ELSEVIEK

Lingua 109(1999)81-107 www.elsevier.nl/locate/lingua

Some consequences of Attract F* Masao Ochi* U- 1145. Department

of Linpistics, University Stows. CT 06269-l 14.5. USA

of Connecticut.

Received 1 February 1998; revised version 21 January 1999

Abstract This paper explores some consequences of Chomsky‘s (1995) Attract F theory. In particular, I propose a way to reconcile Attract F with some apparent evidence for Move. My analysis is based on a particular view of category movement proposed in Chomsky (1993, according to which when the category moves, two separate chains are formed: CHFF and CHcAr.

CHFF is formed via Attract, whereas CH cAT (i.e., the generalized pied-piping chain) is required only for PF convergence. Crucially, the CHCATis not formed via Attract; indeed, it does not even involve feature checking. I argue that this second chain bears the properties of Move. The idea is that once the formal feature F is attracted away from the category a, a is defective with respect to the PF interface. It therefore moves to the position where its missing feature F is located in order to remedy its own deficiency (hence Move rather than Attract). I will show that this hypothesis, within the theory of Attract, offers a way to account for reconstruction effects discussed by Barss (1986), which are often regarded as empirical evidence for Move. The proposed analysis also captures the lack of CED effects in languages such as Japanese, when applied in conjunction with the analysis of Takahashi (1994). I also motivate the necessity for certain cases of acyclic merger, which leads us toward a derivational characterization of the grammar. 0 1999 Elsevier Science B.V. All rights reserved. Ke_ywords: Attract:

pied-piping:

CED; Coordination; Cyclicity: Derivational Japanese; Locality; Relativized minimality

constraints;

Generalized

’ I would like to thank Brian Agbayani, ieljko BoSkoviC, Robert Freidin, Hideki Maki, Diane LilloMartin, Nobuhiro Miyoshi, Daiko Takahashi, two anonymous reviewers of Lingua, and especially Howard Lasnik for valuable comments. I also benefited from discussions with the audiences at University of Connecticut, the Sixteenth West Coast Conference on Formal Linguistics (University of Washington), and Challenges of Minimalism (University of Ottawa), where part of the material was presented. Finally, my sincere thanks go to Tien-Hsin Hsin for proofreading the entire manuscript. * E-mail: [email protected] 0024-3841/99/$ - see front matter 0 1999 Elsevier Science B.V. All rights reserved. PI: SOO24-3841(99)00012-l

82

M. Ochi I Lingua 109 (1999) H-107

1. Introduction The aim of this paper is to explore some consequences of Chomsky’s (1995 : ch. 4) proposal that what is actually affected by a given movement operation is a feature rather than an entire category. Chomsky further suggests that what triggers movement is a morphological requirement of the target rather than the category which moves: a target K with a formal feature to be checked off attracts some relevant feature(s). This Attract F theory, despite its conceptual elegance and empirical coverage for certain constructions, raises non-trivial questions for the theory of Universal Grammar. For instance, what is the nature of feature movement? Is it subject to the same set of constraints as category movement? Also, how are previous accounts of the Condition on Extraction Domain (CED) effects (cf. Huang, 1982) such as Takahashi (1994), which are based on Move, to be maintained under Attract? This paper attempts to answer such questions regarding the very nature of movement within the minimalist program. The following claims will be made in this paper. First, building on Chomsky’s ( 1995 : 265) proposal that overt category movement is a two step process producing two separate formal chains (CHR and CH,,,), I suggest that one of the chains (CH,,r) possesses the characteristic of Move (see section 3). In the context of such a hybrid theory of movement, incorporating both Attract and Move, the apparent evidence against Attract F vanishes. Secondly, certain cases of acyclic merger will be motivated in the proposed analysis, thus leading us to a derivational characterization of grammatical constraints (see section 4). The organization of this paper is as follows. Section 2 reviews Takahashi’s (1994) approach to CED effects based on the theory of Move. After pointing out conceptual problems this account faces under the theory of Attract, section 3 proposes a way to maintain Takahashi’s (1994) insight under the theory of Attract F. This section also demonstrates how the proposed analysis accounts for the (lack of) island effects in English and Japanese. Section 4 examines the nature of Takahashi’s Uniformity Corollary on Adjunction (see section 2.1) and proposes a way to derive its effects in a principled manner. In this connection, two lines of approach allowing acyclic merger will be introduced and discussed: Lebeaux (1988) and BoSkoviC and Lasnik (to appear). In section 5, additional consequences of the analysis are summarized. The conclusion is given in section 6.

2. Previous approaches to movement and locality 2.1. Takahashi (1994) Various attempts have been made within the principles-and-parameters approach to provide a principled account of the island effects shown below (cf. Huang, 1982; Chomsky, 1986a). (2a) and (2b) show Adjunct Condition and Subject Condition effects, respectively. (3) is an instance of a Wh-island Condition effect.

83

M. Ochi I Lingua 109 (1999) 81-107

( 1)

Whati does John think [that Mary likes ti] ? (2) a. ?*WhaG did John cry [after Mary bought t,]? b. ?*What, did [a picture of ti] irritate John? (3) ?*Whati did John wonder [whether Mary bought ti]?

(Adjunct Condition) (Subject Condition) (Wh-island)

Under the minimalist approach (cf. Chomsky, 1993), Takahashi (1994) argues that such island effects are derived through the interaction of the following two principles : (4) Uniformity Corollary on Adjunction (UCA) Adjunction is impossible to a proper subpart of a uniform group, where a uniform group is a non-trivial chain or a coordination (Takahashi, 1994: 25). (5) Shortest Movement Condition (SMC) Make the shortest movement (cf. Chomsky and Lasnik, 1993). Before examining how these principles exclude (2)-(3), let us follow Takahashi in motivating the UCA in (4) and the SMC in (5). I will discuss the UCA first. One thing to notice is that the statement in (4) includes disjunction. As will be argued below, however, this disjunction is well-motivated within the minimalist framework. The UCA is essentially based on the following idea, which is due to Chomsky (1991, 1994). (6) Chains are uniform. Given (6), Takahashi (1994: 20) suggests that uniformity is violated if some element adjoins only to some (but not all) members of a chain. Such a suggestion makes sense especially under the conception of movement as a copying operation, Suppose that the category o has formed a non-trivial chain as a result of movement, as shown in (7a). Now, suppose that an element p, which is contained in ~1,adjoins to the head of the chain al. This adjunction is shown in (7b). (7) a. (a,, 02) b. (1, P ], aJ1~a,) Apparently, uniformity is not observed in (7b). The UCA derives a ban on adjunction to subjects (cf. Chomsky, 1986a), if we assume the VP-internal subject hypothesis (cf. Kuroda, 1988; Fukui and Speas, 1986), according to which the subject in English raises from a VP-internal position and heads a non-trivial chain, which is then subject to the UCA (4). Hence, if an element contained in the subject adjoins to the subject, the UCA is violated. One question is: what exactly goes wrong if uniformity is violated? Takahashi does not elaborate on this issue. One conceivable answer is the following. At PF, chains created via movement are subject to deletion of copies (i.e., the non-head members of a chain).’ Assuming copy deletion, we ’ See Nunes (1995) for an attempt to deduce copy deletion from Kayne’s (1994) Linear Correspondence Axiom (LCA).

84

M. Ochi I Lingua 109 (1999) 81-107

might say that PF cannot delete non-head members of a chain if uniformity is not observed. In (7a), for instance, cr2is deleted at PF under identity with a,. PF cannot perform such an operation in (7b), since the two members of the chain are not identical. Then, an illegitimate PF object results, in the sense that the articulatory and perceptual (A-P) interface cannot interpret it. Thus, ‘uniformity’ required on a nontrivial chain may be reducible to a bare output condition imposed by the A-P interface. How about the uniformity required for coordination? One crucial ingredient for Takahashi’s (1994) analysis is the proposal of Davidson (1967) and Higginbotham (1985) that adjuncts involve coordination. For instance, the example in (8a) has the semantic representation in (8b), which is roughly paraphrased as ‘there is an event such that it was a walking by John and it is slow’, where adjuncts such as slowly are analyzed as predicated of events. Takahashi (1994: 24) suggests that this mapping from syntax to semantics is transparently obtained by assuming the LF representation in (8~) in which the sisters, VP, and the adjunct slowly, are predicated of the event argument which is generated under Infl. (8) a. John walks slowly. b. 3e [walk (John, e) & slow (e)] c. [IP Job [r 1 (e) [vpl[VP*ti walks1s~ow~YIII Takahashi (1994) argues that the UCA (4) coupled with the assumption that adjuncts involve coordination, derives the ban on adjunction to adjuncts (cf. Chomsky, 1986a) in a principled manner. The UCA states that adjunction is not possible to a subpart of a uniform group, such as a coordination. If an adjunct is one of the conjuncts within a coordination structure, as suggested above, then, adjunction to an adjunct violates the UCA. Thus, it is crucial for Takahashi’s analysis that adjuncts be regarded as part of coordination. 2 Yet as Howard Lasnik (p.c.) points out, this categorization holds only under semantic considerations. In purely syntactic terms, adjunction structures need to be distinguished from coordinated structures.3 Thus, in order for the above argument to go through, we would have to regard this aspect of ‘uniformity’ as a requirement on mapping to LF (or Post-LF). If ‘uniformity’ on (semantic) coordinated structure is lost in LF representations, then presumably the conceptual-intentional (C-I) interface cannot interpret such an object. In this sense, the effect the UCA imposes on coordination might be reducible to a bare output condition by the C-I interface. To summarize, it may well be that the disjunction in (4) can be attributed to distinct output conditions by two interfaces, A-P and C-I.

’

Coordinate Structure Constraint (CSC) effects (cf. Ross, 1967) will be discussed in section 5.2.

3 Progovac (1998) is an exception in this regard. She argues that adverbial adjuncts such as slowly involve syntactic as well as semantic coordination. For instance, walks slowly is analyzed as in (i), where the whole phrase is headed by a phonologically null conjunction head (which she calls &), which takes two conjuncts, the VP walks and the adverb slowly as its specifier and complement, respectively (see section 5.2. for discussion of coordinated structures). (i) John Ml fcp [vr walks] [W & [AdvPslowly]]]

M. Ochi I Lingua 109 (1999) 81-107

85

Let us turn to the SMC (5). Takahashi (1994) suggests that the SMC forces the movement of an element to proceed by adjoining to every maximal projection on its way to the target. 4 There is a rather strong empirical argument, originally due to Barss (1986), that movement indeed drops by intermediate positions, in accordance with the spirit of the SMC. (9a) is grammatical. although as (9b) shows, the anaphor himseEf is not licensed in its original position. According to Belletti and Rizzi’s ( 1991) analysis, which claims that Condition A can be met anywhere in the derivation, the anaphor is licensed in the intermediate stage of the derivation shown in (lo), where the wh-phrase containing himself lands in the specifier position of the embedded CP.” That the anaphor is licensed in this configuration is supported by the grammaticality of (11). (9)

a. Which picture of himselfi does John, think that Mary likes t? b. “Johni thinks that Mary likes a picture of himself,. (cf. Barss, 1986) (10) _ does John, think [cp [which picture of himselfilj that Mary likes tj] (11) Johni wonders [cp [which picture of himselfi], Mary likes $1. The following example further illustrates the point. ( 12) ? ? Which picture of himselfi does Johni wonder [whether Mary likes t] ? This example has the status of a typical Wh-island violation, but is no worse than that. The status of (12) suggests that the anaphor himself is indeed licensed in this example, just as in (9a). Takahashi (1994) claims that this fact provides strong empirical support for the SMC. The anaphor is licensed in (12) because the whphrase moves in a cyclic fashion as required by the SMC, adjoining to the embedded CP (and the matrix VP) among other sites, thereby creating a configuration for the licensing of himself, Let us now return to the examples in (2)-(3). The sentences in (2) are excluded by (4) and (5) in the following way. Consider first (2a), which shows an Adjunct Condition effect. Recall that if an element is adjoined to an adjunct, then the UCA is violated as the adjunct clause is a subpart of a uniform group (i.e. coordination in this case). Thus, the derivation shown in (13a), in which the movement of what makes

Crucially, for Takahashi (1994), the SMC is defined from the viewpoint of the element undergoing movement (i.e., Move). Once Attract (cf. Chomsky, 1995: ch. 4) is adopted, however, the SMC needs to be redefined from the viewpoint of the target. See section 2.2 on this important issue. ’ For Barss (I 986), the anaphor himself in (9a) satisfies Condition A, as its antecedent John ‘chain binds’ it in a local domain. The following (simplified) definition of chain binding is taken from Saito (1989: 186). (i) X chain binds Y = df X and Y are coindexed, and a. X c-commands Y, or b. X c-commands a trace of Z, where Z = Y or Z contains Y. The exact choice between the accounts of Belletti and Rizzi (1991) and Barss (1986) does not matter for our discussion.

86

M. Ochi I Lingua 109 (1999) 81-107

use of an adjunction to the adjunct clause, violates the UCA (4).6 If, on the other hand, what does not adjoin to the adjunct clause on the way to its target, as shown in (13b), then the SMC (5) is violated, and example (2a) is instead excluded in this manner.7~8 (13)

a. Whati did [rp t i [John [vp t i [vp cv [cp “t, [CP after [P ti [IP MW [VP t, [VP bought ~]]]]]]]]]] UCA (4) violated b. Whati did [rp ti [John [vp ti [vp cry [cp after [lp “ti [lp Mary [vp tl [vp bought tJ]]]]]]]] SMC (5) violated

One question is why extraction of an element out of the other conjunct, namely (a segment of ) VP, does not lead to a violation of the UCA or the SMC. Consider the grammatical example in (14a). If the movement of what involves adjunction to the VP, as in (14b), then the UCA is violated. Takahashi (1994: 70) suggests that movement operations may not adjoin to a segment of a category. Rather, adjunction targets a full category, its segment being invisible for adjunction operations. Hence, according to Takahashi (1994), what moves by adjoining to the full category VP (i.e., VP& and no violation of the UCA or the SMC occurs in (14~). (14) a. Whati did John [buy ti [after Mary left]]? b. Whati did John [VPZ t, [VPZ [VP, *ti [vpl buy ti ][after Mary left]]]]? C. Whati did John [ VPZ ti 1~~2 [VP, buy ti I[after MW leftIll? Although Takahashi’s suggestion is not implausible, it begs the question of why the movement operation has this character. In section 4, I will attempt to give a more

6 Takahashi (1994) does not use * to indicate a violation of the UCA, as I do in (13a); I use it simply for expository purposes. Note also that it is crucial under Takahashi’s framework that traditional ‘substitution’ into the specifier position of XP is regarded as involving adjunction to XP (or X’) (cf. also Kayne, 1994; Saito and Fukui, 1998). Thus, in (13a), movement of what through the projection of the CP headed by after involves adjunction to this CP (or C’). See Takahashi (1994: section 2.2) for discussion. ’ Takahashi (1994: 76) suggests that his account of Adjunct Condition effects extends to cover the Complex NP constraint (CNPC) violations such as (i), which involve extraction out of a relative clause, if we assume that relative clauses are adjuncts. (i) ?*What did you meet [the man [who wrote t]]? Takahashi (1994: 77) further argues that this account would apply to other cases of CNPC violations such as (ii), which involve extraction from the complement clause of the noun, assuming with Stowell (198 1: 200) that complement clauses of (derived) nouns are appositive and hence are adjuncts rather than complements (cf. also Chomsky, 1986a: fn. 30). (ii) would then also involve extraction out of an adjunct domain. (ii) ?*What did you hear [the rumor [that Mary bought t]]? See Takahashi (1994: section 3.2.7) for more discussion. * As Takahashi (1994: 68) acknowledges, the UCA does not bar movement of a conjunct itself. (i) a. *Who did you see t and Mary? b. *Who did you see Mary and t? Here movement of who does not involve adjunction to a subpart of a uniform group. I will not deal with this type of Coordinate Structure Constraint (CSC) violation in this paper.

M. Ochi I Lingua 109 (1999) 81-107

87

principled explanation for this asymmetry with respect to extraction from the two conjuncts. As for the Subject Condition violation in (2b), recall the role which the VP-internal Subject Hypothesis plays. According to this hypothesis, the subject in English heads a non-trivial chain, which must observe the UCA (4). Thus, adjunction to the subject is banned. (15a-b) demonstrate the interaction of the UCA (4) and the SMC (5) in ruling out (2b):. the derivation in (15a) violates the UCA; conversely, if the movement of what does not make use of adjunction to the subject (so that the UCA is observed), the SMC (5) is violated. In either case, (2b) is ruled out.’ (15) a. What, did LIP t, LIP [NP *ti [NP a picture of ti]]j [vr tj irritate John]]] UCA (4) violated b. What, did [rp ti [rp [NPa picture of *ti]J [vr tj irritate John]]] SMC (5) violated Now, let us turn to Wh-island effects in (3). Consider a step in a derivation shown in (16a), in which what adjoins to the matrix VP from the position adjoined to the embedded IP. This step violates the SMC, since the movement here does not adjoin to (or move through the specifier of) CP. (16) a. ... [vr what, Lvp... [c. whether [c, C cIpt, lip . . . b. ... [vp what, [vp ... [cp ti’ [cp whether [c, C [rp ti ]rp . . . Still another derivation shown in (16b) does not violate the SMC, however, given the definition of Equidistance in Chomsky (1993). According to the notion of Equidistance, the position adjoined to CP and spec-CP are equidistant from the position oft in (16b), since the CP-adjoined position and the spec-CP position are in the same minimal domain. The definitions of domain and minimal domain are shown below: (17) The domain of a head A (DOM (A)) is the set of nodes contained in the least full category maximal projection dominating A that are distinct from and do not contain A. (18) The minimal domain of a head A (MIN (A)) is the smallest subset K of DOM (A) such that for any G, G a member of DOM (A), some B, a member of K, reflexively contains it. Thus the derivation shown in (16b) does not violate the SMC. Nor does it violate the UCA, since the CP here is neither a member of a non-trivial chain nor a member of a coordinated structure. (3) would then be predicted to be grammatical, contrary to fact.

9 There is an alternative derivation which does not violate either the SMC or the UCA. Suppose that the movement of what takes place before the subject NP moves to the specifier position of IP. Then, nothing prevents the adjunction of whar to the subject NP, since the latter has not formed a non-trivial chain at this point of derivation. See Takahashi (1994) and Collins (1994) for discussion on how Economy rules out this derivation independently.

88

M. &hi I Lingua 109 (1999) 81407

Takahashi’s (1994: 60) solution is to modify (17) in such a way that the CPadjoined position and the spec-CP position are not equidistant from the position of t in (16b).‘O His revised definition of domain is shown below. (19) The domain of a head A (DOM (A)) is the set of nodes dominated by the least full category maximal projection dominating A that are distinct from and do not contain A. It follows from this revision that the minimal domain of a head is limited to its specifier, its complement, and a head adjoined to it. I’ Crucially, the CP-adjoined position and the spec-CP position are not in the same minimal domain. As a result, the derivation illustrated in (16b) is ruled out as a violation of the SMC, in accordance with the facts. There is one crucial respect in which Wh-island effects differ from other island effects under Takahashi’s account. While other islands are accounted for by the interaction of the two principles, the UCA and the SMC, the UCA is irrelevant for Wh-island effects. The latter is accounted for solely by the SMC.r2 This divorce between the two types of islands will be even clearer in section 3. 2.2. Chomsky (1995): Attract F More recently, however, Chomsky (1995: ch. 4) advances the hypothesis that movement is triggered solely by the need for the target K to check off its formal feature(s) by attracting the closest relevant feature F. Let us call this approach Attract F, which is defined as in (20). (20) Attract F K attracts F if F is the closest feature that can enter into a checking relation with a sublabel of K. (Chomsky, 1995: 297). One immediate theoretical consequence of Attract F is that the spirit of the SMC is directly incorporated into the definition of the operation Attract. From the viewpoint of the target K, what matters is that it attracts the closest relevant feature F. Hence, under Attract F, the SMC (or Minimal Link Condition for Chomsky, 1995) is redefined as follows.

‘” See Takano (1994) for an alternative approach to Wh-island condition effects under the minimalist framework. ” Takahashi uses French participial agreement facts as an empirical support for this modification. See Takahashi (1994: 59) for details. I2 This situation is reminiscent of the analysis presented by Huang (1982). For Huang, islands such as the Subject Condition and Adjunct Condition are due to the CED, but Wh-island effects do not fall under the CED, since the CP whose specifier position is filled by a wh-phrase is typically a complement and hence is properly governed. Therefore, Subjacency was independently called for in order to explain Whisland effects.

M. Ochi

I Linp.m 109 (1999) 81-107

89

(21) Minimal Link Condition (MLC) K attracts a only if there is no /3, p closer to K than a, such that K attracts p (Chomsky, 1995: 311). It is important to note that under this system, nothing forces the attracted feature F to make a shortest move (i.e. adjoin to every XP on its way to the target K). Hence, the feature F is attracted in one step to the position of K. As a result, Takahashi’s (1994) UCA in (4) virtually loses its force under Attract F, given that movement induced by Attract does not make use of intermediate adjunctions. As Chomsky (1995: section 4.5.5) claims, Wh-island effects (or more generally, the Relativized Minimality effects in the sense of Rizzi, 1990) follow naturally from Attract F. Consider the following example. (22)

? “Whati did John wonder [cp whether Mary bought ti] ?

The matrix interrogative C attracts the closest relevant feature in accordance with the nature of Attract noted above. The wh-feature of what is not the closest relevant feature; rather, whether’s wh-feature is the closest, with the result that the following derivation blocks that in (22): (23)

“Whetheri did John wonder t, Mary bought what?

This derivation, in which the relevant feature of whether is attracted to the matrix C, does not run afoul of the locality requirement of Attract F. Nonetheless, Chomsky (1995; cf. also Maki, 1995) suggests that this structure is not interpretable because the matrix C, which indicates a wh-question, and whether, a yes-no question operator, are not semantically compatible. In short, the .4ttract F hypothesis captures the effect of the Wh-island Condition straightforwardly.‘” However, other island effects such as Adjunct Conditions do not follow immediately under Attract F. Recall that those island effects are accounted for through the interaction of the UCA (4) and the SMC (5) under Takahashi’s account, but neither is relevant for the theory of movement under Attract. Further, recall that some data involving anaphor licensing show that movement drops by intermediate positions on its way to the final landing site, thus providing support for the SMC (5). I repeat the relevant example below. (24) Which picture of himself, does John, think that Mary likes t? The fact that the anaphor is licensed in this example is accounted for if the movement of the wh-phrase containing the anaphor himself moves successive cyclically, ” Problematic cases are those like the example in (i). It is not clear what prevents a single wh-phrase from checking strong features of C more than once. Descriptively, an element moved into an operator position (such as specifier of CP) cannot be further attracted. See Maki (1995) for discussion. (i) *Who, do you wonder [cp t, [t, bought what]]?

90

M. Ochi I Lingua 109 (1999) 81-107

in accordance with the spirit of the SMC (5). Under the theory of Attract, it is not obvious how this fact can be accommodated. Thus there seem to be empirical difficulties for Attract.

3. Proposal: Two movement hypothesis My proposal is based on a particular view of category movement, suggested by Chomsky (1995 : 265). Assuming that what triggers movement is the morphological requirement that some feature of a head be checked, Chomsky claims that the optimal way to satisfy such a requirement is to affect features - only features - via movement. He suggests further that generalized pied-piping is required by properties of the phonological component. If, for instance, formal features (FFs) of a lexical item a are affected (attracted), then the FFs of a and the remnant of the category a are not pronounceable. In Chomsky’s terms, “isolated features and other scattered parts of words may not be subject to [PF] rules, in which case the derivation is canceled” (Chomsky, 1995: 262-263).14 Given these considerations, Chomsky assumes that when the whole category moves, (at least) two chains are created, as shown in (25). When the relevant feature F of the category a is attracted by the target K (Attract F), the whole set of formal features (FF) of a is carried along, forming the chain CH, in (25a).i5 Subsequently, if the operation is overt, generalized pied-piping is required for PF convergence (because the phonological component cannot cope with a lexical item whose features are scattered). This pied-piping chain is shown in (25b). (25) a. CHFF = (FF[F], tFF) b. CHCAT= (a, t,) Notice that the two chains in (25) have quite different characteristics. CHFF is created by Attract. On the other hand, it is clear that CHCATis not formed via Attract: in fact, it does not even involve feature checking. Thus, we find a formal chain which does not result from Attract under Chomsky ‘s (1995) framework. Then, what is the driving force for the formation of CHCAT? Chomsky does not elaborate on this question, merely stating that CH,,r is necessary (solely) for PF convergence: a lexical item whose features are scattered is not pronounceable. Building on Chomsky’s idea that PF convergence alone is relevant for the formation of CH,,,, I suggest that the driving force for the formation of CHCATlies in the PF deficiency of the category a whose formal feature is attracted away from it.i6 That is, the category a, being defective for the PF interface (i.e., being unpronounceable with its features scat-

I4 Thus when PF considerations are irrelevant, as is the case with covert movement, there is no need for the pied-piping. Consequences of this reasoning will be discussed shortly. See sections 3.1 and 3.2. Is I leave aside the question of whether or not a single feature can be affected by a movement operation, stranding the rest of the formal features. I6 I am grateful to Howard Lasnik (pc.) for this specific point.

M. Ochi I Lingua 109 (1999) 81-107

91

tered), moves to the position where its missing feature is located so that it can then be pronounced. Under this analysis, it then follows that CH,,, is motivated by (something like) ‘Greed’: a moves to overcome its own PF inadequacies. Suppose that this chain therefore has the property of Move (as opposed to Attract), thus being subject to the SMC (5) as defined from the viewpoint of the element undergoing movement. In this sense, the proposed analysis advocates a hybrid theory of movement, incorporating both Attract and Move. I7 Specifically, feature movement is driven by the deficiency of the target (hence Attract), while (generalized) pied-piping is motivated solely by the (PF) deficiency of the category undergoing movement (hence Move): the locus of the deficiency determines the ‘type’ of operation (or movement) involved. Let us consider this idea in more detail by looking at the following example : (26) a. What did you eat? b. [C Erryou [vp eat what1

I [Ql

[... FF:,,at ...I

CH, LcpwhatfcC LIP tLpyou I

C.

I

[VP t [VP Cat

III

t111111 CHc,q

Assuming that the interrogative C in English has a strong feature to be checked off, an overt movement is involved here. More specifically, two chains are formed as shown in (26b) and (26~). I assume that CHFF in (26b) is a two-member chain, in accordance with the spirit of Attract: all that is required here is that the target C attracts the ‘closest’ relevant feature, and hence the feature (or FFs) of what does not move in a successive cyclic fashion. In addition to CHFF, since this operation is overt in English, CHcAT is necessary to ensure PF convergence. According to this hypothesis, the category whar in (26), being defective for the PF interface, moves to the position where its missing feature is located so that it can be interpreted by PF rules: otherwise, the derivation would crash at PF. Let us reexamine (24) under the proposed analysis, repeated below. (27) Which picture of himselfi does John, think that Mary likes t? As shown in (28a), the matrix C attracts the closest relevant feature, the wh-feature of which. Then, the whole phrase which picture of himself, being defective for the PF interface, moves to the position where the attracted features of which are located. If this pied-piping movement is successive cyclic, in accordance with the SMC (5), then Barss’s (1986) account (cf. discussion in 2.2) is maintained under Attract.

”

See Maki (1995) for an earlier attempt

to develop

a hybrid

account

of movement.

92

M. Ochi I Lingua

109 (1999) 81-107

(28) a. [C [John thinks that Mary likes [which picture of himself]] I

[.. FL ..] 1C&F

[Ql I

b. [which picture of himself [C [Ip t [Ip John [vp t [vp thinks [cp t [c. that LIIII IIp t lip Mary [vp t [VPlikes t111111111111 --I I I CH,,, In the following subsections, I will demonstrate that this hypothesis enables us to explain the fact that CED effects are observed in English, but not in Japanese. 3.1. Adjunct Condition effects As reported in the literature (cf. Nishigauchi, 1986), English wh-movement, but not its Japanese counterpart, is constrained by the CED such as Adjunct Condition. (29) a. ?*Whati did John cv [after Mary bought ti]? b. John-wa [Mary-ga nani-o katta ato] kaetta no? John-TOP Mary-NOM what-ACC bought after left Q ‘?*What, did John leave [after Mary bought ti]?’ Let us consider the Japanese case first. Assuming that the wh-feature of nani ‘what’ is attracted at LF, only CH, is formed. Crucially, CHcAr need not be formed (hence cannot be formed by economy), since PF considerations are irrelevant for LF movement.” There is no problem with this attraction, since the MLC as defined in (21) is observed: there is no closer relevant feature than that of nani ‘what’ from the viewpoint of the target C. Hence, a two-member chain is formed and the derivation converges without any problem. (30)

[John-TOP [a Mary-NOM what-ACC bought after] left C] (a = adjunct clause) I

I

I.. FF,,,, ..I

‘s See section 3.2 for evidence that Japanese wh-in-situ constructions in fact involve movement. Note that according to Watanabe (1992), examples with nuni ‘what’ involve an overt null operator movement. This analysis may achieve the same desired result: since phonology is irrelevant for null operators, the pied-piping of the whole category should be unnecessary (cf. Takahashi, 1997). See Ishii (1997: section 5.3.2) and Maki and Ochi (to appear) for exploration of the hypothesis that wh-feature movement in Japanese takes place overtly. In this section, I assume that Japanese wh-movement takes place covertly. Also, the reader is referred to Watanabe (1992: 59, fn. 46) (also Maki, 1995: section 2.3), where an empirical problem with Watanabe’s approach, which allows the null operator to be base-generated in a position distant from the location of nani ‘what’, is discussed.

M. Ochi

I Lingua 109 (1999) 81-107

93

This view of movement accounts for the absence of Adjunct Condition (or more generally the CED) effects in Japanese wh-movement.‘” Let us now turn to the English case. As the relevant movement is overt in English, two chains are formed. As far as CH,r is concerned, no problem arises. As shown in (31a), the feature of what is the closest from the viewpoint of the matrix C, and attraction is therefore successful. However, the remnant movement causes a violation of the UCA (4) if it obeys the SMC (5), as shown in (3 1b). If the remnant movement observes the UCA by not adjoining to the adjunct clause, then the SMC is violated (31~). (3 1) a. [cp F [rp John cry [, after Mary bought what]]]

I [ .. FFwhat...I

IQ1 1 b. Lcpwhat ic, C ],P t ]rr John [VP

J CHFF t [vp cry

[,

*t L after

[IP t LIP Mary

[VP t [VP

L_____lllU”L

bought t]]]]]]]]]]]] (a = adjunct) UCA (4) violated C&AT c.

kp

what

ic, C Lp t Lp

John Lvpt [VP

cry

L after

LIP *t LIP Mary

[VP t [VP

uI-2-I

bought t]]]]]]]]]]]] (a = adjunct) SMC (5) violated (-&AT Thus, under the proposed account, a CED effect obtains when the pied-piping chain (i.e. CHc,r) is formed across an adjunct domain.2” Note that to the extent that the proposed analysis relies on Takahashi’s (1994) UCA (4) it faces the same problem as Takahashi’s (cf. section 2.1): why is it that extraction out of (a segment of) the VP is allowed although this segment too constitutes a proper subpart of a uniform group, namely, a coordination? I will come back to this important issue in section 4. ”

Similarly, there is no Subject Condition effect with argument wh-in-situ in Japanese, as (i) shows. (i) [John-ga nani-o katta koto]-ga minna-0 odorokaseta no? John-NOM what-ACC bought fact-NOM everyone-ACC surprised Q ‘?*What did [the fact that John bought t] surprise everyone?’ This fact is consistent with our analysis. The interrogative C attracts the closest relevant feature, namely, that of nani ‘what’ and hence the attraction is successful. Yet, as occasionally noted in the literature (cf. Kayne, 1984; Lasnik and Saito, 1992, to name a few), Japanese lacks Subject Condition effects even for overt movement (such as scrambling). Thus, the grammaticality of (i) is consistent with, but does not necessarily count as evidence for, the proposed analysis. The real question is why Subject Condition effects are absent altogether in Japanese. According to Takahashi (1994: 65), for instance, the absence is due to the fact that the subject in this language may optionally stay within VP in overt syntax, which means that it does not always form a non-trivial chain. Then, even overt extraction out of such a domain is possible without violating either the UCA or the SMC. X) During the writing of this paper, it was brought to my attention that Agbayani (to appear) is independently proposing a very similar analysis of the CED effects based on Chomsky’s (1995) two chain hypothesis, although the two analyses differ somewhat in the manner in which the idea is executed. See Agbayani (to appear) for more details.

94

M. Ochi I Lingua 109 (1999) 81-107

3.2. Wh-island effects The fact that the Wh-island effect is detected in Japanese as well as in English follows from the definition of Attract, as in Chomsky (1995) and Maki (1995). (32) a. ?*What, did John wonder [whether Mary bought ti]? no? katta kadooka] siritai b. ??John-wa [Mary-ga tram-0 John-TOP Mary-NOM what-ACC bought whether want-to-know Q ‘?*What does John want to know [whether Mary bought t]? Since the Q-feature of whetherlkadooka is closer to the matrix C than that of the whphrase whatlnani, it is impossible for the latter to be attracted by the matrix C.21 (33) a. [C [tr John wonder [whether Mary bought what]]]

rb

I [FFI

&I

b. John-TOP [Mary-NOM what-ACC bought whether] want-to-know Q

[FIFI

I IQ1

Further, the discussion here would lead us to suppose that wh-in-situ in English does not involve covert movement. As observed by Baker (1970), examples like the following are ambiguous with respect to the interpretation of the in-situ wh-phrase. The fact that what can take matrix scope would not be expected if it is subject to covert movement. (34) Who wonders where John bought what? I assume with Tsai (1994: 58) and Reinhart (1995) that (argument) wh-in-situ in English is licensed via unselective binding. The lack of feature movement of what in (34) is in fact expected under Attract. Since the morphological requirement of the matrix C is satisfied by attracting the closest wh-feature (of who), there is no need for the C to attract another wh-feature.22,23

2’ Similarly, the following example from Nishigauchi (1986) demonstrates the effect of the Wh-island condition. nani-o tabeta ka] sitteiru no? (i) Kimi-wa [dare-ga Q know Q you-TOP who-NOM what-ACC ate ‘Do you know who ate what?’ Not ‘Who is the person x such that you know what x ate?’ Not ‘What is the thing x such that you know who ate x?’ This example is interpreted only as a matrix yes-no question, and neither wh-phrase can take matrix scope. 22 The same situation holds in Japanese as well. (ia) is a typical Wh-island configuration. In contrast, (ib) shows that presence of an additional wh-phrase outside the island improves the example. This fact

M. Ochi I Lingua 109 (1999) 81-107

9s

To summarize this section, I have proposed that movement operation involves both Move and Attract in the sense that two chains, each of different nature, are involved when the category moves, an idea which stems from Chomsky’s (1995) view of category movement. Note that in the proposed analysis, the Wh-island Condition and the CED-type islands (Adjunct Condition and Subject Condition) are clearly given a separate treatment. While the former concerns only CHFF (i.e. the definition of Attract), the latter arise only when CH cAT is formed across an adjunct domain. Furthermore, the analysis outlined here has the following consequence for the theory of feature movement under Attract. The locality of pure feature movement is looser than that of category movement (i.e., feature movement plus generalized pied-piping). As we saw, feature movement itself is subject only to the Relativized Minimality type islands (cf. Rizzi, 1990) such as Wh-island: it is immune to the CED. On the other hand, overt movement has a pied-piping chain (CH,,,), which is sensitive to the CED-type islands (due to the interaction of the UCA (4) and the SMC (5)).

4. Toward a derivational characterization

of the grammar

In this section, I reevaluate Takahashi’s UCA and propose a way to derive its effects without stipulation. The outcome of the discussion leads us to a derivational characterization of grammatical constraints. is expected under Attract. The matrix C attracts the closest relevant feature, which is the feature of dare ‘who’ in (ib). The lack of Wh-island effects in (ib) shows that the feature of nani ‘what’ need not undergo movement at all. Rather, it is licensed via unselective binding. (i) a. ??John-ga Mary-ni [Taro-ga nani-o katta kadooka] tazuneta no’? John-NOM Mary-DAT Taro-NOM what-ACC bought whether asked Q ‘?*What did John ask Mary [whether Taro bought t]‘? b. John-ga dare-ni [Taro-ga nani-o katta kadooka] tazuneta no? John-NOM who-DAT Taro-NOM what-ACC bought whether asked Q ‘Who did John ask t [whether Taro bought what]?’ (cf. Watanabe, 1992) 33 According to Huang (1982), Chinese (argument) wh-in-situ is not sensitive to any type of island, including Wh-island. (i) Adjunct island John [zai Mary mai-le sheme yihou] likai-le’! John [at Mary buy-ASP what after] leave-ASP ‘?*What did John leave after Mary bought?’ (ii) Wh-island Ni xiang-zhidao [shei mai-le sheme] ? you wonder who buy-ASP what a. ‘Who is the person x such that you wonder what x bought?’ b. ‘What is the thing x such that you wonder who bought x‘?’ This fact suggests that Chinese interrogative C, unlike its English and Japanese counterparts, does not attract a wh-feature at any point in the derivation, and (argument) wh-in-situ is simply licensed by unselective binding. Thus, cross-linguistically, an interrogative complementizer may attract a wh-feature overtly as in English (due to the presence of a strong feature), covertly as in Japanese (due to the lack of a strong feature), or may not attract at any level as in Chinese (due to the lack of the relevant feature to begin with). This set of distinctions is consistent with the view that language variation is limited to properties of lexical items, in particular, functional elements (cf. Fukui, 1995, among others).

M. Ochi I Lingua 109 (1999) 81407

96

4.1. Acyclic merger Let us now return to the issue raised in section 2.1 with respect to Takahashi’s (1994) analysis of Adjunct Condition effects. Assuming that VP and an adjunct clause comprise a coordination(-like) structure, the question that arose was why extraction out of the adjunct clause yields a violation of the UCA (or the SMC) while extraction out of the other conjunct, (a segment of) the VP, does not. (35) Whati did John [buy ti [after Mary left]]? Takahashi’s solution, which resorts to a distinction between a full category and a segment of a category, begs the question of why this should be so. I suggest instead that this asymmetry follows directly from the following claim: (36) Acyclic merger of certain items is allowed. There are at least two lines of work in the literature which would give us this effect. One line of approach (cf. Lebeaux, 1988; Chomsky, 1993) is motivated by the following well-known asymmetry with respect to reconstruction effects between complements and adjuncts (cf. Freidin, 1986). (37a), in which the R-expression John is contained in an adjunct within the fronted wh-phrase, is fine with John and he being coreferential. In contrast, (37b) is judged to be ungrammatical under the relevant reading, where John is part of the complement clause within the fronted whphrase.24 (37) a. [Which claim that Johni made] did he, deny? b. *[Which claim that John, was asleep] did he, deny? Lebeaux (1988: 151) claims that this contrast is accounted for by assuming the following. (38) a. Adjuncts can be inserted acyclically. b. Condition C applies throughout the derivation. According to Lebeaux, (37b) violates Condition C before wh-movement takes place. Its derivation is illustrated below. Note that the clause that John was asleep must be inserted cyclically as it is analyzed by Lebeaux (1988) as a complement to N. (39) a. he deny which claim that John was asleep * Condition C violation b. Wh-movement which claim that John was asleep did he deny t

24

This judgment

is questioned

in the recent literature,

a fact which will be discussed

shortly.

h4. Ochi I Lingua

109 (1999)

81-107

91

In contrast, since adjuncts such as relative clauses need not be introduced cyclically, it is possible for that John made in (37a) to be merged into the structure after whmovement has taken place. There is no Condition C violation in the derivation illustrated below. (40) a. [he deny which claim] [that John made] b. Wh-movement of which claim [which claim did he deny t] [that John made] c. Acyclic merger of the relative clause which claim that John made did he deny t I adopt Lebeaux’s analysis at this point, and will turn shortly to the other line of approach - which I will ultimately adopt - in BoSkoviC and Lasnik (to appear). Let us now return to the example in (39, repeated below. (41) What, did John [buy ti [after Mary left]]? Once Lebeaux’s analysis is adopted, the following derivation should be possible: the adjunct clause is inserted after wh-movement has taken place. (42) a. Attract F ]C LIPJohn [“Pbuy what111

I

[.. &,a, ..I C&F

[Ql -

b. CHcAT formed h-r what [did [ip t ]IPJohn Lvpt

L

I

[VPbuy tllllll L----J C&AT

c. Merger of the adjunct clause [cp what [did LIPt IIpJohn Lvpivpt lvpbuy t I[after Mary left1111111 The idea is that when what is extracted out of the matrix VP, the adjunct clause is not merged with the VP yet. Derivationally speaking, therefore, the category movement of what does not involve adjunction to a proper subpart of a uniform group, although the resulting representation may look like a violation of the UCA. The discussion here therefore suggests that grammatical constraints like the UCA should be characterized derivationally as opposed to representationally.‘s,Z6

?s However, the problem would still remain if uniformity imposed by tures is checked at LF, as suggested in section 2.1. The structure of (41) ble (42c), and thus the UCA should be violated in the LF representation; to a proper subpart of coordination. I must leave this issue open. x See also Murasugi and Saito (1994) for an independent argument should be characterized in derivational terms (their argument is based

the UCA on coordinated strucentering into LF would resemwhat (or its copy) is adjoined that a constraint like the WA on some curious properties of

98

M. Ochi I Lingua 109 (1999) N-107

Note that this derivational line of approach continues to capture Adjunct Condition violations (and is able to do so without appeal to a distinction between a full category and a segment of this category). (43) ?*Whati did John

CY

[after Mary bought ti]?

Assuming that the operation Attract requires a c-command relation between an attractor and an attractee, the adjunct clause must be merged with the rest of the structure in order for what to be attracted by the interrogative C in this example. Hence extraction of what (or formation of CH,,, in connection to its movement) from within an adjunct always results in a violation of either the UCA (4) or the SMC (5). 4.2. Some questions and discussions Thus far, I have adopted Lebeaux’s (1988) analysis in order to offer an alternative to one potential problem for Takahashi’s analysis. However, the empirical ground on which Lebeaux’s analysis is based has been called into question in recent literature (cf. Kuno, 1997; Lasnik, 1998; Postal, 1997). Recall that his analysis is based on the argument/adjunct asymmetry shown in (37), repeated below as (44). (44) a. [Which claim that Johni made] did he, deny? b. *[Which claim that Johni was asleep] did he, deny? While an asymmetry certainly exists between the two examples above, Lasnik (1998) supplies a crucial fact regarding a sentence such as (44b): it is pragmatically odd; in order to be felicitous, such a question must be uttered in a situation in which there is more than one claim according to which John was asleep, and he denied some but not all of them.27 Once the relevant pragmatic factors are controlled, the contrast disappears, according to Lasnik. For example, questions such as (45) are acceptable, on a par with (44a).28 (45) a. [Which piece of evidence that John, was guilty] did he, successfully refute? b. [How many arguments that John’s, theory was correct] did hci publish? (Lasnik, 1998) In fact, this set of facts is not surprising if Stowell (198 1) is correct in claiming that what is analyzed as a clausal complement to N (such as that John was guilty in scramblingin Japanese). Note that the derivational view adopted in the text would lead us to adopt a particular analysis of the coordinated structure. See section 5.2 for discussion. ” The example is thus odd even if he and John do not corefer, according to Lasnik. r8 Here are some additional examples from Kuno (1997) illustrating the same point, (i) a. [Whose allegation that John, was less than truthful] did he, refute vehemently? b. [Whose claim that the Senator, had violated the campaign finance regulations] did he, dismiss as politically motivated?

M.

Ochi

I Lingua

109 (1999)

81-107

99

45a) is in fact an appositive clause, which is an adjunct. However, we then lose the empirical basis for Lebeaux’s (1988) analysis. To the extent that this kind of objection is on the mark, it undermines Lebeaux’s (1988) claim that adjuncts can be inserted acyclically. A more extensive evaluation of the relevant data may be necessary to determine the validity of the empirical basis of Lebeaux’s (1988) account, which is beyond the scope of this paper. Nonetheless, the discussion above indicates that resorting to Lebeaux’s (1988) approach may not be the right answer for present purposes. In order to maintain our analysis incorporating acyclic Merger while avoiding the empirical problems of Lebeaux (1988), let us now turn to an alternative line of proposal put forth in BoSkoviC and Lasnik (to appear). This latter work is primarily concerned with certain redundancies within Chomsky’s (1995) system with respect to strict cyclicity. There are at least two distinct devices which induce the strict cycle for Chomsky (1995). One is the Extension Condition, which requires that both Merge and Move take place at the root of the tree; this condition has the effect of excluding acyclic operations. In addition to this specific condition, a similar result is obtained by means of the following definition of feature strength:29 (46) Suppose that the derivation D has formed X containing c1with a strong feature F. Then D is canceled if c1is in a category not headed by a (Chomsky, 1995: 234).

‘9 As pointed out by Ishii (1997) among others, the condition as stated in (46) entails that strong features of a head located at the root of the structure need not be checked. For instance, the C in (ib) need not have its strong feature checked off, thus wrongly predicting that (ia) is grammatical. (i) a. *John bought what? b. [cp C John bought what] Under an analysis allowing certain instances of acyclic merger, it may be possible to restate (46) as follows. (ii) Suppose that the derivation D has formed Z containing a with a strong feature F. Then D is canceled if F is not checked off by the very next operation. This restatement correctly excludes (ia), assuming that Spell-Out counts as an operation (this assumption is necessary to block a derivation in which Spell-Out applies once (ib) is constructed, in which case the relevant feature-checking takes place covertly). Of course, Chomsky has a specific reason not to adopt something like (ii) over (46). He is concerned with cases like (iii). (iii) [Ip John, probably [r has [vp t, left]]]. If prohubly is merged with Infl before John is, as Chomsky assumes, (ii) would predict that the derivation is canceled: any strong feature(s) of Infl will not be checked off by the next operation after Infl is introduced into the structure. But given Lebeaux’s (1988) claim that adjuncts can be inserted acyclically, it should be possible to move John first, thus satisfying (ii), and subsequently insert prohahly. (iv) a. [Ip John, [r has [vp t, left]] (checking off the strong feature(s) of Infl) b. [Ip John, probably [r has [vp ti left]] (acyclic insertion of probably) This acyclic insertion might enable us to maintain something like (ii), which is empirically preferable to (46). Still, there are other questions with (ii). For example, if a head has more than one strong feature which must be checked off against two distinct lexical items (for example, Tense in some language may have a strong D-feature and a strong V-feature, thus attracting a DP and a V overtly), then it is not obvious how those features can be checked off by the very next operation after such a head is merged into the tree. I will not pursue this issue any further. Note that BoSkoviC and Lasnik’s (to appear) analysis is crucially based on (46), not (ii), as will be clear shortly.

M. Ochi I Lingua

100

109 (1999)

81-107

Like the Extension Condition, (46) also has the effect of preventing strong features from being checked off by acyclic operations. As BoSkoviC and Lasnik (to appear) observe, the effects of the Extension Condition and (46) thus overlap in certain contexts. As we saw in section 3.2, Wh-island effects follow from the definition of Attract under Chomsky (1995). In an example such as (47), the matrix C fails to attract the wh-feature of what, since there is a closer relevant feature, namely, that of whether: (47) ?*Whati did John wonder [whether Mary bought ti]? Notice that if whether (or the embedded interrogative C) can be inserted acyclically, then it should be possible for (the wh-feature of) what to be attracted by the matrix C before whether (or the embedded interrogative C ) is introduced into the tree, and we would thus lose the account of the Wh-island effects. As BoSkoviC and Lasnik point out, both the Extension Condition and (46) exclude this derivation, and a redundancy arises in this respect. Pointing out several conceptual and empirical problems with the Extension Condition, 3o BoSkoviC and Lasnik propose to do away with it and define the cycle solely in terms of (46). One consequence of this move is the following generalization: (48) Acyclic merger of an element E is in principle possible if E contains no strong features.3’ Returning now to the derivation in (42) repeated below as (49), (48) also provides the necessary ingredient for our analysis. Assuming that insertion of an adjunct clause is not triggered by a need to check off a strong feature, the adjunct clause after Mary left can be introduced into the tree acyclically. (49) a. Attract F [C [IP John [VP

I [Ql U

buy

what111

[.. FF1hat-1 CH,

so For instance, head-movement and LF movement are problematic for the Extension Condition (cf. also Chomsky, 1995: 327, for discussion). r’ BoSkoviC and Lasnik (to appear) assume that theta roles are formal features that trigger operations like Merge and Move. Further, following BoSkoviC and Takahashi (1998), it is assumed that the theta features in English are strong (as opposed to languages like Japanese, in which theta features are not strong). Under this line of assumption, (46) bans acyclic insertion of arguments in English. See BoSkoviC and Lasnik (to appear) for more detail. ‘* As BoSkoviC and Lasnik (to appear) note, their proposal has the additional welcome consequence of reconciling Chomsky’s two movement hypothesis (which I adopted throughout this paper) with a set of facts from languages such as French discussed by BoSkoviC (to appear), which at first blush appear to contravene Chomsky’s proposal. Although I cannot go into details here for lack of space, this aspect of BoSkoviC and Lasnik’s (to appear) analysis constitutes an independent reason to adopt it over Lebeaux (1988). See BoSkoviC and Lasnik (to appear) for details.

M. &hi

I Lingua

109 (1999)

81-107

101

b. CH,,, formed fcp what [did [IP t CIPJohn [VPt [VPbuy tlll111 I I I CH,,, c. Merger of the adjunct clause [cp what [did [rp t [,p John [vp t [vp buy t l-after Mary leftlllllll To summarize, I first motivated the necessity of acyclic merger in order to provide a principled explanation for the fact that extraction out of the (segment of) VP is allowed, despite the fact that it exactly resembles an adjunct clause in constituting a proper subpart of a uniform group, namely, a coordination structure. I reviewed two lines of approach in the literature which allow certain cases of acyclic merger. For the reason discussed above, the analysis of BoSkoviC and Lasnik (to appear) is chosen over that of Lebeaux (1988). The fact remains, however, that the insight as formulated in Lebeaux (1988) suffices equally well to capture the particular facts under discussion.

5. Further consequences of the analysis In this section, I will discuss some consequences of the proposed analysis for the theory of grammar. 5.1. On the content

of formal

features

The proposed analysis bears crucially on the issue concerning the content of formal features (FF) of a lexical item (Ll). There are two different views regarding whether the FF-bundle of a lexical item includes features relevant for binding and control (among others). Chomsky (1995: section 4.4.5) claims that the property of Ll relevant for binding and control resides within the FF of Ll. Lasnik (1995), on the other hand, argues that such properties are not part of the FF of Ll. Thus, for Chomsky, feature movement suffices to create new binding/control possibilities. Lasnik (1995) argues against this approach, claiming that if only FF of a lexical item Ll is moved, the properties relevant for binding etc. remain below. For Lasnik, therefore, all parts of Ll must be moved (i.e., generalized pied-piping must take place) in order for new binding/control configurations to be created. Keeping these two contrasting views in mind, let us consider the following example 33 (50) is grammatical, with the anaphor taking the matrix subject as its antecedent (cf. Barss, 1986). (50) Himself,, John, thinks that Mary likes t,. cf. *John, thinks that Mary likes himself,.

13

1 thank ieljko

BoSkoviC

(p.c.) for

the following

discussion.

M. Ochi I Lingua 109 (1999) 81-107

102

(51) illustrates the derivation of (50) under the proposed analysis. Suppose for the sake of discussion that what drives the movement here is the need for a functional head, call it Top, to check off its strong [+ Top] feature. CHrr is formed as a result of Attract, as shown in (5 1a). Further, CH,,, is required for PF convergence, as in (5 lb). (51) a. [Top [John thinks that Mary likes himself]] I

I

WI J CHFF

[+Topl I b-

himself [Top [rp t [rp John [vp t [VP thinks [cp t [cp that [IP t IIIIIIL iIp Mary lIvpt Ivp likes t111111111111 [TOPP

d

-

C&AT

The fact that the anaphor is licensed in (50) is predicted to be impossible under Chomsky’s (1995) view. As shown in (51a), raising of the FFs of himself takes place in one step due to the nature of Attract. If the FFs are relevant for anaphor licensing as Chomsky (1995) argues, then (50) is predicted to be ungrammatical, since the relevant property of the anaphor himself never occupies a position which is close enough to John (i.e. intermediate landing sites under Barss’s, 1986, analysis). On the other hand, Lasnik’s (1995) view is consistent with the grammaticality of (50). The features relevant for anaphor licensing, whatever they may be, are affected only by the pied-piping chain CH,,,. This chain is formed in a successive cyclic fashion, which means that the movement stops by some intermediate position close to John, hence creating the necessary configuration for licensing himself.j4 Thus, to the extent that the proposed hypothesis is justified, it lends strong empirical support for Lasnik’s (1995) view concerning the content of formal features.35 5.2. Feature movement and coordinate structures The proposed analysis also has several consequences for the syntax of coordinated structures. First, our analysis suggests that Attract F itself is not sensitive to the Coordinate Structure Constraint (CSC). Recall from section 3.1 that in Japanese, the feature of nani ‘what’ can be attracted out of the adjunct clause, which was analyzed, following Takahashi (1994), as a conjunct of a coordinated structure. I repeat the relevant data and its structure below for ease of reference.

14 For example, the following data from Lasnik and Saito (1992: 1 IO) show that the anaphor himself in (ib) can take John as its antecedent when this anaphor is fronted within the embedded clause. (i) a. *John thinks that Mary likes himself. b. John thinks that himself, Mary likes t. ” See BoSkoviC (1997) for additional arguments in favor of Lasnik’s (1995) view.

h4. Ochi

i Lingua 109 (1999) 81-107

103

(52) a. John-wa [Mary-ga nani-o katta ato] kaetta no? John-TOP Mary-NOM what-ACC bought after left Q ‘ ?*Whati did John leave [after Mary bought t,] ? ’ b. [John-TOP [Mary-NOM what-ACC bought after] left C] I

I

[.. FF,,,, ..I

[Ql

Thus, the effect of the CSC must be related to the formation of CHc-r (see below). Second, our discussion of acyclic merger in section 4 leads us to adopt a particular structure of the coordination. Specifically, we are led to adopt the structure in (53a), in which the two conjuncts are in the specifier and complement positions of the Boolean Phrase (cf. MUM, 1987; Zoemer, 1995), over (53b), in which the Boolean phrase (BP) consisting of a Boolean head and the second conjunct is adjoined to the first conjunct (cf. Munn, 1993). (53)

a. [aP John [a, and Mary]] b. lNPI [NP~Johnlb and Mary11

(53a) must be adopted because, given the structure in (53b), extraction out of the first conjunct NP, would be expected to be possible before the BP (which consists of the Boolean and and the second conjunct NP) is acyclically merged with it, contrary to fact. (54)

*Who did you buy [a picture oft] and a candy? cf. Who did you buy [a picture of t] ?

Following Munn (1987) and Zoemer (1995), let us therefore adopt the structure in (53a), in which the Boolean head (& for Zoemer) takes two conjuncts, one in its specifier and one in a complement position. If we assume that the Boolean head has strong (theta) features which must be checked off by merging with its arguments,36 then both conjuncts must be cyclically inserted (assuming BoSkoviC and Lasnik’s, to appear, analysis). Then (54) is analyzed as follows. First, as shown in (55), the interrogative C attracts the closest relevant feature, namely, that of who. Nothing blocks such an operation. A problem arises, however, with respect to the category movement of who. If it obeys the SMC, thus adjoining to the first conjunct NP, then the UCA is violated, as shown in (56a). If the movement avoids a violation of the UCA by skipping the step involving adjunction to the conjunct NP headed by picture, then the SMC is violated as in (56b).

JO As mentioned in footnote 3 I, BoSkoviC and Lasnik (to appear) assume that (1) theta roles are formal features triggering operations like Merge and Move, and (2) the theta features in English are strong. Consequently, arguments are prevented from being inserted acyclically in English, given (46).

M. Ochi I Lingua 109 (1999) H-107

104

(55)

C [,r you [vr buy [aP [a picture of who] [B’and [a candy11111

[... k,,,

A

...I

L 1C&F (56) a. Who did [ip t [,r you [vr t [vr buy LBpt [aP INP*t LNPa picture of tl [B,and a

I

I

I

I

I

CH,,,

candy]]]]]]]] UCA violated b. Who did [tp t hp you [vp t rvp buy [sp t LBpLNPa picture of *tl [B’and a

II

I

I

I

CHcAT

candy]]]]]]] SMC violated Finally, I will briefly discuss an implication of the proposed analysis for thereconstructions3’ Chomsky (1986b) suggests that expletive there is replaced at LF by its associate, which is motivated by the agreement facts in those constructions: the verb agrees with an indefinite NP which is not in the subject position, as shown in the pairs in (57a-b) and (57c-d). (57) a. b. c. d.

There is a man in the room. A man is in the room. There are men in the room. Men are in the room.

Under the Attract F hypothesis, the covert movement of the associate of there is reinterpreted as covert feature movement of the associate to the AGRs head which hosts there within its projection (see Chomsky, 1995: 4.4.5; Lasnik, 1995 for relevant discussion). Let us assume with Lasnik (1995) that there has a Case feature but lacks agreement features. At LF, the AGRs head attracts the phi-features of the associate NP to check off its own uninterpretable phi-features. Now, as observed by Munn (1993) and Sobin (1997), there-constructions exhibit a curious agreement pattern when the associate of there is a conjoined NP. When conjoined NPs occupy a subject position, as in (58), the verb agrees with the whole conjunct, thus being realized as plural. However, in there-constructions with a conjoined associate NP, the verb be tends to agree with the first conjunct, as shown in (59). (58) a. b. (59) a. b.

Three men and a woman *is/are in the room. A man and three women *is/are in the room. There *is/are three men and a woman in the room. There is/*are a man and three women in the room.

The agreement pattern shown in (59) is exactly as expected under the proposed account. The AGRs head attracts the closest phi-features. Given the analysis I ” I thank Nobuhiro Miyoshi for pointing out the relevance of this construction to me. Also, I owe much of the following discussion to BoSkovS’s (1997: section 4.3.3.3.) analysis of there-constructions (although for him, there lowers to the position of its associate rather than the associate raising to there).

M. Ochi I Lingua 109 (1999) 81-107

105

adopted regarding the Boolean Phrase (BP) above, it is reasonable to say that the first conjunct, which is in the spec of BP, is closer than the second conjunct in the complement position of Boolean head; the former c-commands the latter.38 Thus, it is the formal features (more specifically, phi-features) of the first conjunct that the AGRs head attracts. The fact that the verb agrees with the first conjunct is thus accounted for.39 Note that this attraction is successful despite the fact that the relevant feature is attracted out of the coordinated structure, which is consistent with the proposed analysis. (60a) and (60b) show the structures of (59a) and (59b) respectively (leaving aside the exact location of the verb hr, etc.). (60) a. There AGRs be [ar three men [a, and a woman] I [... FF ...I

I

I

[... pl ...I [... sg ...I CHFF b. There AGRs be [ap a man [sS and three women] I [... FF ...I

u

I

I

[... sg ...I

[... pl ...I

CHFF

6. Conclusion This paper has attempted to provide solutions to certain issues which arise under the theory of Attract F. Building on Chomsky’s (1995) conception of movement, 1 argued that category movement consists of two chains, each of different nature, and one of them has properties of Move. Thus, the proposed analysis advocates a hybrid theory of movement incorporating Move and Attract. It was further argued that feature movement, as it takes place via Attract, is not subject to Huang’s (1982) CED. According to the proposed analysis, CED effects show up only when the pied-piping chain CHc,r is formed out of certain domains. This accounts for the lack of CED effects with Japanese wh-movement: since the relevant movement is covert in Japanese, CHCAT is not necessary. The proposed analysis has additional consequences for the organization of the grammar. For instance, it provides empirical support for Lasnik’s (1995) view regarding the content of FF-bundles of lexical items, i.e., that features relevant for binding and control are not part of formal features. Also, my analysis has several consequences for the syntax of coordinate structures; in particular, the curious agreement pattern in existential constructions with a con” For instance, the following examples show that a bound pronoun is licensed when it is in the second conjunct and its binder (every sfudent) is in the first conjunct, suggesting that the first conjunct c-commands the second, but not vice versa (cf. BoSkoviC, 1997: 88). (i) a. Every student and his mother left. b. *His mother and every student left. I9 The same reasoning should apply to the examples in (58). Presumably, overt movement of the first (or the second) conjunct is excluded by the CSC (see footnote 8), and hence the whole Boolean Phrase must be moved, although the precise analysis of this type of example needs to be worked out.

106

M. Ochi I Lingua 109 (1999) 81-107

joined NP associate follows from the particular view of feature movement advocated in this paper. I also appealed to acyclic merger in order to preserve the essence of Takahashi’s (1994) analysis of Adjunct Condition effects. In this connection I introduced two analyses, Lebeaux (1988) and BoSkoviC and Lasnik (to appear), the latter of which I ultimately adopted. If the discussions in this paper are on the right track, we are then led to a strongly derivational characterization of the grammar (cf. Epstein et al., 1998).

References Agbayani, B., to appear. Generalized pied-piping and island effects. Proceedings of the North Eastern Linguistic Society 28, GLSA, University of Massachusetts, Amherst. Baker, C.L., 1970. Notes on the description of English questions: The role of an abstract question morpheme. Foundations of Language 6, 197-219. Barss, A., 1986. Chains and anaphoric dependence: On reconstruction and its implications. Ph.D. dissertation, MIT. Belletti, A. and L. Rizzi, 1991. Notes on psych-verbs, O-theory, and binding. In: R. Freidin (ed.), Principles and parameters in comparative grammar, 132-162. Cambridge, MA: MIT Press. BoSkoviC, i., 1997. The syntax of non-finite complementation: An economy approach. Cambridge, MA: MIT Press. BoSkoviC, A., to appear. Sometimes in SpecCP, sometimes in-situ. In: R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by step: Essays on minimalism in honor of Howard Lasnik. Cambridge, MA: MIT Press. BoSkoviC, 2. and H. Lasnik, to appear. How strict is the cycle? Linguistic Inquiry. BoSkovic, 2. and D. Takahashi, 1998. Scrambling and last resort. Linguistic Inquiry 29, 347-366. Chomsky, N., 1986a. Barriers. Cambridge, MA: MIT Press. Chomsky, N. 1986b. Knowledge of language. New York: Praeger. Chomsky, N., 1991. Some notes on economy of derivation and representation. In: R. Freidin (ed.), Principles and parameters in comparative grammar, 417-454. Cambridge, MA: MIT Press. Chomsky, N., 1993. A minimalist program for linguistic theory. In: K. Hale and S. Keyser (eds.), The view from building 20, l-52. Cambridge, MA: MIT Press. Chomsky, N., 1994. Bare phrase structure. MIT Occasional Papers in Linguistics 5, MITWPL. Chomsky, N., 1995. The minimalist program. Cambridge, MA: MIT Press. Chomsky, N. and H. Lasnik, 1993. Principles and parameters theory. In: J. Jacob, A. von Stechow, W. Stemefeld and T. Vennemann (eds.), Syntax: An international handbook of contemporary research. Berlin: de Grutyer. Collins, C., 1994. Economy of derivation and the generalized proper binding condition. Linguistic Inquiry 25, 45-62. Davidson, D., 1967. The logical form of action sentences. In: N. Rescher (ed.), The logic of decision and action, 81-95. Pittsburgh, PA: University of Pittsburgh Press. Epstein, S., E. Groat, R. Kawashima, and H. Kitahara, 1998. A derivational approach to syntactic relations. New York: Oxford University Press. Freidin, R., 1986. Fundamental issues in the theory of binding. In: B. Lust (ed.), Studies in the acquisition of anaphora, 151-188. Dordrecht: Reidel. Fukui, N., 1995. Theory of projection in syntax. Stanford, CA: CSLI Publications. Fukui, N. and M. Speas, 1986. Specifiers and projections. MIT Working Papers in Linguistics 8, 128-I 72. MITWPL. Higginbotham, J., 1985. On semantics. Linguistic Inquiry 16, 547-594. Huang, C.-T.J., 1982. Logical relations in Chinese and the theory of grammar. Ph.D. dissertation, MIT. Ishii, T., 1997. An asymmetry in the computation of phrase structure and its consequences. Ph.D. dissertation, University of California, Irvine.

M. Ochi

I Lingua 109 (1999) RI-107

107

Kayne, R., 1984. Connectedness and binary branching. Dordrecht: Foris. Kayne, R., 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Kuno, S., 1997. Binding theory in the minimalist program. Unpublished ms., Harvard University. Kuroda, S.-Y.. 1988. Whether we agree or not: A comparative syntax of English and Japanese. In: W. Poser (ed.), Papers from the second international workshop on Japanese syntax. Stanford, CA: CSLI Publications. Lasnik, H., 1995. Last Resort and Attract F. In: L. Gabriele, D. Hardison and R. Westmoreland (eds.), Proceedings of the sixth annual meetings of the formal linguistics society of mid-America, 62-81. Indiana University. Lasnik. H., 1998. Some reconstruction riddles. A talk given at 22nd Penn Linguistics Colloquium. Lasnik, H. and M. Saito, 1992. Move a. Cambridge, MA: MIT Press. Lebeaux, D., 1988. Language acquisition and the form of the grammar. Ph.D. dissertation, University of Massachusetts. Amherst. Maki, H., 1995. The syntax of particles. Ph.D. dissertation. University of Connecticut. Maki, H. and M. Ochi, to appear. Scrambling of wh-phrases and the Move-F hypothesis. Japanese/Korean Linguistics 8. Stanford, CA: CSLI Publications. Munn, A., 1987. Coordinate structure and X-bar theory. McGill Working Papers in Linguistics 4(l). 121-140. Munn, A., 1993. Topics in the syntax and semantics of coordinate structures. Ph.D. dissertation. University of Maryland. Murasugi, K. and M. Saito, 1994. Adjunction and cyclicity. Proceedings of the Thirteenth West COW Conference on Formal Linguistics, 302-317. Stanford, CA: CSLI Publications. Nishigauchi, T., 1986. Quantification in syntax. Ph.D. dissertation, University of Massachusetts, Amherst. Nunes, J., 1995. The copy theory of movement and linearization of chains in the minimalist program. Ph.D. dissertation, University of Maryland. Postal, P., 1997. Strong crossover violations and binding principles. A talk given at the Eastern States Conference on Linguistics, 1997, Yale University. Progovac, L., 1998. ‘Avoid conjunction’, adjunction, and the ‘coordination of likes constraint’. In: Z. BoSkoviC, S. Franks and W. Snyder (eds.), Annual workshop on formal approaches to Slavic linguistics: The Connecticut meeting 1997, 252-266. Ann Arbor, MI: Michigan Slavic Publications. Reinhart, T., 1995. Interface strategies, OTS Working Paper, Utrecht University. Rizzi, L., 1990. Relativized minimality. Cambridge, MA: MIT Press. Ross, J.R., 1967. Constraints on variables in syntax. Ph.D. dissertation, MIT. Saito, M., 1989. Scrambling as semantically vacuous A’-movement. In: M. Baltin and A. Kroch (eds.). Alternative conceptions of phrase structure, 182-200. Chicago, IL: University of Chicago Press. Saito, M. and N. Fukui, 1998. Order in phrase structure and movement. Linguistic Inquiry 29, 439474. Sobin, N., 1997. Agreement, default rules, and grammatical viruses. Linguistic Inquiry 28, 318-344. Stowell, T., 198 I. Origins of phrase structure. Ph.D. dissertation, MIT. Takahashi, D., 1994. Minimality of movement. Ph.D. dissertation, University of Connecticut. Takahashi, D., 1997. Move F and null operator movement. The Linguistic Review 14. 181-196. Takano, Y.. 1994. Scrambling, relativized minimality and economy of derivation. Proceedings of the Thirteenth West Coast Conference on Formal Linguistics. 3855399. Stanford, CA: CSLI Publications. Tsai, W.-T.D., 1994. On economizing the theory of A-bar dependencies. Ph.D. dissertation, MIT. Watanabe, A., 1992. Wh-in-situ, subjacency, and chain formation. MIT Occasional Papers in Linguistics 2, MITWPL. Zoemer, E., 1995. Coordination: The syntax of &P. Ph.D. dissertation, University of California, Irvine.