taken to have undergone “covert” movement due to the string-vacuity of movement involved. It is predicted under this mechanism that when non-string-vacuous movement is involved in producing the chain (DP3, DP2) in (50), the latter chain is sensitive to locality conditions. This is borne out by the following RNR example, which gives rise to an RRC violation: 152

(52) *Joss said [that he was going to donate claimed [that she would donate

to the library] yesterday, and Jamie

to the museum] last week, a large

collection of ancient texts.

(ibid.:355)

Here the shared DP is moved out of the embedded finite clause in each conjunct in an ATB fashion, and the movement involved in producing the second chain is not string-vacuous, hence taken as an “overt” movement. This induces a violation of the RRC. Going back to the case of NGC, we now see why the top copy of the resulting chain cannot be pronounced in this construction: this would induce a violation of the relative clause island. On the other hand, when the bottom copy is pronounced, as in (27), it does not induce such a violation according to (51).

3.2

Minimality

Given the present mechanism of producing an A-chain in Japanese, we can answer the question why A-movement in this language can violate the NIC: the movement involved may be scrambling and hence can skip a Case-position in which the moved phrase could have checked its nominative Case. One might then expect that such A-movement may also violate the SSC. That this is not the case is shown by Hiraiwa (2001b) with the SOR construction. Hiraiwa observes that when a multiple nominative construction is embedded in an SOR construction, minimality (or the SSC) is respected concerning Case checking, as shown below:

153

(53) a.

John-ga [Mary-ga

me-ga

warui to]

John-Nom Mary-Nom

eye-Nom bad

Comp

omoikondeita. believed

‘John believed that Mary had a bad eyesight.’ b.

John-ga

[Mary-o

John-Nom Mary-Acc c.

*John-ga [Mary-ga John-Nom Mary-Nom

me-ga

warui to]

eye-Nom bad me-o

Comp

warui to]

eye-Acc bad

Comp

omoikondeita. believed omoikondeita. believed (Hiraiwa 2001b:73)

As shown in (53b), the outer subject in the embedded clause can bear accusative Case, whereas the inner subject cannot, as shown in (53c). Hiraiwa accounts for this pattern of facts by Chomsky’s (2000) Defective Intervention Constraint (henceforth, DIC), which in effect requires that the element closest to a probe for a relevant feature be chosen as a goal. (53c) violates the DIC since Mary-ga is closer to the probe in the matrix vP than me-o ‘eye-Acc’. The present mechanism of producing A-chains is not compatible with Hiraiwa’s (2001b) approach since the movement involved is scrambling, hence should be immune from such a minimality effect as induced by the DIC. We thus need a representational condition that applies after Case-checking is done. Recall that we have adopted such a condition in the previous chapter, which is reproduced below: (54) Crossing Constraint on Features (CCF) Given a chain C = (α1, … αi, … αn),

154

*αi [X]

…

β …

αi+1

[X]

unless feature [X] is carried along as a free ride. Under the present assumptions, (53c) will have the following structure:12 (55) John-ga [VP [CP [TP [vP Mary-ga me-o warui]] to] omoikondei]ta In this structure, the A-chain of me-o violates the CCF, since it is produced across Mary-ga and both me-o and Mary-ga carry Case-features. On the other hand, (53b) has the following structure: (56) John-ga [VP [CP [TP [vP Mary-o me-ga warui]] to] omoikondei]ta The reason why this structure does not induce a violation of the CCF is naturally attributed to the fact that each of the two A-chains crosses only one member of the other. Suppose that a Case-feature is carried not by a particular member of an A-chain but rather by a chain itself. It will then follow that the CCF is violated with respect to Case-features only if an A-chain is produced across another in such a way that the movement involved in producing the former chain crosses all members of the other. Interestingly, the NGC construction shows a pattern that appears to violate the CCF. Miyagawa (1993) observes that when a stative predicate such as kirai na ‘dislike’ that takes a nominative object appears in this construction, all four possibilities are fully 12

In this structure, the bottom copy of the chain of Mary-ga needs to be pronounced

according to (43).

155

realized, as shown below: (57) a.

John-ga

piza-ga

kirai na koto

John-Nom pizza-Nom dislike

(Nom-Nom)

fact

‘the fact that John dislikes pizza’ b. John-no piza-ga

kirai na koto

(Gen-Nom)

John-Gen pizza-Nom dislike fact c.

John-no

piza-no

kirai na koto

(Gen-Gen)

John-Gen pizza-Gen dislike fact d.

John-ga

piza-no

kirai na koto

John-Nom pizza-Gen dislike

fact

(Nom-Gen) (Miyagawa 1993:228)

Under the present assumptions, (57d) will have the following structure: (58) [DP [TP [vP John-ga piza-no kirai na]] koto] The chain of piza-no in this representation should violate the CCF, since it is produced across the A-chain of John, and both A-chains carry Case-features. I suggest that successive-cyclic scrambling is involved in deriving (57d), as shown below: (59) a.

[DP [TP [TP [vP John piza kirai na]]] koto] (scrambling)

b. [DP [TP [TP [vP John-ga piza-no kirai na]]] koto] (assignment of genitive and nominative Cases to piza and John) Given this derivation, it can be claimed that in (59b), the lower chain link of piza-no does not violate the CCF, since both the occurrence of piza-no and that of John-ga in 156

the embedded Spec-TP are equidistant from the lower occurrence of piza-no in the sense of Chomsky (1993), so that the movement involved in producing the lower chain of pizza-no does not cross the A-chain of John-ga in a strict sense. I further suggest that the possibility that a genitive-marked DP passes by the embedded Spec-TP is attributed to the special verbal inflectional morphology, called rentai-kei (the adnominal form; for example, the adnominal form of kirai da ‘dislike’ is kirai na) in the traditional Japanese literature, following the insight of Hiraiwa (2001a), who in fact claims that a genitive-marked DP is licensed by this special inflection within the adnominal clause rather than the above DP. Suppose that when more than one DP occupies the specifier position of a Case-checking head, these DPs need to be Case-compatible, where Case-compatibility is defined as follows: (60) Case Compatibility a.

α and β are Case-compatible iff (i) they are licensed by the same Case-checking head; and (ii) they are clause-mates.

b.

α is licensed by a Case-checking head if (i) it enters into Case-checking with the latter or (ii) the latter is a potential Case-checking head for α.

With the assumption that the adnominal form of a predicate is derived by merging the predicate with a tense form carrying a D-feature, piza-no and John-ga, which both occupy the embedded Spec-TP in (59b), are regarded as Case-compatible since they are clause-mates and John-ga enters into nominative Case-checking with the embedded T, hence licensed by the latter and piza-no is also licensed by this T due to the fact that the

157

latter, carrying a D-feature, is a potential Case-checking head for piza-no.13 Note that the same strategy will not be used for such an SOR case as (53c), since Mary-ga and me-o cannot be Case-compatible in the sense given in (60): in (55), me-o cannot stop by the embedded Spec-TP on its way because the embedded T is not a potential Case-checking head for me-o. The clause-mate condition in the definition of Case-compatibility given in (60) is motivated by the impossibility of NGC in the embedded context of an adnominal clause, as observed by Watanabe (1996): (61) a.

[Mary-ga

[John-ga

Mary-Nom John-Nom

katta

to]

omotteiru] hon

bought Comp think

book

‘the book that Mary thinks John bought’ b. *[Mary-ga [John-no katta

to]

omotteiru] hon

Mary-Nom John-Gen bought Comp think

book

(Watanabe 1996:390)

Under the present assumptions, (61b) may have the following derivation: 13

Given the derivation in (59), it might be expected that piza-no is pronounced in the

position of the embedded Spec-TP, but this is not borne out: (i) a.

[kodomo-no toki-kara child-Gen

John-ga

piza-no

kirai na] koto

time-from John-Nom pizza-Gen dislike fact

‘the fact that John has liked pizza since in his childhood’ b. *[kodomo-no toki-kara piza-no John-ga child-Gen

time-from pizza-Gen John-Nom

kirai na] koto dislike

fact

In (ib), the genitive-marked object piza-no is put after the adverbial phrase modifying the adnominal clause and before the nominative-marked subject John-ga. The unacceptability of (ib) will then indicate that the genitive-marked object cannot have its intermediate copy pronounced.

158

(62) a.

[DP [TP [TP [vP Mary [CP [TP [vP John kat]ta] to] omotteiru]]] hon] (scrambling)

b. [DP [TP [TP [vP Mary-ga [CP [TP [vP John-no kat]ta] to] omotteiru]]] hon] (assignment of genitive and nominative Cases to John and Mary) In (62b), the occurrences of John-no and Mary-ga in the upper Spec-TP of the adnominal clause are not Case-compatible since they are not clause-mates, which thus excludes the derivation given in (62). Hiraiwa (2001a) and Ochi (2001) attribute the impossibility of NGC in such a case as (61b) to what Hiraiwa calls the complementizer blocking effect, first noted by Inoue (1976). Such an effect is exemplified by the following examples: (63) a.

[rainen

sinzyu-ga/-no

yasui] kanousei

next year pearl-Nom/-Gen cheap possibility ‘the possibility that pearl will be cheap next year’ b. [[rainen

sinzyu-ga/*-no

yasui]

next year pearl-Nom/-Gen cheap

toiu]

kanousei

Comp

possibility (Ochi 2001:262)

The fact that a genitive-marked subject is impossible in (63b), unlike in (63a), indicates that the overt complementizer toiu blocks NGC. Given this effect, the unacceptability of (61b) might be given the same account, as Hiraiwa and Ochi in fact do: the overt complementizer to blocks NGC. However, as Hiraiwa (2001a) observes, NGC 159

“becomes licit when the higher subject Mary in [61b] is ‘evacuated’ by some operation (e.g. passivization),” (p. 102) as shown below: (64) [[John-no John-Gen

katta

to]

omow-areteiru] hon

bought Comp is-thought

book

‘the book that it is thought that John bought’ In this example, the genitive-marked subject is allowed despite the fact that the overt complementizer to is present. Under the present assumptions, the acceptability of (64) follows immediately, since no DP intervenes in the produced A-chain of John-no, as shown below: (65) [DP [TP [vP [CP [TP [vP John-no e1 kat]ta] to] omow-areteiru]] hon1] Hiraiwa (2001a) notes a further case in which NGC is blocked in the embedded context of an adnominal clause (I changed his original examples slightly so as to show the relevant contrast more clearly): (66) a.

[[kodomo-no child-Gen

toki-kara John-ga

piza-ga

kirai da

time-from John-Nom pizza-Nom dislike

to] Comp

omow-areteiru] riyuu is-thought

reason

‘the reason that it is thought that John has disliked pizza since his childhood’ b. *[[kodomo-no child-Gen

toki-kara John-ga piza-no

kirai da to]

time-from John-Nom pizza-Gen

dislike

omowareteiru] riyuu is-thought

reason 160

Comp

c.

[kodomo-no child-Gen

toki-kara John-ga piza-no

kirai na] riyuu

time-from John-Nom pizza-Gen

dislike

reason

‘the reason that John dislikes pizza’ (66b) shows a case in point, and Hiraiwa attributes the unacceptability of this example to a violation of the DIC due to the intervening phrase John-ga, on the assumption that piza-no is licensed by the higher T of the adnominal clause. This account is further supported by the fact, unnoticed by Hiraiwa (2001a), that when John-ga in (66a) is changed into John-no, it does not give rise to unacceptability: (67) [[kodomo-no toki-kara John-no piza-ga child-Gen

time-from John-Gen pizza-Nom

kirai da to] dislike

Comp

omowareteiru] riyuu is-thought

reason

In this case, no DIC violation will arise in licensing the genitive-marked subject John-no.14 The present approach accounts for the (un)acceptability of both (66b) and (67) in a similar way. In (67), no DP intervenes in the produced A-chain of John-no, which thus satisfies the CCF. (66b) will have the following structure after successive-cyclic scrambling of piza takes place for evading a violation of the CCF:

14

A DIC violation may arise, however, with respect to licensing the nominative object

piza-ga if its licenser is a finite T: in such a case, John-no will intervene in the probing of the nominative object by the embedded T. This problem will not arise in our Agree-less framework, since the relevant condition CCF is characterized in such a way that it prohibits a phrase carrying a feature X from crossing the whole chain of another phrase carrying X.

161

(68) *[DP [TP [vP [CP kodomo-no toki-kara [TP [TP [vP John-ga piza-no kirai da]]] to] omow-areteiru]] riyuu] In this structure, the occurrences of piza-no and John-ga in the embedded Spec-TP are not Case-compatible, since even though they are clause-mates, piza-no is in no way licensed by the T head whose Spec it occupies, in the sense given in (60b): it does not enter into Case-checking with the latter nor is the latter potential Case-checking head for pizza-no (the verbal form in question is kirai da, which takes syuusikei, the clause-ending form rather than the adnominal form kirai na). Thus the derivation in (68) gives rise to a violation of the Case-compatibility condition, and if piza were directly raised into the Spec-DP, the resulting chain would violate the CCF due to the intervening nominative subject John-ga.15 Further support for the present approach comes from the data Ochi (2001) provides for the complementizer blocking effect: (69) (*)[Taroo-ga [sinzyu-no Taroo-Nom pearl-Gen

kotosi

yasui

this year cheap

to]

omotteiru] riyuu

Comp think

‘the reason that Taroo thinks that pearl is cheap this year’ 15

reason (Ochi 2001:263)

It is predicted under the present approach that if both subject and object in (66a) are

marked with genitive Case, then it will give rise to a legitimate sentence. The prediction seems to be borne out, though the relevant example requires a subtle judgment: (i)

[[kodomo-no toki-kara John-no child-Gen

piza-no kirai da to]

omow-areteiru]

time-from John-Gen pizza-Gen dislike Comp is-thought

riyuu reason (i) seems much better than (66b).

162

Though Ochi takes this example as unacceptable on a par with (61b), there seems to be a significant difference in acceptability between the two sentences: (69) seems much better than (61b).16 The difference is naturally attributed to the availability of SOR. Thus, the genitive-marked subject in (69) can be replaced with one bearing accusative Case, but this is not possible with (61b), as illustrated below: (70) a.

[Taroo-ga [sinzyu-o Taroo-Nom pearl-Acc

b. *[Mary-ga

[John-o

kotosi

yasui

this year cheap katta

to]

to]

omotteiru] riyuu

Comp think

reason

omotteiru] hon

Mary-Nom John-Acc bought Comp think

book

The difference in acceptability between these examples will arise from that of the type of the predicates of the accusative-marked subjects: such subjects demand stative predicates. One might object that the difference between (61b) and (69) comes rather from the type of the adnominal clauses involved: the former involves a relative clause whereas the latter involves what Ochi (2001) calls a gapless clause. The following pair of examples shows that the type of the adnominal clauses involved is irrelevant for the acceptability of (61b) and (69) but rather that our reasoning in terms of the availability of SOR is on the right track:

16

Ochi also provides the following example in his note 17:

(i)(*) [Taroo-no Taroo-Gen

[sinzyu-no pearl-Gen

kotosi

yasui

this year cheap

to]

omotteiru] riyuu

Comp think

reason

Again, Ochi takes this example as unacceptable, but I find it as acceptable as (69).

163

(71) a. ?[John-ga

[Mary-no

John-Nom Mary-Gen

tokui da to] good

omotteiru] kyooka

Comp think

subject

‘the subject that John thinks that Mary is good at’ b. ?[John-ga

[Mary-o

tokui da to]

John-Nom Mary-Acc good

omotteiru] kyooka

Comp think

subject

(71a) is fairly acceptable despite the fact that it involves a relative clause, and importantly its SOR counterpart given in (71b) is as acceptable as (71a). Under the present approach, the correlation of the possibility of genitive-marked subject with that of accusative-marked subject is captured on the assumption that the legitimate A-chain of a genitive-marked subject can be created by this DP passing through the accusative Case position that licenses the accusative-marked subject. Thus (69) may have the following derivation: (72) [DP [TP [TP [vP Taroo-ga [VP [CP [TP [vP sinzyu-no kotosi yasui]] to] omotteiru]]]] riyuu] In this derivation, sinzyu-no passes through the Spec-VP position in which it could check accusative Case. Suppose that the clause-mate condition required for Case compatibility is taken in such a way that the DPs involved must be in the same clause in the previous step of the derivation. Given this, the occurrences of sinzyu-no and Taroo-ga in the higher Spec-TP are Case-compatible, since they are clause-mates in the sense that their next lower occurrences belong to the same clause and they are licensed by the T head whose Spec they occupy in the sense given in (60b). One might expect, following this argument, that not only nominative object but 164

also accusative object can undergo conversion to a genitive-marked one, but this expectation is not fulfilled: (73) a.

[John-ga

sono hon-o

katta]

riyuu

John-Nom that book-Acc bought reason ‘the reason that John bought that book’ b. *[John-ga

sono hon-no katta]

riyuu

John-Nom that book-Gen bought reason Given the present approach, (73b) would have the following derivation, which obeys both the CCF and the Case compatability condition: (74) [DP [TP [TP [vP John-ga sono hon-no kat]ta]] riyuu] The unacceptability of (73b) then suggests that accusative-genitive conversion in such a case is prohibited for a non-structural reason. I suggest that such a case involves inherent Case-marking, unlike SOR cases, in the sense that Case-marking is tied with θ-role assignment. Suppose that inherent Case marking is defined as follows: (75) Given α, β, where α is a DP and β is a inherent Case-marker, α is marked by β with inherent Case iff it is marked by β with θ-role. Note that this condition is defined bi-directionally, so that not only inherent Case-marking requires θ-assignment but also the latter requires the former. Further, suppose that accusative Case is inherent in Japanese, except for the cases of SOR, in which the verbs that host such raising checks accusative Case structurally, as has been assumed above. Given this, sono hon ‘that book’ in (73b) cannot be marked with 165

genitive Case, since in that case, it would not receive the relevant θ-role from the verb kau ‘buy’ according to (75). From the above discussion on Case-checking/marking, we have reached the following conclusion: (76) Nominative as well as genitive Case is structurally licensed whereas accusative Case is either inherently marked or structurally licensed in Japanese.

3.3

The Correlation of Pronunciation and LF Interpretation in A-Chains

We have seen in Section 3.1.2 that Abe’s (2016) proposal regarding pronunciation of an A-chain, which is reproduced below, is supported in the most transparent way by cases of SOR in Japanese. (77) In the case of an A-chain, any member can be the target for pronunciation. The relevant examples are reproduced below: (78) John-ga [mada John-Nom still

Mary-o

kodomo da to]

Mary-Acc child

omotta.

be Comp thought

‘John thought that Mary was still a child.’ (79) Yamada-wa Tanaka-o Yamada-Top Tanaka-Acc

orokanimo [tensai da

to]

stupidly

Comp thought

genius be

omotteita.

‘Yamada stupidly thought that Tanaka was a genius.’ (78) illustrates a case where the bottom copy of the resulting A-chain is pronounced and (79) a case where the top copy is pronounced. In the case of NGC, we have seen Abe’s 166

(2016) argument that the bottom copy of the resulting A-chain must be pronounced. This is because the A-chain involved necessarily crosses an island created by the adnominal clause, and hence according to (51), reproduced below, the bottom copy needs to be pronounced. (80) Locality conditions such as the RRC and the island conditions apply only to “overt” movement. In this section, we consider how the determination of pronunciation of an A-chain is related to that of its LF interpretation. Abe (2016) proposes (82), which he calls the P-L Match Condition, on the assumption given in (81): (81) a.

The semantic feature [SF] indicates that its carrier can participate in LF interpretation.

b. The phonetic feature [PF] indicates that its carrier must be pronounced. (82) [SF] and [PF] must be carried by the same member of an A-chain. Abe motivates this proposal by examining English ECM constructions. The make-out type ECM construction provides the most straightforward evidence for this proposal. Recall that in this construction, the optionality of pronunciation is instantiated by the word order of make, out and the ECM subject, as shown below: (83) a.

John made Mary out to be a fool.

b. John made out Mary to be a fool. These sentences have the schematic representations given below: (84) a.

John [vP v+make [VP Mary tmake+out [TP to be a fool]]] 167

b. John [vP v+make [VP tmake+out [TP Mary to be a fool]]] (84a) illustrates a case where the top copy of Mary is pronounced whereas (84b) illustrates a case where the lower copy is pronounced. Keeping this in mind, let us consider the following examples: (85) a.

The mathematician made every even number out not to be the sum of two primes.

(Lasnik 1999:198)

b. The mathematician made out every even number not to be the sum of two primes.

(ibid.:201)

The every-phrase in (85a) must take scope over negation, whereas that in (85b) can take scope under negation as well as the wider scope. This is straightforwardly accounted for by the P-L Match Condition: in (85a), the every-phrase has its top copy pronounced in the matrix Spec-VP, which thus serves for the scope interaction with the negative not. Since it belongs to the matrix clause, it necessarily takes scope over not. In (85b), on the other hand, the every-phrase has the copy in the embedded Spec-TP pronounced. Thus, the scope ambiguity between the every-phrase and not is attributed to the fact that they belong to the same clause, just as they give rise to scope ambiguity in a simple clause such as the following: (86) Every even number is not the sum of two primes. Likewise, evidence from licensing of anaphors and NPIs gives support to the P-L Match Condition. Consider the following examples: (87) a.

The lawyer made no witnesses out to be idiots during any of the trials.

b.?*The lawyer made out no witnesses to be idiots during any of the trials. 168

(88) a.

The DA made the defendants out to be guilty during each other’s trials.

b.?*The DA made out the defendants to be guilty during each other’s trials. (Lasnik 1999:202) When the top copies of the A-chains involved are pronounced as in (87a) and (88a), they can bind into the higher clauses, but when the lower copies are pronounced as in (87b) and (88b), they cannot. Let us now consider a normal case of the ECM construction such as (32), repeated below: (89) John believes Mary to be honest. In such a case, the ECM subject Mary can be pronounced either in the matrix Spec-VP, as in (90a) or in the embedded Spec-TP, as in (90b): (90) a.

John [vP v+believes [VP Mary tbelieves [TP to be honest]]]

b. John [vP v+believes [VP tbelieves [TP Mary to be honest]]] Recall, however, that we have followed Abe and Hornstein (2012) in assuming that string-vacuous movement is prohibited, which is derived from (43), repeated below: (91) The head of a chain produced by Move cannot be pronounced unless it has an effect on PF output. This condition prohibits (90a), in which the top copy of Mary is pronounced, since the last step of the A-movement is string-vacuous. This gives rise to a problem with the P-L Match Condition, since the evidence suggests that both the copy of Mary in the matrix Spec-VP and that in the embedded Spec-TP can participate in LF interpretation, which thus indicates that both 169

representations in (90) should be legitimate. The following examples, taken from Lasnik and Saito (1991), show that the upper copy is active in licensing of anaphors and NPIs: (92) a. ?The DA proved [the defendants to be guilty] during each other’s trials. b.?*The DA proved [that the defendants were guilty] during each other’s trials. (93) a. ?The DA proved [none of the defendants to be guilty] during any of the trials. b.?*The DA proved [that none of the defendants were guilty] during any of the trials.

(Lasnik and Saito 1991:328-9)

The following example, taken from Lasnik (1999), shows that the lower copy can serve for scope interaction: (94) I proved every Mersenne number not to be prime.

(Lasnik 1999:199)

This sentence is ambiguous with respect to the scope interaction of the every-phrase and not. This fact follows from the assumption that the copy of the every-phrase in the embeddded Spec-TP interacts scopally with not. In order to solve the discrepancy noted above, Abe (2016) proposes that the PF condition (91) be reformulated as a sort of PF adjustment rule: (95) Given a chain C = (α1, … αn), if any pair of (αi, αj) in which αi carries [PF] has no effect on PF output, then adjust this pair as follows: make αj carry [PF]. Given that the P-L Match Condition applies at the point of Transfer or Spell-Out, the following derivation is legitimate: (96) a.

John [vP v+believes [VP Mary tbelieves [TP Mary to be honest]]] [SF][PF] ⇩ apply (95) 170

(satisfying (82))

b.

John [vP v+believes [VP Mary tbelieves [TP Mary to be honest]]] [SF]

[PF]

In this way, the top copy of an ECM subject can be active for LF interpretation even though the lower copy is pronounced.

3.3.1

Cases of Subject-to-Object Raising

In this subsection, I demonstrate that the present mechanism of pronunciation and LF interpretation in A-chains also accounts for relevants facts of the SOR case in Japanese. Sakai (1998) argues for the overt A-movement analysis of this construction. One piece of evidence that Sakai provides is given below: (97) a.?*Rie-wa [karera-ga

muzitu

da to]

otagai-no

syoogen-niyotte

Rie-Top they-Nom innocent be Comp each other-Gen testimony-by sinziteiru. believe ‘Rie believes that they are innocent by each other’s testimonies.’ b. Rie-wa Rie-Top

[karera-o muzitu da to]

otagai-no

syoogen-niyotte

they-Acc innocent be Comp each other-Gen testimony-by

sinziteiru. believe In (97a), karera-ga ‘they-Nom’, which is located in the embedded subject position, cannot c-command outside of the embedded clause, hence unable to license otagai ‘each other’, which is contained in the matrix PP. The acceptability of (97b), then, 171

indicates that the accusative-marked subject karera-o ‘they-Acc’ is not located within the embedded clause but rather is raised into the matrix object position, so that it licenses otagai. This fact is captured under the present approach with the following representation, in which the top copy of the produced A-chain of karara-o carries both [PF] and [SF]: (98) Rie-wa [VP karera-o [CP [TP [vP kareka-o muzitu da]] to] [otagai-no [SF][PF] syoogen-niyotte] sinziteiru] Note that since karera has undergone string-vacuous movement, the PF adjustment rule (95) moves the [PF] carried by the top copy of the A-chain into the bottom copy in the PF component. Sakai (1998) also provides an example of the sort that suggests that the bottom copy of the A-chain involved is active for LF interpretation: (99) Masao-ga

[dare-o

baka da to]-mo

omottei-nai.

Masao-Nom who-Acc stupid be Comp-also think-not ‘Masao thinks that nobody is stupid.’

(Sakai 1998:489)

Here, dare-o functions as what Kuroda (1965) calls indeterminate, and is licensed by the particle -mo, which gives rise to the meaning equivalent to that of the negative polarity item dare-mo. Given the standard assumption that such an indeterminate must be c-commanded by the particle -mo, the acceptability of (99) indicates that dare-o behaves as if it stayed within the embedded clause. This is straightforwardly captured under the present approach with the following representation, in which the bottom copy 172

of the A-chain involved carries both [PF] and [SF]: (100) Masao-ga [VP dare-o [CP [TP [vP dare-o baka da]] to]-mo omottei]nai [SF][PF] Sakai (1998), who tries to argue for the overt A-movement approach to the Japanese SOR construction, claims that the acceptability of (99) can be accounted for on the assumption that licensing of an indeterminate by the particle -mo can be done through reconstruction. Kobayashi and Maki (2002), however, oppose this solution, arguing that if this were the case, it could not account for the fact that when a matrix modifier is inserted after the accusative subject so as to force the overt A-movement of this subject, the latter is no longer licensed by the particle -mo attached to the embedded CP, as shown below: (101) *Masao-ga

dare-o1

kokoro-no soko-kara [t1 baka da to]-mo

omottei-nai.

Masao-Nom who-Acc heart-Gen bottom-from stupid be Comp-also think-not ‘Masao thinks from the bottom of his heart that nobody is stupid.’ Under the present approach, the unacceptability of this sentence follows immediately: since the top copy of the A-chain involved carries its [PF], it also carries its [SF] according to the P-L Match Condition, so that it cannot be licensed by -mo. Kobayashi and Maki (2002) observe that the same pattern holds for the following examples: (102) a.

John-wa [dare-o

hannin daroo ka to]

kangaeteiru.

John-Top who-Acc culprit be-will Q Comp think ‘John wonders who will be the culprit.’ 173

b. *John-wa dare-o1

kasikokumo [t1 hannin daroo

John-Top who-Acc wisely

ka to]

culprit be-will

kangaeteiru.

Q Comp think

‘John wisely wonders who will be the culprit.’ (Kobayashi and Maki 2002:214) (102a) is an example provided by Sells (1990), where a wh-phrase is licensed by the Q-morpheme ka in the embedded clause headed by the complementizer to. This clearly indicates that the accusative-marked wh-phrase dare-o ‘who-Acc’ is licensed in the position occupied by its bottom copy. (102b) then shows that when the top copy of dare-o is pronounced, it is no longer licensed by the Q-morpheme ka, as predicted by the present approach. Relevant scope facts also give support to the present approach. Consider the following examples: (103) a.

John-wa [subete-no hito-ga

atama-ga ii

to]-wa

omowa-nakat-ta.

John-Top every-Gen person-Nom head-Nom good Comp-Top think-not-Past ‘John did not think that everyone was smart.’ b. John-wa [subete-no John-Top every-Gen

hito-o

atama-ga

ii

to]-wa

person-Acc head-Nom good Comp-Top

omowa-nakat-ta. think -not-Past (103a) only allows the reading where subete-no hito ‘everyone’ takes scope under negation, and this is expected since the universal quantifier is located in the embedded subject position whereas negation applies to the matrix clause. (103b), on the other hand, 174

allows not only this reading but also the reading where the opposite scope order holds. This is exactly what the present approach predicts, under which the following two representations can be assigned to (103b): (104) a.

John-wa [VP subete-no hito-o [CP [TP [vP subete-no hito-o atama-ga ii]] to]-wa [SF][PF] omowa]nakat-ta

b. John-wa [VP subete-no hito-o [CP [TP [vP subete-no hito-o atama-ga ii]] to]-wa [SF][PF] omowa]nakat-ta In (104a), the bottom copy of the every-chain carries its [PF] and [SF], which thus represents the reading where the every-phrase takes scope under negation. In (104b), by contrast, it is the top copy that carries both features, which thus gives rise to the reading where the every-phrase takes scope over negation. This account is confirmed by the fact that when a matrix adverbial is inserted after the accusative-marked every-phrase, only the latter reading is obtained, as shown below: (105) John-wa John-Top

subete-no hito-o1

orokanimo [t1 atama-ga ii

every-Gen person-Acc stupidly

to]-wa

head-Nom good Comp-Top

omowa-nakat-ta. think-not-Past ‘John stupidly did not think that everyone was smart.’ This only allows the reading where the every-phrase takes scope over negation. This is exactly as predicted under the present approach, since the word order dictates that the 175

top copy of the every-chain be pronounced and the P-L Match Condition then requires that it be active for LF interpretation. Thus, this scope fact follows on a par with the fact that in a simple clause such as the following, the object universal quantifier takes scope over negation: (106)

John-wa

subete-no hito-o

orokanimo sira-nakat-ta.

John-Top every-Gen person-Acc stupidly

know-not-Past

‘John stupidly did not know anyone.’

3.3.2

Cases of Nominative-Genitive Conversion

In this subsection, I demonstrate that the correlation of pronunciation and LF interpretation in A-chains observed in the English ECM construction is also observed in the Japanese NGC construction. Let us first consider the case in which string-vacuous movement is involved in producing an A-chain of a genitive-marked subject: (107) a.

[Mary-no Mary-Gen

sukina]

syoonen

like

boy

‘the boy Mary likes’ b. [DP [TP [vP Mary-no sukina]] syoonen] In this case, the bottom copy of Mary is pronounced due to the PF adjustment rule (95). It is predicted that in such a case, either copy can participate in LF interpretation. That the top copy is active for LF interpretation is shown by Miyagawa (1993) with such data as the following, which involve scope interaction of a genitive QP with the 176

licensing N head riyuu ‘reason’: (108) a.

[Daremo-ga

paatii-ni kita]

everyone-Nom party-to came

riyuu-o

osiete.

reason-Acc

tell

(*every > reason)

‘Tell me the reason everyone came to the party.’ b. [Daremo-no

paatii-ni kita]

everyone-Gen party-to came

riyuu-o

osiete.

reason-Acc

tell

(every > reason)

In (108a), the universal quantifier bearing nominative Case cannot take scope over riyuu, and this is expected since it is located in the Spec-TP in the adnominal clause, hence structurally lower than the nominal head riyuu. On the other hand, the fact that the universal quantifier bearing genitive Case can take scope over riyuu indicates that the top copy of this quantifier is active for this scope interaction, as indicated below: (109) [DP daremo-no [TP [vP daremo-no paatii-ni kita]] riyuu] [SF][PF] Note that the P-L Match Condition (82) requires that [SF] and [PF] be borne by the same phrase, hence forcing the top copy of daremo in (109) to bear both features. After Spell-Out, [PF] is transferred to the bottom copy by the PF adjustment rule (95). There is also evidence that a genitive-marked subject can take scope in the position occupied by the bottom copy of the A-chain involved. Consider the following examples: (110) a.

[subete-no

hito-ga

ko-nakat-ta]

riyuu

everyone-Gen person-Nom come-not-Past reason ‘the reason everyone didn’t come’ 177

b. [subete-no

hito-no

ko-naka-tta]

riyuu

everyone-Gen person-Gen come-not-Past reason In (110b), the genitive-marked subject subete-no hito-no ‘everyone-Gen’ can take scope under negation, just like the nominative-marked subject in (110a). This indicates that (110b) can have the following representation, in which the bottom copy is pronounced, hence involved in scope interaction with negation: (111) [DP subete-no hito-no [TP [vP subete-no hito-no ko-naka-tta]] riyuu] [SF][PF] Let us now consider the case where non-string-vacuous movement is involved in producing an A-chain of a genitive-marked subject. As Nakai (1980) observes, it is possible to place an overt phrase in front of a genitive-marked subject, thus clearly indicating from the word order that the subject is inside the adnominal clause: (112) [paatii-ni party-to

Mary-no kita]

riyuu

Mary-Gen came

reason

‘the reason Mary came to the party’ It is predicted under the present approach that in such a case, a genitive-marked QP cannot take scope outside the adnominal clause, since the relevant word order indicates that the bottom copy of the QP is pronounced and hence according to the P-L Match Condition, it must also be the one that is involved in scope interaction. This is in fact borne out with the following sentences: (113) a.

[Daremo-no

paatii-ni kita]

everyone-Gen party-to came

riyuu-o

osiete.

reason-Acc

tell

178

(every > reason)

b.

[Paatii-ni daremo-no

kita]

party-to everyone-Gen came

riyuu-o

osiete.

reason-Acc

tell

(*every > reason)

Unlike (113a), (113b) does not allow the reading where daremo ‘everyone’ takes scope over the head noun riyuu ‘reason’, as predicted. The following example shows that a genitive-marked quantifier phrase can take scope within the adnominal clause when preceded by an overt phrase: (114) [paatii-ni party-to

subete-no hito-no

ko-naka-tta]

riyuu

(not > every)

every-Gen person-Gen come-not-Past reason

‘the reason everyone didn’t come to the party’ Thus, when the relevant word order clearly indicates that the bottom copy of a genitive-marked phrase is pronounced, it must also be the one that is involved in LF interpretation in accordance with the P-L Match Condition. Hiraiwa (2001a) objects that the wide socpe reading of a genitive-marked subject quantifier over the head noun riyuu ‘reason’, as observed in (108b) and (113a), does not necessarily confirm such an A-movement approach to NGC as taken by Miyagawa (1993) and Ochi (2001), which our approach follows, and rather that such a fact can be accounted for by assuming that a genitive-marked quantifier is simply base-generated in Spec-DP. Although this objection does not seem untenable for the particular cases we have dealt with, i.e., those which involve such a head noun as riyuu,17 there are cases

17

In the Appendix, I suggest the same possibility to those cases, discussed by Ochi

(2001), which appear to constitute counterexamples to the generalization that overt movement is impossible in the NGC construction.

179

that seem to be immune to such an objection. Consider the following examples, taken from Sakai (1994): (115) a.

[Daremo-ga

kuru] koto-o dareka-ga

everyone-Nom

nozondeiru.

come fact-Acc someone-Nom expect

‘That everyone will come, someone expects.’ b. [Daremo-no

kuru] koto-o dareka-ga

nozondeiru.

everyone-Gen come fact-Acc someone-Nom expect

(Sakai 1994:186)

Sakai (1994) observes that while daremo ‘everyone’ cannot take scope over dareka ‘someone’ in (115a), such a reading is available to (115b). This fact can be accommodated, under the present A-movement approach, on a par with the fact that a quantifier that occupies the Spec of a DP can take scope outside that DP, as illustrated below: (116) Daremo-no everyone-Gen

hon-o

dareka-ga

yonda.

book-Acc someone-Nom read

‘Everyone’s book, someone read.’ Here daremo ‘everyone’ can take scope over dareka ‘someone’. Notice that the base-generation approach does not seem to explain the wide scope reading of the genitive-marked quantifier in (115b), since it is unlikely that the head noun koto ‘fact’, which does not have much semantic content, is able to host such a universal quantifier as daremo in its Spec. To confirm the present approach, let us note that when an overt phrase precedes a genitive-marked subject in such a configuration as in (115b), the wide scope reading in question disappears, as shown below: 180

(117) a.

[Daremo-no

paatii-ni kuru] koto-o dareka-ga

nozondeiru.

everyone-Gen party-to come fact-Acc someone-Nom

expect

‘That everyone will come to the party, someone expects.’ b. [Paatii-ni daremo-no

kuru] koto-o dareka-ga

nozondeiru.

party-to everyone-Gen come fact-Acc someone-Nom expect (117a) allows the wide scope reading of daremo, just like (115b). In (117b), by contrast, the wide scope reading in question is almost impossible to obtain, as predicted. Further evidence for the A-movement approach comes from such facts as the following, originally noted by Miyagawa (1993): (118) a.

[daremo-no

Mary-no sukina] riyuu

everyone-Gen

Mary-Gen like

(every > reason)

reason

‘the reason that everyone likes Mary’ b. [Mary-ga Mary-Nom

daremo-no

sukina]

everyone-Gen like

riyuu

(*every > reason)

reason

‘the reason that Mary likes everyone’ c.

[Mary-no Mary-Gen

daremo-no

sukina]

everyone-Gen like

riyuu

(every > reason)

reason

The fact that the wide scope reading of daremo over riyuu is possible with (118a) whereas such a reading is impossible with (118b) is exactly as predicted under the present approach. Miyagawa (1993) observes that in such a case as (118c), where not only the subject but also the object of the adnominal clause is marked with genitive Case, the wide scope reading of the genitive-marked object is possible. Notice that the 181

availability of such a reading will not be accounted for by the base-generation approach suggested by Hiraiwa (2001a); it is almost impossible to make sense of the claim that both Mary-no and daremo-no are base-generated in Spec-DP. Under the present approach, (118c) will have the following derivation: (119) [DP Mary-no [DP daremo-no [TP [vP Mary-no daremo-no sukina]] riyuu]]

Here, both the subject Mary-no and the object daremo-no undergo movement to Spec-DP, so that their genitive Case is properly licensed. Recall that under the present assumptions, the bottom copies of Mary-no and daremo-no need to be pronounced, so that the apparent violation of the island condition by the movement involved can be circumvented, according to (80), reproduced below: (120) Locality conditions such as the RRC and the island conditions apply only to “overt” movement. Suppose that the bottom copy of daremo-no carries its [PF]. Then, according to the P-L Match Condition, the very same copy must carry its [SF] as well, but in that case, daremo-no should take scope under the head noun riyuu. Thus, we want to claim that (118c) can have a derivation like that given in (96), where both [PF] and [SF] are carried by the top copy of the produced A-chain, but due to the string-vacuity of the movement involved, the PF adjustment rule (95) forces [PF] to move into the bottom copy. The derivation in (119) will cause a problem in serving to this end, however. In order to determine whether each of the two chains involves string-vacuous movement, 182

we need to know which copy of the other chain is pronounced or carries its [PF]. It seems that when more than one A-chain is overlapped, as in (119), it gives rise to a configuration in which indeterminacy arises with respect to string-vacuity. Putting aside further examination of what this leads us to, I suggest an alternative derivation of (118c) in terms of what Takano (2002) calls oblique movement, a movement operation that targets α and adjoins it to β, which does not dominate α. Thus, given this operation, (118c) can have the following derivation: (121) [DP [DP Mary-no daremo-no] [TP [vP [DP Mary-no daremo-no] daremo-no sukina]] riyuu] (oblique movement) In this derivation, daremo-no is first adjoined to Mary-no in Spec-vP via oblique movement, and then the resulting DP undergoes movement to Spec-DP. Notice that both applications of movement are clearly string-vacuous, so that it is possible for the top copy of daremo-no to carry its [PF] as well as [SF] and then to make its [PF] transferred to the bottom copy via the PF adjustment rule (95); more precisely, in the chain ([DP Mary-no daremo-no], [DP Mary-no daremo-no]), the top copies of Mary-no and daremo-no may carry their [PF] as well as their [SF], and then after Spell-out, their [PF] are transferred to their bottom copies due to the phonetic adajency between the two members of the chain in question. Given that daremo-no can c-command out of the whole DP, it c-commands the head noun riyuu, hence taking scope over it. It is predicted that if an overt phrase is inserted to the left of daremo-no in (118c), then its wide-scope reading becomes unavailable. This is in fact borne out by the 183

following examples: (122) a.

[Mary-no Mary-Gen

daremo-no

kodomo-no

everyone-Gen child-Gen

koro-kara

sukina]

riyuu

period-from

like

reason

‘the reason that Mary has liked everyone since her childhood’ b. [Mary-no Mary-Gen

kodomo-no

koro-kara

daremo-no

sukina]

child-Gen

period-from

everyone-Gen like

riyuu reason

(122a) is just like (118c) in that it allows the reading where daremo-no takes scope over riyuu, as expected. In (122b), the order of daremo-no and the adverbial phrase kodomo-no koro-kara ‘since her childhood’ is flipped, and this example does not have the wide scope reading of daremo-no. This is exactly what we predict, since the word order demands that the bottom copy of daremo-no carry its [PF] throughout the derivation, and as a result, the P-L Match Condition demands that the same bottom copy carry its [SF] as well.

3.4

Cases of A-Movement Involving More Than One θ-Role

We have not much discussed the Japanese SSR so far. This is because, as briefly noted in fn. 9, it is not an easy task to examine which copy of the produced A-chain is pronounced in this construction due to the fact that no Case alternation is involved in it. Consider the relevant examples again: (123) a.

John-ga1

saikin [t1 motto

benkyoosuru yoo-ni]

natta.

John-Nom

recently more

study

became

‘Recently, John has come to study harder.’ 184

Comp

b. [Nagai long

aida

John-ga

benkyoosuru yoo-ni]

natta.

period

John-Nom

study

became

Comp

‘John has come to study for a long time.’ (123a) clearly illustrates a case where the top copy of the produced A-chain is pronounced. (123b), on the other hand, appears to illustrate a case where the bottom copy is pronounced, but in this case, there is a possibility that the nominative subject John-ga is Case-checked with the embedded T. Thus, the fact that the long-distance A-movement, as exemplified in (123a), checks the same Case as the alleged short-distance A-movement, as exemplified in (123b), makes it difficult to check whether both these examples involve one and the same A-chain. Uchibori (2000, 2001) claims, contrary to what is said above, that in such an SSR case as (123a), the embedded T does not have the ability to check nominative Case. She bases this claim on the fact that the embedded clause of this construction does not tolerate past tense forms, as shown below: (124) *John-ga1

saikin [t1 motto

John-Nom recently more

benkyoo-sita yoo-ni]

natta.

study-Past

became

Comp

However, this argument is not so convincing, as there are other cases in Japanese where nominative subject is possible even in a clause that does not allow past tense forms. One such case is given below: (125) a.

[Mary-ga Mary-Nom

benkyoosuru yoo-ni]

John-ga

study

John-Nom home tutor-Acc

Comp

‘John hired a home tutor for Mary to study.’ 185

katei-kyoosi-o

yatotta. hired

b. *[Mary-ga Mary-Nom

benkyoo-sita yoo-ni]

John-ga

study-Past

John-Nom home tutor-Acc

Comp

katei-kyoosi-o

yatotta. hired

In (125a), the purpose clause is headed by the complementizer yoo-ni, one with the same form as that used in such an SSR construction as (123a), and it licenses a nominative subject. The unacceptability of (125b) indicates that the purpose clause in question does not allow a past tense form. Thus, the correlation with respect to the availability of nominative Case and the possibility of past tense forms breaks down here. Thus, there is no good reason to suppose that nominative Case is unavailable in the embedded clause of such an SSR construction as (123a). In this section, we consider another type of SSR, one that involves movement into θ-position, to see if it shows the same properties as the normal type of A-movement with respect to locality, pronunciation, and LF interpretation. Abe (2014) argues, along the lines of Hornstein’s (1999, 2001) analysis of obligatorty control, that when a null subject takes an argument in the next clause up as its antecedent, the anaphoric relation involved is best captured as an instance of A-movement. Thus, the following example has the derivation given in (127):18 (126) John1-ga [e1 atama-ga warui John-Nom head-Nom bad

to]

omotteiru/itteiru.

Comp think

/say

‘John1 thinks/says that e1 is stupid.’

18

Abe’s (2014) instantiation of the A-movement analysis is different from that given in

(127): it is pro rather than John-ga that undergoes A-movement. See Abe (2014) for why he takes this option.

186

(127) [TP John-ga [vP John-ga [VP [CP [TP John-ga [vP John-ga [VP atama-ga warui]]] to] … In this derivation, the second step of movement is movement into θ-position, so that the whole A-chain involves two members carrying different θ-roles. Under the present approach to A-movement, it is most natural to regard each step of movement in (127) as an instance of scrambling and nominative Case is assigned to the resulting A-chain afterwards. Abe (2014) provides a variety of evidence for this A-movement analysis of null subjects. One such evidence is concerned with the obligatoriness of the bound variable reading; consider the following example: (128) Daremo1-ga [e1 atama-ga everyone-Nom

head-Nom

warui to]

omotteiru/itteiru.

bad

think

Comp

/say

‘Everyone1 thinks/says that e1 is stupid.’ In this case, e1 is forced to be interpreted as a variable bound by a universal quantifier; hence (128) means that everyone thinks/says of himself that he is stupid. Compare this sentence with the following, in which e is replaced by an overt pronoun that cannot be interpreted as a bound variable by its nature: (129) Daremo1-ga

[karera1-ga atama-ga warui

everyone-Nom they-Nom head-Nom bad

to]

omotteiru/itteiru.

Comp think

/say

‘Everyone1 thinks/says that they1 are stupid.’ In this sentence, karera ‘they’ refers to the set denoted by daremo ‘everyone’; hence this sentence is interpreted as such that each person thinks/says that everyone is stupid. 187

Interestingly, Abe (2014) further observes that when the anaphoric relation between a null argument and its quantificational antecedent does not satisfy the SSC, the bound variable reading is not forced, as shown below: (130) Daremo1-ga

[Mary-ga [e1 atama-ga warui to]

everyone-Nom Mary-Nom head-Nom bad

itteiru

Comp say

to]

omotteiru.

Comp think

‘Everyone1 thinks that Mary says that e1 is stupid.’ This sentence has not only the bound variable reading of e1 but also what may be called the referential reading, where it refers to the set denoted by daremo ‘everyone’, and hence can be interpreted in the same way as the corresponding sentence with e1 replaced by the overt pronoun karera: (131) Daremo1-ga

[Mary-ga

everyone-Nom

Mary-Nom

itteiru to]

omotteiru.

say

think

Comp

[karera1–ga

atama-ga

warui

to]

they-Nom

head-Nom

bad

Comp

‘Everyone1 thinks that Mary says that they1 are stupid.’ The above observation receives a natural account under the A-movement analysis of null subjects in such a case as (126): an A-chain of a QP forces its member occupying θ-position to act as a variable of that quantifier. Further, given that such an A-chain is subject to a minimality condition like CCF, reproduced below: (132) Crossing Constraint on Features (CCF) Given a chain C = (α1, … αi, … αn),

188

*αi [X]

…

β …

αi+1

[X]

unless feature [X] is carried along as a free ride. it will follow that in the configuration in which the anaphoric relation of a null argument with its antecedent violates the SSC, the anaphoric relation in question cannot be established via A-movement. Thus, (130) would have roughly the following derivation if the null subject were associated with its antecedent via A-movement: (133) [TP daremo-ga [vP daremo-ga [VP [CP [TP Mary-ga [vP Mary-ga [VP[CP[TP daremo-ga

[vP daremo-ga [VP atama-ga warui]]] to] itteiru]]] to] omotteiru]]]

In this derivation, the second step of A-movement of daremo-ga ‘everyone-Nom’ will violate the CCF since it crosses the Spec-TP position occupied by Mary, which carries nominative Case. Abe (2014) claims that in such a case, the null argument undergoes null operator-like movement to evade a violation of a minimality condition and that the resulting chain functions like one headed by a null operator, thereby giving rise to both the referential and bound variable readings of the null argument. As further evidence for the present A-movement analysis of null subjects, Abe (2014) provides the following examples, where a null subject appears in an adjunct clause in (134) and in a relative clause in (135): (134) Daremo1-ga [e1 sono piano-o everyone-Nom

that

piano-Acc

gokai-made

hakondeiru

aida]

fifth floor-to is-carrying

while

189

kanozyo-no

koto-o

kangaeteita.

girl friend-Gen thing-Acc thought ‘Everyone1 thought of his1 girl friend while e1 was carrying that piano to the fifth floor.’ (135) Daremo1-ga [e1 sokode everyone-Nom

there

atta] gakusei-to

kekkonsi-tagatteiru.

saw student-with want-to-marry

‘Everyone1 wants to marry a student who e1 saw there.’ These sentences allow the referential reading of the null subjects, so that they may have the same reading as the ones in which the null subjects are replaced by the overt pronoun karera-ga ‘they-Nom’. Abe (2014) claims that this fact immediately follows under the A-movement analysis of null subjects, since in these cases, the null subjects cannot be associated with their antecedents via A-chains without violating the island conditions. Hence, these null subjects must undergo null operator-like movement instead, thereby either functioning as a variable bound by its antecedent or simply referring to the set denoted by it. Recall, however, that we have assumed the following condition, which is reproduced from (120): (136) Locality conditions such as the RRC and the island conditions apply only to “overt” movement. Notice that both of the examples in (134) and (135) would involve string-vacuous movement if daremo-ga and the embedded null subject were associated via an A-chain. Thus, according to the PF adjustment rule in (95), reproduced below: 190

(137) Given a chain C = (α1, … αn), if any pair of (αi, αj) in which αi carries [PF] has no effect on PF output, then adjust this pair as follows: make αj carry [PF]. the bottom copy of each A-chain would be pronounced, hence the movement involved being regarded as an instance of “covert” movement. It should follow from (136), then, that such movement is immune to island conditions, contrary to fact. It is reasonable to ascribe this unexpected property to the fact that the A-chains produced in (134) and (135) would involve movement into θ-position. Thus, something like the following should hold true, though at the present stage of understanding it is not clear why:19 (138) Movement into θ-position is sensitive to island conditions, irrespective of whether it is string-vacuous or not.

3.4.1

“Backward Binding” in the Case of Subject-to-Subject Raising

Recall that we have made the following assumption: (139) In the case of an A-chain, any member can be the target for pronunciation. Given the A-movement approach to null subject shown above, it is predicted that any copy of such an A-chain as given in (127), reproduced below, should be pronouceable: (140) [TP John-ga [vP John-ga [VP [CP [TP John-ga [vP John-ga [VP atama-ga warui]]] to] … In particular, if either of the lower copies of John-ga is pronounced, that will instantiate a case similar to so-called backward control, though the A-chain involved in this case 19

See Abe and Hornstein (2012) for the observation that movement of adjuncts is also

sensitive to island conditions irrespective of whether it is string-vacuous or not.

191

does not express a control relation. Let us thus refer to such a case as an instance of “backward binding.” It has been well known since Polinsky and Potsdam (2002) that universally, control relations allow not only forward control, the standard one in which the controller is pronounced, but also backward control, in which the controllee is pronounced, and that the existence of the latter relation in some languages is taken as strong evidence for a movement theory of control such as proposed by Hornstein (1999, 2001). This is because the PRO-based theory will not be able to provide any obvious means to capture backward control. Under this theory, a control relation is expressed by PRO and its antecedent for forward control, as schematically represented below: (141) a.

DP1 V [PRO1 …]

b. DP V DP1 [PRO1 …] Given these representations, it is far from clear how backward control is expressed in terms of PRO and its antecedent, since simply flipping the occurrences of PRO and its antecedent, as shown below, causes Principle C violations: (142) a.

PRO1 V [DP1 …]

b. DP V PRO1 [DP1 …] On the other hand, as Boeckx, Hornstein and Nunes (2010) claim, Hornstein’s (1999, 2001) A-movement theory of control provides a straightforward way of capturing backward control under the single cycle hypothesis according to which overt vs. covert movement is expressed in terms of which copy of a produced chain is pronounced. Thus, under this hypothesis, backward control is simply a case where the 192

tail of a resulting chain is pronounced, as schematically represented below: (143) a.

V [DP …]

b. DP V [DP …] In this subsection, I demonstrate that such an A-chain as given in (140) manifests backward binding, hence giving further support to the A-movement approach to null subject. First, let us consider what is the right structure for such a Japanese sentence as the following: (144) John-ga

Mary-ni

atta to

John-Nom Mary-Dat saw Comp

itta. said

Confining our consideration to the reading of this sentence in which the matrix and embedded subjects refer to the same person, there are standardly two possible ways to assign a structure to this sentence, as shown below: (145) a.

John1-Nom [pro1 Mary-Dat saw Comp] said

b. *pro1 [John1-Nom Mary-Dat saw Comp] said (145a) represents a structure in which John-ga ‘John-Nom’ is taken as the matrix subject whereas in (145b) it is taken as the embedded subject. It has been assumed that the structure in (145b) is excluded as a violation of Condition C of the binding theory. However, given the present A-movement theory of null subject, there is a new way of analyzing (144); that is, taking this sentence as an instance of backward binding, as indicated below: (146) [John-Nom Mary-Dat saw Comp] said 193

Recall that we have been assuming the PF adjustment rule (137), which forces a chain link produced via string-vacuous movement to have the bottom copy pronounced. Given this rule, (146) should be the only legitimate structure for (144). This analysis is supported by the fact that an adjunct that modifies the embedded clause can precede the subject that was ambiguously interpreted, as illustrated below: (147) a.

[Asu

John-ga

tomorrow John-Nom

daigaku-e

iku to]

university-to go

Comp

itta. said

Lit. ‘e1 said that John1 would go to the university tomorrow.’ b.

[Kyookai-de John-ga church-in

Mary-ni

John-Nom Mary-Dat

puropoozu-suru] to propose

omotteita.

Comp thought

Lit. ‘e1 thought that John1 would propose to Mary in a church.’ The modifiers asu ‘tomorrow’ in (147a) and kyookai-de ‘in church’ in (147b) disambiguate the position that John-ga occupies; that is, it occupies the embedded subject position. Nonetheless, John can also be interpreted as the matrix subject. Thus, these sentences clearly instantiate cases of backward binding. One might object to this claim, noting that there is actually one more structure that can be assigned to (144), one that is derived from (145b) by moving the whole embedded clause to the left of pro, as shown below: (148) *[John1-Nom Mary-Dat saw Comp]2 pro1 t2 said In this structure, John-Nom is outside the c-command domain of pro, hence not inducing a Condition C violation. Thus, if (148) were a possible structure for (144), the sentences in (147) would be analyzed in the same way. Notice, however, that the 194

application of scrambling in (148) is string-vacuous and that it has been claimed since Hoji (1985) that string-vacuous scrambling must be prohibited. Under the present analysis, in such a case, the bottom copy of the resulting chain is pronounced, hence inducing a Condition C violation. We can give further support to our claim by examining the relevant conditions that license backward binding. Recall that we have seen above that, as stated in (138), movement into θ-position is sensitive to island conditions, irrespective of whether it is string-vacuous or not. It is then predicted that the backward binding analysis for such a sentence as (144) will become impossible when the anaphoric relation involved crosses an island. This is in fact borne out; compare the (a)- with the (b)-sentences in the following pairs: (149) a.

Mary-ga

John-kara

uketotta tegami-o

yaburi-suteta.

Mary-Nom

John-from

received letter-Acc

tear-threw away

‘Mary1 tore up and threw away a letter [e1 received from John].’ b. *[John-kara Mary-ga John-from Mary-Nom

uketotta] tegami-o

yaburi-suteta.

received letter-Acc

tear-threw away

‘Lit. e1 tore up and threw away a letter [Mary1 received from John].’ (150) a.

John-ga Mary-ni

puropoozusuru mae-ni yubiwa-o katta.

John-Nom Mary-Dat

propose

before ring-Acc bought

‘John1 bought a ring before e1 proposed to Mary.’ b. *[Mary-ni John-ga Mary-Dat John-Nom

puropoozusuru mae-ni] yubiwa-o katta. propose 195

before ring-Acc bought

’Lit. e1 bought a ring before John1 proposed to Mary.’ (151) a.

Mary-ga

asu

dooyatte eki-e

Mary-Nom

tomorrow how

ikeba ii no ka wakara-nakat-ta.

station-to go

Q know-not-Past

‘Mary1 didn’t know how e1 should go to the station tomorrow.’ b.?*[Asu

Mary-ga

dooyatte eki-e

tomorrow Mary-Nom how

ikeba ii no ka] wakara-nakat-ta.

station-to go

Q know-not-Past

‘Lit. e1 didn’t know how Mary1 should go to the station tomorrow.’ The contrast in acceptability between (a)- and (b)-sentences in each pair suggests that the bold-faced subjects in the (a)-sentences must be taken as the matrix ones. The unacceptability of the (b)-sentences with the intended readings immediately follows on the assumption that movement into θ-position does not take place across an island, since the only remaining structure we can assign to each (b)-sentence is something like (145b), thus inducing a Condition C violation. Notice that if we admitted such a structure as (148) that involves string-vacuous scrambling, then we would not be able to account for the unacceptability of the (b)-sentences in (149)-(151). Let us now examine whether the correlation observed with a normal A-chain between pronunciation and LF interpretation is also observed with those A-chains that include a link produced by movement into θ-position. Recall that when a null subject takes a QP as its local antecedent, it is forced to be interpreted as a bound variable, as shown in (128), reproduced below: (152) Daremo1-ga [e1 atama-ga everyone-Nom

head-Nom

warui to]

omotteiru/itteiru.

bad

think

196

Comp

/say

‘Everyone1 thinks/says that e1 is stupid.’ Let us now compare this sentence with a backward binding case such as the following: (153) [Sono toozi daremo-ga that

atama-ga

time everyone-Nom head-Nom

yokatta

to]

was-good Comp

omotteiru/itteiru. think

/say

‘Lit. e1 thinks/says that everyone1 was smart at that time.’ It will be predicted under the present approach that daremo in (152) takes scope over not only the matrix but also the embedded clause, whereas daremo in (153) takes scope over only the embedded clause. This is because these two sentences should have the following representations, given what has been assumed so far:20 (154) a.

[TP [CP [TP daremo-ga atama-ga warui] to] omotteiru/itteiru] [SF][PF]

b. [TP [CP [TP daremo-ga atama-ga warui] to] omotteiru/itteiru] [SF][PF] (by PF adjustment rule (137)) (155) [TP [CP [TP sono toozi daremo-ga atama-ga yokatta] to] …] [SF][PF] This prediction is not borne out, however: in both (152) and (153), the embedded subjects must be interpreted as variables bound by daremo, hence this universal

20

In each representation, the chain of daremo-ga should have four occurrences of this

universal quantifier, but the representations are simplified in such a way that only two occurrences are represented, one in the matrix clause and the other in the embedded clause.

197

quantifier necessarily taking scope over the matrix clause. It is natural to conjecture that this fact has something to do with the special property of the A-chains involved: they involve more than one θ-role. The following generalization seems to hold true: (156) In an A-chain C, [SF] cannot be carried by a member of C that is lower to another member carrying a θ-role. This generalization might follow from an LF condition for deriving appropriate semantic representations, to the following effect: (157) If an A-chain C is involved in an operator-variable chain, then the member(s) carrying a θ-role must be interpreted as a variable. Given this condition, the only legitimate representations for (152) and (153) must be (158) and (159), respectively: (158) [TP [vP [CP [TP [vP daremo-ga …]] ] …]] [SF][PF]

(159) [TP [vP [CP [TP sono toozi [vP daremo-ga [SF]

[PF]

…]] ] …]] In each representation, the two occurrences of daremo-ga in the embedded and matrix Spec-vP are interpreted as variables, according to (157), and hence [SF] needs to be carried by the one in the matrix Spec-TP; otherwise, the occurrence of daremo-ga in the matrix Spec-vP would end up being a free variable. Notice that (159) clearly violates the 198

P-L Match Condition. This indicates that condition (157) overrides this requirement. Supporting evidence for this claim can be provided regarding scope interaction of QPs. Compare the following examples: (160) a.

Daremo-ni

dareka-ga

kisusita.

everyone-Dat someone-Nom kissed ‘Everyone, someone kissed.’ b. [Daremo-ni

dareka-ga

kisusita to]

itteita.

everyone-Dat someone-Nom kissed Comp said ‘Lit. e1 said that everyone, someone1 kissed.’ As seen in Section 2.4, when object QP moves across subject QP by scrambling, the former can take scope over the latter, as shown in (160a). Interestingly, (160b) does not allow such a reading if dareka ‘someone’ serves as not only the embedded subject but also the matrix subject. This is exactly what we predict, since in such a case, dareka must take scope over the matrix clause, as shown below: (161) [TP [vP [CP [TP daremo-ni [vP dareka-ga [SF]

[PF]

…]] ] …]] Thus, the fact that the wide scope reading of daremo ‘everyone’ is not available to (160b) follows from the clause-boundedness of QP. This is confirmed by the following sentence, where the top copy of the chain of dareka-ga is pronounced: (162) Dareka-ga someone-Nom

[daremo-ni kisusita to] everyone-Dat

kissed 199

itteita.

Comp said

‘Someone1 said that everyone, e1 kissed.’ This sentence allows only the reading in which dareka takes scope over daremo, as predicted. Finally, further support for the present analysis comes from the fact that an anaphor sitting in the matrix VP can be licensed by the embedded subject, if the latter is taken as a case of backward binding. Consider the following example: (163) [Asu

daremo-ga

daigaku-e

iku to]

tomorrow everyone-Nom university-to go

zibun-no sensei-ni

itta.

Comp self-Gen teacher-Dat said

‘Lit. e1 told self1’s teacher that everyone1 would go to the university.’ Under the present analysis, zibun ‘self’ is bound and hence licensed by the unpronounced matrix subject daremo ‘everyone’, which carries the [SF] of the produced A-chain in which the member in the embedded subject position is pronounced.

3.4.2

“Backward Binding” in the Case of Subject-to-Object Raising

We have seen in the previous subsection that A-chains involving more than one θ-role exhibit a property different from normal A-chains: [SF] cannot be carried by a member of an A-chain that is lower to another member carrying a θ-role, as stated in (156). In this subsection, we examine whether there is any case in the Japanese SOR that involves such an A-chain, by exploiting this property as a clue for this examination. There have been many researchers who claim that in the Japanese SOR, the accusative-marked “subject” is in fact the base-generated object of the higher verb that 200

binds the embedded null subject (standardly identified as pro). 21 Thus, under this analysis, (3b), for instance, will have the following structure: (164) John-ga

Bill1-o [CP [TP pro1 baka da] to]

John-Nom Bill-Acc

stupid be Comp

omotteiru think

As has often been noticed in the literature, it is not at all clear whether such an analysis is in fact necessary or simply dispensed with for such a sentence as (3b). However, given the possibility of recapturing such an analysis as one involving A-movement into θ-position, a new way of approaching this question is now available; that is, to investigate whether there is any case of backward binding in the Japanese SOR. If the answer is yes, then it clearly indicates that such a “base-generation” analysis as shown in (164) is necessary, though it must now be replaced with an A-movement analysis of the sort we having been arguing for. Thus, (164) will have the following structure instead: (165) John-ga [VP [CP [TP [vP baka da]] to] omotteiru]

Here, movement of Bill-o into the matrix VP is an instance of movement into θ-position. If the bottom copy of the resulting chain is pronounced, it instantiates a case of backward binding, which is most clearly illustrated by a sentence in which overt material intervenes between the two members of the A-chain. Kobayashi and Maki (2002) make a relevant observation according to which even 21

See Kobayashi and Maki (2002) for the references of such an analysis as well as the

problems that will arise if all instances of the Japanese SOR are analyzed in this way.

201

though the accusative subject of the raising construction in question stays in situ, it can bind into an adverbial phrase that is attached to the higher clause, as shown below: (166) Rie-wa [kaminoke-no ippon-ni itaru made Rie-Top hair-Gen otagai1-no

one-to

as far as

karera1-o muzitu da

to]

they-Acc innocent be Comp

syoogen-niyotte sinziteiru.

each other-Gen testimony-by

believe

‘Rie believes them1 to be every inch innocent based upon each other1’s testimonies.

(Kobayashi and Maki 2002:221)

In this sentence, the accusative subject karera-o ‘they-Acc’ stays in situ, as is clear from the fact that it is preceded by the italicized adverbial phrase that modifies the embedded clause. Despite this, it is able to bind into the adverbial phrase in the matrix clause that includes otagai ‘each other’. Kobayashi and Maki take this fact as indicating that Agree is involved in licensing anaphors, so that the matrix v, which enters into an Agree relation with the accusative subject, can serve as a licenser of the reciprocal. However, this does not comport with what we have found out under the present approach: [SF] and [PF] must be carried by the same member of an A-chain unless the chain involves more than one θ-role. Thus, it is more likely that (166) instantiates a case of backward binding. Further support for the existence of such a case can be provided with SOR examples involving scope interaction. Consider the following example: (167) John-wa John-Top

[umaretuki naturally

subete-no hito-o

atama-ga ii

to]-wa

every-Gen person-Acc head-Nom good Comp-Top 202

omowa-nakat-ta. think-not-Past ‘John did not think that everyone was smart by nature.’ This sentence allows not only the reading where the matrix negation takes scope over everyone but also the one where the scope order is reversed. This would be unexpected if subete-no hito-o ‘everyone-Acc’ were taken to undergo a normal type of A-movement; in that case, the every-phrase should take scope under negation in accordance with the P-L Match Condition. On the other hand, if the every-phrase is taken to undergo movement into the internal θ-position of the matrix verb omow ‘think’, then it can take scope in that θ-position, thereby having negation under its scope. Since the bottom copy of the every-phrase chain is pronounced in (167), it must be an instance of backward binding, according to the present analysis. Yamashita (2005) also makes a relevant observation similar to Kobayashi and Maki’s (2002) with passivized cases of the SOR construction, as shown below:22 (168) a.

Aya-ga

zibun-no

Aya-Nom self-Gen

ryoosin-ni(yotte) [madamada kodomo da

to]

parents-by

Comp

still

child

be

omow-areteiru. is-thought

22

I slightly changed Yamashita’s (2005) original examples by adding -niyotte, another

variant expressing ‘by’, with parentheses, since it is often claimed that the Japanese -niyotte passive reflects the standard properties of passives more directly than the -ni passive.

203

‘Aya is thought by self’s parents to be still a child.’ b.

Zibun-no ryoosin-ni(yotte) [madamada Aya-ga

kodomo da

to]

self-Gen

child

Comp

parents-by

still

Aya-Nom

be

omow-areteiru. is-thought (169) a.

Aya-ga

zibun-no ryoosin-ni(yotte) [imadani

Aya-Nom self-Gen parents-by

still

kawaii to] cute

Comp

omow-areteiru. is-thought ‘Aya is thought by self’s parents to be still cute.’ b. Zibun-no ryoosin-ni(yotte) [imadani Aya-ga self-Gen

parents-by

still

kwaii to]

omow-areteiru.

Aya-Nom cute Comp is-thought

The acceptability of the (a)-sentences in (168) and (169) is not surprising since the antecedent of zibun, i.e., Aya, appears in the matrix subject position. In the corresponding (b)-sentences, Aya appears in the embedded subject position instead. What is surprising here is that Aya can still serve as the antecedent of zibun. If we assumed that Aya underwent a normal type of A-movement from the embedded to the matrix subject position in the (b)-sentences, then we would expect that these sentences were unacceptable due to the P-L Match Condition. Thus, the acceptability of these sentences clearly indicates that they instantiate cases of backward binding, so that (168b), for instance, has the following derivation:

204

(170) [TP [vP [VP zibun-no ryoosin-ni(yotte) [… Aya-ga …] …]]]

Here the movement of Aya-ga into the matrix Spec-VP is an instance of movement into the internal θ-position of the matrix verb omow ‘think’. Hence, either the occurrence of Aya-ga in the matrix Spec-VP or that in the matrix Spec-TP can serve as the antecedent of zibun (given the subject orientation property of this reflexive, it is more likely that the one in the matrix Spec-TP serves as such). Recall that when we discussed Abe’s (2014) arguments for the A-movement analysis of null subject in Japanese, we observed that the resulting A-chains obey such a minimality condition as the CCF and island conditions. Recall further that when the anaphoric relation between a null argument and its antecedent does not satisfy such locality conditions, the null argument undergoes null operator-like movement. Given this, it is expected that the SOR construction under consideration should have cases that are analyzed in a way similar to (164); that is, the accusative-marked DP is base-generated in the internal θ-position of the higher verb and the pro bound by this DP undergoes null operator-like movement in the embedded clause, as shown below:23 (171) John-ga Bill1-o [CP pro1 [TP pro1 baka da] to] omotteiru

23

According to the movement theory of null arguments proposed by Abe (2014), the

pro that have undergone null operator-like movement should undergo further movement to the position occupied by its antecedent. I ignore this final step of movement in the text just for simplification.

205

According to Abe (2014), this null operator-like movement of pro is the last resort strategy, hence available only when A-movement is prohibited for an independent reason. Following this reasoning, the particular case given in (171) should be unavailable and be replaced by the A-movement strategy, given in (165), repeated below: (172) John-ga [VP [CP [TP [vP baka da]] to] omotteiru]

Hence, we need to look for cases in which such an A-movement strategy is prohibited by locality conditions such as the CCF and island conditions. Oka (1988) provides such a case: (173) Bill-ga Bill-Nom

Mary-o

[John-ga

Mary-Acc John-Nom

horeteiru to]

omotteiru.

is-in-love Comp think

‘Bill thinks of Mary that John is in love with her.’

(Oka 1988:192)

Here, Mary-o ‘Mary-Acc’ is interpreted as the object of horeteiru ‘be in love’, but does not directly serve as such, since this verb takes a dative DP as its object, as shown below: (174) John-ga

Mary-ni

horeteiru.

John-Nom Mary-Dat is-in-love ‘John is in love with Mary.’ Oka (1988) proposes that such an SOR construction as in (173) has a structure like the topic construction embedded in the SOR environment. This proposal comports well with the present approach to the SOR construction, since in this case, A-movement of 206

Mary-o would be prohibited by the CCF and hence operator movement of pro is chosen instead, as shown below: (175) Bill-ga Mary1-o [CP pro1 [TP John-ga pro1 horeteiru] to] omotteiru

That A-movement is unavailable in this case is confirmed by the fact that Mary-o cannot be pronounced in the complement position of the embedded verb horeteiru ‘is-in-love’ instead of the matrix object position in (173), as shown below: (176) *Bill-ga [John-ga Bill-Nom John-Nom

Mary-o

horeteiru

Mary-Acc is-in-love

to]

omotteiru.

Comp think

‘Bill thinks of Mary that John is in love with her.’ If A-movement were available for (173), then nothing would prevent the bottom copy of Mary-o from being pronounced, as in (176), contrary to fact.24 24

Oka (1988) claims that the pro analysis for such a construction as (173) is supported

by the subjacency test; that is, it is free from the effects of subjacency. He provides the following examples to verify his claim: (i) (*)Bill-ga Mary1-o [Tom-ga [e2 e1 horeteiru] otoko2-o kiratteiru to] Bill-Nom Mary-Acc Tom-Nom

is-in-love man-Acc hate

omotteiru.

Comp think

‘Bill thinks of Mary that Tom hates the man who is in love with her.’ (Oka 1988:193) However, I agree with Kobayashi and Maki (2002) on their claim that this sentence is fairly degraded and far worse than the corresponding topic construction, given below: (ii)

Mary1-wa Tom-ga [e2 e1 horeteiru] otoko2-o kiratteiru. Mary-Top Tom-Nom

is-in-love man-Acc hate

‘As for Mary, Tom hates the man who is in love with her.’ This suggests that null operator-like movement of pro is involved in this construction rather than simply base-generating pro, along the lines of Abe’s (2014) theory of null

207

Finally, let us briefly consider whether there is any case of backward binding in the NGC construction. As in the SOR cases, it has often been noticed that there are cases that involve base-generation of genitive-marked subjects in this construction, those which Hiraiwa (2001a) calls instances of pseudo-NGC. Ochi (2001) provides a relevant example: (177) John-no, John-Gen

[totemo atarasii], [kinoo very

new

untensiteita]

yesterday was-driving

‘John’s very new car that he was driving yesterday’

kuruma car (Ochi 2001:267)

Here, John-no precedes the relative clause totemo atarasii ‘very new’ and hence is clearly outside the relative clause in which it serves as the agent of untensiteita ‘was driving’. Ochi observes that (177) lacks the reading in which John is merely the agent of this predicate and has to be interpreted as the possessor of kuruma ‘car’ as well. This indicates, Ochi argues, that “overt raising structure is not available, and only the ‘control structure’ is allowed with this example.” (p. 267-8) Thus, under this analysis, (177) will have the following structure: (178) [DP John1-no [TP totemo atarasii] [TP pro1 kinoo untensi-teita] kuruma] Here, John-no is base-generated in Spec-DP, hence serving as the possessor of kuruma. Further, it binds pro, thereby interpreted as the agent of untensi-teita as well. Given this possibility of base-generation of genitive-marked subjects, one might expect that the NGC construction may instantiate cases of backward binding. But recall that this construction does not allow overt A-movement since it involves an adnominal arguments.

208

clause that constitutes an island. Recall further that movement into θ-position is sensitive to island conditions, irrespective of whether it is overt or covert. Then the prediction should be that the NGC construction does not allow backward binding. This is in fact borne out by such an example as the following: (179) [totemo very

atarasii] [kinoo new

John-no

untensi-teita] kuruma

yesterday John-Gen was-driving car

‘the very new car that John was driving yesterday’

(Ochi 2001:269)

Here, the genitive-marked subject John-no is pronounced inside the relative clause. It would be expected that (179) had the same possessor plus agent reading as (177) if it had the structure of backward binding. But, as Ochi correctly observes, such a reading is unavailable to (179); rather, it merely has the agent only reading. This clearly indicates that the NGC construction does not allow backward binding, as predicted. To sum up this chapter, I have argued that the apparent violation of the NIC manifested in the A-chains involved in the SSR, SOR, and NGC constructions can be attributed to the fact that these A-chains are produced by scrambling. I have then demonstrated that the properties of the resulting A-chains with respect to pronunciation and LF interpretation are best captured by the Agree-less approach, advocated recently by Hornstein (2009), according to which the overt vs. covert distinction is made in terms of which member of a given chain is pronounced. In particular, I have shown, along the lines of Abe (2016), that pronunciation and LF interpretation correlate with each other in the A-chains involved in the NGC and SOR constructions; that is, the member of a given A-chain that is pronounced must be the one that serves for LF 209

interpretation. Finally, I have demonstrated that there is a special type of A-movement involved in the SSR and SOR constructions, namely one that involves a step of movement into θ-position. I have shown that, as predicted under the Agree-less approach, the resulting A-chain allows the option of having its bottom copy pronounced, a case of backward binding.

Appendix: Is Overt Raising Possible with the NGC Construction? This appendix examines whether the following generalization, put forth by Ochi (2001), is well-established: (180) While relative clauses resisit overt raising of a genitive subject, gapless clauses do allow such overt raising. Ochi provides the following examples as supporting evidence for the former part of this generalization: (181) a.

totemo

mazusii, (seihu-niyotte)

zyuubun-na tyuui-no

very

poor

enough

government-by

haraw-aretei-nai

tiiki

is-paid-not

area

attention-Gen

‘the area to which enough attention is not paid (by the government)’ b. *zyuubun-na tyuui-no1, enough

totemo

attention-Gen very

haraw-aretei-nai

tiiki

is-paid-not

area

mazusii, (seihu-niyotte) t1 poor

government-by

(Ochi 2001:268) 210

As for supporting evidence for the latter part of (180), Ochi provides the following examples: (182) a.

kompyuutaa-ga

keisansita, zyuubun-na tyuui-no

computer-Nom

calculated enough

tiiki-ni-mo

haraw-aretei-nai

area-Dat-also is-paid-not

dono

attention-Gen any

kanousei possibility

‘the possibility that enough attention is not paid to any area which the computer calculated’ b. zyuubun-na tyuui-no1, enough

attention-Gen

tiiki-ni-mo

kompyuutaa-ga

keisan-sita,

computer-Nom

calculated

t1

dono any

harawarete-inai kanousei

area-Dat-also is-paid-not

possibility

(ibid.:282)

Though Ochi claims that the acceptability of (182b) shows that overt raising of a genitive subject is possible in gapless clauses, the contrast between (181b) and (182b) does not seem so sharp; to me (182b) is just as bad as (181b). Thus, the above data alone will not suffice to establish the generalization given in (180). Ochi provides further examples to support his claim: (183) a. *rubii ka sinzyu-ga1,

[kono kompyuutaa-ga

ruby or pearl-Nom

this computer-Nom

yasuku

naru]

cheap

become possibility

keisansita] [ti kotosi calculated

this year

kanousei

‘the possibility that

rubies or pearls become cheap this year which this 211

computer calculated’ b. rubii ka sinzyu-no1, ruby or pearl-Gen

[kono kompyuutaa-ga this computer-Nom

yasuku

naru]

cheap

become possibility

keisan-sita] [t1 kotosi calculated

this year

kanousei

Ochi takes the acceptability of (183b), in contrast with that of (183a), as a piece of evidence for overt movement of a genitive subject in gapless clauses. Further, he observes that (183b) only allows the reading where rubii-ka sinzyu ‘ruby or pearl’ takes scope over kanousei ‘possibility’, attributing this fact to the fairly standard assumption that there is no scope reconstruction with A-movement. However, there is another possibility of analyzing (183b): to take it as an instance of pseudo-NGC in the sense of Hiraiwa (2001a), in which the genitive subject is simply base-generated in Spec-DP. Under this analysis, the scope fact follows immediately. Hiraiwa (2001a) provides the following examples as independent evidence for this analysis: (184) a.

nihon-ga

kome-o/*-ga/*-no

yunyuusuru kanousei

Japan-Nom

rice-Acc/-Nom/-Gen import

possibility

‘the possibility that Japan imports rice’ b. kome-no nihon-ga rice-Gen Japan-Nom

yunyuusuru

kanousei

import

possibility

(Hiraiwa 2001a:118)

(184b) could not be derived by overt movement of kome-no ‘rise-Gen’ out of the adnominal clause, Hiraiwa argues, since in that case, it would be taken as an instance of accusative-genitive conversion, as seen from (184a). Given that such conversion is 212

generally prohibited (cf. (73)), kome-no in (184b) must be base-generated in Spec-DP. Ochi further notes that in the gapless clause of kanousei ‘possibility’, “the order in which the gapless clause precedes the genitive phrase yields very low acceptability,” (p. 266) providing the following examples: (185) a.

[Sinzyu-no kotosi

yasuku naru]

pearl-Gen this year cheap

kanousei-ga

80% izyoo

da.

become possibility-Nom 80% over

be

‘The possibility that pearls become cheap this year is over 80%.’ b. *[Kotosi

yasuku

this year cheap

naru]

sinzyu-no kanousei-ga

80% izyoo da.

become pearl-Gen possibility-Nom 80% over be

The low acceptability of (185b) follows under the overt movement approach, Ochi claims, since in this sentence, the trace of sinzyu-no ‘pearl-Gen’ remains unbound. He contrasts (185b) with the following example: (186) [kinoo yesterday

untensiteita]

John-no

kuruma

was-driving

John-Gen car

‘John’s car that he was driving yesterday’

(Ochi 2001:265)

In this example, which involves a relative clause, John-no is base-generated in Spec-DP, hence interpreted as the possessor of kuruma ‘car’. On top of that, Ochi claims that with this example, “only one reading obtains in which John is both an agent and a possessor,” (p. 265) and that this reading is derived by the representation in which John binds the subject pro inside the relative clause. However, it is not clear whether the agent+possesor reading is actually possible with (186); the more natural reading is the one in which the possessor of the car is John and the agent of driving that car is 213

someone else. Given that John is preceded by the subject pro in the structure of (186), it is rather natural to attribute this fact to the difficulty of backward anaphora. But notice that once this is established, the low acceptability of (185b) can be given the same account. Overall, the generalization given in (180) does not seem to be well-established. To the extent that the present approach is independently well-motivated, the more natural generalization will be that no overt movement is possible at all with a genitive subject in the NGC construction.

214

4

Scrambling for Relativization: String-Vacuity and Island Effects

It has been standardly assumed since Kuno’s (1973) influential work that Japanese relativization does not involve movement, at least in any significant way. A main reason comes from the fact that it does not show island sensitivity (the relevant examples are reproduced below from Chapter 1): (1) a.

[[e1 kawaigatteita] was-fond-of

inu-ga

sinde-simatta]

kodomo1

dog-Nom dying-ended-up

child

‘Lit. the child1 who [the dog [e1 was fond of] died]’ b.

[[e1 kiteiru]

yoohuku-ga

is-wearing suit-Nom

yogoreteiru]

sinsi1

is-dirty

gentleman

‘Lit. the gentleman1 who [the suit [that e1 is wearing] is dirty]’ (2) a.

[[e1 sinda died

node]

minna-ga

kanasinda]

hito1

because everyone-Nom was-distressed person

‘Lit. the person1 who, because e1 died, everyone was distressed’ b.

[[pro e1 hara ippai belly-full

tabeta-ra] geri-o

site-simatta]

okasi1

ate-when diarrhea-Acc doing-ended-up-with cookies

‘Lit. cookies1 which, when (we) had stuffed ourselves with e1, (we) ended up with diarrhea’ (1) illustrates cases of relative clause island violations and (2) cases of adjunct clause

215

island violations. Based upon the observation that those languages that allow island violations in relativization also allow “pronoun drop,” Perlmutter (1972) claims that relativization in these languages involves deletion of an ordinary pronoun rather than movement of a relative pronoun, like in English. In current terms, this amounts to positing pro in the gap position of a relative clause in such a language. Granting that, however, Ishii (1991) remarks that “the absence of Subjacency effects does not immediately deny the existence of empty operator movement in Japanese relative clauses.” (p. 3)1 In fact, Ishii argues that “Japanese relative clauses can (and in fact must) involve movement when Subjacency is not violated.” (ibid.)2 This chapter aims to demonstrate, along the lines of Ishii (1991), that contrary to the standard view on Japanese relativization, this construction involves movement; more specifically, it involves an overt operator scrambled to a designated position and the overt operator is then deleted in the PF component a la Chomsky (1977). In particular, building upon Abe and Hornstein’s (2012) theory of movement, according to which string-vacuous movement is immune to island conditions, I argue that some cases of Japanese relativization, such as those given in (1) and (2), which appear to violate

1

Ishii (1991) uses this logic, following Saito’s (1985) claim on Japanese topicalization:

though this construction does not show island sensitivity, it does not follow from this fact that it does not involve movement at all. In fact, Saito demonstrates that PP-topicalization does involve movement. 2

See Ishii (1991) for the evidence for his empty operator movement analysis of

Japanese relativization, which is concerned with reconstruction effects of reflexive binding and weak crossover effects.

216

island conditions, do not induce island violations due to the fact that they involve string-vacuous scrambling of relative operators. One of the most important consequences of this analysis is that Japanese relativization involves leftward movement.

4.1

Inoue’s (1976) Generalization

Though Kuno (1973) provides examples of Japanese relativization that violate island conditions, as given in (1) and (2), it is not the case that this construction always tolerates such violations, as Kuno himself notes with such examples as the following (the judgments expressed with ? are Kuno’s): (3) a. ?[[Bill-ga e1 korosita Bill-Nom killed

toiu] zihaku-o

keisatu-ga

mada urazukesitei-nai]

that confession-Acc police-Nom yet

substantiate-not

onna1 woman ‘Lit. the woman1 who [the police have not substantiated yet [Bill’s confession that he killed e1]]’ b. ?[[e1 syuppansita] kaisya-ga published

kazi-de

company-Nom fire-by

yakete-simatta]

hon1

was-burned-down book

‘Lit. the book1 which [the company [that published e1] was burned down by a fire]’

(Kuno 1973:240)

There is a significant contrast in acceptability between these examples and those given in (1) and (2). 217

Inoue (1976) makes the following generalization regarding relativization out of relative clauses ((4) is adapted from Hasegawa (1981:283)): (4) Inoue’s Generalization: Relativizing a phrase in a relative clause is allowed if (i) the phrase is the subject of the relative clause and (ii) the head of that relative clause serves as the subject of the higher clause. Notice that (1a, b) observe this generalization whereas (3b) does not, since the gap associated with the head of the whole relative clause is not a subject but rather an object. Inoue provides more examples that do not conform to this generalization: (5) a. *[Bill-ga [e1 kaita] hon-o Bill-Nom

yakusiteiru] gakusya1

wrote book-Acc is-translating scholar

‘Lit. the scholar1 who [Bill is translating the book [that e1 wrote]]’ (Inoue 1976:178) b. *[[sono gakusya-ga e1 okutta] that scholar-Nom sent

syoten-ga

yaketa]

hon1

bookstore-Nom burned-down book

‘Lit. the book1 that [the bookstore [that the scholar sent e1 to] burned down]’ (ibid.:179) (5a) does not conform to the generalization since the relative clause that contains the gap associated with the head of the whole relative clause does not modify the subject of the higher clause. Nor does (5b), since the gap in question does not serve as a subject.3 3

Inoue (1976) provides an example of a more involved case of relativization out of

relative clauses:

218

Inoue’s generalization is easily extendable to cases of relativization out of pure complex NPs. The relevant generalization is something like the following: (6) Relativizing a phrase in a pure complex NP is allowed if (i) the phrase is the subject of the appositive clause of the complex NP and (ii) the head of that appositive clause serves as the subject of the higher clause. Thus, compare the following examples: (7) a.

[[e1 Mary-o

korosita toiu] zihaku-ga

Mary-Acc killed

sinyoo-deki-nai] otoko1

that confession-Nom

unreliable

man

‘Lit. the man1 who [[the confession that e1 killed Mary] is unreliable]’ (i) [[[e1 kaita] hon-o

syuppansita] honya-ga

wrote book-Acc published

hasansita]

gakusya1

publisher-Nom went-bankrupt

scholar

‘the scholar1 who [the publisher [that published the book [that e1 wrote] went bankrupt]]’

(Inoue 1976:178)

In (i), the phrase that is relativized into the highest head, whose original position is marked with e1, is doubly embedded within relative clauses. Inoue takes this example as acceptable, suggesting that the generalization she proposes should be modified in such a way that (i) a phrase that is relativized out of a relative clause must be the subject of the relative clause and (ii) the head of a relative clause that contains that phrase (which does not have to be the immediately higher relative clause) must serve as the subject of the higher clause. In (i), e1 serves as the subject of the immediately dominating relative clause and, although that relative clause modifies the head that serves as an object, namely, hon-o ‘book-Acc’, e1 is contained in the higher relative clause that modifies the head that serves as a subject, namely, honya-ga ‘publisher-Nom’, hence conforming to the modified generalization. I am not sure about the acceptability of such an involved case as in (i), which is a lot harder to judge than those simpler examples given in the text. For this reason, I leave such cases aside. See Hasegawa (1981) for relevant discussion.

219

b. *[[Bill-ga e1 korosita toiu] zihaku-ga Bill-Nom killed

that confession-Nom

sinyoo-deki-nai]

onna1

unreliable

woman

‘Lit. the man1 who [[the confession that Bill killed e1] is unreliable]’ c. *[keisatu-ga [e1 police-Nom

Mary-o

korosita toiu] zihaku-o

mada

Mary-Acc

killed

yet

that confession-Acc

urazukesitei-nai] otoko1 substantiate-not man ‘Lit. the man1 who [the police have not substantiated yet [the confession that e1 killed Mary]]’ In (7a), the phrase that is relativized out of the pure complex NP, whose original position is marked with e1, is the subject of the appositive clause of that complex NP and the head of this NP serves as the subject of the higher clause. Hence, this conforms to the generalization in (6), explaining the acceptability of (7a). (7b, c), on the other hand, do not conform to the generalization: in (7b), the gap in question is not the subject of the appositive clause of a pure complex NP and in (7c), the head of the appositive clause that contains the gap in question does not serve as the subject of the higher clause. This yields the unacceptability of (7b, c). Note that Kuno’s example (3a) does not conform to the generalization in question, either, which matches the fact that this sentence is as bad as (7b, c). At this point, one may raise the question why Inoue’s generalization and the one extended to the case of a pure complex NP hold true. It is natural to reason that a key to answering this question has to do with what may be called the left edge effect. Notice 220

that in those cases that conform to these generalizations, the gap that marks the original position of a phrase relativized out of a complex NP, whether it involves a relative or appositive clause, is located at the left edge of the whole relative clause. This reasoning is further confirmed with those cases of relativization out of adjunct clauses; Inoue (1976) provides the following set of data that demonstrates the relevant contrast ((8a) is reproduced from (2a)): (8) a.

[[e1 sinda died

node]

minna-ga

kanasinda]

hito1

because everyone-Nom was-distressed person

‘Lit. the person1 who, because e1 died, everyone was distressed’ b.

[[e1 sinda noni] died

though

daremo kanasima-nakat-ta]

hito1

anyone was-not-distressed

person

‘Lit. the person1 who, though e1 died, no one was distressed’ (9) a. *[[John-ga e1 hikiukere-ba] subete-no zinzi-ga John-Nom undertake-if

(Inoue 1976:183) umaku iku]

every-Gen personnel matter-Nom well

go

sigoto1 job ‘Lit. the job1 that if John will undertake e1, all the personnel matters will go well’ b. *[[John-ga e1 katta-ra] Bill-ga John-Nom buy-if

taipu-raitaa-o

Bill-Nom typewriter-Acc

kau] tokei1 buy clock

‘Lit. the clock1 that if John buys e1, Bill will buy a typewriter’

(ibid.:184)

In (8a, b), the gaps in question are located at the left edge of the whole relative clauses 221

whereas in (9a, b) they are not. This is reflected in the contrast in acceptability between (8a, b) and (9a, b).4 The following sentences show the relevant contrast more sharply 4

There are a number of Japanese native speakers, including me, who find (9a) a lot

better than (9b) (The five Japanese native speakers I consulted with all agree upon this judgment.) This is probably because the former involves a cause-effect relation between the subordinate and main clauses, whereas the latter involves a more elusive conditional relation. Yukiko Ueda (personal communication) pointed out to me that the same effect is observed with a variant of (2b) in which pro is replaced by an overt NP, as shown below: (i) ?[[John-ga e hara ippai John-Nom belly-full

tabeta-ra] geri-o

site-simatta]

okasi1

ate-when diarrhea-Acc doing-ended-up-with cookies

‘Lit. cookies1 which, when John had stuffed himself with e1, he ended up with diarrhea’ This may indicate that when such a cause-effect relation holds between the subordinate and main clauses, the subordinate clause constitutes only a very week island. This conjecture will be supported by the fact that the cleft versions of (9a) and (i) seem fairly acceptable, as shown below (where NL stands for nominalizer): (ii) ?[[John-ga e1 hikiukere-ba] subete-no zinzi-ga

umaku iku]

John-Nom undertake-if every-Gen personnel matter-Nomwell sono

sigoto1-o da.

that

job-Acc be

go

no-wa NL-Top

‘Lit. It is that job1 that if John will undertake e1, all the personnel matters will go well.’ (iii) ?[[John-ga e1 hara ippai tabeta-ra] geri-o Jon-Nom

site-simatta]

no-wa

belly-full ate-when diarrhea-Acc doing-ended-up-with NL-Top

sono

okasi1-o

da.

that

cookie-Acc

be

‘Lit. It is those cookies1 which, when John had stuffed himself with e1, he ended up with diarrhea’ Despite such a factor, the relevant left edge effect seems to hold in relativization out of

222

((10) should be compared with (9b)): (10) ?[[e1 tokei-o

katta-ra] Bill-ga

clock-Acc buy-if

taipuraitaa-o kau] hito1

Bill-Nom typewriter-Acc buy person

‘Lit. the person1 that if e1 buys a clock, Bill will buy a typewriter’ (11) a. *[[John-ga e1 sikatta toki] John-Nom scolded when

Mary-ga

naite-simatta] kodomo1

Mary-Nom

cried

child

‘Lit. the child1 that when John scolded e1, Mary got angry’ b.

[[e1 John-ni sikar-areta

toki]

Mary-ga

John-by was-scolded when

naite-simatta] kodomo1

Mary-Nom cried

child

‘Lit. the child1 that when e1 was scolded, Mary got angry’ Again, these examples show that when the gap of a relativized phrase is located at the left edge, it is immune from an island effect. Further confirmation of the relevance of the left edge effect comes from the observation made by Inoue (1976) that scrambling the complex NP that contains the gap in question in such a way that the latter ends up being located at the left edge improves such an adjunct clause. The five Japanese native speakers I consulted with all find a clear contrast in acceptability between the following examples: (iv) a.

[[e1 kareta node] died

minna-ga

kanasinda]

hana1

because everyone-Nom was-distressed flower

‘Lit. the flower1 which, because e1 died, everyone was distressed’ b.?*[minna-ga [e1 kareta node]

kanasinda]

hana1

everyone-Nom died because was-distressed flower Thus, these data suggest that the left edge effect in question is also relevant for those cases where phrases are relativized out of adjunct clauses.

223

the acceptability of the example involved. Inoue provides the following pairs ((12a) is reproduced from (5a)): (12) a. *[Bill-ga [e1 kaita] hon-o

yakusiteiru]

gakusya1

Bill-Nom wrote book-Acc is-translating scholar ‘Lit. the scholar1 who Bill is translating the book that e1 wrote’ b. ?[ [e1 kaita] hon-o

Bill-ga

yakusiteiru]

wrote book-Acc Bill-Nom (13) a. *[Bill-ga [e1 kaita] hon-ni

gakusya1

is-translating scholar

zyobun-o

noseta]

Bill-Nom wrote book-Dat forward-Acc put

gakusya1 scholar

‘Lit. the scholar1 who Bill put a forward to the book that e1 wrote’ b. [[e1 kaita] hon-ni

Bill-ga

zyobun-o

noseta]

wrote book-Dat Bill-Nom forward-Acc put

gakusya1 scholar (Inoue 1976:178-9)

The improvement in acceptability in the (b)-examples strongly suggests that the left edge effect is relevant for violability of island conditions. The same contrast obtains with respect to those examples that involve violation of pure complex NP and adjunct clause island conditions, as shown below ((14a) is reproduced from (7c)): (14) a. *[keisatu-ga [e1 Mary-o police-Nom

Mary-Acc

korosita toiu] zihaku-o killed

mada

that confession-Acc yet

urazukesitei-nai] otoko1 substantiate-not man ‘Lit. the man1 who the police have not substantiated yet the confession that e1 224

killed Mary’ b. [[e1 Mary-o

korosita toiu] zihaku-o

Mary-Acc killed

mada

that confession-Acc police-Nom yet

urazukesitei-nai]

otoko1

substantiate-not

man

(15) a. *[Bill-ga [e1 musume-o

keisatu-ga

nagutta

Bill-Nom daughter-Acc hit

node]

okotteiru] sensei1

because is-angry teacher

‘Lit. the teacher1 who Bill is angry because e1 hit his daughter’ b. [[e1 musume-o daughter-Acc

nagutta

node]

Bill-ga

okotteiru]

hit

because Bill-Nom is-angry

sensei1 teacher

In the (b)-examples, the phrases that contain the gaps in question are raised leftward, so that those gaps are located at the left edges of the whole relative clauses. Again, the improvement in acceptability of these examples strongly suggests that the left edge effect is at work here. We can summarize the violability of island conditions in Japanese relativization as follows: (16) The Left Edge Effect: A phrase that is relativized out of an island causes acceptability only if its original position ends up being located at the left edge position of the whole relative clause. A relevant question to ask here is whether the notion of left edge is defined in purely syntactic terms or in terms of phonetic strings. There is evidence that the latter is correct. 225

Consider the following examples ((17a) is reproduced from (2b)): (17) a.

[[pro e1 hara ippai belly-full

tabeta-ra] geri-o

site-simatta]

okasi1

ate-when diarrhea-Acc doing-ended-up-with cookies

‘Lit. cookies1 which, when (we) had stuffed ourselves with e1, (we) ended up with diarrhea’ b. [[pro e1 nome-ba] take-if

byooki-ga

naoru]

kusuri1

illness-Nom

is-cured medicine

‘Lit. the medicine1 which, if (we) take e1, (we) recover from (our) illness’ (Inoue 1976:183) In these examples, the gaps marked with e1, which indicate the original positions of the phrases that are relativized out of adjunct clause islands, are not located at the left edge position in its strictly syntactic sense, since the leftmost element is pro. Nonetheless, the examples are almost perfect in their acceptability. This strongly suggests that the notion of left edge relevant here should be understood in terms of phonetic strings: in order to circumvent an island violation, the gap in question must be located leftmost among pronounced elements. But this is not the end of the story. Recall one of Kuno’s (1973) examples, i.e., (3b), repeated below, which does not sound good despite the fact that the gap in question is located at the left edge of the whole relative clause (I changed the mark of acceptability judgment from ? to * to make it consistent with those of the other examples). (18) *[[e2 e1 syuppansita] kaisya2-ga published

kazi-de yakete-simatta]

hon1

company-Nom fire-by was-burned-down book 226

‘Lit. the book1 which the company that published e1 was burned down by a fire’ Notice that in this case, the gap in question is not the leftmost element in purely syntactic terms, since the subject gap that is relativized locally sits to its left. Thus, the gap that is associated with a relative head is somehow visible for determining what is at the left edge among phonetic strings. Similar examples are found in Hasegawa (1981): (19) a.

[[e1 e2 katteita] inu2-ga kept

sinde-simatta] kodomo1-ga

dog-Nom have-died

child-Nom

naiteiru. is-crying

‘Lit. the child1 [who the dog [which e1 kept] has died] is crying.’ b. *[[e2 e1 katteita] kodomo2-ga kept

child-Nom

sinde-simatta] inu1-ga have-died

kanasi-soo da.

dog-Nom look-sad

‘Lit. the dog1 [which the child [who kept e1] has died] looks sad.’ (Hasegawa 1981:281) (20) [[e e nagusameta] tomodati-ga comforted

friend-Nom

sinde-simatta] hito have-died

person

‘Lit. the person1 [who the friend [who e1 comforted] has died]’ ‘Lit. *the person1 [who the friend [who comforted e1] has died]’

(ibid.:282)

The contrast in acceptability between (19a) and (19b) shows that both subject and object gaps in the innermost relative clause counts for the determination of what is at the left edge: In (19a), the subject gap is at the left edge and is associated with the higher relative head without inducing an island effect, whereas in (19b), it is the object gap that is associated with the higher relative clause, hence inducing an island effect. (20) shows the same point: it can be interpreted in only one way, as indicated in the translations, 227

that is, in such a way that the subject gap is associated with the higher relative clause head. All the above examples point to the conclusion that, unlike pro, a gap associated with a relative head is somehow visible for the determination of what is at the left edge. There is independent evidence for supporting this conclusion, which is concerned with so-called wanna-contraction, as exemplified below: (21) a.

Who do you want to see?

b. Who do you wanna see? (22) a.

Who do you want to go out?

b. *Who do you wanna go out? The reason why wanna-contraction is blocked in (22b) is that the trace of who intervenes between want and to, even though it is unpronounced, as shown below: (23) who do you want [TP t to go out] This indicates that a trace left behind by an operator movement is somehow visible for PF operations such as wanna-contraction. Notice that the fact that wanna-contraction is possible in (21b) suggests that PRO does not block this PF operation, as shown below: (24) who do you want [TP PRO to see t] The generalization that comes out of this paradigm of wanna-contraction regarding what counts as visible for PF operations is quite compatible with the above generalization regarding what counts as visible for the determination of what is at the left edge in Japanese relativization, with an additional assumption: Japanese relativization involves operator movement at least when no islands intervene, following Ishii (1991). With this latter assumption, we can reach the following conclusion: 228

(25) Pronominal elements such as pro and PRO are invisible for PF operations and processes while traces left behind by operator movement are visible for such operations and processes. Thus, the left edge effect, given in (16), should be understood to be concerned with a PF process with the proviso given in (25).5

4.2

Proposal

In this section, I address the question why the left edge effect, which is reproduced below, holds true in Japanese relativization. (26) The Left Edge Effect: A phrase that is relativized out of an island causes acceptability only if its original position ends up being located at the left edge position of the whole relative clause. Let us first hypothesize the following:6 (27) Japanese relativization involves scrambling of an overt operator. This leads us to the claim that the scrambling involved in relativization is somehow immune to island conditions when the launching site is or ends up being located at the left edge.

5

Korean relativization out of relative clauses also seems to show the left edge effect

characterized in the text. See Han and Kim (2004) for relevant discussion. 6

I discuss the option of base-generating pro in Japanese relativization in the next

section.

229

4.2.1

String-Vacuous Movement and Island Sensitivity

In deriving the left edge effect given in (26), I crucially assume Abe and Hornstein’s (2012) theory of movement, especially their mechanism of how pronunciation of a chain is determined and how this determination is related to island sensitivity. The main claim of theirs that is adopted here is the following: (28) String-vacuous movement is exempt from island conditions. Abe and Hornstein (2012) argue that string-vacuous movement is always “covert” in that the bottom copy of a movement chain is pronounced, as in (29b), hence prohibiting “overt” string-vacuous movement, as in (29a). (29) a. *XP [ …]

b. [ XP …]

They attribute the reason for such prohibition to the idea expressed by Chomsky (1995) that a feature “enters the numeration only if it has an effect on output.” (p. 294) In his framework, it is a strong feature that triggers overt movement and this condition prohibits this feature from being included in the numeration unless it gives rise to an effect on output. Taking PF output as relevant here, this in effect prohibits “overt” string-vacuous movement, since this movement has no effect on PF output. Abe and Hornstein (2012) characterize the relevant condition as follows: (30) The head of a chain created by Move cannot be pronounced unless it has an effect

230

on PF output. Abe and Hornstein then propose the following: (31) Locality conditions such as island conditions apply only to “overt” movement. From this assumption, (28) follows.7 Given this mechanism of movement, it is rather straightforward to derive the left edge effect of Japanese relativization, given in (26), since when a phrase that is relativized out of an island is located at the left edge of the whole relative clause, the scrambling involved will be string-vacuous. Suppose that Japanese has a wh-phrase-like operator, a la Chomsky (1977), that functions for relativization, though it is doomed to get deleted at PF.8 Suppose further that it undergoes scrambling to Spec-CP in relative clauses, as schematically shown below: (32) [CP WH1 [ … WH1 …]] Head1

Then according to Abe and Hornstein’s (2012) theory of movement, the determination of which copy of the chain (WH1, WH1) is pronounced depends upon whether the 7

See Section 3.1.2 for the evidence for Abe and Hornstein’s (2012) theory of

movement, which is concerned with English right node raising. 8

I have nothing interesting to say about why such a wh-phrase-like operator is

obligatorily deleted in Japanese, unlike, say, in English. The obligatory deletion of this wh-phrase-like operator must take place after the determination of which copy is pronounced in accordance with an effect on PF output, as stated in (30). To posit a null operator instead will probably not be a good idea, since it would make no sense to talk about which copy of a chain is pronounced if the head of a chain is an unpronounced operator.

231

movement involved has an effect on PF output, as schematically shown below: (33) a. b.

[CP WH1 [ … …]] Head1 [CP [WH1 …]] Head1

(33b) illustrates a case that involves string-vacuous movement, hence giving rise to a chain in which the bottom copy is pronounced in accordance with (30). All the examples of Japanese relativization that have demonstrated the left edge effect belong to this case. To take one for illustration, consider (1a), reproduced below: (34) [[e1 kawaigatteita] inu-ga was-fond-of

sinde-simatta]

kodomo1

dog-Nom dying-ended-up

child

‘Lit. the child1 who the dog e1 was fond of died’ Under the present assumptions, this example has the following structure: (35) [CP [TP [DP [CP [TP WH1 kawaigatteita]] inu-ga] sinde-simatta]] kodomo1

It follows from (31) that the movement involved in (35) is immune to island conditions since it is “covert” in the sense that the bottom copy of the chain is pronounced. Further, recall those cases where the left edge effect occurs thanks to scrambling the phrase that includes a phrase to be relativized out of an island. The relevant examples are reproduced below: (36) a. *[Bill-ga [e1 kaita] hon-o

yakusiteiru]

gakusya1

Bill-Nom wrote book-Acc is-translating scholar ‘Lit. the scholar1 who Bill is translating the book that e1 wrote’

232

b. ?[ [e1 kaita] hon-o

Bill-ga

yakusiteiru]

wrote book-Acc Bill-Nom

gakusya1

is-translating scholar

The improvement in the acceptability of (36b) is straightforwardly explained under the present theory of movement, since scrambling the whole object [WH kaita] hon-o ‘the book that WH wrote’ makes the scrambling of WH string-vacuous, hence immune to the relative clause island condition. If the way the left edge effect in question is derived relies crucially on an effect on PF output, as indicated in (30), it should be predicted that any material, be it an argument or an adjunct, should count for the PF effect. This is in fact borne out with cases in which adjuncts serve as interveners for scrambling of WH, though the contrast may not be as clear as those in which arguments are involved. Let us consider the following examples: (37) a. ?[[[e1 itidomo kawaigattei-nakat-ta to] ever gasisita]

cherish-not-Past

mitomer-areru] inu-ga

Comp is-recognized

dog-Nom

kodomo1

died-of-hunger child ‘Lit. the child1 who [the dog [it is recognized that e1 never cherished] died of hunger]’ b.?*[[ itidomo [ e1 kawaigatteita to] ever

cherished

gasisita]

kodomo1

mitomer-are-nai] inu-ga

Comp is-recognized-not dog-Nom

died-of-hunger child 233

‘Lit. the child1 who [the dog [it is never recognized that e1 cherished] died of hunger]’ In (37a), the negative polarity adjunct itidomo ‘ever’ is licensed by the negative morpheme nai (which is in fact pronounced as nak because it is followed by a vowel) in the most embedded clause. Since the relativized WH phrase functions as the subject of this clause, it can be taken as base-generated before itidomo, as marked with e. Thus, the scrambling of WH is string-vacuous in this case, thereby inducing no island effect. In (37b), on the other hand, the negative polarity adjunct itidomo is licensed by the negative morpheme nai, which is attached to the higher verb mitomer-are ‘is recognized’, and hence is located in the higher clause, on the standard assumption that licensing of such a negative polarity item obeys a clause-mate condition in Japanese.9 Thus, the relativized WH phrase, which functions as the subject of the lower clause, necessarily follows the negative polarity adjunct, as marked with e. This intervener thus makes the scrambling of WH non-string-vacuous, thereby inducing an island effect. A more example is given below:

9

Compare the following examples:

(i) a.

Itidomo [kodomo-ga inu-o ever

child-Nom

kawaigatta to]

dog-Acc cherished Comp

mitomer-are-nai. is-recognized-not

‘It is never recognized that the child cherished the dog.’ b.?*[kodomo-ga itidomo child-Nom ever

inu-o

kawaigatta to]

dog-Acc cherished Comp

mitomer-are-nai. is-recognized-not

The unacceptability of (ib) clearly shows that such a negative polarity item as itidomo must be licensed by a clause-mate negative item.

234

(38) ?*[gan-de [e1 zyuunen-kan kawaigatteita] inu-ga cancer-by

ten years-for was-fond-of

sinde-simatta]

dog-Nom dying-ended-up

kodomo1 child ‘Lit. the child1 who [the dog [e1 was fond of for ten years] died of cancer]’ Here, the adjunct gan-de ‘by cancer’ modifies sinde-simatta ‘ended up dying’ whereas the relativized WH phrase serves as the subject of the lower clause, which is thus base-generated after the adjunct. This makes the scrambling of WH non-string-vacuous, hence inducing an island effect.10 10

Kensuke Takita (personal communication) pointed out to me that so-called high

adverbs such as kinoo ‘yesterday’ can be put, without significant degradation, at the top of a relative clause that involves relativization out of a relative clause, as shown below: (i) ?[kinoo yesterday

[ e1 zyuunen-kan kawaigatteita] inu-ga ten years-for cherished

sinde-simatta] kodomo1

dog-Nom dying-ended-up child

‘Lit. the child1 who [the dog [e1 was fond of for ten years] died yesterday]’ Compare this example with (38), which involves a low adverb put at the top of a relative clause of the same configuration. We may account for the fair acceptability of (i) in the following way: I have assumed without any argument that the relative operator WH moves to Spec-CP. Based upon the fact that Japanese relativization does not allow overt complementizer, however, Murasugi (1991) argues that Japanese relative clauses constitute TP rather than CP. If this is the case, then we will need to assume that the relative operator WH is adjoined to TP. Given this, the acceptability of (i) can be accounted for on the assumption that high adverbs can be adjoined to TP; in that case, the scrambling of the relative operator WH involved in (i) can be string-vacuous by being adjoined to the TP of the higher relative clause before kinoo ‘yesterday’ is adjoined to that TP. This line of account is supported by the unacceptability of (38) on the assumption that low adverbs must be adjoined lower than TP.

235

Finally, recall that we have seen that the left edge effect in question is subject to some sort of PF visibility, as stated in (25), reproduced below: (39) Pronominal elements such as pro and PRO are invisible for PF operations and processes while traces left behind by operator movement are visible for such operations and processes. We can make sense of this fact under Abe and Hornstein’s (2012) theory of movement by assuming that the determination of whether a chain produced by movement has an effect on PF output is subject to (39). Since pronominal elements such as pro and PRO are invisible for such determination, they do not act as interveners that make relevant chains non-string-vacuous. A relevant example is reproduced below from (17a): (40) [[pro e1 hara ippai tabeta-ra] geri-o

site-simatta]

okasi1

belly-full ate-when diarrhea-Acc doing-ended-up-with cookies ‘Lit. cookies1 which, when (we) had stuffed ourselves with e1, (we) ended up with diarrhea’ In this example, the relative operator WH undergoes scrambling from the position marked with e1, crossing pro, and lands in the Spec-CP of the whole relative clause. Crossing only pro, this movement is regarded as having no effect on PF output, and hence the bottom copy of the resulting chain must be pronounced according to (30). This makes the chain immune to island conditions. On the other hand, since a trace left behind by scrambling of a relative operator is visible for determining whether a movement operation has an effect on PF output, it makes a relevant chain non-string-vacuous. A relevant example is reproduced below from (18): 236

(41) *[[e2 e1 syuppansita] kaisya2-ga published

kazi-de yakete-simatta]

hon1

company-Nom fire-by was-burned-down book

‘Lit. the book1 which the company that published e1 was burned down by a fire’ Here, the relative operator WH undergoes scrambling from the position marked with e1, and it crosses the subject trace left behind by scrambling of another relative operator WH, which refers to kaisya ‘company’. Since this trace is visible, it causes the chain created by the other movement to have an effect on PF output. Hence, this chain is subject to island conditions, giving rise to the degraded status of (41). In this way, the left edge effect, with the proviso of PF visibility given in (39), is neatly deduced from the movement theory proposed by Abe and Hornstein (2012).11

11

See Abe and Hornstein (2012) for the demonstration that what Nakao (2009) calls

left node raising in Japanese also shows a left edge effect similar to the one discussed in the text and that their theory of movement naturally accommodates this effect. See also Abe and Nakao (2012) for the claim that the same sort of left edge effect is also observed in so-called Japanese right node raising and that it is also nicely captured by Abe and Hornstein’s theory of movement. One may raise the question why trace is visible for PF operations and processes, but not PRO and pro. I have no good answer to this question at the moment but to offer a speculation: Instead of assuming that a chain is produced in terms of Copy, Merge and Delete, let us assume, following Holmberg (2000), that features are not copied when Move is applied but rather they are scattered (see also Section 2.1). Under this conception, whether a given instance of movement is “overt” or “covert” is determined by which copy the [PF] feature resides in. Suppose further that PRO and pro simply lack [PF] features or if they carry such features, they get deleted when the lexical items bearing them enter into the PF component. This will immediately explain why they are invisible for PF operations and processes. As for trace, Abe (2015) proposes that any

237

4.2.2

Sakai’s (1994) Approach

Since the present approach owes much insight to Sakai’s (1994), let me briefly discuss his approach, arguing that it causes some problems in dealing with the data presented so far. Addressing the question why Inoue’s (1976) generalization holds true for Japanese relativization, Sakai argues that in a configuration that allows a phrase to be moved out of a relative clause into the top of the higher relative clause, the movement involved makes its way through independently motivated steps of movement; namely one involved in the NGC construction, reproduced in (42), and the other involved in what Kuno (1973) calls subjectivization, exemplified in (43). (42) NGC: a.

[John-ga

sukina] hon

John-Nom like

book

‘the book John likes.’

member of a chain acts as an intervener for the notion of adjacency relevant to string-vacuous movement, irrespective of whether it itself carries a [PF] feature or not. Abe characterizes such an intervention by an unpronounced copy as follows: (i)

Given a chain C = (α1, … αn) and a sequence of lexical strings … β, αi, γ …, where αi is a member of C that does not carry its [PF] feature, β and γ are not adjacent to each other unless αi is a single-membered chain.

In this way, when a trace intervenes in a chain produced by movement, the resulting two members of that chain are not regarded as adjacent, hence causing an effect on PF output. This dictates that the upper copy be pronounced, which thus makes the chain in question sensitive to islands.

238

b. [John-no sukina] John-Gen like

hon book

(43) Subjectivization: a.

[Zoo-no

hana-ga]

nagai.

elephant-Gen trunk-Nom long ‘An elephant’s trunk is long.’ b. Zoo-ga

hana-ga

elephant-Nom trunk-Nom

nagai. long

Sakai assumes that both constructions involve movement, as shown below: (44) a.

[DP John-no [TP

t sukina] hon]

b. [TP zoo-ga [DP t hana-ga] nagai]

(44a) shows that NGC involves movement from the Spec-TP of the relative clause into the Spec-DP of the whole complex DP. (44b) shows that subjectivization involves movement from the Spec-DP of a subject phrase into the Spec-TP of the whole sentence. Given this, Sakai claims that such an example as (1a), repeated below, which involves a violation of the relative clause island condition, has the derivation given in (46).12

12

Just for ease of discussion, I present the derivation as if the relative head kodomo

‘child’ underwent movement from its original position within the relative clause to its head position, but there is no theoretical commitment here.

239

(45) [[e1 kawaigatteita] inu-ga was-fond-of

sinde-simatta]

dog-Nom dying-ended-up

kodomo1 child

‘Lit. the child1 who [the dog [e1 was fond of] died]’ (46) a.

[DP kodomo1 [TP t1 t2 kawaigatteita] inu2 (movement involved in NGC)

b. [TP kodomo1 [DP t1 [TP t1 t2 kawaigatteita] inu2-ga sinde simatta] (movement involved in subjectivization) c.

[TP t1 [DP t1 [TP t1 t2 kawaigatteita] inu2-ga sinde simatta] kodomo1

This proposal nicely captures Inoue’s (1976) generalization, reproduced below: (47) Inoue’s Generalization: Relativizing a phrase in a relative clause is allowed if (i)

the phrase is the subject of the relative clause and

(ii)

the head of that relative clause serves as the subject of the higher clause.

The fact that a phrase to move must be “the subject of the relative clause” follows from a restriction on NGC, since an object does not undergo such conversion. The fact that “the head of that relative clause serves as the subject of the higher clause” follows from a restriction on subjectivization, since this operation takes place from within a subject DP, but not from within an object DP, as noted by Kuno (1973): (48) a.

John-ga1 [DP t1 kodomo-ga]

sensei-ni sikar-areta.

John-Nom

teacher-by was-scolded

child-Nom

‘John, his child was scolded by the teacher.’

240

b. *John-ga1

sensei-ga [DP t1 kodomo-o]

John-Nom teacher-Nom

child-Acc

sikatta. scolded

‘John, the teacher scolded his child.’

(Kuno 1973:70)

Though Sakai’s (1994) approach is attractive in that it reduces the violability of the relative clause island by a phrase to be relativized to independently motivated operations, there are several problems with this approach. First, it does not really answer the question why the relative clause island is violable in Japanese relativization. Sakai claims that the movement responsible for NGC is the one that can violate the relative clause island, but this simply begs the question: why does this movement not induce an island effect? Second, notice that Sakai’s approach tries to derive Inoue’s (1976) generalization. It can also accommodate the generalization, given in (6), that is concerned with cases of relativization out of pure complex NPs, since these NPs also allow NGC. However, it fails to capture cases of relativization out of adjunct clauses, as illustrated in (8), reproduced below, since adjunct clauses do not support NGC: (49) a.

[[e1 sinda node]

minna-ga

kanasinda]

hito1

died because everyone-Nom was-distressed person ‘Lit. the person1 who, because e1 died, everyone was distressed’ b.

[[e1 sinda noni] died

though

daremo kanasima-nakat-ta]

hito1

anyone was-not-distressed

person

‘Lit. the person1 who, though e1 died, no one was distressed’ Further, there is a case that undermines Sakai’s claim that subjectivization is also responsible for deriving Inoue’s generalization. This is concerned with those examples 241

that show significant improvement in acceptability when a phrase that is relativized out of a relative clause island is contained in an object DP and this latter phrase is shifted by scrambling in such a way that the original position of the relativized phrase ends up being located at the left edge. The relevant examples are reproduced below from (36): (50) a. *[Bill-ga [e1 kaita] hon-o

yakusiteiru]

gakusya1

Bill-Nom wrote book-Acc is-translating scholar ‘Lit. the scholar1 who Bill is translating the book that e1 wrote’ b. ?[ [e1 kaita] hon-o

Bill-ga

yakusiteiru]

wrote book-Acc Bill-Nom

gakusya1

is-translating scholar

Sakai’s approach could not explain the improvement of the acceptability of (50b), since the same effect is not observed with subjectivization cases. Compare (48b) with the corresponding case in which the object DP is scrambled in front of the subject DP, as shown below: (51) *John-ga1 [DP t1 kodomo-o]2 sensei-ga John-Nom

child-Acc

teacher-Nom

t2

sikatta. scolded

‘John, the teacher scolded his child.’ (51) shows no improvement in acceptability when compared with (48b). This makes doubtful the claim that subjectivization is involved in those cases of Japanese relativization that violate the relative clause island condition.13 Finally, recall that we

13

One might claim that this argument does not necessarily exclude the possibility of

deriving some cases of relativization out of relative clauses from their corresponding subjectivization cases. For instance, nothing seems to prevent (1b) from being derived

242

have reached the conclusion that unlike traces left behind by operator movement, pro and PRO are invisible for PF operations and processes, so that an object DP can be relativized out of an island only if the subject DP to its left is pro, as shown in (17), from the following subjectivization case: (i) Sono sinsi1-ga that

[pro1 kiteiru]

gentleman-Nom

yoohuku-ga yogoreteiru.

is-wearing suit-Nom

is-dirty

‘As for that gentleman, the suit he is wearing is dirty.’ If (1b) were derived from the underlying structure for (i), then the WH operator would be able to undergo movement to the Spec-CP of the whole relative clause without crossing any island. This account, however, leaves unanswered the question why (i) is acceptable, to begin with. If we follow Sakai (1994) in assuming that subjectivization involves movement, then we need to say something about why the subjectivized phrase sono sinsi ‘that gentleman’ in (i) can be moved out of the relative clause. On the other hand, if we assume that subjectivization can be established through a coreference relation, as indicated in (i), then it will face the problem of why subjectivization does not take place from an object, as shown in (48b) and the following sentence: (ii) *Sono that

gakusya1-ga

[Bill-ga [pro1 kaita] hon-o

scholar-Nom Bill-Nom

yakusiteiru]

wrote book-Acc is-translating

‘Lit. As for that scholar1, Bill is translating the book that he/she1 wrote.’ Note that the corresponding relativization is also unacceptable, as shown in (50a), reproduced below: (iii) *[Bill-ga [e1 kaita] hon-o Bill-Nom

yakusiteiru] gakusya1

wrote book-Acc is-translating scholar

‘Lit. the scholar1 who [Bill is translating the book [that e1 wrote]]’ Under the present assumptions, the (un)acceptability of (i) and (ii) is immediately explained under the assumption that subjectivization involves movement: (i) involves string-vacuous movement whereas (ii) does not. In other words, the present approach can deal with subjectivization and relativization on a par with respect to island sensitivity. Given this state of affairs, it is superfluous to claim that subjectivization intermediates in deriving relativization out of a relative clause.

243

reproduced below: (52) a.

[[pro e1 hara ippai belly-full

tabeta-ra] geri-o

site-simatta]

okasi1

ate-when diarrhea-Acc doing-ended-up-with cookies

‘Lit. cookies1 which, when (we) had stuffed ourselves with e1, (we) ended up with diarrhea’ b. [[pro e1 nome-ba] take-if

byooki-ga

naoru]

kusuri1

illness-Nom

is-cured medicine

‘Lit. the medicine1 which, if (we) take e1, (we) recover from (our) illness’ This sort of PF effect will be unexpected under Sakai’s approach. Though the present approach follows Sakai’s line of analysis in assuming that Japanese relativization basically involves movement and that a special property of this operation gives rise to the violability of island conditions in this construction, I crucially depart from his analysis in claiming that the crucial factor for the violability of island conditions comes from the left edge effect given in (26) and that this follows from the assumption made by Abe and Hornstein (2012) that string-vacuous movement is exempt from island conditions.

4.2.3

Some Consequences

There are a couple of consequences arising from the above discussions. One has to do with relativization from sentential subjects in Japanese. Kuno (1973) provides the following examples to show that Japanese relativization is immune to sentential subject islands: 244

(53) a.

[[watakusi-ga e1

au

I-Nom

koto/no]-ga muzukasii]

meet that-Nom

difficult

hito1 person

‘Lit. the person1 who that I see e1 is difficult’ b. [[kare-ga e1 kaita

koto]-ga yoku

sir-areteiru]

bun1

he-Nom wrote

that-Nom well

known-is

line

‘Lit. the line1 which that he wrote e1 is well known’

(Kuno 1973:241)

These examples are fairly acceptable despite the fact that the phrases that are relativized out of sentential subjects are not located at the left edges of the whole relative clauses. Under the present theory of movement, this strongly suggests that Japanese relativization is not subject to sentential subject islands, to begin with. This comports well with Saito’s (1985) observation that Japanese scrambling does not show any difference in acceptability no matter whether it takes place out of a sentential object or out of a sentential subject. He provides the following pair:14 (54) a.

Dono hon-o1

Mary-ga

[John-ga t1 katta koto]-o mondai-ni

which book-Acc Mary-Nom John-Nom bought fact-Acc problem-to siteiru

no?

is-making Q

14

Saito (1985) judges both sentences as ??/?*, but I omit these marks in (54) since I

find these sentences fairly good. Strictly, both the sentential object in (54a) and the sentential subject in (54b) constitute complex NPs headed by koto ‘fact’, which seems to induce very weak island effects, if any. The point here is that there is no significant difference in acceptability with respect to scrambling out of these complex NPs, which thus indicates that sentential subjects do not give rise to any extra island effect.

245

‘Lit. Which book1 is Mary calling the fact that John bought t1 into question?’ b. Dono

hon-o1

Mary-ga

[[John-ga t1 katta

which

book-Acc Mary-Nom John-Nom bought

to]

omotteru no?

Comp

think

koto]-ga mondai

da

fact-Nom problem be

Q

‘Lit. Which book1 does Mary think that the fact that John bought t1 is a problem?’

(Saito 1985:272)

Thus, from the present point of view, Kuno (1973) is right in claiming that Japanese relativization is insensitive to sentential subject islands, but he is not quite correct in claiming that it is also insensitive to complex NP and adjunct clause islands. Secondly, the present analysis of Japanese relativization strongly argues against the type of analysis of this construction that involves rightward movement of a relative operator. If the head raising analysis of relativization proposed by Vergnaud (1974) is applied to head-final relative clauses such as in Japanese, it will be forced to postulate rightward movement of a relative head, as schematically shown below: (55) [ … t1 …] Head1

This, however, leaves the left edge effect observed above unaccounted for; no instance of Japanese relativization would involve string-vacuous movement of a relative operator under the head raising analysis. Further, the left edge effect in question will not be properly captured by the kind of analysis proposed by Kayne (1994) according to which head-final relative clauses are 246

derived from basically head-initial underlying structures by applying more than one instance of leftward movement. Thus, under this analysis, a simple Japanese relative clause such as in (56) is derived in the way shown in (57): (56) [t1 John-o

nagutta] otoko1

John-Acc hit

man

‘the man who hit John’ (57) a.

[DP [NP [CP [TP otoko John-o nagutta]]]] ↓

(movement of the relative head otoko ‘man’ to Spec-CP)

b. [DP [NP [CP otoko1 [TP t1 John-o nagutta]]]] ↓ c.

(remnant movement of the embedded TP to Spec-DP)

[DP [TP t1 John-o nagutta] [NP [CP otoko1 tTP]]]

Note that in this derivation, the movement of otoko ‘man’ involves string-vacuous leftward movement and yet the next step of movement “cancels” the string-vacuity of this movement, since overt elements end up intervening in the resulting chain of (otoko1, t1). Abe and Hornstein (2012) argue that the ban on overt string-vacuous movement, characterized in (30), reproduced below, should apply to the output structure of a derivation rather than to each step of movement involved in the derivation. (58) The head of a chain created by Move cannot be pronounced unless it has an effect on PF output. Abe and Hornstein base this argument upon relevant facts about what Nakao (2009) calls left node raising (henceforth, LNR) in Japanese, which shows island insensitivity 247

in a configuration symmetrical to that of right node raising in English which also gives rise to island insensitivity. First, the following Japanese example illustrates a typical instance of LNR: (59) Zibun-no syasin-o self-Gen picture-Acc

[Taroo-wa Taroo-Top

Mary-ni

okuri],

Mary-Dat

send

[Ziroo-wa Ziroo-Top

Susan-ni ___ ageta]. Susan-Dat

gave

‘Self’s picture [Taroo sent

to Mary] and [Ziroo gave

to Susan].’

Here zibun-no syasin ‘self’s picture’ can have a so-called distributive reading, taking different values in the two conjuncts; that is, it can be interpreted as Taroo’s picture in the first conjunct and as Ziroo’s picture in the second. This is one of the typical properties of such a construction as RNR that involves an ATB dependency, as witnessed by the following English RNR: (60) Every man1 loves

, but no man2 wants to marry

, his(1, 2) mother. (Jacobson 1999:167)

Thus, it is not unreasonable to assume that such an LNR as given in (59) involves ATB movement, just like English RNR, at least when it induces a distributive reading. This is further confirmed by the fact that such an LNR shows insensitivity to locality, just like RNR, when a phrase that undergoes ATB movement is located at the left edge of each conjunct. Compare the following Japanese LNR examples:15 15

Here the intended reading of zibun-no syasin-o ‘self’s picture-Acc’ is its distributive

one. All the LNR examples follow this rule; thus, when a given sentence involves

248

(61) *Taroo-wa Taroo-Top

zibun-no syasin-o [vP self-Gen picture-Acc

omotta]] riyuu-o

suteyoo

Mary-Nom

tazune], [vP [Bill2-ga [pro2

thought reason-Acc asked riyuu-o

[Mary1-ga [pro1

Bill-Nom

to]

throw away Comp totte-okoo to] keep

omotta]]

Comp thought

tazuneta].

reason-Acc asked ‘Lit. Taroo, self’s pictures [asked the reason why Mary thought she should throw ___ away] and [asked the reason why Bill thought he should keep (62) ?Taroo-wa Taroo-Top

zibun-no syasin-o [vP

[[pro1

self-Gen picture-Acc

omotta]] riyuu-o

to]

Mary1-ga

throw away Comp Mary-Nom

tazune], [vP [pro2

thought reason-Acc asked riyuu-o

suteyoo

].’

totte-okoo to] keep

[Bill2-ga

omotta]]

Comp Bill-Nom thought

tazuneta].

reason-Acc asked (61) is a case in which the original position of the shared phrase zibun-no syasin-o ‘self’s picture-Acc’ is not located at the left edge in each conjunct. In (62), on the other hand, the embedded clause of omotta ‘thought’ in each conjunct is preposed across its subject, so that the original position of the shared phrase is shifted at the left edge of each vP conjunct. The contrast in acceptability between (61) and (62) immediately follows if we adopt the same ATB movement approach for Japanese LNR as we did for zibun-no

‘self’s

’ as the shared phrase of LNR, the intended reading is always

its distributive one.

249

English RNR in Section 3.1.2. In (61), the shared phrase zibun-no syasin-o undergoes non-string-vacuous movement, hence giving rise to a violation of the relative clause island. In (62), on the other hand, the shared phrase undergoes string-vacuous movement with respect to the first conjunct, and hence is pronounced in its original position within that conjunct. Hence, this ATB movement is immune to the relative clause island of each conjunct within which the shared phrase is originally located. Now compare (63), an LNR example, with its clefted counterpart (64): (63) ?pro zibun-no syasin-o [vP

[[pro1

suteyoo

self-Gen picture-Acc omotta]] riyuu-o

Mary1-ga

throw away Comp

tazune], [vP [pro2

totte-okoo to]

thought reason-Acc asked riyuu-o

to]

keep

Mary-Nom [Bill2-ga

omotta]]

Comp Bill-Nom thought

tazuneta].

reason-Acc asked ‘Lit. pro, self’s pictures [asked the reason why Mary thought she should throw ___ away] and [asked the reason why Bill thought he should keep (64) *[pro [vP [[pro1

tazune], [vP [pro2

suteyoo

to]

Mary1-ga

omotta]] riyuu-o

throw away Comp

Mary-Nom

thought reason-Acc

totte-okoo to]

[Bill2-ga omotta]]

asked

keep

tazuneta]] no-wa

zibun-no syasin-o

da.

asked

self-Gen picture-Acc

be

NL-Top

].’

Comp Bill-Nom thought

riyuu-o reason-Acc

‘It was self’s pictures that pro [asked the reason why Mary thought she should 250

throw

away] and [asked the reason why Bill thought he should keep

].’

(63) has the same configuration as (62) in the relevant respects, and hence its acceptability is explained in exactly the same way as that of (62). (64) is derived from (63) by clefting the shared phrase zibun-no syasin-o ‘self’s picture-Acc’. Hence, this further movement seems to render the derivation ungrammatical. The reason for this is, Abe and Hornstein claim, that the resulting chain produced by clefting this shared phrase has an effect on PF output in the sense that overt elements intervene in the top and bottom copies of this chain. To illustrate this point, Abe and Hornstein adopt an analysis of the Japanese cleft construction of the sort a la Kayne, such as Hasegawa’s (1997), according to which it is derived by first moving a focused phrase out of a given clause and then moving the remnant clause to the topic phrase. The clause in the topic phrase functions as the presupposition. (65) illustrates the derivation of (64). (65) a.

[e]-Topic [TP pro [[pro1 self’s picture-Acc throw away] Mary1-Nom thought] reason-Acc asked and [[pro2 self’s picture-Acc keep] Bill2-Nom thought] reason-Acc asked] NL copular ↓ (ATB movement of self’s picture)

b. [e]-Topic [TP self’s picture-Acc [TP pro [[pro1 t throw away] Mary1-Nom

thought] reason-Acc asked and [[pro2 t keep] Bill2-Nom thought] reason-Acc

asked]] NL copular ↓ (remnant movement of the lower TP to the empty topic phrase) 251

c.

[TP pro [[pro1 t throw away] Mary1-Nom thought] reason-Acc asked and [[pro2 t keep] Bill2-Nom thought] reason-Acc asked]] NL-Topic [self’s picture-Acc] copular

The ATB leftward movement of self’s picture at stage (65b) involves string-vacuous movement in the first conjunct, but a further application of remnant movement from this stage results in the configuration in which the chain of self’s picture created by the ATB movement has a PF effect. Hence, the higher copy in what amounts to the focused position in the cleft construction is pronounced. Since the ATB movement in question ends up affecting PF output (i.e., it is “overt”), it causes a violation of the relative clause island condition. Given this argument, it will be untenable to adopt an analysis of Japanese relativization of the sort Kayne (1994) proposes, as shown in the derivation (57) for example (56); (57) is reproduced below for ease of reference: (66) a.

[DP [NP [CP [TP otoko John-o nagutta]]]] ↓

b. [DP

[NP [CP otoko1 [TP t1 John-o nagutta]]]] ↓

c.

(movement of the relative head otoko ‘man’ to Spec-CP)

(remnant movement of the embedded TP to Spec-DP)

[DP [TP t1 John-o nagutta] [NP [CP otoko1 tTP]]]

Although the movement of otoko ‘man’ involves string-vacuous leftward movement at stage (66b), the next step of remnant movement makes the chain (otoko1, t1) have a PF effect since overt elements end up intervening in this chain. Hence, it would be incorrectly predicted that no cancelation of island violations would take place in the 252

derivation of Japanese relativization due to the remnant movement of TP. Thus, the left edge effect observed above lends strong support to the kind of analysis proposed in this section in which Japanese relativization involves leftward movement of a relative operator. The above discussion leads to the prediction that Japanese clefts do not show the left edge effect observed with Japanese relativization. This is in fact borne out; consider the following examples:16 (67) a.

[[pro e1

tyakuyoositeiru]

basyo-ga

husawasiku-nai]

zubon1

is-wearing

place-Nom

appropriate-not

pants

‘Lit. the pants1 that [the place [where pro is wearing e1] is not appropriate]’ b.?*[[pro e1 tyakuyoositeiru] basyo-ga husawasiku-nai no]-wa sono zubon-o is-wearing

place-Nom appropriate-not NL-Top that pants-Acc

da. be ‘Lit. It is those pants1 that the place [where pro is wearing e1] is not appropriate.’ (68) a.

[[pro e1

nusunda toiu] uwasa-ga stole

16

that

hiromari-kanenai] e1

rumor-Nom spread-likely

picture

Unfortunately, most examples of Japanese relativization that show the left edge effect

involve relativization of subjects, and as is well known, subjects marked with nominative case cannot appear in the focused position of Japanese clefts without inducing quite uncomfortable status of acceptability. That is why I have provided relevant examples that involve object relativization with pro subject in (67) and (68).

253

‘Lit. the picture1 that [the rumor that [pro stole e1] is likely to spread]’ b.?*[[pro e1 nusunda toiu] uwasa-ga stole

hiromari-kanenai no]-wa sono e-o

that rumor-Nom spread-likely

NL-Top that picture-Acc

da. be ‘Lit. It is that picture1 that the rumor that [pro stole e1] is likely to spread.’ (67a) involves relativization of an object out of a relative clause and (68a) relativization of an object out of a pure complex NP. Since the subjects sitting next to the object gaps are pro, the scrambling of WH involved in these constructions is string-vacuous, hence immune to the complex NP islands. (67b) and (68b), the cleft counterparts of these constructions, on the other hand, do exhibit island effects, as predicted, since the resulting chains of the phrases in the focused position have effects on PF output.

4.3

Why Movement in Japanese Relativization?

So far we have been pretending that movement is the only option available for establishing an anaphoric relation between a relative head and its associated gap. But nothing would prevent such an anaphoric relation from being established through base-generation of pro, as schematically shown below: (69) [ … pro1 …] Head1 Notice that if we admitted this option freely, then we would not expect any island effect to emerge in Japanese relativization. Hence, we need to prevent it somehow. Following Ishii’s (1991) claim that the resumptive pronoun strategy is somehow 254

marginal in Japanese relativization, I suggest that the marginality is attributed to the fact that the anaphoric relation between pro and its antecedent would always be backward in Japanese relativization. In fact, replacing a gap associated with a relative head by an overt pronoun does not yield a fully acceptable case of relativization, as shown below: (70) a.??[[Bill-ga

kanozyo1-o korosita toiu] zihaku-o

Bill-Nom she-Acc

killed

keisatu-ga mada

that confession-Acc police-Nom yet

urazukesitei-nai] onna1 substantiate-not woman ‘Lit. the woman1 who [the police have not substantiated yet [Bill’s confession that he killed her1]]’ b.??[[sore1-o syuppansita] kaisya-ga it-Acc

published

kazi-de

company-Nom fire-by

yakete-simatta]

hon1

was-burned-down book

‘the book1 which [the company [that published it1] was burned down by a fire]’ Compare these examples with their topicalization counterparts: (71) a.

Sono onna1-wa that

[Bill-ga pro1/kanozyo1-o korosita toiu] zihaku-o

woman-Top Bill-Nom

she-Acc

keisatu-ga

mada urazukesitei-nai].

police-Nom

yet

killed

that confession-Acc

substantiate-not

‘That woman1, the police have not substantiated yet Bill’s confession that he killed her1]]’ b.

Sono

hon1-wa [pro1/sore1-o

syuppansita] kaisya-ga

that

book-Top

published

it-Acc

255

kazi-de

company-Nom fire-by

yakete-simatta]. was-burned-down ‘That book1, the company that published it1 was burned down by a fire.’ These examples involve forward anaphora in which the pronouns may be overt or covert, and sharply contrast with those in (70) in their acceptability. I thus assume that base-generation of pro is not exploited without causing marginality across-the-board in Japanese relativization and hence that this option does not cover up the island effects caused by movement crossing islands in this construction, except for those cases in which the movement involved has no PF effect. Oka (1988) tries to find a case of Japanese relativization where base-generation of pro is excluded for an independent reason. He exploits Chao and Sells’s (1983) observation that when a relative head is a quantified NP in English, this disallows the resumptive pronoun strategy, as shown below: (72) a.

I’d like to meet the linguist1 [that Mary couldn’t remember if she had seen him1 before].

b. *I’d like to meet every linguist1 [that Mary couldn’t remember if she had seen him1 before].

(Chao and Sells 1983:49)

Oka then reasons that under the assumption that this prohibition also carries over to Japanese relativization, those examples of this construction that involve quantified NPs as relative heads should instantiate genuine cases where movement is involved. This leads to the prediction that such a case gives rise to island effects in the most lucid way. Oka claims that this prediction is in fact borne out, providing the following examples: 256

(73) a.

[Bill-ga [John-ga e1 nagutta to] Bill-Nom John-Nom hit

omotteiru]

Comp think

dono onna-mo1 every woman

‘every woman1 who Bill thinks that John hit e1’ b. *[Bill-ga

[John-ga e1 nagutta

Bill-Nom John-Nom

hit

node]

okotteiru] dono

because is-angry

onna-mo1

every

woman

‘every woman1 who Bill is angry because John hit e1’ c. *[John-ga [e2 e1 nagutta] otoko2-o ketobasita] John-Nom

hit

man-Acc kicked

dono

onna-mo1

every

woman

‘every woman1 who John kicked the man who hit e1’

(Oka 1988:204)

Interestingly, Oka (1988) provides an example of Japanese relativization that involves a wh-island, marking it with ? for its acceptability: (74) ?[Bill-ga

[John-ga e1 nagutta

Bill-Nom John-Nom

hit

kadooka] sitteiru] dono

onna-mo1

whether know

woman

every

‘every woman1 who Bill knows whether John hit e1’ The fair acceptability of this example suggests that movement of a relative operator is in fact a species of scrambling, given that the latter operation also shows a very weak wh-island effect, as seen in Section 3.1.1; compare (74) with the following scrambling sentence: (75) ?Dono every

onna-o-mo1 Bill-ga

[John-ga t1 nagutta

woman-Acc Bill-Nom John-Nom hit

kadooka] sitteiru. whether know

‘Every woman1, Bill knows whether John hit t1.’ Granting Oka’s argument, we should examine whether the left edge effect 257

discussed above still holds even if a relative head is changed into a quantificational NP. Basically, no significant change in acceptability is observed, as illustrated below: (76) a.

[[e1 Kawaigatteita] inu-ga was-fond-of

sinde-simatta]

dog-Nom dying-ended-up

dono

kodomo-mo1

every

child

gakkarisiteita. was-disappointed ‘Lit. Every child1 who the dog e1 was fond of died was disappointed.’ b. [[e1 Kiteiru]

yoohuku-ga

is-wearing suit-Nom

yogoreteiru] dono sinsi-mo1 nihonzin da. is-dirty

every gentleman Japanese be

‘Lit. Every gentleman1 who the suit that e1 is wearing is dirty is Japanese.’ (77) a. *[Bill-ga [e1 kaita] hon-ni

zyobun-o

noseta]

Bill-Nom wrote book-Dat forward-Acc put gengogaku-ga

dono

gakusya-mo1

every

scholar

senmon da.

linguistics-Nom specialty be ‘Lit. Every scholar1 who Bill put a forward to the book that e1 wrote makes a specialty of linguistics.’ b. [[e1 Kaita] wrote

hon-ni

Bill-ga

zyobun-o

noseta]

book-Dat Bill-Nom forward-Acc put

gakusya-mo1 gengogaku-ga

senmon da.

scholar

specialty be

linguistics-Nom

(78) a. *[Bill-ga [e1 musume-o

nagutta

Bill-Nom daughter-Acc hit

node]

every

okotteiru] dono

because is-angry 258

dono

every

sensei-mo1 teacher

hanseisiteiru. is-repentant ‘Lit. Every teacher1 who Bill is angry because e1 hit his daughter is repentant.’ b. [[e1 Musume-o

nagutta

daughter-Acc hit sensei-mo1

hanseisiteiru.

teacher

is-repentant

node]

Bill-ga

okotteiru] dono

because Bill-Nom is-angry

every

(76) shows that the left edge effect observed in the case of relativization of a phrase out of a relative clause still holds even if the associated relative head is a quantificational NP. (77) and (78) show that the improvement in acceptability of those cases where a gap that marks the original position of a relative operator is shifted to the left edge thanks to scrambling of the containing phrase does not change even with a quantificational relative head. All of these are exactly what is predicted under the present theory, and to the extent that Oka (1988) is right in assuming that the resumptive pronoun strategy is excluded in the case where the relative head is a quantificational NP in Japanese relativization, they instantiate the left edge effect in the most transparent way, thereby giving strong support to the present theory of movement based upon Abe and Hornstein’s (2012). Oka (1988) provides another case of Japanese relativization that excludes the option of base-generation of pro, one where an adjunct is relativized, as illustrated below:

259

(79) [John-ga e1

kita]

mura1

John-Nom

came

village

‘the village from which John came’ On the assumption that pro must be an NP or, to put it in Oka’s terms, that pro must be one that receives Case, Oka reasons that in such an example as (79), the gap that functions as the source role of kita ‘came’ cannot be pro, and hence that such an example instantiates a genuine case of relativization that involves movement of a relative operator. He then provides an example that shows that such a case in fact exhibits an island effect: (80) a.

[Bill-ga

[John-ga e1

Bill-Nom John-Nom

kita

to]

omotteiru] mura1

came

Comp

think

village

‘the village1 from which Bill thinks that John came e1’ b. *[Bill-ga [e1 kita] Bill-Nom came

otoko-o sagasiteiru]

mura1

man-Acc is-looking-for village

‘the village1 from which Bill is looking for a man who came e1’ (Oka 1988:209-210) (80a) demonstrates that relativization of an adjunct can be unbounded in principle, and (80b) shows that it is subject to the relative clause island. Relevant to our present concern is whether this type of relativization also shows the left edge effect in question. The fact is that it does not, as shown below: (81) a. *[Bill-ga [pro e1 kita] Bill-Nom

came

riyuu-o

kakusiteiru]

mura1

reason-Acc

is-hiding

village

260

‘the village1 from which Bill is hiding the reason why he came e1’ b. *[[pro e1 kita] came c.

riyuu-o

reason-Acc Bill-Nom

[[pro e1 sukina] riyuu-o like

Bill-ga

Bill-ga

reason-Acc Bill-Nom

kakusiteiru]

mura1

is-hiding

village

kakusiteiru]

mura1

is-hiding

village

‘the village1 which Bill is hiding the reason why he likes e1’ (82) a. *[Bill-ga [pro e1 wazawaza Bill-Nom

all-the-way

kita

noni]

kakusiteita]

mura1

came

though

was-hiding

village

‘the village1 from which Bill was hiding (this fact) although he had come all the way e1’ b. *[[pro e1 wazawaza

kita

all-the-way came c.

noni]

Bill-ga

though

Bill-Nom was-hiding

village

Bill-ga

mura1

[[pro e1 wazawaza otozureta noni] all-the-way visited

though

kakusiteita]

kakusiteita]

Bill-Nom was-hiding

mura1

village

‘the village1 which Bill was hiding (this fact) although he had taken the trouble to visit e1’ (81a) and (82a) show cases that violate the relative clause and adjunct clause island conditions, respectively. That the (b)-examples are as bad as the (a)-examples shows that scrambling the phrases that contain the gaps of relative operator movement to the left edge does not help to improve the acceptability of these examples. This strongly indicates that the left edge effect in question is somehow inoperative in those cases in which adjuncts are relativized. The (c)-examples confirm this, since they minimally 261

differ from the (b)-examples in that the phrases that are relativized are arguments. I have no real answer for the reason why adjuncts behave differently from arguments with respect to the left edge effect (but see the discussion below). However, it is interesting to note here that the same sort of argument-adjunct asymmetry is also observed with English RNR, a case that Abe and Hornstein (2012) base their movement theory upon. They provide the following examples: (83) a.

John got fired

and Bill had his salary reduced

, because he talked

back/because of office politics. b. *Mary helped the person who got fired had his salary reduced

and comforted the person who

, because he talked back/because of office

politics.

(Abe and Hornstein 2012:201)

(83a) shows that adjunct phrases/clauses such as because

can serve as shared

phrases for RNR. The unacceptability of (83b) indicates that unlike shared arguments, shared adjuncts cannot evade island effects even though they undergo ATB movement from the right edge of each conjunct. This means under Abe and Hornstein’s theory of movement that string-vacuous ATB movement of an adjunct, though it ends up producing a chain in which the bottom copy is pronounced, is sensitive to island conditions. Note that whatever principle derives this fact will also take care of the fact that relativizing an adjunct shows island sensitivity no matter whether it involves string-vacuous movement or not in Japanese relativization. 17 Thus, it is not 17

The argument-adjunct asymmetry discussed here with respect to island sensitivity is

reminiscent of Huang’s (1982) claim that unlike arguments, which are sensitive to

262

unreasonable to conclude that the argument-adjunct asymmetry observed in Japanese relativization does not really cause a fatal problem; on the contrary, it might indicate that the present movement theory of Japanese relativization is on the right track. To sum up this chapter, I have argued that Japanese relativization involves scrambling of a relative operator on the basis of the left edge effect on island sensitivity, according to which the association of a relative head with the corresponding gap is free from island effects when the gap is located at the left edge of the relative clause. Assuming Abe and Hornstein’s (2012) theory of movement, I have shown how the left edge effect is derived from this theory with one crucial assumption: the movement involved in Japanese relativization is leftward. With this assumption, those cases that exhibit the left edge effect involve string-vacuous leftward movement, hence immune to island conditions according to Abe and Hornstein’s theory of movement. The present work has a significant implication on the directionality of movement. Given that no direct evidence for leftward movement in Japanese relativization seems available to children learning Japanese, it must be the case that UG principles are involved in determining the directionality of movement in Japanese relativization. Thus this will lend support to the so-called asymmetrical structure hypothesis, first proposed by Kayne (1994), according to which only leftward movement is allowed in UG or the

islands only in the case of overt movement, adjuncts show island sensitivity no matter whether they undergo overt or covert movement. It is thus expected that the data Huang deals with are also accommodated under the theory that will property deal with the argument-adjunct asymmetry discussed in the text.

263

hypothesis proposed by Saito (1995) and Fukui (1993), among others, that the direction of movement is constrained by the head-parameter: a movement operation should create a structure that is consistent with the value of the head parameter in a given language, which thus allows Japanese to have leftward movement only.18 Finally, let me briefly discuss the question why “covert” movement is not subject to island effects and why such island insensitivity is limited to movement of arguments; we might call this set of questions Huang’s problem (see fn. 17). This is also related to what is called island repair by ellipsis, which is instantiated by what Chung et al. (1995) call the merger type of sluicing. The following examples show that this construction exhibits island insensitivity: (84) a.

They want to hire someone who speaks a Balkan language, but I don’t remember which language.

(Merchant 2001:87)

b. Ben will be mad if Abby talks to one of the teachers, but she couldn’t remember which.

(ibid.:88)

Abe and Hornstein (2012) argue that their theory of movement that incorporates the ban on “no effect on PF output” given in (30) can capture the island insensitivity of this construction as well. They claim that the sluice of (85), for instance, has the derivation given in (86). (85) She’s reading something, but I can’t imagine what.

18

We are led to the same conclusion when we consider Japanese right dislocation in the

next chapter.

264

(86) a.

I can’t imagine [CP C[+wh] [TP she’s reading ]

b. I can’t imagine [CP C[+wh] [TP she’s reading what] (86a) represents the structure in which what constitutes a two-membered chain. That the two members of what are both put with angled brackets indicates that no decision is made yet with respect to which member is pronounced. The surface form indicated in (86b) is derived from (86a) by deleting the embedded TP except the wh-phrase what. Since the two members of what are adjacent to each other thanks to deletion, the bottom member must be pronounced according to the PF condition (30). This, then, explains the island insensitivity of sluicing, as shown in (84), since the island conditions are operative only to “overt” movement for arguments. Interestingly, unlike wh-arguments, wh-adjuncts do exhibit island sensitivity when they function as remnant wh-phrases in sluicing, as noted by Merchant (2001): (87) a.

She’s practicing her serve so that she’ll be able to hit the ball in a certain deadly way, but her trainer won’t tell us in what way/??how.

b. He wants to interview someone who works at the soup kitchen for a certain reason, but he won’t reveal yet ?what reason/*why.

(Merchant 2001:129)

These facts follow under the assumption that adjuncts show island sensitivity even when they undergo “covert” movement. Now a question remains as to how Huang’s problem is explained. Here I can only

265

offer a speculation. 19 Abe and Hornstein (2012) characterize a mechanism for instantiating the island insensitivity of “covert” movement as follows: if a given movement step violates a locality condition (e.g., the RRC or island condition), mark the moved phrase with ‘*’. At PF evaluate the chain as in (88): (88) a.

‘*’ is removed when the bottom copy of the chain of its carrier is pronounced.

b. If the chain retains ‘*’, it leads to ungrammaticality. The island repair by (88a) may be understood in terms of Ross’s (1967) characterization of island effects: such effects arise only with chopping rules, which amounts to saying that only “real” gaps, i.e., unpronounced gaps, are subject to island conditions. This will take care of the first part of Huang’s problem. As for the second part, namely, the question why adjuncts do not show such an island repair, suppose that such a repair strategy as given in (88) also holds at the LF side. Following the idea of Rizzi (1990), we might characterize such a strategy as the following: (89) a.

‘*’ is removed when the bottom copy of the chain of its carrier is “referential.”

b. If the chain retains ‘*’, it leads to ungrammaticality. “Referential” phrases basically mean those phrases that function as arguments and can refer to their antecedents; we might identify such phrases as pro. Then, (89a) dictates that when a “gap” of a chain is “referential,” the island violation goes away. Since adjuncts are not referential, the island violation is retained. This takes care of the fact that once adjuncts induce an island violation, it never goes away. 19

I owe the basic idea of the speculation to Norbert Hornstein (personal

communication). See also Abe (2015) for relevant discussion.

266

5

Scrambling for Right Dislocation: Bi-Clausal Analysis

In this chapter, I discuss so-called right dislocation (henceforth, RD) in Japanese, which appears to involve rightward movement, as illustrated below: (1) a.

John-wa

Mary-o

hihansita yo.

John-Top

Mary-Acc

criticized

‘John criticized Mary.’ b.

John-wa e1 hihansita yo, Mary-o1.

Haraguchi (1973) claims that (1b) is an instance of Japanese RD in which Mary-o is moved rightward from the underlying order given in (1a). This proposal, however, is not compatible with what has been claimed regarding how phrase structure is constrained. Saito (1985), Fukui (1993) and Saito, and Fukui (1998) argue that the direction of adjunction is constrained by X’-theory: to put it in Fukui’s (1993) terms, the value of the head parameter should be preserved in derived structures in such a way that an adjunction operation should create a structure that is consistent with the value of the head parameter in a given language.1 According to this constraint, a phrase must be adjoined to a category on the side opposite to that of its head. Thus, a head-final language such as Japanese is allowed to have leftward adjunction and is never allowed

1

See Abe and Hoshi (1997, 1999) for arguments from English and Japanese gapping to

support Saito’s (1985) constraints on the direction of adjunction sites.

267

to have rightward adjunction. This nicely captures the fact that Japanese is a strictly head-final language, with no phrase put after a head, but it leaves the RD construction as an exception if Haraguchi’s (1973) claim is correct. Alternatively, Kayne (1994), Takano (1996, 1998), and Fukui and Takano (1998) propose that hierarchical structure unambiguously determines the order of terminal symbols, thereby denying the existence of symmetrical structures. According to Kayne, hierarchy and precedence correspond roughly in such a way that what precedes is structurally higher than what follows. He further argues that the universal basic structure is S(pecifier)-H(ead)-C(omplement), as given below, and that various word orders exhibited by different languages are derived from this basic word order by applying movement operations to heads and phrases.2 (2)

XP ZP

X’ X

YP

According to this asymmetrical structure theory, rightward movement is universally prohibited, since if it were allowed, it would create a structure in which what follows is structurally higher than what precedes. Hence, the RD construction appears to raise a problem to such a theory.3 2

Though Takano (1996, 1998) and Fukui and Takano (1998) also argue for this

asymmetrical structure theory, they claim, contrary to Kayne, that the universal basic word order is S-C-H. 3

In fact, by examining a construction in Turkish similar to Japanese RD, Kural (1997)

argues that this language is underlyingly head-final and has the option of rightward movement. I hope that this Turkish construction can be reanalyzed in terms of an

268

In this chapter, I argue for the restricted adjunction theory or the asymmetrical structure theory by offering an alternative to Haraguchi’s (1973) analysis of the Japanese RD construction. I argue, following Tanaka (2001), that this construction involves a bi-clausal structure that involves leftward scrambling of a postverbal phrase and ellipsis in the second clause.

5.1

Bi-Clausal Analysis of Japanese Right Dislocation

The Japanese RD construction, as illustrated in (1b), is strictly a root phenomenon (cf. Haraguchi (1973) and Kuno (1978a)):4 (3) *John-ga

Mary-ni watasita,

John-Nom Mary-Dat handed

sono hon-o

koto

that book-Acc fact

‘the fact that John handed that book to Mary’ Here sono-hon-o ‘that book-Acc’ is shifted after the verb watasita ‘handed’ in the appositive clause of koto ‘fact’, which gives rise to total unacceptability. Further, as originally observed by Haraguchi (1973), overt resumptive pronouns are possible with the Japanese RD construction: (4) John-wa kanozyo1-o John-Top her-Acc

nagutta yo,

Mary1-o.

hit

Mary-Acc

‘John hit her, that is, Mary.’

alternative of the sort proposed in this chapter. 4

In what follows, phrases put postverbally will be italicized in English translations

unless they are clearly indicated in translations themselves.

269

Thus, this construction looks like English RD, which Emonds (1976) describes as derived by a rule that “remove NP’s from their ordinary position in sentences, set them off by comma, and replace them with pronouns,” (p. 32) as illustrated below: (5) a. b.

It really bothers me, John’s big cigar. She won’t do anything rash, John’s sister.

(Emonds 1976:33)

This construction is also a root phenomenon, as shown below: (6) a. *I didn’t say that it bothered me, riding in the back seat, on the trip out. b. *I predicted that her attempt to do something daring, John’s sister, would end in disaster.

(ibid.:34)

Thus, one might consider it most plausible to analyze the Japanese RD construction on a par with the English RD, so that (1b) has the following structure: (7) John-wa pro1 hihansita yo, Mary-o1. However, it has been observed (see Simon (1989), Rosen (1996), Tanaka (2001), among others) that this construction exhibits island sensitivity, a typical property of movement, as shown below: (8) Watasi-wa [John-ga e1 I-Top

John-Nom

sono hon-o

kureta

that book-Acc gave

to]

itta yo,

Comp

said

watasi-no musume-ni1. I-Gen

daughter-Dat

‘I said that John gave that book to my daughter.’ (9) ?*Watasi-wa [John-ga e1 kureta] I-Top

John-Nom gave

hon-o

yonde-simatta yo, watasi-no

book-Acc have-read 270

I-Gen

musume-ni1. daughter-Dat ‘I have read the book John gave to my daughter.’ (10) ?*Watasi-wa [John-ga e1 tabeta I-Top

John-Nom

ate

node]

kare-no hahaoya-o

because he-Gen

mother-Acc

sikarituketa yo, keeki-o1. scolded

cake -Acc

‘Because John ate cake, I scolded his mother.’ The acceptability of (8) shows that the Japanese RD construction is not clause-bound. (9) shows that this construction is sensitive to a relative clause island, and (10) shows that it is sensitive to an adjunct clause island. In order to account for the island sensitivity of this construction, one might propose, following Haraguchi (1973), that the Japanese RD construction is derived from a head-final underlying structure by applying rightward movement to a constituent so as to put it postverbally, as schematically shown below: (11) a.

SOV

b. [S t V] O

(rightward movement of O)

This analysis, however, immediately faces the problem of how to deal with the fact that resumptive pronouns can appear in the gaps of this construction, as shown in (4), since this possibility suggests that pro can also appear in the gap position of this construction, and hence should show no island sensitivity. To put it in other words, the problem comes down to how to explain the apparently contradictory properties of the Japanese

271

RD construction: while the island sensitivity of this construction suggests a movement approach, the availability of resumptive pronouns suggests a base-generation approach. Further, as Tanaka (2001) points out, the fact that the Japanese RD construction does not exhibit clause-boundedness, as shown in (8), indicates that it should be leftward movement rather than rightward movement that is involved in this construction, if Ross’s (1976) RRC applies universally.5 To account for these properties, Tanaka (2001) proposes the bi-clausal analysis of the Japanese RD construction according to which (1b) is derived as follows:6 (12) a.

[TP John-wa pro1 hihansita yo], [TP Mary-o1 [TP John-wa t1 hihansita yo]] ↓

b.

(leftward scrambling of Mary-o)

[TP John-wa pro1 hihansita yo], [TP Mary1-o [TP John-wa t1 hihansita yo]] (deletion)

In this derivation, Mary-o is scrambled and adjoined to the top of the second clause and then the rest of this clause is deleted under identity with the first clause. Since scrambling is involved in this derivation, this bi-clausal analysis correctly captures the fact that this RD constructions shows island sensitivity, as illustrated in (9) and (10). Further, since the gap in the first clause that is associated with the scrambled phrase is identified as pro, this analysis also captures the fact that an overt resumptive pronoun

5

See Abe and Hoshi (1997, 1999) for the claim that English gapping involves

rightward movement, thus obeying the RRC, whereas Japanese gapping involves leftward movement, hence free from this constraint. 6

See also Kuno (1978b) for a bi-clausal analysis of the Japanese RD construction that

Tanaka’s (2001) analysis is based upon.

272

can appear in the gap position in question in this RD construction. This bi-clausal analysis can be regarded as a plausible instantiation in structural terms of Kuno’s (1978a) claim that the Japanese RD construction involves “a process that adds afterthoughts to the end of a sentence.” (p. 61-62) In this sense, it will be misleading to regard the relation of the two clauses in (12) as that of conjunction, since these two clauses basically repeat the same proposition. Thus, it is more appropriate to regard them as involving clause-repetition. Further, It is reasonable to claim that this clause-repetition property of this construction that holds for expressing afterthoughts makes it a root phenomenon.7 Notice that this bi-clausal analysis of Japanese RD is compatible with both the restricted adjunction theory and the asymmetrical structure theory discussed at the beginning of this chapter, according to which only leftward movement is permissible in a head-final language such as Japanese. Thus, under this bi-clausal analysis, Japanese RD need not be treated as an exception. It is predicted under this analysis that island effects hold even if resumptive pronouns are inserted into the gaps of this construction. Cecchetto (1999) observes that this is in fact the case: (13) Watasi-wa [John-ga I-Top

John-Nom

(kanozyo1-ni) sono hon-o her-Dat

watasi-no

musume1-ni.

I-Gen

daughter-Dat

that

book-Acc

kureta to]

itta yo,

gave Comp

said

‘I said that John gave that book to my daughter.’

7

See Tanaka (2001) for a more detailed discussion along this line of thought. 273

(14) ?*Watasi-wa [John-ga

(kanozyo1-ni) kureta] hon-o

I-Top

John-Nom her-Dat

watasi-no

musume1-ni.

I-Gen

daughter-Dat

gave

yonde-simatta yo,

book-Acc have-read

‘I have read the book John gave to my daughter.’ (15) ?*Watasi-wa [John-ga (kanozyo1-o) nagutta I-Top

John-Nom her-Acc

hit

node]

kare-no hahaoya-o

because he -Gen mother-Acc

sikarituketa yo, watasi-no musume1-o. scolded

I-Gen

daughter-Acc

‘Because John hit my daughter, I scolded his mother.' It is usually assumed that in a configuration in which an operator and its gap form a chain by movement across an island, then replacement of the gap by a resumptive pronoun makes the chain free from an island effect. This is quite often illustrated by the resumptive pronoun strategy for English wh-question formation across islands, as shown below: (16) a.

the man who1 they think that if Mary marries *(him1), then everyone will be happy.

b. I wonder who1 they think that if Mary marries *(him1), then everyone will be happy.

(Chomsky 1982:11)

According to the standard account, when the resumptive pronoun him appears as the variable of who in these examples, who is base-generated in Spec-CP position, forming a wh-operator-variable chain without recourse to movement, thereby circumventing an island violation. Given this, it will be a total mystery under a mono-clausal analysis

274

such as the one given in (11) why islands effects do not go away when resumptive pronouns are inserted into the gap positions in the Japanese RD construction. Under the bi-clausal analysis, on the other hand, the examples in (13)-(15) have the following schematic structure: (17) [TP ... kanozyo1-ni/o ...] [TP watasi-no musume-ni/o1 [TP ... t1 ... ]] In this structure, movement takes place in the second clause irrespective of whether a resumptive pronoun is inserted into the gap in the first conjunct. Hence it follows that island effects do not go away even if resumptive pronouns are inserted in the Japanese RD construction. A question immediately arises as to what guarantees that watasi-no musume-ni/o involves movement in the second clause in (17). If it were possible that the resumptive pronoun strategy could be involved in the second clause, then we would lose our account for the island effects observed in (14) and (15). We can answer this question by assuming that it is scrambling that is involved in this construction, as Saito (1985) notes that it does not allow resumptive pronouns, as shown below: (18) a. *Mary-o1

John-ga [pro kanozyo-o1

Mary-Acc John-Nom

she-Acc

nagutta

to]

itta.

hit

Comp

said

‘Mary, John said that he hit her.’ b. ?Mary1-wa John-ga [pro kanozyo1-o nagutta to] Mary-Top John-Nom

she-Acc

hit

itta.

Comp said

(18a) illustrates the fact that scrambling of Mary does not allow the resumptive pronoun kanozyo-o to be inserted into its trace position. It contrasts with (18b), where Mary is

275

attached to by the topic marker -wa rather than the accusative case marker. This shows that unlike scrambling, topicalization of the sort that does not accompany case markers allows resumptive pronouns. Given this fact, it follows that a resumptive pronoun cannot be inserted into the gap position of the scrambled phrase watasi-no musume-ni/o in the second clause of (17) and hence that the Japanese RD construction is sensitive to islands, no matter whether a resumptive pronoun is inserted in the first clause or not. This leads us to an interesting prediction: if the postverbal phrases in (14) and (15) are changed into ones bearing the topic marker -wa, the island effects should go away. This prediction is borne out. First of all, such phrases can appear in postverbal position, as shown below:8 (19) John-ga (kanozyo1-o) nagutta yo, John-Nom she-Acc

hit

Mary1-wa. Mary-Top

‘Mary, John hit her.’ Then compare each pair of the following examples: (20) a.?*[John-ga (kanozyo1-o) nagutta John-Nom her-Acc

hit

node]

kare-no hahaoya-o

because he-Gen

mother -Acc

sikarituketa yo, Mary1-o. scolded

Mary-Acc

‘Because John hit Mary, (I) scolded his mother.’ b. ?[John-ga

(kanozyo1-o) nagutta

John-Nom she-Acc 8

hit

node]

kare-no hahaoya-o

because he-Gen

mother-Acc

See Yamashita (2011) for the claim that postverbal NPs with topic marker -wa can

function as thematic topic.

276

sikarituketa yo, Mary1-wa. scolded

Mary-Top

(21) a.?*[John-ga (sore1-o) kasita] John-ga

it-Acc

Chomsky-no

lent

hito-ga

taihos-arete-simatta yo,

person-Nom

having-been-arrested

hon-o.

Chomsky-Gen book-Acc ‘The person John lent Chomsky’s book to has been arrested.’ b. ?[John-ga John-ga

(sore1-o) kasita] it-Acc

Chomsky-no

lent

hito-ga

taihos-arete-simatta yo,

person-Nom

having-been-arrested

hon-wa.

Chomsky-Gen book-Top When the (b)-examples are compared with the (a)-examples, we see that there is a dramatic improvement in acceptability when the postverbal phrases are changed into ones bearing topic marker -wa. Thus, this confirms our prediction: when a postverbal NP bears the topic marker -wa, it allows the resumptive pronoun strategy in the elided clause, hence free from island effects, irrespective of whether a resumptive pronoun shows up in the first clause. Given that scrambling shows a very weak wh-island effect, as seen in Section 3.1.1, it is also predicted that the Japanese RD construction patterns alike in this respect. Haraguchi (1973), Kuno (1978a) and Tanaka (2001) independently provide examples of the Japanese RD construction that show its insensitivity to wh-islands; consider the following example, adapted from Haraguchi (1973):

277

(22) Kimi-wa [watasi-ga kinoo e nani -o you-Top I-Nom

tabeta ka] sitteimasu ka, asoko-de.

yesterday what-Acc ate

Q know

Q there

‘Do you know what I ate there yesterday?’ This strongly suggests that the movement involved in the Japanese RD construction is scrambling. Further, we have seen in the last chapter that when the movement involved in a given construction is string-vacuous, it is immune to island conditions. This property also carries over to the Japanese RD construction. Consider the following examples: (23) a.

[e1 Kawaigatteita] was-fond-of

inu-ga

sinde

dog-Nom dying

simatta yo,

Mary1-ga.

ended-up

Mary-Nom

‘The dog that Mary was fond of died.’ b.

[e1 Kiteiru]

yoohuku-ga

is-wearing suit-Nom

yogoreteiru yo, John1-ga. is-dirty

John-Nom

‘The suit that John is wearing is dirty.’ (24) a.?*Bill-ga [e1 kaita] hon-o

yakusiteiru yo,

Bill-Nom wrote book-Acc is-translating

Chomsky1-ga. Chomsky-Nom

‘Bill is translating the book that Chomsky wrote.’ b.?*[e1 syuppansita] kaisya-ga published

kazi-de yakete-simatta yo, Chomsky-no

company-Nom fire-by was-burned-down Chomsky-Gen

hon1-o. book-Acc ‘The company [that published Chomsky’s book] was burned down by a fire.’

278

The Japanese RD examples in (23) are fairly acceptable despite the fact that the gaps corresponding to the postverbal NPs are placed within relative clauses. The acceptability of these examples is what the present bi-clausal analysis predicts, since the scrambling involved in deriving these RD examples is string-vacuous, as shown in the following derivation for (23a):9 (25) a.

[TP [pro1 kawaigatteita] inu-ga sinde simatta yo], [TP [Mary-ga kawaigatteita] inu-ga sinde simatta yo] ↓

b.

(leftward scrambling of Mary-ga)

[TP [pro1 kawaigatteita] inu-ga sinde simatta yo], [TP [TP [Mary-ga kawaigatteita] inu-ga sinde simatta yo]] ↓

c.

(deletion of the second clause)

[TP [pro1 kawaigatteita] inu-ga sinde simatta yo], [TP [TP [Mary-ga kawaigatteita] inu-ga sinde simatta yo]]

On the other hand, the degraded acceptability of the examples in (24) is attributed to the fact that the scrambling involved in deriving them is not string-vacuous. Recall that a trace acts as visible for calculating string-vacuity, so that in (24b), the subject trace of the relative head intervenes in the chain produced by scrambling Chomsky-no hon-o

9

According to Abe and Hornstein’s (2012) mechanism of pronunciation, the bottom

copy of Mary-ga must be pronounced due to the string-vacuity of the scrambling involved, as in (25b). As a consequence of this, we need to assume that non-constituent deletion is possible, as in (25c). See Abe (2015) for the in-situ analysis of sluicing, according to which the remnant wh-phrase of this construction has its bottom copy pronounced due to the string-vacuity of the movement involved and deletion applies to TP except this bottom copy of the remnant wh-phrase.

279

‘Chomsky’s book-Acc’ in the second clause, hence making the scrambling non-string-vacuous. Further, compare (24a) with its variant, in which the complex NP object that includes the gap associated with the postverbal NP is shifted to the left edge: (26) [e1 kaita] hon-o

Bill-ga

wrote book-Acc Bill-Nom

yakusiteiru yo,

Chomsky1-ga.

is-translating

Chomsky-Nom

‘Bill is translating the book that Chomsky wrote.’ (26) is much better than (24a) in its acceptability. This immediately follows under the present analysis, since the leftward scrambling of Chomsky-ga in the second clause is now string-vacuous thanks to shifting the whole complex NP object to the left of Bill-ga. The above facts then give strong support to the present bi-clausal analysis of this construction according to which the movement involved in the Japanese RD construction is leftward.

5.2

The Status of the Gap in Japanese Right Dislocation

Under the present bi-clausal analysis of Japanese RD, we have posited pro in the gap position of the first clause on the basis of the fact that a resumptive pronoun can appear in that position. However, we need further consideration of the status of the gap of the first clause. As observed by Haraguchi (1973), when an NP put postverbally is interpreted as indefinite, a resumptive pronoun cannot be used, as shown below: (27) John-ga (*sore-o) katta yo, Chomsky-no John-Nom it-Acc bought

hon-o.

Chomsky-Gen book-Acc

‘John bought a book by Chomsky.’

280

This might indicate that there has to be an indefinite null pronoun in the first clause of such a Japanese RD construction as (27). In fact, as Carlo Cecchetto (personal communication) pointed out to me, indefinite NPs can also be used as “resumptive” in the first clause of the Japanese RD construction: (28) John-ga John-Nom

hon-o

katta yo, Chomsky-no

book-Acc bought

Chomsky-Gen

hon-o. book-Acc

‘John bought a book, (that is) a book by Chomsky.’ Hoji (1998) demonstrates that such an indefinite null argument is attested quite generally in Japanese. Consider the following dialogue: (29) A: John-ga

kuruma-o

John-Nom car-Acc

aratta. washed

‘John washed a car.’ B: Bill-mo e aratta. Bill-also washed ‘Bill did too.’ In B’s utterance, the null object can refer to a car as well as the car John washed. Furthermore, null arguments in Japanese allow so-called quantificational readings when they take QP antecedents. Let us consider the following dialogue: (30) A: John-ga

subete-no/hotondo-no sandai-no kuruma-o aratta.

John-Nom every-Gen/most-Gen/ three-Gen car-Acc washed ‘John washed a car.’

281

B: Bill-mo e aratta. Bill-also washed ‘Bill did too.’ In this dialogue, the missing object of B’s utterance can mean every car/most cars/three cars and in this reading, the set of cars expressed by these QPs can be different from that expressed by the antecedent QPs in A’s utterance. It is now standard to assume, following Oku (1998) and Kim (1999), that ellipsis may be involved in null arguments in Japanese. Under the LF copying analysis of ellipsis, adopted by Oku (1988), for instance, the indefinite reading of the null object in (29B) is obtained by simply copying its antecedent kumuma-o ‘car-Acc’ in (29A) into the null object position. Likewise, the quantificational readings of the null object in (30B) are obtained by copying the relevant antecedents into the null object position. Given this ellipsis analysis of null arguments, one might assume that in those instances of Japanese RD construction that involve postverbal QPs, such as those given below, the null gaps obtain their semantic content through LF copying:10 (31) a.

John-ga e1 aratta yo, kuruma1-o. John-Nom washed

car-Acc

‘John washed a car.’ b.

10

John-ga e1

aratta yo, subete-no/hotondo-no/sannin-no

kuruma1-o.

John-Nom

washed

car-Acc

every-Gen/most-Gen/three-Gen

See Takita (2011) for the claim that null argument ellipsis is in fact involved in

Japanese RD.

282

‘John washed every car/most cars/three cars.’ Notice, however, that under the bi-clausal analysis of the Japanese RD construction, the two clauses involved are basically in the relation of repetition of the same proposition with the second clause added as an afterthought. Notice, further, that in some cases, the second clause adds some more semantic content to the first, as seen in (28), repeated below: (32) John-ga John-Nom

hon-o

katta yo, Chomsky-no

book-Acc bought

Chomsky-Gen

hon-o. book-Acc

‘John bought a book, (that is) a book by Chomsky.’ Here, the type of a book John bought is further specified by adding Chomsky-no ‘Chomsky-Gen’ in the second clause. Given this possibility, it is not implausible to assume that in such RD cases as in (31), the null object simply denotes something unspecified and the semantic content of this object is further specified in the second clause. Under this assumption, (31a) will be interpreted as follows: (33) ∃x [John washed x] & it was a car that John washed. In fact, as Hoji (1998) observes, B’s utterance of (29B) can be interpreted as such that Bill also washed something, which thus independently demonstrates that the semantic values of null arguments can be unspecified. As indicated in (33), it is reasonable to regard the postverbal phrase of the Japanese RD construction as functioning like the pivot of a cleft sentence. Thus I assume that a [Focus] feature is assigned to the postverbal phrase after it is scrambled to the left edge of the second clause, so that the derivation of the RD example (1b), for instance, proceeds as follows: 283

(34) a.

[TP John-wa e hihansita yo], [TP Mary-o1 [TP John-wa t1 hihansita yo]] ↓

(leftward scrambling of Mary-o)

b. [TP John-wa e hihansita yo], [TP Mary-o1 [TP John-wa t1 hihansita yo]] ↓ c.

[Focus]

(assignment of [Focus] to Mary-o)

[TP John-wa e hihansita yo], [TP Mary1-o [TP John-wa t1 hihansita yo]] [Focus]

(deletion)

Even though nothing seems to exclude the possibility that in (31a, b), the gap marked with e obtains its semantic value from the postverbal QP by LF copying, this copying operation seems redundant in the context of Japanese RD and hence I assume in what follows that the gap of this construction has its semantic value unspecified, as indicated in (34).11

11

Carlo Cecchetto (personal communication) raises the question of what is missing

when the phrase put postverbally is an adjunct: (i)

John-ga

kubi-ni natta yo, anna riyuu-de.

John-Nom was-fired

that

reason-for

‘John was fired for that reason.’ It is plausible to claim that in this case, nothing is missing in the first clause, so that this sentence has the following structure before deletion applies: (ii)

[TP John-ga kubi-ni natta yo], [TP anna riyuu-de [TP John-ga kubi-ni natta yo]]

The relevant configuration becomes slightly different when the adjunct put postverbally modifies the embedded clause in such a sentence as (iiia), which will be assigned the structure given in (iiib) under the bi-clausal analysis: (iii) a.

Mary-ga

[John-ga

Mary-Nom John-Nom

kubi-ni natta to] was-fired

Comp said

‘Mary said that John was fired for that reason.’

284

itta yo, anna riyuu-de. that

reason-for

There is in fact supporting evidence for this line of analysis. Haraguchi (1973) and Kuno (1978a) independently observe that in the Japanese RD construction, a wh-phrase cannot be included in the postverbal constituent, as illustrated below: (35) John-wa kizuita no, Mary-ga (*dare-to) John-Top noticed Q

Mary-Nom who-with

koi-ni otita

koto-ni.

love-in fell

fact-Dat

‘Did John notice that Mary fell in love?’ ‘*Who did John notice that Mary fell in love with?’ Under the present analysis, this RD example without the wh-phrase dare-to ‘who-with’ will be interpreted as follows: (36) Is there any x that John noticed? & was it the fact that Mary fell in love that John noticed? On the other hand, when the wh-phrase dare-to is included in the postverbal NP, (35) gives rise to a nonsense statement: (37) #Is there any x that John noticed? & who was it that John noticed that Mary fell in love with? This explains the fact that a wh-phrase cannot be included in the postverbal constituent b.

[TP Mary-ga [John-ga kubi-ni natta to] itta yo], [TP anna riyuu-de1 [TP Mary-ga [John-ga t1 kubi-ni natta to] itta yo]]

First of all, (iiia) seems to me fairly degraded with the intended reading. If it is acceptable at all, it will be necessary to claim that in (iiib), what is at stake for identity is the notion of “nondistinctness,” so that the existence of the trace of anna riyuu-de ‘for that reason’ in the second clause does not prevent the two clauses in question from being regarded as non-distinct. I am greatly indebted to Carlo Cecchetto for discussing the matters taken up in this footnote.

285

of the Japanese RD construction.

5.3

Evidence from Quantifier Scope Interaction

In this section, I argue for the present bi-clausal analysis of the Japanese RD construction by presenting evidence involving quantifier scope interaction between postverbal QPs and others. We have seen in Section 2.4 that the surface order of QPs affects their scope order in Japanese. Compare the following examples: (38) a.

Sannin-no

otoko-ga

hutari-no onna-o

kinoo

tazuneta.

three-Gen

man-Nom

two-Gen woman-Acc

yesterday visited

‘Three men visited two women yesterday.’ b.

Hutari-no onna-o1 two-Gen

(3>2)

sannin-no otoko-ga t1 kinoo

woman-Acc three-Gen man-Nom

tazuneta.

yesterday visited

‘Two women, three men visited yesterday.’

(3> <2)

In (38a), sannin-no otoko ‘three men’ takes scope over hutari-no onna ‘two women’, whereas in (38b), either QP can take scope over the other. This shows that a QP that undergoes scrambling, crossing another QP, makes the scope order ambiguous. Let us describe these facts in terms of such interpretive rules as follows:12 (39) a.

[ ... Q1 ... Q2 ] -> Q1 > Q2

b. [ ... Q1 ... Q2 ... t1 ] -> Q1 > < Q2 where Q1 c-commands Q2 and Q2 c-commands t1 and order is irrelevant. The same pattern of facts is also observed with the IO-DO and DO-IO orders (cf. Hoji 12

See Section 2.4 for a more accurate explanation of the relevant scope facts. 286

(1985)): (40) a.

John-ga

sannin-no otoko-ni hutari-no onna-o

syookaisita.

John-Nom

three-Gen man-Dat two-Gen woman-Acc introduced

‘John introduced two women to three men.’ b.

John-ga

hutari-no onna-o1

(3>2)

sannin-no otoko-ni t1 syookaisita.

John-Nom two-Gen woman-Acc three-Gen man-Dat

introduced (3> <2)

With the assumption that the IO-DO order reflects the underlying word order in Japanese, these facts follow from the interpretive rules given in (39): since (40a) reflects the underlying order and hence involves no movement, sannin-no otoko ‘three men’ takes scope over hutari-no onna ‘two women’ according to (39a), whereas in (40b), hutari-no onna-o crosses sannin-no otoko by scrambling, and hence the sentence becomes ambiguous according to (39b). With this background in mind, let us examine how QPs interact in the Japanese RD construction. Consider the following pair of sentences: (41) a.

Sannin-no

otoko-ga e

kinoo

tazuneta yo,

three-Gen

man-Nom

yesterday visited

‘Three men visited two women yesterday.’ b.

e Hutari-no two-Gen

onna-o

kinoo

tazuneta yo,

woman-Acc

yesterday visited

‘Three men visited two women yesterday.’

hutari-no onna-o. two-Gen woman-Acc (3> <2) sannin-no otoko-ga. three-Gen man-Nom (3> <2)

(41a) is an RD counterpart of (38a) in which the object hutari-no onna-o ‘two women’ 287

is put postverbally, while (41b) is another counterpart of (38a), in which the subject sannin-no otoko-ga ‘three men’ is put postverbally. The scope ambiguity observed in (41a) will be accounted for by the interpretive rules given in (39), no matter whether we take the rightward movement analysis or the bi-clausal analysis. (41a) has the following schematic structures under these analyses: (42) a.

Rightward movement analysis: [[three men ... t1 ... ] two women1]

b.

Bi-clausal analysis: [ ... ], [two women1 [three men ... t1 ... ]]

Whichever structure is taken will induce the scope ambiguity between the two QPs involved according to interpretive rule (39b). On the other hand, it is not immediately clear why (41b) is ambiguous with respect to the scope interaction between the two QPs in question, in particular, why it has the reading of 2>3, under the rightward movement analysis. The same pattern of facts obtains with the scope interaction between IO and DO, as illustrated below: (43) a.

John-ga

sannin-no otoko-ni e syookaisita yo, hutari-no

John-Nom three-Gen man-Dat

introduced

two-Gen

‘John introduced two women to three men.’ b.

onna-o. woman-Acc (3> <2)

John-ga e hutari-no onna-o

syookaisita yo, sannin-no

otoko-ni.

John-Nom two-Gen woman-Acc

introduced

man-Dat

‘John introduced two women to three men.’

three-Gen

(3> <2)

In (43a), the underlyingly lower DO is put postverbally whereas in (43b), the 288

underlyingly higher IO is put postverbally. Again, the question is why (43b) has the reading of 2>3. One might claim that the availability of this unexpected reading has something to do with the linear order of the two QPs involved: even though hutari-no onna ‘two women’ is structurally lower than sannin-no otoko ‘three men’ in both (41b) and (43b), the former precedes the latter and this may contribute to the availability of the reading of 2>3. However, this is not the case with Japanese clefts, as shown below: (44) [John-ga t1

hutari-no onna-o

syookaisita

no]-wa

sannin-no

John-Nom

two-Gen woman-Acc

introduced

Comp-Top

three-Gen

otoko-ni1

da.

man-Dat

be

(3>2)

‘It was three men that John introduced two women to.’ This sentence allows only the reading in which sannin-no otoko ‘three men’ takes scope over hutari-no onna ‘two women’. Thus, it is unlikely that the linear order of the two QPs in such RD examples as in (41b) and (43b) affects their scope order. Under the bi-clausal analysis, on the other hand, there is a natural way to account for the availability of the 2>3 readings in (41b) and (43b): the direct object hutari-no onna-o can serve as a topic of the second clause. Thus, these sentences can have the following structures under the bi-clausal analysis:13

13

Just for ease of presentation, I am using trace to mark the original position from

which scrambling takes place in the second clause in (45a, b). Since the scrambling of otoko-ga ‘man-Nom’ is string-vacuous in (45a), the bottom copy of the resulting chain

289

(45) a.

[TP e hutari-no onna1-o kinoo tazuneta yo], [CP Øtopic(1) [TP sannin-no otoko-ga2 [TP t2 pro1 kinoo tazuneta yo]]]

b.

[TP John-ga e hutari-no onna1-o syookaisita yo], [CP Øtopic(1) [TP sannin-no otoko-ni2 [TP John-ga t2 pro1 syookaisita yo]]]

In each of these representations, the zero topic refers to hutari-no onna ‘two women’ and it binds the null object, identified as pro, in the second clause. (41b) and (43b) are derived from these representations by deleting the lower TP in the second clause under identity with the corresponding TP in the first clause. These representations represent the 2>3 readings according to interpretive rule (39a). The plausibility of this analysis is confirmed by the following examples, in which the zero topics are replaced by overt ones: (46) a. ?Hutari-no onna1-o two-Gen

kinoo

tazuneta yo,

woman-Acc

yesterday visited

sannin-no

otoko-ga

(kinoo

three-Gen

man-Nom

yesterday visited

b. ?John-ga

hutari-no onna1-o

John-Nom two-Gen woman-Acc

hutari-no onna1-wa two-Gen woman-Top

tazuneta yo).

syookaisita yo, hutari-no

onna1-wa

introduced

woman-Top

two-Gen

(John-ga) sannin-no otoko-ni (syookaisita yo). John-Nom three-Gen man-Dat

introduced

should be pronounced according to Abe and Hornstein’s (2012) mechanism of pronunciation. See fn. 9 for relevant discussion. The same comment applies to the representations given in (53).

290

These sentences are awkward due to the repetition of phrases and yet are still reasonably acceptable. As expected, they have the 2>3 readings, which thus confirms that the representations in (45) represent these readings. In this way, the scope ambiguity of (41b) and (43b) is explained under the assumption that a zero topic is optionally available for the second clause under the bi-clausal analysis. The present analysis in terms of zero topics is further supported by the behaviors of pronominal variable binding and anaphor binding in the Japanese RD construction. It is well known that anaphors and pronouns that function as variables must be c-commanded by their antecedents, as shown below: (47) a.

Dono

tyosya-mo1

Taroo-ni

soitu1-no

hon-o

okutta.

every

author

Taroo-Dat

that person-Gen

book-Acc sent

‘Every author1 sent Taro his/her1 book.’ b.?*Sono1 its

tyosya-ga

Taroo-ni

dono

hon-mo1 okutta.

author-Nom

Taroo-Dat

every

book

sent

‘Its1 author sent Taro every book1.’ (48) a.

Seito-tati-ga

Taroo-ni

otagai-no

student-pl.-Nom

Taroo-Dat each other-Gen

tomodati-o

syookaisita.

friends-Acc

introduced

‘The students introduced each other’s friends to Taroo.’ b.?*Otagai-no

tomodati-ga

each other-Gen friend-Nom

Taroo-ni

seito-tati-o

syookaisita.

Taroo-Dat

student-pl.-Acc

introduced

‘Each other’s friends introduced the students to Taroo.’ Given that subject asymmetrically c-commands object, the above data are 291

straightforwardly accounted for by the c-command requirement on pronominal variable binding and anaphor binding. The same pattern of facts also obtains in the relation between IO and DO, as shown below: (49) a.

John-wa dono

hito-ni-mo1

soitu1-no

syasin-o

okutta.

John-Top every

person-Dat

that person-Gen

picture-Acc

sent

‘John sent his/her1 pictures to every person1.’ b.?*John-wa sono1

dono

hon-mo1 okutta.

author-Dat

every

book

John-Top its

tyosya-ni

sent

‘John sent every book1 to its1 author.’ (50) a.

John-wa seito-tati-ni

otagai-no

tomodati-o

syookaisita.

John-Top student-pl.-Dat

each other-Gen

friend-Acc

introduced

‘John introduced each other’s friends to the students.’ b.?*John-wa otagai-no John-Top each other-Gen

tomodati-ni

seito-tati-o

syookaisita.

friend-Dat

student-pl.-Acc

introduced

‘John introduced the students to each other’s friends.’ Given that IO asymmetrically c-commands DO, the above data are explained in exactly the same way as those in (47) and (48). Using these syntactic properties as a diagnostic, Inagaki (1998) demonstrates that in the Japanese RD construction, what precedes behaves as if it is structurally higher than what follows. Consider the following examples: (51) a.

Taroo-ni

dono

hon-mo1 okutta yo, sono1

tyosya-ga.

Taroo-Dat

every

book

author-Nom

sent 292

its

‘Its1 author sent every book1 to Taro. b.

Taroo-ga

dono

hon-mo1 okutta yo, sono1

tyosya-ni.

Taroo-Nom

every

book

author-Dat

sent

its

‘Taro sent every book1 to its1 author.’ (52) a.

Taroo-ni

seito-tati-o

Taroo-Dat student-pl.-Acc

syookaisita yo, otagai-no

tomodati-ga.

introduced

friend-Nom

each other-Gen

‘Each other's friends introduced the students to Taroo.’ b.

Taroo-ga

seito-tati-o

syookaisita yo, otagai-no

Taroo-Nom

student-pl.-Acc introduced

tomodati-ni.

each other-Gen friend-Dat

‘Taroo introduced the students to each other’s friends.’ The acceptability of these data is unexpected under the rightward movement analysis. This is because in each sentence, the bound pronoun or the anaphor is embedded in either the subject or the indirect object that is underlyingly higher than its antecedent in direct object position and undergoes rightward movement, hence having no chance to be c-commanded by its antecedent. Under the bi-clausal analysis, on the other hand, the above data are immediately explained by postulating a zero topic. On this assumption, sentences (51a) and (52a), for instance, have the following representations: (53) a.

[TP e Taroo-ni dono hon-mo1 okutta yo], [CP Øtopic(1) [TP [sono1 tyosya-ga]2 [TP t2 Taroo-ni pro1 okutta yo]]]

b.

[TP e Taroo-ni seito-tati-o1 syookaisita yo], [CP Øtopic(1) [TP [otagai1-no tomodati-ga]2 [TP t2 Taroo-ni pro1 syookaisita]]]

In these representations, the zero topics refer to dono hon-mo ‘every book’ in (53a) and 293

to seito-tati ‘student-pl.’ in (53b) and bind the null objects in the second clauses. The surface forms are derived from them by deleting the lower TPs of the second clauses under identity with the corresponding TP of the first clause. In these representations, the bound pronoun sono in (53a) and the anaphor otagai in (53b) are properly c-commanded by their antecedents, i.e., the zero topics, which thus explains the acceptability of (51a) and (52a). This explanation is confirmed by the acceptability of their counterparts in which the zero topics are replaced by overt ones:14 (54) a. ?Taroo-ni

dono

Taroo-Dat every (Taroo-ni

hon-mo1 sono1 tyosya-ga

book

book

sent

every

its

author-Nom

okutta yo).

Taroo-Dat b. ?Taroo-ni

hon-mo1 okutta yo, dono

sent

seito-tati-o

syookaisita yo,

Taroo-Dat student-pl.-Acc introduced

seito-tati-wa

otagai-no

student-pl.-Top each other-Gen

tomodati-ga

(Taroo-ni

syookaisita yo).

friend-Nom

Taroo-Dat

introduced

Again these sentences are awkward due to the repetition of phrases and yet reasonably acceptable, which thus confirms that the bound pronoun and the anaphor are properly licensed in the representations given in (53). Recall that we have seen in Section 2.4 that long-distance scrambling does not

14

In (54a), the topic marker -wa cannot be attached to dono hon-mo ‘every book’ in the

second clause due to the idiosyncrasy of Japanese particles even though this phrase can function as a topic.

294

change scope order. Consider the following examples: (55) a.

Dareka-ga

[sensei-ga

hutari-no onna-ni

kisusita to]

sinziteiru.

someone-Nom teacher-Nom two-Gen woman-Dat kissed Comp believe ‘Someone believes that the teacher kissed two women.’ b.

Hutari-no onna-ni1 two-Gen

dareka-ga

[sensei-ga t1

kisusita to]

woman-Dat someone-Nom teacher-Nom kissed

Comp

sinziteiru. believe ‘Two women, someone believes that the teacher kissed.’ (55b) is derived from (55a) by applying long-distance scrambling to the embedded object hutari-no onna-ni ‘two women-Dat’. Although this QP crosses dareka ‘someone’ in the matrix subject position, this does not make the sentence ambiguous regarding the scope order of these two QPs, unlike such cases of clause-internal scrambling as (38b). I have argued that this is due to the clause-boundness condition on QR; that is, the [Scope] feature carried by a QP must be satisfied in the closest TP. Thus, in (55b), the [Scope] feature carried by the scrambled phrase hutari-no onna-ni must be satisfied in the embedded TP-adjoined position, so that it takes scope under the matrix subject QP dareka ‘someone’. We have further seen that when the matrix and the embedded subjects are swapped in (55b), as given below: (56) Hutari-no two-Gen

onna-ni1

sensei-ga

woman-Dat

teacher-Nom someone-Nom kissed

295

[dareka-ga t1 kisusita to] Comp

sinziteiru. believe ‘Two women, the teacher believes that someone kissed.’ the scrambled phrase hutari-no onna-ni can take scope over dareka. This is because these two QPs have their [Scope] features satisfied in the same embedded TP projections, hence yielding the scope ambiguity. Under the present bi-clausal analysis of Japanese RD, it is predicted that the same pattern of facts should obtain regarding the scope interaction between the postverbal QP and the matrix/embedded subject QP. This is in fact borne out, as shown below: (57) a.

Dareka-ga

[sensei-ga t1 kisusita to]

someone-Nom teacher-Nom kissed

sinziteiru yo,

Comp believe

hutari-no onna-ni1.

(*2>some)

two-Gen woman-Dat ‘Someone believes that the teacher kissed two women.’ b.

Sensei-ga

[dareka-ga t1 kisusita to]

teacher-Nom someone-Nom kissed

sinziteiru yo,

Comp believe (OK2>some)

hutari-no onna-ni1. two-Gen woman-Dat ‘The teacher believes that someone kissed two women.’

These RD examples have the following schematic structures under the bi-clausal analysis: (58) a.

[TP …], [TP hutari-no onna-ni [TP dareka-ga [CP [TP …] ] …]] 296

b. [TP …], [TP hutari-no onna-ni [TP … [CP [TP dareka-ga …] ] …]] In these structures, hutari-no onna-ni has undergone long-distance scrambling to the top of the second clause, hence giving rise to exactly the same structures as those of (55b) and (56). Thus, the fact that (57a, b) shows the same pattern of scope facts as (55b) and (56) lends strong support to the present bi-clausal analysis of Japanese RD, in which the postverbal phrase undergoes leftward scrambling in the second clause.

5.4

Proper Binding Condition Violations

In this section, I discuss a Japanese RD case that apparently violates the Proper Binding Condition (henceforth, PBC), which requires that traces be bound. Saito (1985) observes that Japanese scrambling obeys this condition, as illustrated below: (59) *[Mary-ga t1 aisiteiru to]2 Mary-Nom

love

Bill-o1

John-ga t2 itta (koto)

Comp Bill-Acc John-Nom

said fact

‘Lit. That Mary loves t1, Bill1, John said.’ This sentence is derived by applying long-distance scrambling to Bill-o first and then applying clause-internal scrambling to the complement clause of itta ‘said’, putting it to the left of the scrambled phrase Bill-o. This violates the PBC, since the trace of Bill-o is unbound. The Japanese RD construction behaves as if postverbal constituents were right-adjoined with respect to the PBC; that is, when more than one constituent is put postverbally, what follows behaves as if it were structurally higher than what precedes with respect to this condition. Let us consider the following examples, taken from 297

Inagaki (1998): (60) a.

Taroo-ga e itta yo,

[Hanako-ga e1 yonda to],

sono hon-o1.

Taroo-Nom

Hanako-Nom read Comp

that

said

book-Acc

‘Lit. Taroo said that Hanako read t1, that book1.’ b.(*)Taroo-ga e Taroo-Nom

itta yo,

sono hon-o1,

said

that

[Hanako-ga e1

book-Acc Hanako-Nom

yonda to]. read Comp (Inagaki 1998:33)

(60a) sounds acceptable only when a pause is put between the two postverbal phrases as well as after the verb itta yo ‘said’. Inagaki judges (60b) as acceptable, which is the case only when no pause is put between the two postverbal phrases, so that sono hon-o ‘that book’ is taken as a phrase scrambled to the top of the following clause. If a pause is put between them in a way parallel to (60a), then the sentence sounds unacceptable. This would follow straightforwardly under the rightward movement analysis. (60a) will be derived by first moving the complement clause of itta ‘said’ rightward and then moving sono hon-o to the right of this complement clause. No violation of the PBC ensues in this derivation. (60b), in contrast, will induce a PBC violation under this analysis, since the trace of sono hon-o is not bound. Under the bi-clausal analysis, there is a plausible way of accounting for the data in (60). Tanaka (2001) proposes that such RD data involve tripartite clausal structures with double application of deletion. Under this proposal, these sentences will have the

298

following structures before deletion applies:15 (61) a.

[TP Taroo-ga e itta yo], [TP [Hanako-ga e yonda to]1 ], [TP sono hon-o2 >]

b. *[TP Taroo-ga e itta yo], [TP sono hon-o1 ], [TP Hanako-ga t1 yonda to]2 >] (61a), when fully uttered, does not sound like an unacceptable sentence, apart from its awkwardness due to frequent repetition of phrases. Note that in this structure, the postverbal constituents undergo leftward movement in the second and third clauses, hence no violation of the PBC. (61b), in contrast, is unacceptable because the second clause cannot be properly interpreted and further the trace of sono hon-o in the third clause is unbound, violating the PBC. This analysis in terms of tripartite clausal structure is further supported by the following examples: (62) a.

Taroo-ga e itta yo, [e hutari-no onna-o

kinoo

Taroo-Nom

yesterday visited

said

two-Gen woman-Acc

tazuneta to],

sannin-no otoko-ga.

Comp (3> <2)

three-Gen man-Nom ‘Lit. Taroo said that t1 visited two women yesterday, three men1.’ b.

15

Taroo-ga e

itta yo,

[Hanako-ga e hutari-no onna-o

syookaisita

Taroo-Nom

said

Hanako-Nom two-Gen woman-Acc

introduced

In (61) and the subsequent representations, the angled brackets indicate deletion sites. 299

to],

sannin-no otoko-ni.

Comp

three-Gen man -Dat

(3> <2)

‘Lit. Taroo said that Hanako introduced two women t1, to three men1.’ The acceptability of these RD sentences could be accounted for in the same way as that of (60a) under the rightward movement analysis, but this analysis faces the problem of why these sentences allow the reading in which hutari-no onna ‘two women’ takes scope over sannin-no otoko ‘three men’. According to this analysis, sannin-no otoko-ga/ni is moved to the end of the sentence from a position higher than that occupied by hutari-no onna-o, and hence the former should necessarily take scope over the latter according to the interpretive rules given in (39), repeated below: (63) a.

[ ... Q1 ... Q2 ] -> Q1 > Q2

b. [ ... Q1 ... Q2 ... t1 ] -> Q1 > < Q2 where Q1 c-commands Q2 and Q2 c-commands t1 and order is irrelevant. Under the bi-clausal analysis, on the other hand, we can rely on the zero topic strategy so that a zero topic is inserted in the third clause of the tripartite clausal structure assumed for such RD constructions as in (62). According to this analysis, these RD sentences will have the following representations: (64) a.

[TP Taroo-ga e itta yo], [TP [e hutari-no onna1-o kinoo tazuneta to]2 ], [CP Øtopic(1) [TP sannin-no otoko-ga3 >]]

b.

[TP Taroo-ga e itta yo], [TP [Hanako-ga e hutari-no onna1-o syookaisita to]2 , [CP Øtopic(1) [TP sannin-no otoko-ni3
t3 pro1 syookaisita to]2 >]] In each of these representations, a zero topic appears in the top of the third clause, referring to hutari-no onna and binding the null object in that clause. Since the zero topic is structurally higher than sannin-no otoko-ga/ni in these representations, they represent the reading in which hutari-no onna takes scope over sannin-no otoko. Further evidence comes from pronominal variable binding and anaphor binding. We have seen in the last section that anaphors and pronouns that function as variables must be c-commanded by their antecedents, as shown in (47)-(50). Keeping this in mind, let us consider the following RD sentences: (65) a.

Taroo-ga e itta yo, [e Hanako-ni

dono

hon-mo1 okutta

to],

Taroo-Nom

every

book

Comp

said

sono1

tyosya-ga.

its

author-Nom

Hanako-Dat

sent

‘Lit. Taroo said that t1 sent every book2 to Hanako, its2 author1.’ b.

Taroo-ga e itta yo,

[Hanako-ga e dono

hon-mo1 okutta

to],

Taroo-Nom

Hanako-Nom every

book

Comp

said

sono1

tyosya-ni.

its

author-Dat

sent

‘Lit. Taroo said that Hanako sent every book2 t1, to its2 author1.’ (66) a.

Taroo-ga e itta yo, [e Hanako-ni

seito-tati1-o

Taroo-Nom

student-pl.-Acc introduced

said

Hanako-Dat

301

syookaisita

to], Comp

otagai1-no

tomodati-ga.

each other-Gen

friend-Nom

‘Lit. Taroo said that t1 introduced the students to Hanako, each other’s friends1.’ b.

Taroo-ga e itta yo,

[Hanako-ga e seito-tati1-o

Taroo-Nom

Hanako-Nom student-pl.-Acc introduced

said

otagai1-no

tomodati-ni.

each other-Gen

friend-Dat

syookaisita

to], Comp

‘Lit. Taroo said that Hanako introduced the students t1, to each other’s friends1.’ The acceptability of these sentences cannot be accounted for under the rightward movement analysis, since at no point of the derivation involved can the anaphor or the bound pronoun be c-commanded by its antecedent in each sentence. Under the bi-clausal analysis, we can assign the following representations to these sentences, exploiting zero topics: (67) a.

[TP Taroo-ga e itta yo], [TP [e Hanako-ni dono hon-mo1 okutta to]2 ], [CP Øtopic(1) [TP sono1 tyosya-ga3 >]]

b.

[TP Taroo-ga e itta yo], [TP [Hanako-ga e dono hon-mo1 okutta to]2 ], [CP Øtopic(1) [TP sono1 tyosya-ni3 >]]

302

(68) a.

[TP Taroo-ga e itta yo], [TP [e Hanako-ni seito-tati1-o syookaisita to]2 ], [CP Øtopic(1) [TP otagai1-no tomodati-ga3 >]]

b.

[TP Taroo-ga e itta yo], [TP [Hanako-ga e seito-tati1-o syookaisita to]2 ], [CP Øtopic(1) [TP otagai1-no tomodati-ni3 >]]

In these representations, the bound pronouns and the anaphors in the third clauses are properly bound by the zero topics that function as their antecedents, which thus explains the acceptability of the sentences in (65) and (66).

5.5

Condition C Effects

Let us now consider how condition C effects emerge in Japanese RD constructions and how they are accommodated under the bi-clausal analysis. Condition C requires that an R-expression not be bound. Thus, in the following sentences, (69) a. *He1 put his cigars in Ben1’s box. b. *She1 tickled Dr. Levin with Zelda1’s feather. the R-expressions Ben and Zelda are bound by the pronouns he and she, respectively, thereby violating Condition C. Reinhart (1976) observes an interesting contrast with respect to reconstruction effects of this condition, as shown below: (70) a. *In Ben1’s box, he1 put his cigars. b. *With Zelda1’s feather, she1 tickled Dr. Levin. (71) a.

In the ivory box that Ben1 brought from China, he1 put his cigars. 303

b. With the feather that Zelda1 inherited from her late peacock, she1 tickled Dr. Levin. Saito (1985) gives the following generalization on reconstruction effects of Condition C: if an R-expression is c-commanded by a pronoun that is coreferential to it in the underlying structure and a phrase that dominates the R-expression escapes the c-command domain of the pronoun by movement, then the resulting structure is free from a Condition C violation only if the R-expression is “deeply embedded” in the moved phrase. Here, a typical case of “deep embedding” is one in which an element is embedded within a relative clause. In (70), since the R-expressions Ben and Zelda are not “deeply embedded” within the moved phrases, Condition C effects hold. These effects disappear when Ben and Zelda are “deeply embedded,” as shown in (71). The following sentences illustrate that the same pattern of facts also holds for Japanese: (72) a. *Kare1-wa Mary-ni [anata-ga John1-ni he-Top

Mary-Dat you-Nom John-Dat

kaita

tegami]-o miseta.

wrote

letter-Acc showed

‘He1 showed Mary the letter you wrote to John1.’ b. [Anata-ga John1-ni kaita you-Nom John-Dat wrote

tegami]-o kare1-wa

Mary-ni

miseta.

letter-Acc he-Top

Mary-Dat showed

‘The letter you wrote to John1, he1 showed Mary.’ c.?*John1-no tegami-o kare1-wa

Mary-ni

John-Gen letter-Acc he-Top

miseta.

Mary-Dat showed

‘John1’s letter, he1 showed Mary.’ (72a) illustrates a typical case of Condition C violation in which a pronoun c-commands 304

the R-expression it is coreferential to. (72b, c) illustrate cases where a phrase that dominates an R-expression escapes the c-command domain of a pronoun that the R-expression is coreferential to by scrambling. Since the R-expression John is “deeply embedded” in the scrambled phrase in (72b), it does not induce a Condition C violation. In (72c), in contrast, John is not “deeply embedded” in the scrambled phrase, hence inducing a Condition C violation. Keeping this generalization in mind, let us see how Condition C effects emerge in Japanese RD constructions. Consider the following sentence: (73) *Kare1-wa Mary-ni he-Top

Mary-Dat

miseta yo, [anata-ga

John1-ni kaita

tegami]-o.

showed

John-Dat wrote

letter-Acc

you-Nom

‘He1 showed Mary the letter you wrote to John1.’ The unacceptability of this sentence will be unexpected under the rightward movement analysis, since if the postverbal complex NP is right-adjoined to a position higher than the subject, then (73) should behave in the same way as (72b) according to Saito’s (1985) generalization on reconstruction effects of Condition C. One might object that the postverbal NP is in fact adjoined to a position lower than the subject under the rightward movement analysis, so that such rightward movement does no serve to escape the c-command domain of kare. That this is not the right account of the unacceptability of (73) is indicated by the fact that Condition C effects emerge even if the pronoun is a non-subject in such a configuration as in (73): (74) *Mary-wa Mary-Top

kare1-ni miseta yo, [anata-ga

John1-ni kaita

tegami]-o.

he-Dat

John-Dat wrote

letter-Acc

showed

you-Nom 305

‘Mary showed him1 the letter you wrote to John1.’ It is unlikely that under the rightward movement analysis, the position that the postverbal DO is right-adjoined to is structurally lower than that occupied by the IO. On the other hand, there is a straightforward account for the unacceptability of (73) and (74) under the bi-clausal analysis. The underlying structures of these sentences are given below: (75) *[TP kare1-wa Mary-ni e miseta yo], [TP [anata-ga John1-ni kaita tegami]-o2 [TP kare1-wa Mary-ni t2 miseta yo]] (76) *[TP Mary-wa kare1-ni e miseta yo], [TP [anata-ga John1-ni kaita tegami]-o2 [TP Mary-wa kare1-ni t2 miseta yo]] If these underlying structures are taken as actually uttered sentences, they are unacceptable with the intended coreference readings. The unacceptability of these sentences is similar in nature to that of the following sentence, which has a coordinate structure: (77) *Mary-wa Mary-Top

kare1-ni atte, [anata-ga John1-ni

kaita

tegami]-o

he-Dat

wrote

letter-Acc

met

you-Nom John-Dat

yaburi-suteta. tore-up ‘Mary met him1 and tore up the letter you wrote to John1.’ Thus, it is natural to claim that whatever constraint prohibits the intended backward coreference in (77) carries over to the account of the unacceptability of (75) and (76), hence also accounting for the unacceptability of their elliptic versions, namely, (73) and 306

(74). A word is in order with respect to the restriction on backward anaphora of pronouns. Notice that this restriction needs to be formulated in such a way that it does not exclude backward anaphora of pronouns in Japanese RD constructions such as (4), repeated below: (78) John-wa John-Top

kanozyo1-o

nagutta yo,

Mary1-o.

her-Acc

hit

Mary-Acc

‘John hit her, that is, Mary.’ What is a crucial difference between the cases where backward anaphora is possible and those where it is impossible? Intuitively, in the former cases, the antecedent of the pronoun functions as the “pivot” of the proposition expressed by a given RD construction. Thus, the speaker, when uttering (78), is talking about Mary as the pivot of the proposition added as an afterthought with the elided part functioning as its background, and in such a case kanozyo ‘her’ in the first clause can refer backward to Mary. Such use of backward anaphora is independently attested in a Japanese example like the following: (79) Hora look

kare1

da. John1-ga

kita yo.

he

be

came

John-Nom

‘Look! That’s he. John is coming.’ In this sentence, it is natural to consider that John functions as the pivot of the proposition expressed by (79) and hence kare ‘he’ can refer backward to John. Given this restriction on backward anaphora of pronouns, the unacceptability of (73) and (74) 307

as well as that of (75) and (76) can be attributed to the fact that in these RD sentences, what the pronoun refers backward to does not function as the pivot of the proposition expressed by each sentence; rather it is the whole phrase put postverbally that functions as such.

5.6

The VP Scrambling Approach to Multiple Right Dislocation

In Section 5.4, we have seen cases of Japanese RD that involve multiple RD, for which I have argued on the basis of the behaviors of this construction with respect to the PBC that they involve tripartite clausal structures. In this section, we consider other possible structures for multiple RD and some restrictions on them to examine what follows as consequences from this consideration. Let us first consider a simple Japanese RD construction that involves multiple RD: (80) John-wa sinsetunimo katte-agetanda yo, Mary-ni John-Top kindly

buy-gave

aisu-o.

Mary-Dat ice cream-Acc

‘John kindly bought Mary, an ice cream.’ Under the present analysis, this sentence can be assigned the following tripartite clausal structure: (81) [TP John-wa sinsetunimo e e katte-agetanda yo], [TP Mary-ni1 ], [TP aisu-o2 >] Recall our assumption made in Section 5.2 that under the bi-clausal analysis of Japanese RD, a [Focus] feature is assigned to the scrambled phrase since it functions as the pivot 308

of the proposition expressed in the clause it belongs to. Thus, in (81), the scrambled phrases Mary-ni in the second clause and aisu-o in the third are assigned [Focus] features. Recall further that in Section 2.6, I have argued that a multiple scrambling case such as (82) has the structure given in (83) if scrambling is applied separately to Mary-ni and aisu-o: (82) Mary-ni1 Mary-Dat

aisu-o2

John-ga

sinsetunimo t1 t2 katte-ageta.

ice cream-Acc John-Nom kindly

buy-gave

‘Lit. To Mary, an ice cream, John kindly bought.’ (83) [TP Mary-ni1 [TP [vP aisu-o2 [vP John-ga sinsetunimo t1 t2 katte-age]]ta]] On the assumption that no more than one phrase can be adjoined to a given phrase, Mary-ni and aisu-o are adjoined to different categories, one to TP and the other to vP in (83). Given these assumptions, (81) should be modified into the following: (84) [TP John-wa sinsetunimo e e katte-agetanda yo], [TP Mary-ni1 ], [TP aisu-o2 >tanda yo>] It is most natural to claim that this structure reflects the intonation contour of (80) in which a pause is put right after Mary-ni with falling intonation. One might assign a more simple structure to (80): (85) [TP John-wa sinsetunimo e e katte-agetanda yo], [CP Mary-ni1 [CP [TP aisu-o2 [Focus] 309

[Focus]

]]]] This representation consists of two clauses, and Mary-ni and aisu-o ‘ice cream-Acc’ undergo scrambling in the second clause, adjoining to different categories in accordance with the restriction of one adjunction per category.16 However, (85) violates the CCF, reproduced below: (86) Crossing Constraint on Features (CCF) Given a chain C = (α1, … αi, … αn), *αi [X]

…

β …

αi+1

[X]

unless feature [X] is carried along as a free ride. This constraint is independently motivated by the fact that double application of operator movement within a single clause is prohibited, as shown in the following English examples (those in (87) are taken from Lasnik and Saito (1991) and (88a) is taken from Lasnik and Saito (1992):17

16

Recall the assumption made in Section 2.6 that a [Focus] feature needs to be licensed

in a position where it takes scope, that is, in TP or a higher projection. 17

Saito and Fukui (1998) argue that English heavy NP shift shares its basic properties

with scrambling. One piece of evidence they provide for this claim is that multiple heavy NP shift is in fact possible. They give the following example: (i)

John told t1 t2 yesterday [a most incredible story]1 [to practically everyone who was willing to listen]2.

This might suggest that there are some dialectal or idiolectal variations among native speakers. I must leave further investigations on this point for future research.

310

(87) a. *John built t1 t2 yesterday [with a hammer]2 [the house that he will live in]1. b. *John built t1 t2 yesterday [the house that he will live in]1 [with a hammer]2. (88) a.??On the table2, this book1 John put t1 t2. b.??This book1, on the table2 John put t1 t2. Given that both topicalization and heavy NP shift involve licensing of the [Focus] features of the topicalized and rightward shifted phrases (cf. Abe (forthcoming)), then the unacceptability of the sentences given in (87) and (88) follows from the CCF. Given the VP scrambling approach to multiple scrambling adopted in Section 2.6, we can assign another structure to (80): (89) [TP John-wa sinsetunimo e e katte-agetanda yo], [TP [VP Mary-ni aisu-o2 tV]1 [Focus] ] Since the whole VP is assigned a [Focus] feature in this representation, it will not induce a violation of the CCF. If (89) is in fact an appropriate structure for (80), it is natural to claim that this structure reflects the intonation contour of (80) in which no pause is put right after Mary-ni, hence Mary-ni and aisu-o constituting a single intonation unit. Given this VP scrambling analysis of multiple RD, we have a couple of predictions: Koizumi (1995, 2000) argues that the fact that more one one phrase can appear in the focus position of Japanese clefting, as illustrated below, is attributed to the

311

availability of remnant VP preposing in this language:18 (90) [John-ga John-Nom

ageta

no]-wa [Mary-ni

ringo-o]

gave

NL-Top Mary-Dat apple-Acc

da. be

‘Lit. It was [an apple to Mary] that John gave.’ Under this VP preposing analysis, this sentence has the following structure: (91) [TP John-ga [VP t] age+ta no]-wa [VP Mary-ni ringo-o tV] da As evidence for this analysis, Koizumi claims that it properly derives the clause-mate condition on the multiple phrases appearing in the focus position of Japanese clefting, as illustrated below: (92) *[Mary-ga t1 [John-ga t2 katta Mary-Nom

to]

itta no]-wa [Nancy-ni1 ringo-o2]

John-Nom bought Comp said NL-Top Nancy-Dat apple-Acc

da. be

‘Lit. It was [Nancy1, apples2] that Mary told t1 that John had bought t2.’ Koizumi (1995, 2000) claims that the clause-mate condition observed above is attributed to the locality condition on head movement of the V that feeds remnant VP movement. Thus, in order to derive the sentence in (92), the embedded V katta ‘bought’ must be raised at least to the matrix T, so that the remnant VP to be preposed can contain both the embedded object ringo-o ‘apple-Acc’ and the matrix object Nancy-ni ‘Nancy-Dat’. Koizumi claims that such head movement is prohibited since it crosses a tensed clause boundary. Given this restriction on remnant VP preposing, it is predicted that multiple RD in 18

See Abe (2012) for the argument that it is in fact VP scrambling that is responsible

for preposing multiple focus phrases in Japanese clefting.

312

Japanese is also subject to the clause-mate condition. Cecchetto (1999) observes that this is in fact the case (the examples in (93) and (94) are adapted from Cecchetto (1999:67): (93) a. *e [Bill-ga e aisiteiru to] Bill-Nom love

omotteiru yo, sono mura-o

Comp think

that

John-ga.

village-Acc John-Nom

‘John thinks that Bill loves that village.’ b. *John-ga e [Mary-ga e John-Nom Mary-Nom

aisiteiru to] love

itta yo, sono mura-o

Comp said

that

Bill-ni

village-Acc Bill-Dat

‘John told Bill that Mary loved that village.’ (94) a.?*e [Bill-ga e Bill-Nom

aisiteiru to] love

omotteiru yo, John-ga

Comp think

b.?*John-ga e [Mary-ga e aisiteiru to] John-Nom Mary-Nom

love

sono mura-o.

John-Nom that village-Acc itta yo, Bill-ni

Comp said

sono mura -o.

Bill-Dat that village-Acc

The examples in (94) are the same as those in (93) except that the two postverbal phrases are flipped in their order. Though the former sentences are somewhat better than the latter, they are still degraded. This fact follows immediately if these RD examples are assigned structures like (89), ones involving VP scrambling of the postverbal phrases. On the other hand, if they are assigned tripartite clausal structures, such as the one given in (84), then no such restriction as the clause-mate condition should hold. (93a) and (94a), for instance, will have the following tripartite clausal structures:

313

(95) a.

[TP e [Bill-ga e aisiteiru to] omotteiru yo], [TP sono mura-o1 ], [TP John-ga2 ]

b.

[TP e [Bill-ga e aisiteiru to] omotteiru yo], [TP John-ga1 ], [TP sono mura-o2 ]

Nothing seems to go wrong with these representations.19 In fact, it seems that if a pause is put between the postverbal NPs in each example in (93) and (94), then the acceptability improves significantly.20 19

One might claim that (95a) will be ruled out independently since the third clause of

this representation, if uttered as an independent sentence, is unacceptable. Given the assumption made in Chapter 2 that a scrambled phrase carries a [Focus] feature whenever possible, sono-mura-o ‘that village-Acc’ in this fully uttered sentence is forced to carry a [Focus] feature, hence inducing a violation of the CCF. However, this account will not apply to (95a) since in this case sono mura-o does not carry a [Focus] feature due to the fact that it is part of the target of deletion. Alternatively, it can be claimed that nothing will require that the deleted TP of the third clause of (95a) has sono mura-o scrambled long-distance rather than just stay in situ; even in the latter case, the identity requirement for deletion will be met. For these reasons, it is reasonable to conclude that (95a) is a legitimate representation. 20

It seems to me that those RD examples in (94) are somewhat worse than those in (93)

314

Another prediction that arises from the VP scrambling approach to multiple RD in Japanese is concerned with the scope interaction between indirect and direct object QPs when they are both put postverbally. Recall that we have observed in Section 2.6 that when indirect and direct object QPs are both scrambled to the top of a sentence in this order, then the indirect object QP takes scope over the direct object QP but not the other way around, as shown in (2.165), reproduced below: (96) Hutari-no two-Gen

onna-ni1

sannin-no

woman-Dat three-Gen

otoko-o2 John-ga t1 t2

syookaisita.

man-Acc John-Nom

introduced

‘Lit. To two women, three men, John introduced.’ I have argued there that the reason why this sentence does not have the reading of 3>2 is attributed to the fact that the possible derivations for this reading will be doomed to violate either the CCF or the ECDII, which is reproduced below: (97) The Economy Condition on Dependency II (ECDII) If αi is a member of the chain (α1, ... αi, ...), then αi+1 must be the closest possible dependent for αi. (98) α is a possible dependent for β if γ is dependent upon β, and α and γ have the same feature. The relevant derivations are reproduced below:

under the intended intonation pattern. This might have something to do with a certain restriction on the way the filler-gap relation is established when a given RD sentence involving a tripartite clausal structure is parsed.

315

(99) a. *[TP hutari-no onna-ni1 [vP sannin-no otoko-o2 [vP John-ga t1 [Focus]

[Scope]

t2 syookaisi]]ta]

[Scope]

b. *[CP hutari-no onna-ni1 [CP [TP sannin-no otoko-o2 [TP dareka-ga3 [vP t3 [Focus]

[Focus][Scope]

t1

t2

[Scope]

syookaisi]ta]]]] In both representations, when hutari-no onna-ni ‘two women-Dat’ is scrambled, it has left behind its [Scope] feature, and hence they represent the reading of 3>2. However, (99a) violates the ECDII since the scramble direct object QP sannin-no otoko ‘three men’ is regarded as a possible dependent for hutari-no onna-ni due to the fact that it carries a [Scope] feature and is closer to the top member of hutari-no onna-ni than its real dependent. In (99b), the [Focus] feature borne by the indirect object QP hutari-no onnan-ni crosses that borne by the direct object QP, hence induing a violation of the CCF. The only ways to derive (96) without inducing a violation of either the CCF or the ECDII is either to make the indirect object QP carry along its [Scope] feature when we apply literal multiple scrambling, as in (100a), or to derive (96) via VP scrambling, as in (100b): (100) a.

[TP hutari-no onna-ni1 [vP sannin-no otoko-o2 [vP John-ga t1 t2 syookaisi]]ta] [Focus][Scope]

b.

[Scope]

[TP [vP [VP hutari-no onna-ni sannin-no otoko-o tV]1 [vP John-ga t1 [Scope]

[Scope]

syookaisi+v]]ta] In (100a), the scrambling of the indirect object QP hutari-no onna-ni ‘two women-Dat’ 316

has produced a focus chain and hence its [Scope] feature can be regarded as carried along as a free ride, hence not violating the CCF. (100a, b) both represent the reading available to (96), namely that of 2>3. Given that in the case of RD, postverbal phrases necessarily bear [Focus] features, VP scrambling is the only option to derive multiple RD sentences without violating the relevant conditions. It is then predicted that the RD counterpart of (96) should also have only the reading in which the indirect object QP takes scope over the direct object QP. Abe (2012) observes that this is in fact the case: (101) John-ga t1 t2 syookaisita yo, hutari-no John-Nom

introduced

two-Gen

onna-ni1

sannin-no

otoko-o2.

woman-Dat three-Gen

man-Acc

‘John introduced three men to two women.’

(Abe 2012:81)

Under the present assumptions, this sentence has the following structure: (102) [TP John-ga e e syookaisita yo], [TP [VP[Focus] hutari-no onna-ni sannin-no otoko-o [Scope]

[Scope]

tV]1 ] In this representation, the indirect object QP hutari-no onna-ni asymmetrically c-commands the direct object QP sannin-no otoko-o, hence taking scope over the latter. Here again, things are different when a pause is put right after hutari-no onna-ni in (101), in which case the reading of 3>2 is much easier to obtain. Under the present analysis, this pausing reflects the fact that (101) is parsed as having a tripartite structure: (103) [TP John-ga e e syookaisita yo], [TP hutari-no onna-ni1
syookaisita yo>], [TP sannin-no otoko-o2
[Scope]

t1 t2 syookaisita yo>>>] In this representation, sannin-no otoko-o asymmetrically c-commands hutari-no onna-ni, hence taking scope over the latter. The fact that (101) also has the reading of 2>3 even with a pause right after hutari-no onna-ni is attributed to the availability of the zero topic strategy, according to which this sentence has the following structure: (104) [TP John-ga e e syookaisita yo], [TP hutari-no onna-ni1 ], [CP Øtopic(1) [TP sannin-no otoko-o2
[Focus][Scope]

syookaisita yo>]] Here the zero topic, which refers to hutari-no onna, asymmetrically c-commands sannin-no otoko, which thus represents the reading of 2>3. The same point can also be made concerning the scope interaction between subject and object QPs, which probably shows the relevant contrast clearer. Consider the following RD example: (105) Kinoo e e yesterday

tazuneta yo,

sannin-no otoko-ga

hutari-no onna-o.

visited

three-Gen man-Nom

two-Gen woman-Acc

‘Three men visited two women yesterday.’ This sentence, when uttered without a pause between the two postverbal phrases, has only the reading in which the subject QP takes scope over the object QP. This follows 318

from the present VP scrambling approach to multiple RD: in this case, it must be vP that undergoes scrambling since the scrambled phrase needs to include the subject QP, as shown below: (106) [TP kinoo e e tazuneta yo], [TP [vP[Focus] sannin-no otoko-ga hutari-no onna-o tv]1 [Scope]

[Scope]

tV]1 ] On the other hand, when (103) is uttered with a pause right after the subject QP, the opposite reading is much easier to obtain. This is because in this case, (103) may have the following tripartite structure: (107) [TP kinoo e e tazuneta yo], [TP sannin-no otoko-ga1 ], [Focus][Scope] [TP hutari-no onna-o2 >>] [Focus][Scope]

[Scope]

Thus, the scope patterns observed with the two QPs put postverbally lend strong support to the VP (or vP) scrambling approach as well as the tripartite structure approach to multiple RD in Japanese. To sum up this chapter, I have argued for the bi-clausal analysis of the Japanese RD construction, according to which the second clause functions as an afterthought for the first clause, filling the gap of the latter with semantic content. According to this analysis, the postverbal phrase undergoes leftward scrambling in the second clause, with the other materials deleted under identity with those in the first clause, hence serving as a filler of the gap of the first clause. This analysis properly captures the island 319

sensitivity of this construction since it involves leftward scrambling; more importantly, this island sensitivity persists even if the gap of the first clause is replaced by an overt resumptive pronoun. This analysis also properly captures the quantifier scope interaction between the postverbal QP and another QP put preverbally. In particular, the unexpected scope patterns observed in this construction is naturally attributed to the zero topic strategy, which makes it possible for a structurally lower QP to take scope over the postverbal QP. We have also discussed those RD cases that appear to violate the PBC under the bi-clausal analysis, those consisting of the postverbal clause containing an unbound trace and the other postverbal phrase that has moved out of that clause, producing that unbound trace. I have argued, following Tanaka (2001), that these cases involve tripartite structures in which the two postverbal phrases are scrambled to the top of the second and third clauses, respectively, with the other materials deleted under identity with those of the preceding clauses. Given this possibility, no actual violation of the PBC is induced with those RD cases that appear to violate this condition. We have also discussed those RD cases that seem to show reconstruction effects of Condition C. I have argued that such effects are only apparent and that they are more appropriately ascribed to the ban on backward anaphora under the bi-clausal analysis. Finally, I have discussed cases of multiple RD constructions, arguing that apart from the possibility of analyzing them as having tripartite structures, they must involve VP scrambling that raises the postverbal phrases to the left periphery of the second clause under the bi-clausal analysis. This analysis properly captures why the clause-mate condition holds between the two postverbal phrases and how the two 320

postverbal QPs interact scopally. Once the bi-clausal analysis of Japanese RD is well-established, this construction is no more a challenge to the restricted adjunction theory or the asymmetrical structure theory mentioned at the beginning of this chapter, which both demand that Japanese, a head-final language, never allow rightward movement. This is a welcome result from the point of view of language acquisition: how do Japanese children come to know the right structure for the RD construction? The answer should be that since rightward movement is unavailable to Japanese, this construction should be analyzed differently and that the only available option is a bi-clausal analysis in which the postverbal phrase undergoes leftward movement.

321

6

Conclusions

In this manuscript, I have argued that scrambling may have semantic effects mainly in either one of the following two ways: (i) a chain produced by scrambling is reinterpreted as an operator-variable chain or (ii) it serves as a mediator of various constructions. Discussing the first case in Chapter 2, I have argued that the reinterpretation for an operator-variable chain occurs whenever it is possible and that when it is prohibited, the chain produced by scrambling is regarded as a semantically vacuous chain, hence collapsed into a single-membered one or absorbed into a larger chain. Addressing an issue regarding the possibility of LF “undoing” of long-distance scrambling of wh-phrases, I have argued that long-distance scrambling of a wh-phrase produces a focus chain unless this choice leads the derivation to crash. This analysis regards Takahashi’s (1993) wh-movement-like case as unmarked and Saito’s (1989) radical reconstruction case as the one in which the assignment of a [Focus] feature to a wh-phrase scrambled long-distance leads the derivation to crash and hence the alternative derivation without [Focus] feature assignment is chosen. Only in the latter case does long-distance scrambling produce a semantically vacuous chain. I have also argued that those QPs scrambled long-distance interact with other QPs in the predicted ways under our hypothesis that they produce focus chains. In particular, I have refuted Oka’s (1989) claim that chains produced by long-distance scrambling are unlicensed and hence must be undone, by arguing that the fact that QPs scrambled long-distance

322

cannot take scope in the scrambled positions is naturally explained by the clause-boundedness condition on QR. As for clause-internal scrambling, I have argued, following Abe (1993), that clause-internal scrambling can be too short to license [Focus] feature assignment, hence not establishing a focus chain. In that case, the resulting chain is taken as semantically vacuous or ends up being part of a larger chain. Finally, I have shown good consequences of our claim that scrambling produces a focus chain whenever possible by examining the behaviors of multiple scrambling of wh-phrases and QP. I have argued for Koizumi’s (1995, 2000) VP scrambling approach to such cases, having reached the conclusion that VP scrambling is always involved in the case of long-distance multiple scrambling while literal multiple scrambling is available for limited cases of clause-internal multiple scrambling. In arguing for our theory of scrambling, we have crucially relied on two minimality conditions that constrain the chains produced by scrambling: (1)

Crossing Constraint on Features (CCF) Given a chain C = (α1, … αi, … αn), *αi [X]

…

β …

αi+1

[X]

unless feature [X] is carried along as a free ride. (2)

The Economy Condition on Dependency II (ECDII) If αi is a member of the chain (α1, ... αi, ...), then αi+1 must be the closest possible dependent for αi.

It is beyond the scope of this manuscript to discuss the exact nature of these minimality 323

conditions (see Abe (1993, forthcoming) for further discussion), but they seem to suggest that two kinds of minimality condition are necessary, which are schematically shown below: (3)

a. *αi, … β, … α j [F] [F] b. *αi, … β, … α j [F]

[F]

One type, schematically shown in (3a), prohibits β from intervening in the chain link (αi, α j) where αi and β share the same feature. The CCF is one instantiation of this type of minimality condition. The other type, schematically shown in (3b), prohibits β from intervening in the chain link (αi, α j) where αj and β share the same feature. The ECDII is one instantiation of this type of minimality condition. To the extent that our theory of scrambling is successful, it gives support to the claim made by Abe (1993, forthcoming) that two kinds of minimality condition are independently necessary. In Chapters 3 through 5, I have discussed those cases in which chains produced by scrambling serve as “mediators” of various constructions. I have mainly argued for the following hypothesis, which builds upon Abe’s (2012) claim that no overt operator movement exists in Japanese and that what appears to be an instance of such movement in fact involves scrambling: (4)

No phrase is attracted in Japanese.

Abe (2012) suggests that this property has to do with the defectiveness of functional categories in the sense of Fukui (1986). To support hypothesis (4), I have demonstrated 324

in Chapter 3 that what appears to be an instance of A-movement in Japanese also involves scrambling. I have discussed three constructions, those of nominative-genitive conversion (NGC), subject-to-subject raising (SSR) and subject-to-object raising (SOR), which all manifest apparent violations of the NIC, and have attributed them to the fact that the movement involved in deriving these constructions is in fact scrambling. Given that the NIC effects on A-movement are captured by MCL, which prohibits a given application of movement from skipping possible feature checking/licensing positions, scrambling is not subject to MCL since it does not involve any feature checking/licensing. On the other hand, these three constructions are sensitive to the SSC and this fact follows from the fact that scrambling is subject to the CCF, given in (1). Thus, the fact that these three constructions are subject to the SSC, but not the NIC, lends strong support to our claim that scrambling is involved in producing A-chains in Japanese. Further, I have demonstrated that the A-chains produced by scrambling show properties with respect to pronunciation and LF interpretation that are best captured by the Agree-less approach, advocated recently by Hornstein (2009), according to which the overt vs. covert distinction is made in terms of which member of a given chain is pronounced. In particular, I have shown, along the lines of Abe (2016), that pronunciation and LF interpretation correlate with each other in the A-chains involved in the NGC and SOR constructions; that is, the member of a given A-chain that is pronounced must be the one that serves for LF interpretation. As further support for the Agree-less approach, I have argued that there is a special type of A-movement involved 325

in the SSR and SOR constructions, namely one that involves a step of movement into θ-position and that the resulting A-chain allows the option of having its bottom copy pronounced, a case of backward binding. In Chapter 4, I have argued that Japanese relativization involves scrambling of a relative operator on the basis of the left edge effect on island sensitivity, according to which the association of a relative head with the corresponding gap is free from island effects when the gap is located at the left edge of the relative clause. Assuming Abe and Hornstein’s (2012) theory of movement, I have shown how the left edge effect is derived from this theory with the assumption that the movement involved in Japanese relativization is leftward. With this assumption, those cases that exhibit the left edge effect involve string-vacuous leftward movement, hence immune to island conditions according to Abe and Hornstein’s theory of movement. In Chapter 5, I have argued against the rightward movement approach to the Japanese Right Dislocation construction, defending instead the bi-clausal analysis proposed by Tanaka (2001), according to which the second clause functions as an afterthought for the first clause, filling the gap of the latter with semantic content. According to this analysis, the postverbal phrase undergoes leftward scrambling in the second clause, with the other materials deleted under identity with those in the first clause, hence serving as a filler of the gap of the first clause. This analysis properly captures the island sensitivity of this construction since it involves leftward scrambling; more importantly, this island sensitivity persists even if the gap of the first clause is replaced by an overt resumptive pronoun. 326

The present claims made regarding Japanese relativization and right dislocation have a significant implication on the directionality of movement. Given that no direct evidence for leftward movement in Japanese relativization and right dislocation seems available to children learning Japanese, it must be the case that UG principles are involved in determining the directionality of movement in these Japanese constructions. This will thus lend support to the so-called asymmetrical structure hypothesis, first proposed by Kayne (1994), according to which only leftward movement is allowed in UG. Alternatively, it will lend support to the hypothesis proposed by Saito (1995) and Fukui (1993), among others, that the direction of movement is constrained by the head-parameter in such a way that a movement operation should create a structure that is consistent with the value of the head parameter in a given language. According to this hypothesis, Japanese is allowed only leftward movement.

327

References

Abe, Jun. 1993. Binding conditions and scrambling without A/A' distinction. Ph.D. dissertation, University of Connecticut. Abe, Jun. 1997. LF undoing effects of scrambling. In Scrambling, ed. by Shigeo Tonoike, 35-59. Tokyo: Kurosio. Abe, Jun. 2012. Scrambling and operator movement. Lingua 122: 66-91. Abe, Jun. 2014. A movement theory of anaphora. Berlin: Mouton de Gruyter. Abe, Jun. 2015. The in-situ approach to sluicing. Amsterdam: John Benjamins. Abe, Jun. 2016. String-vacuity and LF interpretation in A-chains: Cases of ECM and nominative-genitive conversion. The Linguistic Review 33: 199-238. Abe, Jun. forthcoming. Minimalist syntax for quantifier raising, topicalization and focus movement: A Search and Float approach for Internal Merge. Berlin: Springer. Abe, Jun and Norbert Hornstein. 2012. ‘Lasnik-effects’ and string-vacuous ATB movement. In Ways of structure building, ed. by Myriam Uribe-Etxebarria and Vidal Valmala, 169-205. Oxford: Oxford University Press. Abe, Jun and Hiroto Hoshi. 1997 Gapping and P-stranding. Journal of East Asian Linguistics 6: 101-136. Abe, Jun and Hiroto Hoshi. 1999. Directionality of movement in ellipsis resolution in English and Japanese. In Fragments: Studies in ellipsis and gapping, ed. by

328

Shalom Lappin and Elabbas Benmamoun, 193-226. Oxford: Oxford University Press. Abe, Jun and Chizuru Nakao. 2012. String-vacuity under Japanese right node raising. In Proceedings of the Fifth Formal Approaches to Japanese Linguistics Conference, ed. by Matthew A. Tucker, Anie Thompson, Oliver Northrup and Ryan Bennett, 1-14. Cambridge, MA: MITWPL. Aoyagi, Hiroshi. 1994. On association with focus and scope of focus particles in Japanese. In Formal Approaches to Japanese Linguistics 1, MIT Working Papers in Linguistics 24, ed. by Masatoshi Koizumi and Hiroyuki Ura, 23-44. Cambridge, MA: MITWPL, MIT. Boeckx, Cedric, Norbert Hornstein and Jairo Nunes. 2010. Control as movement. Cambridge: Cambridge University Press. Bošković,

Željko.

1997.

Superiority

effects

with

multiple

wh-fronting

in

Serbo-Croatian. Lingua 102: 1-20. Bošković, Željko and Daiko Takahashi. 1998. Scrambling and last resort. Linguistic Inquiry 29: 347-366. Cecchetto, Carlo. 1999. Optionality and directionality: A view from leftward and rightward scrambling in Japanese. In Researching and verifying an advanced theory of human language, Report (3) for Grant-in-Aid for COE Research. Graduate School of Language Sciences, Kanda University of International Studies. Chao, Wynn and Peter Sells. 1983. On the interpretation of resumptive pronouns. In Proceedings of NELS 13, ed. by Peter Sells and Charles Jones, 47-61. Amherst, 329

MA: GSLA, University of Massachusetts. Chomsky, Noam. 1964. Current issues in linguistic theory. The Hague: Mouton. Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press. Chomsky, Noam. 1976. Conditions on rules of grammar. Linguistic Analysis 2: 303-351. Chomsky, Noam. 1977. On wh-movement. In Formal Syntax, ed. by Peter W. Culicover, Thomas Wasow and Adrian Akmajian, 71-132. New York: Academic Press. Chomsky, Noam. 1980. On binding. Linguistic Inquiry 11: 1-46. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Chomsky, Noam. 1982. Some concepts and consequences of the theory of government and binding. Cambridge, MA: MIT Press. Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press. Chomsky, Noam. 1991. Some notes on economy of derivation and representation. In Principles and parameters in comparative grammar, ed. by Robert Freidin, 417-454. Cambridge, MA: MIT Press. Chomsky, Noam. 1993. A minimalist program for linguistic theory. In View from Building 20: Essays in linguistics in honor of Sylvain Bromberger, ed. by Ken Hale and Samuel J. Keyser, 1-52. Cambridge, MA: MIT Press. Chomsky, Noam. 1995. Categories and transformations. In The minimalist program, by Noam Chomsky, 219-394. Cambridge, MA: MIT Press. Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Step by step: Essays in 330

honor of Howard Lasnik, ed. by Roger Martin, David Michaels and Juan Uriagereka, 89-155. Cambridge, MA: MIT Press. Chomsky, Noam. 2004. Beyond explanatory adequacy. In Structure and beyond: The cartography of syntactic structures, vol.3, ed. by Adriana Belletti, 104-131. Oxford: Oxford University Press. Chomsky, Noam. 2008. On phases. In Foundational issues in linguistic theory: Essays in honor of Jean-Roger Vergnaud, ed. by Robert Freiden, Carlos P. Otero and Maria Luisa Zubizarreta, 133-166. Cambridge, MA: MIT Press. Chomsky, Noam and Howard Lasnik. 1993. The theory of principles and parameters. In Syntax: An international handbook of contemporary research, ed. by Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld and Theo Vennemann. Berlin: Walter de Gruyter. Chung, Sandra, William A. Ladusaw and James McCloskey. 1995. Sluicing and Logical Form. Natural Language Semantics 3: 239-282. Emonds, Joseph E. 1976. A transformational approach to English syntax: Root, structure-preserving, and local transformations. New York: Academic Press. Epstein, Samuel D. 1992. Derivational constraints on A’-chain formation. Linguistic Inquiry 23: 235-259. Fiengo, Robert, C.-T. James Huang, Howard Lasnik, and Tanya Reinhart. 1988. The syntax of wh-in-situ. In Proceedings of the Seventh West Coast Conference on Formal Linguistics, ed. by Hagit Borer, 81-98. Stanford, CA: CSLI Publications. Fukui, Naoki. 1986. A theory of category projection and its applications. Ph.D. 331

dissertation, MIT. Fukui, Naoki. 1993. Parameters and optionality. Linguistic Inquiry 24: 399-420. Fukui, Naoki and Margaret Speas. 1986. Specifiers and projection. In MIT Working Papers in Linguistics 8, ed. by Naoki Fukui, T.R. Rapoport and E. Sagey, 128-172. Fukui, Naoki and Yuji Takano. 1998. Symmetry in syntax: Merge and demerge. Journal of East Asian Linguistics 7: 27-86. Goto, Nobu. 2011. Interplay between Agree and Move. Tohoku: Essays and Studies in English Language and Literature 44, 39-68. Sendai, Japan: Tohoku Gakuin University. Han, Chung-hye and Jong-Bok Kim. 2004. Are there “double relative clauses” in Korean? Linguistic Inquiry 35: 315-337. Haraguchi, Shosuke. 1973. Remarks on dislocation in Japanese. Ms., MIT. Hasegawa, Nobuko. 1981. A Lexical Interpretive theory with emphasis on the role of subject. Ph.D. dissertation, University of Washington. Hasegawa, Nobuko. 1997. A copular-based analysis of Japanese clefts: Wa-clefts and ga-clefts. In Researching and verifying an advanced theory of human language, Report for Grant-in-Aid for COE Research, 15-38. Graduate School of Language Sciences, Kanda University of International Studies. Higginbotham, James and Robert May. 1981. Questions, quantifiers and crossing. The Linguistic Review 1: 41-80. Hiraiwa, Ken. 2001a. On nominative-genitive conversion. In A few from Building E-39, MIT Working Papers in Linguistics 39, ed. by Elena Guerzoni and Ora 332

Matsushansky, 66-124. Cambridge, MA: MITWPL. Hiraiwa, Ken. 2001b. Multiple Agree and the Defective Intervention Constraint in Japanese. In Proceedings of the 1st HUMIT Student Conference in Language Research (HUMIT 2000), MIT Working Papers in Linguistics 40, ed. by Ora Matsushansky, Albert Costa, Javier Martín-González, Lance Nathan and Adam Szczegielniak, 67-80. Cambridge, MA: MITWPL. Hoji, Hajime. 1985. Logical form constraints and configurational structures in Japanese. Ph.D. dissertation, University of Washington. Hoji, Hajime. 1987. Japanese clefts and chain binding/reconstruction effects. Presented at WCCFL VI. Hoji, Hajime. 1998. Null object and sloppy identity in Japanese. Linguistic Inquiry 29: 127-152. Holmberg, Anders. 2000. Scandinavian stylistic fronting: How any category can become an expletive. Linguistic Inquiry 31: 445-483. Hornstein, Norbert. 1999. Movement and control. Linguistic Inquiry 30: 69-96. Hornstein, Norbert. 2001. Move!: A minimalist approach to construal. Oxford: Blackwell. Hornstein, Norbert. 2009. A theory of syntax: Minimal operations and universal grammar. Cambridge: Cambridge University Press. Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. Ph.D. dissertation, MIT. Inagaki, Daisuke. 1998. Hierarchy and linearity: A case study of postverbal construction 333

in Japanese. Ms., Kyorin University. Inoue, Kazuko. 1976. Henkei bumpoo to nihongo, Vol 1. [Transformational grammar and Japanese]. Tokyo: Taishuukan. Ishii, Yasuo. 1991. Operators and empty categories in Japanese. Ph.D. dissertation, University of Connecticut. Jacobson, Pauline. 1999. Towards a variable-free semantics. Linguistics and Philosophy 22: 117-185. Johnson, Kyle. 1991. Object positions. Natural Language and Linguistic Theory 9: 577-636. Kayne, Richard. 1985. Principles of particle constructions. In Grammatical representation, ed. by Jacqueline Guéron, Hans-Georg Obenauer, and Jean-Yves Pollock, 101-140. Dordrecht: Foris. Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Kim, Soowon. 1999. Sloppy/strict identity, empty objects, and NP ellipsis. Journal of East Asian Linguistics 8: 255-284. Kobayashi, Keiichiro and Hideki Maki. 2002. A non-exceptional approach to exceptional Case-marking in Japanese. English Linguistics 19: 211-238. Koizumi, Masatoshi. 1995. Phrase structure in minimalist syntax. Ph.D. dissertation, MIT. Koizumi, Masatoshi. 2000. String vacuous overt verb raising. Journal of East Asian Linguistics 9: 227-285. Kuno, Susumu. 1973. The structure of the Japanese language. Cambridge, MA: MIT 334

Press. Kuno, Susumu. 1976. Subject raising. In Syntax and semantics 5: Japanese generative grammar, ed. by Masayoshi Shibatani, 17-49. New York: Academic Press. Kuno, Susumu. 1978a. Japanese: A characteristic OV language. In Syntactic typology: Studies in the phenomenology of language, ed. by Winfred P. Lehmann, 57-138. Austin: University of Texas Press. Kuno, Susumu. 1978b. Danwa-no bunpoo [The grammar of discourse]. Tokyo: Taishuukan-shoten. Kural, Murat. 1997. Postverbal constituents in Turkish and the Linear Correspondence Axiom. Linguistic Inquiry 28: 498-519. Kuroda, S.-Y. 1965. Generative grammatical studies in the Japanese language. Ph.D. dissertation, MIT. Kuroda, S.-Y. 1971. Remarks on the notion of subject with reference to words like also, even or only. In the Annual Bulletin of the Research Institute of Logopedics and Phoniatrics 4, 127-152. Tokyo: University of Tokyo. Lasnik, Howard. 1999. Chains of arguments. In Working minimalism, ed. by Samuel David Epstein and Norbert Hornstein, 189-215. Cambridge, MA: MIT Press. Lasnik, Howard and Mamoru Saito. 1991a. On the subject of infinitives. In Papers from the 27th Regional Meeting of the Chicago Linguistic Society, ed. by Lise M. Dobrin, Lynn Nichols, and Rosa M. Rodriguez, 324-343. Chicago Linguistic Society. Lasnik, Howard and Mamoru Saito. 1991b. Curious correlations between configurations 335

licensing (or failing to license) HNPShift and those for gapping. Note, University of Connecticut. Lasnik, Howard and Mamoru Saito. 1992. Move α: Conditions on its application and output. Cambridge, MA: MIT Press. Mahajan, Anoop K. 1990. The A/A-bar distinction and movement theory. Ph.D. dissertation, MIT. Merchant, Jason. 2001. The syntax of silence: Sluicing, Islands, and the theory of ellipsis. Oxford: Oxford University Press. Miyagawa, Shigeru. 1993. LF Case-checking and Minimal Link Condition. In Case and agreement II, MIT Working Papers in Linguistics 19, ed. by Collin Phillips, 213-254. Cambridge, MA: MITWPL. Murasugi, Keiko. 1991. Noun phrases in Japanese and English: A study in syntax, learnability and acquisition. Ph.D. dissertation, University of Connecticut. Nakai, Satoru. 1980. A reconsideration of ga-no conversion in Japanese. Papers in Linguistics 13: 279-320. Nakao, Chizuru. 2009. Island repair and non-repair by PF strategies. Ph.D. dissertation, University of Maryland. Nishida, Mizuki. 1999. Multiple scrambling without multiple adjunction. Ms., Nagoya University. Nishigauchi, Taisuke. 1990. Quantification in the theory of grammar. Dordrecht: Kluwer Academic Publishers. Ochi, Masao. 2001. Move F and GA/NO conversion in Japanese. Journal of East Asian 336

Linguistics 10; 247-286. Oka, Toshifusa. 1988. Abstract Case and empty positions. Tsukuba English Studies 7, 187-227. University of Tsukuba. Oka, Toshifusa. 1989. On the SPEC of IP. Ms., MIT. Oka, Toshifusa. 1993a. Minimalism in syntactic derivation. Ph.D. dissertation, MIT. Oka, Toshifusa. 1993b. Shallowness. In Papers on Case & Agreement II, MIT Working Papers in Linguistics 19, ed. by Collin Phillips, 255-320. Cambridge, MA: MITWPL. Oku, Satoshi. 1998. A theory of selection and reconstruction in the minimalist perspective. Ph.D. dissertation, University of Connecticut. Perlmutter, David. 1972. Evidence for shadow pronouns in French relativization. In The Chicago which hunt: Papers from the Relative Clause Festival, ed. by Paul M. Levi, Judith N. Phares and Gloria C Peranteau, 73-105. Chicago: Chicago Linguistic Society. Polinsky, Maria and Eric Potsdam. 2002. Backward control. Linguistic Inquiry 33: 245-282. Postal, Paul. 1974. On raising. Cambridge, MA: MIT Press. Reinhart, Tanya. 1976. The syntactic domain of anaphora. Ph.D. dissertation, MIT. Rizzi, Luigi. 1990. Relativized minimality. Cambridge, MA: MIT Press. Rosen, Eric Robert. 1996. The postposing construction in Japanese. Master Thesis, University of British Columbia. Ross, John. 1967. Constraints on variables in syntax. Ph.D. dissertation, MIT. 337

Sabbagh, Joseph. 2007. Ordering and linearing rightward movement. Natural Language and Linguistic Theory 25: 349-401. Saito, Mamoru. 1985. Some asymmetries in Japanese and their theoretical implications. Ph.D. dissertation, MIT. Saito, Mamoru. 1989. Scrambling as semantically vacuous A’-movement. In Alternative conceptions of phrase structure, ed. by Mark A. Baltin and Anthony S. Kroch, 182-200. Chicago: The University of Chicago Press. Saito, Mamoru. 1992. Long distance scrambling in Japanese. Journal of East Asian Linguistics 1: 69-118. Saito, Mamoru. 1994. Additional-WH effects and the adjunction site theory. Journal of East Asian Linguistics 3: 195-240. Saito, Mamoru and Naoki Fukui. 1998. Order in phrase structure and movement. Linguistic Inquiry 29: 439-474. Sakai, Hiromu. 1994. Complex NP Constraint and Case-conversions in Japanese. In Current topics in English and Japanese, ed. by Masaru Nakamura, 179-203. Tokyo: Hituzi Syobo. Sakai, Hiromu. 1998. Raising asymmetry and improper movement. In Japanese/Korean linguistics: Volume 7, ed. by Noriko Akatsuka, Hajime Hoji, Shoichi Iwasaki and Susan Strauss, 481-497. Los Angeles: The Center for the Study of Language and Information Publications. Sano, Masaki. 2001. On the scope of some focus particles and their interaction with causatives, adverbs, and subjects in Japanese. English Linguistics 18: 1-31. 338

Sauerland, Uli and Paul Elbourne. 2002. Total reconstruction, PF movement, and derivational order. Linguistic Inquiry 33: 283-319. Sells, Peter. 1990. Is there subject-to-object raising in Japanese? In Grammatical relations: A cross-theoretical perspective, ed. by Katarzyna Dziwirek, Patrick Farrel and Errapel Mejias-Bikandi, 445-457. Stanford, CA: CSLI Publications. Simon, Mutsuko Endo. 1989. An analysis of the postposing construction in Japanese. Ph.D. dissertation, University of Michigan. Sloan, Kelly. 1991. Quantifier-Wh interaction. In More papers on wh-movement, MIT Working Papers in Linguistics 15, ed. by Lisa Cheng and Hamida Demirdash, 219-237. Cambridge, MA: MITWPL. Sohn, Keun-Won. 1994. Adjunction to argument, free ride and a minimalist program. In Formal approaches to Japanese linguistics 1, MIT Working Papers in Linguistics 24, ed. by Masatoshi Koizumi and Hiroyuki Ura, 315-334. Cambridge, MA: MITWPL. Stjepanović, Sandra. 1995. Short-distance movement of wh-phrases in Serbo Croatian matrix clauses. Ms., University of Connecticut. Tada, Hiroaki. 1993. A/A-bar partition in derivation. Ph.D. dissertation, MIT. Taglicht, Josef. 1984. Message and emphasis: On focus and scope in English. London: Longman. Takahashi, Daiko. 1993. Movement of wh-phrases in Japanese. Natural Language and Linguistic Theory 11: 655-678. Takano, Yuji. 1996. Movement and parametric variation in syntax. Ph.D. dissertation, 339

University of California, Irvine. Takano, Yuji. 1998. Object shift and scrambling. Natural Language and Linguistic Theory 16: 817-889. Takano, Yuji. 2002. Surprising constituents. Journal of East Asian Linguistics 11: 243-301. Takita, Kensuke. 2011. Argument ellipsis in Japanese right dislocation. In Japanese/Korean linguistics, vol. 18, ed. by William McClure and Marcel den Dikken. Stanford, CA: CSLI Publications. Tanaka, Hidekazu. 2001. Right-dislocation as scrambling. Journal of Linguistics 37: 551-579. Tanaka, Hidekazu. 2002. Raising to object out of CP. Linguistic Inquiry 33: 637-652. Terada, Michiko. 1987. Unaccusativity in Japanese. In Proceedings of North East Linguistic Society 17, ed. by Joyce McDonough and Bernadette Plunkett, 619-640. Amherst, MA: GLSA, University of Massachusetts. Tsai, Wei-Tien. 1994. On economizing the theory of A’-dependencies. Ph.D. dissertation, MIT. Uchibori, Asako. 2000. The syntax of subjunctive complements: Evidence from Japanese. Ph.D. dissertation, University of Connecticut. Uchibori, Asako. 2001. Raising out of CP and C-T relations. In Proceedings of formal approaches to Japanese linguistics 3, MIT Working Papers in Linguistics 41, ed. by María Cristina Cuervo, Daniel Harbour, Ken Hiraiwa, and Shinichiro Ishihara, 145-162. Cambridge, MA: MITWPL. 340

Vergnaud, Jean-Roger. 1974. French relative clauses. Ph.D. dissertation, MIT. Watanabe, Akira. 1996. Nominative-genitive conversion and agreement in Japanese: A cross-linguistic perspective. Journal of East Asian Linguistics 5: 373-410. Watanabe, Akira. 2000. Absorption: Interpretability and feature strength. In Researching and verifying an advanced theory of human language, COE research report (4), 253-296. Chiba, Japan: Graduate School of Language Sciences, Kanda University of International Studies. Wexler, Ken and Peter Culicover. 1980. Formal principles of language acquisition. Cambridge, MA: MIT Press. Yamashita, Hideaki. 2005. A probe-goal analysis of binding relations in Japanese. Ms., Yokohama National University. Yamashita, Hideaki. 2011. An(other) argument for the “repetition” analysis of Japanese right dislocation: Evidence from the distribution of thematic topic -wa. In Japanese/Korean linguistics, vol. 18, ed. by William McClure and Marcel den Dikken. Stanford, CA: CSLI Publications. Yatsushiro, Kazuko. 1996. On the unaccusative construction and nominative Case licensing. Ms., University of Connecticut.

341