<<

AND RELATIVIZATION IN MINIMALIST

HIROKI EGASHIRA Tokyo Metropolitan University

This paper reanalyzes topicalization and relativization in the frame- work of the put forward recently by Chomsky (1995) and proposes a unified approach to these two operations. We analyze topicalization as morphologically driven movement, depending, on the circumstances where topicalization is applicable. We further extend this analysis to relativization, and clarify a driving force for movement of a relative pronoun.*

1. Introduction Lasnik and Saito (1992) (henceforth L&S) analyze the topicalization construction in terms of adjunction. Although this analysis is in- genious, there are some topicalization constructions that it cannot deal with. In this paper, we will first reexamine their analysis and point out some problems. We will then propose an alternative analysis of topicalization in the framework of the Minimalist Program put forth recently by Chomsky (1995). We will further extend the proposed analysis of topicalization to relativization, and propose that relativiza- tion undergoes the same operation as topicalization.

2. Lasnik and Saito's Analysis and Its Problems L&S argue that the topicalization construction as in (1a) below is

* This paper is based on my presentation at the 14th National Conference of the English Linguistic Society of Japan held at Kwansei Gakuin University on November 16-17, 1996. I am. grateful to Heizo Nakajima, Ken-ichi Takami, Hiroshi Hasegawa, Shigeo Tonoike, Ichiro Hirata, Kensuke Takonai, and Kosuke Tanaka for their long and patient discussion with me. I also thank Koji Fujita and Jun Abe for their comments and criticisms on the presentation. I am also indebted to two anonymous EL reviewers for insightful comments and suggestions. It goes without saying that all remaining errors are mine. English Linguistics 14 (1997) 28-51 -28- (C)1997 by the English Linguistic Society of Japan TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 29

derived by adjunction of a topic element to TP (=IP), =as shown in (1b): (1) a. This book, I really like. b. [TP This booki [TP I really like ti]] This analysis is based on the fact that in an embedded clause, a topic element appears between the complementizer that occupying the head of CP and the subject occupying the Spec of TP, as observed in (2): (2) I believe [CP that [TP the booki [TP I gave ti away to some friends]]] (L&S (1992: 76)) Behind this analysis, as pointed by Rochemont (1989), lies the ad- junction condition; namely, adjunction is possible only to a maximal projection that is a nonargument (Chomsky (1986)). The TP node, whether it is a matrix TP or an embedded one, is a nonargument maximal projection since it is not semantically-selected by a lexical category. Although this analysis correctly predicts the distribution of a topic element, a closer look at other topicalization constructions reveals that L&S's analysis has both theoretical and empirical problems. Let us begin with its empirical problems. There are several in- stances that resist adjunction of a topic element to the TP node in spite of the fact that the adjunction condition is observed. Observe the following sentence involving an infinitival clause: (3) a. My friends tend to support more liberal candidates. b. *My friends tend [CP [TP more liberal candidatesi [TP to support ti]]] (Hooper & Thompson (1973: 485)) In (3b) the of the support in (3a) is topicalized and adjoined to the infinitival TP, which is a nonargument maximal - jection. However, the derived sentence is ungrammatical. Next, compare the following two sentences: (4) a. I believe [that you will like the book] b. It's important [that he study the book more carefully] (Hooper & Thompson (1973: 485)) The bracketed clause in (4a) is declarative and that in (4b) is sub- junctive. However, they are identical in the sense that they consist of the projection of C (complementizer) and that of T.1 Then, L&S's analysis would predict that topicalization should be applicable in both

1 Although in (4b) T is not lexicallyrealized, it is natural to assumethat there is a null element in T because in the British Englishshould generallyappears. 30 ENGLISH LINGUISTICS, VOLUME 14 (1997) embedded clauses, because the embedded TP in (4) is a nonargument maximal projection. However, topicalization for (4a) is grammatical, while that for (4b) is not: (5) a. I believe [CPthat. [TPthis booki [TPyou will really like ti]]] b. *It's important [CP that [TP the booki [TP he study ti more carefully]]] (Hooper & Thompson (1973: 485)) Another problem with L&S's analysis concerns imperatives. Imai & Nakajima (1972) and Imai et al. (1989) argue that imperatives consists of the projection of T, whose feature is specified as [Imp], and that a null subject occupies the Spec of TP: (6) [TP pro T[Imp][VP study the Minimalist Program developed recently by Chomsky]] Given this structure, it is natural to assume that it is possible to topicalize the object of the verb study and adjoin it to the nonargument TP. However, the application of topicalization in (6) gives rise to ungrammaticality: (7) *[TP [the Minimalist Program developed recently by Chom- sky]i [TP pro T[Imp][VP study ti]]] Let us turn to a theoretical problem with L&S's analysis. Basically, the adjunction operation is executed freely. In other words, this operation requires no driving force. However, in the Minimalist Program, all movement operations are restricted to morphologically driven movement. If this assumption is correct (and we believe it), L&S's analysis would end up encountering a serious problem, since their analysis of topicalization requires no driving force. In this section we have seen that L&S's TP adjunction analysis of topicalization encounters both empirical and theoretical problems; empirically, there are several constructions that resist topicalization in spite of the fact that the adjunction condition is observed, and theoretically, movement by adjunction would be called into question in the framework of the Minimalist Program. Given this state of affairs, we can conclude that L&S's analysis loses its force, and that some modifications of the analysis of the topicalization construction seems to be called for.

3. An Alternative Analysis Let us first compare the grammatical (1a) with the ungrammatical TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 31

(3b), (5b) and (7), all repeated below: (8) This book, I really like. (9) a. *My friends tend more liberal candidates to support. b. *It's important that the book he study more carefully. c. *The Minimalist Program developed recently by Chom- sky study. The crucial difference between (8) and (9) seems to lie in the property of T, to whose maximal projection a topic element is adjoined. The property of T in (8) is finite, while that of T in (9) in non-finite. Thus, we can generalize that topicalization is possible if T is finite, and make the following assumptions: (10) a. A finite T is assigned a [Top(ic)] feature, which is a strong feature to be checked when it is drawn from the Lexicon to the Numeration. b. D(eterminer) of a topicalized element is always as- signed an Interpretable [Top] feature. The abstract morphological feature [Top], which we assume in (10), can be found in topicalization in Illongo: (11) a. dala sang babayi sang bata. bring AGT woman OBJ child 'A woman brought a child.' b. Nag-dala ang babayi sang bata. AGT-bring TOP woman OBJ child ' The woman brought a child.' c. Gin-dala sang babayi ang bata OBJ-bring AGT woman TOP child 'The/A woman brought the child.' (Schachter (1973: 24)) According to Schachter (1973), (b) and (c) in (11) are topicalization sentences derived from (11a). In this language, when an element is topicalized, the topicalized element is marked with the topic marker ang as shown in (11b) and (11c). This suggests that in some lan- guages, a [Top] feature is morphologically realized. Therefore, it can be assumed that the overt realization of the [Top] feature in Illongo holds covertly for English even when morphological evidence is lacking. It should be emphasized here that the assumptions in (10) claim that topicalization is derived by the morphological checking operation. The topicalization in Illongo in (11) also supports this claim. Let us see (11) again. When the subject is topicalized as in (11b), the prefix nag, 32 ENGLISH LINGUISTICS, VOLUME 14 (1997) an agent marker, is attached to the verb dala. The same prefixation is also observed in the object topicalization: when the object is topicalized as seen in (11c), the object-marking prefix gin is attached to the verb dala. This kind of prefixation cannot be observed in (11a), where no elements are topicalized. This agreement phenomena suggest that a topic element covertly raises to T which has overtly hosted a verb at the Spell-Out for the feature checking as seen below: (12) a. [TP ang-babayi [Top]i [T' [T[Top] [V nag-dala]] [VP ti sang bata]]] b. [TP ang-bata[Top]i [T' [T[Top] [V gin-dala]] [VP sang babayi ti]]] We can, therefore, conclude that these agreement (or prefixation) phenomena observed in this language support the claim in (10). Furthermore, the assumption in (10a) can explain the ungrammati- cality of (9) because it claims that a strong [Top] feature is assigned to a finite T: T in (9) is not finite. With the assumption in (10) in mind, let us see how the topicaliza- tion construction in (8) above, repeated below, can be derived:2 (13) This booki, I (really) like ti. At an earlier stage of the derivation of (13), we have (14), where the subject I is attracted from the VP internal subject position to the Spec of T by the strong [D] feature that T has:3 (14) [TP I[D]i [T' T [D][Top] [VP ti like [DP [D this [Top]] [NP book]]]]] In order for the strong [Top] feature to be checked, another checking domain of T has to be projected. Chomsky (1995), adopting the analysis by Reinhart (1981) and Ura (1994), assumes that when a feature belonging to a head is not erased after being checked by an element raising to a checking domain, another checking domain is projected by the still activating feature. Give this argument, DP this book, whose head D is assigned a [Top] feature, can move to the second Spec of T

2 In this paper, we restrict our attention to the topicalization where DP's are topicalized. As for the topic construction in which PPs are topicalized as in (i) below, we assume that Interpretable [Top] feature assigned to D percolates up to the whole category PP dominating DP: (i) Mary found the book [PP[Top] in [DP the[Top] living room]] Though APs, VPs, and CPs can also be fronted, we consider these constructions not a typical topicalization construction since they do not yield the interpretation of "aboutness" (or "as for"). 3 We do not adopt here the light verb v for expository convenience. TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 33 since the strong [Top] feature is not checked and is thus not erased. Then we have (15) where the [Top] feature that T has is checked: (15) a. [TP [DP [D this[TOP]] [NPbook]]]i [T' I[D] [T' T[D] [TOP][VP like ti]]]] b.

Notice that in (15) the [Top] feature of T is deleted and is further erased while that of the topicalized element remains intact. The proposed analysis at least has three consequences. One con- sequence relates to the relation between the position and intonation of a topic element, another to the interpretation of the topicalization construction, and the other to the impossibility of multiple topicali- zation. As is well-known, a topicalized element has a comma (or topic) intonation. We can assume that a topicalized element in the second specifier is given a comma intonation in PF.4 As argued above, the [Top] feature that a topicalized element has remains intact since this feature is plainly Interpretable. This means that the sentence is interpreted as topicalization because a plainly Interpretable feature enters into interpretation in LF. Let us turn to the third consequence. Fukui (1993) argues that a multiple topicalization construction gives rise to ungrammaticality: (16) a. ??That booki, to Maryj, John handed ti tj. b. ??To Maryj, that booki, John handed ti tj. In (16a, b) the direct object the book and the indirect object to Mary are topicalized and the sentences are marginal or ungrammatical. This can be explained in terms of deletion and erasure of the strong [Top]

4 An anonymous reviewer has pointed out that not only a topic element but also sentence-initial adverbs have a comma intonation. As pointed out in footnote 2, we do not consider this kind of sentence as typical topic construction in the sense of the "aboutness" interpretation. Thus, we assume that a comma intonation of this construction is assigned for another reason. 34 ENGLISH LINGUISTICS, VOLUME 14 (1997) feature of T. In (16) T's strong [Top] feature is deleted and is further erased after being checked by the first topic element (i.e. to Mary in (16a) and that book in (16b)). As a result, there is no strong [Top] feature that can attract a second topic element. Hence, the ungrammat- icality of (16) results.5 At this point, we need to discuss the order of a subject and a topic element. The reversed order, i.e. subject-topic, must be prevented from being generated: (17) *[TPI[D] [T' the book[Top] [T' T[D] [Top] [VP (really) like]]]] In (17) the topic element the book moves to the first Spec of TP and the subject I to the second Spec of TP for feature checking. Although the derivation is convergent since the strong [D] and [Top] features of T are checked off, the derived sentence in (17) is ungrammatical. We must ensure that the subject moves to the first Spec and the topic to the second Spec. Notice here that we cannot appeal to the Minimal Link Condition (henceforth MLC). The subject generated VP internally is closer to the node T than the object of the verb which is a topic as seen in (18): (18) [TP T[D][Top] [VP Subj[D] V Obj[Top]]] However, according to the feature-based Minimalist theory, both the subject and the object are the same distance from T: the closest feature for T's [D] is subject's [D] and the closest feature for T's [Top] feature is object's [Top]. Thus MLC permits the object to move to the first Spec of T as seen in (17) above. However, adopting Maximal Checking Condition (henceforth MCC), given in (19) below, proposed by Fujita (1996), enables us to derive the correct word order in the topicalization construction: (19) Maximal Checking Condition (MCC): Attracting an element that checks more formal features of

5 An anonymous reviewer has pointed out, such an unwanted multiple topicaliza- tion construction as in (i) will be generated in our framework: (i) [TP I[Top] T bought [DP [Top] this book] [DP [Top] yesterday]] In (i) three Interpretable [Top] features are assigned to three DPs. This means that the other two DPs as well as the fronted DP are interpreted as a topic of the sentence, which is not the case. This possibility is to be excluded by assuming that an Interpretable [Top] feature is one to one correspondence to a strong [Top] feature assigned to T. This assumption seems natural since a sentence always has a topic even if an element is not fronted. As for a sentence in which no element is fronted, i.e. the subject topicalization, see 4.1 below for details. TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 35

the target is more economical. (Fujita (1996: 120)) Let us briefly review this condition. Suppose that we have the following configuration as seen in (20) below. In (20) τ is the target of the. movement of either α or β, both of which are in the same domain, and that τhas two features: [F1] and [F2]. Further α has

[F1] and β has [F1] and [F2]:

(20) [τ[F1][F2] [α[F1] [β[F1][F2]]]] MCC permits β to move to τ across α since the movement of β can check more formal features of τ than that of α. The same explanation can apply to the topic construction in (18), where T is assigned a [Nom(inative)] feature as well as [D] and [Top]: (21) [TP T[D][Nom][Top] [VP Subj[D][Nom] V Obj[Top]]] The movement of the subject checks more features of T, i.e. [D] and [Nom], than that of the object: (22) a. [TP Subj[D][Nom] [T' T[D][Nom][Top] [VP t V Obj[Top]]]] b. *[TP Obj[Top] [T' T[D][Nom][Top] [VP Subj[D][Nom(inative)] V t]]]6 Thus, we have the desired word order, i.e. topic-subject. In this section, we have seen how the topicalization construction is derived in the Minimalist program. Our analysis seems to be more comprehensive than L&S's analysis in that the former does not overgenerate ungrammatical topicalization constructions that the latter wrongly rules in. In the next section, we will discuss some conse- quences of our analysis.

4. Consequences This section will explore some other consequences of our analysis. One consequence relates to the short subject topicalization and the other to the distribution of a topic element in interrogatives.

4.1. Short Subject Topicalization As is well discussed (cf. Quirk et al. (1985: 726)), the subject is a topic of the sentence. However, as noted by L&S, if the subject, a

6 As an anonymousreviewer pointed out, T's [D] feature as well as [Top] feature attracts the object to the first Spec of TP rather than the subject. Assumingthat T's nominative case must be checked at the overt syntax, this derivation is to be excluded in terms of the feature mismatch: the cannot check T's nominativecase. 36 ENGLISH LINGUISTICS, VOLUME 14 (1997) topic of the sentence, has a comma intonation, the resulting sentence degrades, as seen in (23a): (23) a. *John, left. b. [TPJohn [TPt [VP left]]] This has led L&S to conclude that the subject never undergoes short subject topicalization, as in (23b) above. Their argument is also based on the syntactic fact that the subject-object asymmetry can be found in the effects of topicalization on anaphor binding: (24) a. *John thinks that Mary likes himself. b. John thinks that himself Mary likes t. In (24a) the anaphor himself is not bound in its governing category, the embedded TP; hence ungrammaticality of (24a). In (24b), on the other hand, the governing category is extended to the matrix TP by virtue of the application of topicalization to the anaphor himself. We expect that the same effect is to be found in the subject topicalization. However, this is not the case, as demonstrated in (25): (25) a. *John thinks that himself likes Mary. b. *John thinks that himself, t likes Mary. The ungrammatical status of (25a) is the same as that of (24a). If the governing category for the topicalized anaphor in (25b) could be extended to the matrix TP in the same way as (24b), (25b) should be grammatical as well. This difference has led L&S to conclude that there is no short subject topicalization. Although L&S explain this fact in terms of a head-government requirement, this way of explana- tion does not hold in the Minimalist Program since the notion of government is liable to be dispensed with. We must explain the above fact in another way. In contrast with L&S's analysis, the analysis proposed here explains more naturally the fact pointed out above. Suppose that at an earlier stage of the derivation of (23a) we have (26), where T is assigned [D] and [Top], which are strong, and the subject John also has these plainly Interpretable features: (26) [TP T[D][Top] [VP John[D][Top] left]] At the next stage of the derivation, the subject John is raised to the Spec of TP due to the strong [D] feature that T has. Thus we have (27), where T's strong [D] feature is deleted and is further erased: (27) [TP John[D][Top]i [T' T[D] [Top] [VP ti left]]] Note that the [Top] feature that the subject has is pied-piped as a free rider. In this configuration it is possible for T's [Top] feature to be TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 37 checked as well as T's [D] without subject DP's moving any further. Therefore, we never have such a derivation (or representation) as (23b), repeated here as (28) below: (28) *[TP John[D][Top]i [T' ti [T' T[Top] [VP ti left]]] In (28) the subject moves to the higher specifier, namely to the second specifier of TP and T's [Top] feature undergoes checking. Although (28) is a convergent derivation as well as (27), we can exclude it in terms of the Economy condition, requiring that among convergent deri- vations the derivation with fewest operations excludes other competing derivations. Before leaving this section, let us observe an implication of this section. According to the discussion, the subject of the sentence is a topic in spite of the lack of the movement to the second Spec of TP. This suggests that a sentence always has a topic, and consequently that a finite T and either the subject or elements other than the subject, entering into the Numeration, are inherently assigned a strong [Top] feature and an Interpretable [Top] feature, respectively. If this reasoning is correct, we can safely conclude that the postulation of a [Top] feature is not stipulative but well-motivated.

4.2. The Distribution of a Topic Element in Interrogatives Topicalization is applicable not only in declaratives but also in (wh-) interrogatives: (29) a. In the living room, what did Mary find? (Rochemont (1989: 147)) b. John, does Mary like? (ibid.) It is generally assumed that interrogatives consist of the projection of C, to whose head and specifier an auxiliary and a wh-phrase move, respectively. Then, it follows that in (29) a topic element appears in front of CP in interrogatives. If a topic element appears in front of TP and a wh-phrase or an auxiliary moves across the topic element, the resulting sentence is ungrammatical: (30) a. *What does in the living room Mary find? b. *Does John Mary like? The contrast observed in (29) and (30) has led Rochemont (1989) and Grimshaw (1993) to generalize that a topic element is adjoined to the propositional domain: to TP in declaratives and to CP in interrogatives: (31) a. [TPThe booki [TPI really like ti]]. b. [CPIn the living roomi [CPwhat did [TPMary find ti]]]. 38 ENGLISH LINGUISTICS, VOLUME 14 (1997) c. [CPJohni does [TPMary likes ti]]. Our analysis can give an explanation to the generalizationmade by Rochemontand Grimshaw. To see this, let us analyze(29a) first. At an earlier stage of the derivationof (29a), we have (32), where T with [D] and [Top]merges with VP already formed, attractingthe subject to its specifierposition: (32) [TP Mary [T' T[Top] (did) [VPfind what [PP in the living room]]]]7 At the next stage of the derivation, C with a strong [Q] is merged to TP alreadyformed and consequentlyT movesto C:8 (33) [CP [C T[Top] (did) C[Q]] [TP Mary find what [PP[Top] in the livingroom]]] Note here that in (33) a [Top] feature that T has is pied-piped as a free rider. This means that the checkingdomain of T's [Top] feature is projected within the projection of CP. At the final stage of the derivation in the overt syntax, the wh-phrase what and the topic element in the living room move to the first and second Spec of CP, respectively:9 (34) [CP [PP[Top] In the living room]j [C' whati [C T[Top] (did) C[Q]] [TPMary find ti tj]]]. The same analysiscan be appliedto the matrixyes-no interrogatives with a topic, as demonstratedin (29b) above, repeated here as (35): (35) John, does Mary like t? The object John is topicalizedto the Spec of CP after movementof T with a strong [Top]feature to C:

7 We assume with Rizzi (1993) that the auxiliary verb do is generated in T. 8 The question arises why T's [Top] feature in (33) above is not checked before the merger of C with a strong [Q] feature: (i) [CP C[Q] [TP [PP[Top] in the living room]i [T' Mary [T' T[Top] [VP find what ti]]]]]. The derivation in (i) above observes the Strict Cycle condition by Chomsky (1995: 234). However, the derived sentence is ungrammatical as seen in (30a) above. I leave the question open why the checking of T's [Top] feature can be delayed in the interrogative topicalization. 9 The reversed word order as in (i) below is not possible: (i) [CP Whati [C' [PP[Top] in the living room]j [C T[Top] (did) C[Q]] [TP Mary find tij]]]? One plausible way to derive the desired word order is to assume that the feature- checking of the hosting head, namely C in this case, takes precedence over that of the adjoined head, namely T. I left open why this holds. TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 39

(36) [CP John[Top]i [C T[Top] (does) C[Q]] [TP Mary like ti]]. In both (34) and (36), the position of the topic element is realized in CP due to the head movement of T to C.10 Given the argument above, we can conclude that our analysis can give an explanation to the generalization made by Rochemont and Grimshaw: in declaratives the checking domain of T with a [Top] feature is projected to the second Spec of TP,11 while in interrogatives, due to the head movement of T with a [Top] feature to C, the checking domain is projected to the second Spec of CP.12 It seems reasonable to conclude from the discussions in this section that a topic movement is triggered by a [Top] feature assigned to T.

5. Relativization as Topicalization In this section we will attempt to extend the analysis proposed in section 3 to relativization. In section 5.1, we will clarify a driving force for movement of a relative pronoun and propose a novel deri- vation and structure for relative clauses with a wh-relative pronoun. In section 5.2 we will see some consequences of our analysis, com- paring our analysis with a previous analysis of relativization.

5.1. Relativization and Topicalization Relativization shows similarities with topicalization in many relevant respects. First of all, as topicalization can involve a resumptive pro- noun in a left-dislocation configuration, as seen in (37) (cf. Grimshaw (1993)), so does relativization, as seen in (38):

10 This way of reasoning can also account for the distribution of a topic in the negative inversion constructions, where a topic appears in front of an inverted negative constituent: (i) Fred said that Susan not once had they seen. (Nakamura (1994: 162)) Due to the head movement of T[Top] to Pol in the sense of Culicover (1991) and Nakajima (1992), the landing site of a topic is projected in the second Spec of PolP: (ii)...[PolP Susani [Pol' not once[Neg]j[Pol T[Top](had)k Pol[Neg]][Tp they tk seen ti ti]]] 11Notice that the declarative C is merged in LF. 12 It is possible to unify the checking domain of John to the second specifier position in (36) if we assume with Grimshaw (1993) that a base-generated empty operator occupies the first Spec of C: (i) [CPJohni [C' OP [C T[Top](does) C[Q]][TP Mary like ti]]]. 40 ENGLISH LINGUISTICS, VOLUME 14 (1997)

(37) a. The man, I can't stand t. b. The man, I can't stand him. (Grimshaw (1993: 24)) (38) a. the man who I saw t b. the man who I saw him (Chomsky (1982: 11)) In the (b) examples of (37) and (38) the resumptive pronoun him occupies the position where the topic and the relative pronoun occupy previously and the (b) sentences are grammatical. Secondly, a resumptive pronoun of topicalization and relativization can occur in an island: (39) a. Syntactic Structures, I wonder [who read it]. b. the man who I wonder [whether he will win the race]. (Lasnik & Uriagereka (1988: 154)) The embedded clause in (39) constitutes an island by application of wh- movement in (a) and occurrence of a wh-complementizer whether in (b). Further extraction, namely topicalization in (a) and relativization in (b), has taken place inside an island, and a resumptive pronoun occurs in the position where a moved element previously occupies. Finally, both operations constitute an island for further extraction (Bold-faced brackets indicate "Island"): (40) a. *I wonder whati [on the tablej [Mary put ti tj]]. b. *To whomi have you found someone [who would speak ti]. In (40a) the prepositional phrase on the table is topicalized in the embedded clause and further wh-movement has taken place in the embedded clause. Likewise, in (40b) to whom is moved out of the embedded clause headed by a relative pronoun. The ungrammaticality of (40) is attributed to an island created by topicalization and relativi- zation. Given the arguments above, we can conclude that relativization and topicalization are similar constructions. With this in mind, we assume (41): (41) A T[Top] in relative clauses attracts a relative pronoun, which is assigned an Interpretable [Top] feature. The assumption in (41) claims that movement of a relative pronoun is triggered by a strong [Top] feature assigned to T. The supporting evidence for a [Top] feature as a driving force for movement of a relative pronoun can be found in relativization in Illongo. Before discussing the issue, let us briefly review topicalization in this language discussed in the previous section. In this language, when an element is topicalized, it is marked with the topic marker ang. Moreover, a TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 41 prefix specifying the topicalized element is attached to the verb: when a subject is topicalized, the prefix nag, which is the agent marker, is attached to the verb as seen in (42a) and when an object is topicalized, the prefix gin, which is the object marker, is attached to the verb as seen in (42b) (see (11) for details): (42) a. Nag-dala ang babayi sang bata. AGT-bring TOP woman OBJ child ' The woman brought a child.' b. Gin-dala sang babayi ang Bata OBJ-bring AGT woman TOP child 'The/A woman brought the child.' (Schachter (1973: 24)) With this inflection of the verb in mind, let us turn to relativization in Illongo: (43) a. babayi nga nag-dala sang bata woman REL AGT-bring OBJ child '(a) woman that brought a child' b. bata nga gin-dala sang babayi child REL OBJ-bring AGT woman '(a) child that a/the woman brought' (Schachter (1973: 24)) In (43) each relative clause is headed by the relative pronoun nga. (43a) shows relativization of a subject and (43b) relativization of an object. Let us concentrate on the inflection of the verb. When the subject is relativized as seen in (43a), the prefix nag, which is the agent marker, is attached to the verb and when the object is relativized, as seen in (43b), the object marking prefix gin is attached to the verb. This is the same process as we have observed in topicalization of this, language. Thus, we can safely conclude that it is natural to assume that relativization is derived by topicalization, more specifically by a strong [Top] feature assigned to T. We must also refer to an Interpretable [Top] feature assigned to a relative pronoun. As is well-known, relative clauses headed by a relative pronoun contain a predication relation holding between a relative pronoun and the clause following the relative pronoun. This means that a relative pronoun is a kind of the subject and the clause following it is the predicate. If so, it is natural to assume that a relative pronoun serves as the function of the "aboutness" of a relative clause. The typical topicalization construction like Gordon, Trudie 42 ENGLISH LINGUISTICS, VOLUME 14 (1997) really loves has an interpretation as About (as for) Gordon, Trudie really loves him. Thus, even if the direct interpretation is not reflected in a relative pronoun itself, we can assume that a relative pronoun is assigned an Interpretable [Top] feature.13 Keeping the assumption in (41) in mind, let us analyze such relative clauses with a wh-relative pronoun as in (44): (44) a. the spy whom I loved b. the spy who loved me We analyze (44a) first. At an earlier stage of the derivation, we have (45), where T is assigned a strong [Top] feature and the subject I is raised to the Spec of TP from the VP-internal position due to T's strong [D]: (45) [TP I[D]i [T' T[D][Top] [VP ti loved whom[Top]]]] At the next stage of the derivation, T's strong [Top] feature attracts the wh-relative pronoun whom with an Interpretable [Top] feature to the second Spec of TP (T's strong [Top] feature is deleted and further erased after being checked.): (46) [TP whom[Top]j [T' Ii [T' T[Top] [VPti loved ti]]]] Let us turn to (44b), repeated here as (47a) below. As seen in (47b), the subject relative pronoun is attracted from the VP-internal position due to T's strong [D] feature and raised to the first Spec of TP: (47) a. the spy who loved me b. [TP who[D][Top]i [T' T[D][Top] [VP ti loved me]]] Note that a [Top] feature that the subject relative pronoun who has is pied-piped as a free rider. In this configuration it is possible for T's [Top] feature to be checked without further movement of who. It is, of course, possible for the relative pronoun who to move to the second Spec of TP for feature checking: (48) *[TP who[D][Top]i [T' ti [T'T[Top] [VP ti loved me]]]] Although (48) is a convergent derivation as well, the admissible deriva- tion is (47b) because it contains fewer steps than (48) (see also (27) and (28)). We thus have the following structures for a subject and object relative clause, as demonstrated in (49) below: (49) a. [TP(SPEC2) whom[Top]i [T'(SPEC1) I[D] [T' T[D][Top] [VP loved

13 I owe this reasoning to Heizo Nakajima (personal communication). TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 43

ti]]]] b. [TP(SPEC1) who[D][Top]i [T' T[D][Top] [VP ti loved me]]] It is worth noting here that the derivation and the derived structure of relative clauses are parallel with those of topicalization constructions. As argued in section 3, an element topicalized from the object position occupies the second Spec of TP, as seen in (50a) below. On the other hand, the subject which is the topic of a sentence moves from the VP- internal subject position and occupies the first Spec of TP, as seen in (50b): (50) a. [TP(SPEC2) This book[Top]i [T'(SPEC1) I[D] [T' T[D][Top] [VP really like ti]]]]. b. [TP(SPEC1) John[D][Top]i [T' T[D][Top] [VP ti left]]]. Worthy of notice, the structure of relative clauses that we have proposed so far is different from the structure that is widely accepted in generative in that in our analysis there is no projection that hosts a relative pronoun over TP, namely CP.14 However, the pro- posed structure of relative clauses discussed here is theoretically supported by the Minimal Structure Principle proposed by Boskovic (1996): (51) The Minimal Structure Principle: Provided that lexical requirements of relevant elements are satisfied, if two representations have the same structure, and serve the same function, then the representation that has fewer projections is to be chosen as the syntactic repre- sentation serving that function. (Boskovic (1996: 290)) This principle claims that the structure containing fewer projections is preferred. It effectively excludes a representation such as (52a), one assumed in , and instead chooses the presentation of our analysis, given in (52b): (52) a. the spy [CP whom [TP I loved t]] b. the spy [TP whom I loved t]] Both relative clauses in (52) serve the same function as a relative clause and (52b) contains fewer projections than (52a). Since Baltin (1982), it is well known that topicalization is inapplicable in relative clauses: (53) ??the man [REL[to whom]j [liberty]i we could never grant ti tj] (Baltin (1982: 17))

14 We restrict ourselves to relative clauses without that, which presumably belongs to C. We will turn to this issue later. 44 ENGLISH LINGUISTICS, VOLUME 14 (1997)

Our analysis claims that a relative pronoun is topicalized by T's strong [Top] feature. One of two elements with [Top], i.e. a topic element or a relative pronoun, raises to the checking domain of T and T's [Top] feature is deleted and is further erased after being checked, so T cannot attract another topic element. Therefore, our analysis never generates such ungrammatical sentences as (53) because a topic and a relative pronoun both have a [Top] feature and either of them checks a strong feature that T has. This explanation is reminiscent of the explanation of ungrammaticality of multiple topicalization discussed in section 3: (54) a. ??That booki, to Maryj, John handed ti tj. b. ??To Maryj, t hat booki, John handed ti tj. We exclude multiple topicalization in the same way: a strong [Top] feature of T is deleted and is further erased by one of the topic elements that raises to the checking domain of T, and consequently there are no features that can attract another topic element. Incidentally, judgment of grammaticality of (53) is equal to that of (54). This equivalent judgment observed in both constructions seems to suggest that our way of explanation is appropriate. Marginal acceptability of (53) and (54) can be attributed to the way of realizing multiple specifies. Chomsky (1995) argues that multiple specifiers are allowed when a feature belonging to a head is not erased after being checked by an element in the first specifier, and due to the still activating feature, another element is attracted and raised to the second specifier. Following his argument, it is possible to reason that those speakers who judge (53) and (54) marginal tolerate erasure of a [Top] feature after being checked by one topic element and consequently allow the second topic to raise, and that those speakers who exclude (53) and (54), for instance, like L&S (1992: 86), do not allow such an operation of delay of erasure of a [Top] feature. As a final remark of this section, we need to discuss the nature of a [Top] feature of relative clauses. In order to derive a relative clause, a (strong) [Top] feature must always be assigned to a head attracting a relative pronoun and to a relative pronoun. It is easy to ensure that a relative pronoun has a [Top] feature if we follow classification of a feature by Chomsky (1995). He argues that there are two kinds of feature: one is an intrinsic feature and the other an optional feature. The former is either listed in a lexical item or determined by listed feature: its prototype is a categorial feature of a lexical item, for TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 45

instance, [D] of DP and [Q] (or [Wh]) of a wh-interrogative pronoun. In other words, an intrinsic feature is the feature that a lexical item has inherently. An optional feature is arbitrary assigned as LI (Lexical Item) enters the Numeration: its prototype is a case feature of DP. A [Top] feature of a relative pronoun falls into a category of Intrinsic feature in that a relative pronoun inherently has a [Top] feature. Thus we ensure that a relative pronoun always has a [Top] feature. Let us turn to the issue of a strong [Top] feature attracting a relative pronoun. The answer how to ensure a [Top] feature of an attracter of a relative pronoun can be found in the relation holding between D and a relative clause. Jackendoff (1977), showing the contrast in (55), argues that there is an interaction between them: (55) a. *the Paris b. the Paris that I love A proper noun cannot co-occur with a determiner. However, it can co-occur with a determiner if a relative clause follows the proper noun as demonstrated in (55b). Taking this observation into account, we assume that D of a relative head selects a relative clause in terms of Head-head agreement holding between them: D of a relative head selects a relative clause that is headed by T with a strong [Top] feature: (56)

Thus, we can make sure that T of relative clauses must always have a strong [Top] feature by virtue of selectional relation holding between D of a relative head and T of a relative clause. As a final stage of the derivation of relative clauses, we assume that a relative TP is extraposed and adjoined to DP: 46 ENGLISH LINGUISTICS, VOLUME 14 (1997)

(57) [DP[D the ti] [NP spy][TP whom I loves]i]15 At this point we should refer to relative clauses without a wh-relative pronoun as seen in (58) below: (58) a. the spy that I loved b. the spy that loved me In this case D of a relative head selects C with a [Top] feature and a null relative pronoun moves to the Spec of CP (we omit here extra- position of CP for expository convenience): (59) a. [DPthe spy [CP OP[Top]i [C' that[Top] [TP I loved ti]]]] b. [DPthe spy [CP OP[Top]i [C' that[Top] [TP ti loved me]]]]16 We have argued in this section that relative clauses with a wh-relative pronoun consist of TP, to whose first specifier and second specifier a subject and object wh-relative pronoun are raised, respectively, by virtue of application of topicalization. This argument exactly reflects the claim made by Kuno (1973), who argues from the functional view that what is relativized in a relative clause is a noun that is a topicalized (in his terms, thematized) noun phrase. This means that the approach proposed in this section is compatible with the functional approach given by Kuno.17 In the next section our radical proposal on the structure of relative clause will be exposed to the comparison with the traditional analysis.

5.2. The Comparison with the Traditional Analysis Traditionally, relative clauses are considered to consist of the pro- jection of C, to whose specifier a relative pronoun moves: (60) a. the spy [CPwhoi [C' C [TPti [VPti loved me]]]] b. the spy [CPwhomi [C' C [TPI [VPloved ti]]]] According to our analysis, relative clauses with a wh-relative pronoun consist of the projection of T, rather than C:

15 As for the extraposition of a relative TP, we left open the question why it moves. One plausible analysis is to assume that D of the relative head has a feature to be checked by the relative clause TP: a relative TP moves to the checking domain of D, i.e. to the extraposed position. It goes without saying that the base- generated position of the relative clause TP is not the checking domain of D. 16 As for relative clauses without a wh-relative pronoun, see Egashira (1997) for details. In Egashira (1997), infinitival relatives are analyzed in the same way as relatives without a wh-relative pronoun. 17 I am thankful to an anonymous reviewer for drawing my attention to Kuno (1973). TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 47

(61) a. the spy [TP(SPEC1) whoi [VP ti loved me]] b. the spy [TP(SPEC2) whomi [T'(SPEC1) I [VP loved ti]]] The derivation and structure discussed throughout this section can give a better explanation to occurrence of a resumptive pronoun. A resumptive pronoun is possible in the object position, but not in the subject position: (62) a. the man whoi I saw himi b. *the man whoi hei saw me Suppose that a resumptive pronoun can occupy the position where a relative pronoun occupies before movement. If this postulation is right, a resumptive pronoun can occur both in the subject position and the object position in the framework of the traditional CP analysis because in this framework both subject and object relative pronouns move to the specifier position of CP and leave a trace behind in the original position, where a resumptive pronoun can occur: (63) a. the man [CP who [C' C [TP I saw him]]] b. *the man [CP who [C' C [TP he saw me]]] In order for the traditional analysis to exclude (63b), it needs to appeal to the stipulation that a resumptive pronoun cannot occupy the subject position. However, this stipulation is not in effect as seen from the possibility of occurrence of a resumptive pronoun in the subject posi- tion as seen in (64): (64) the man whoi I wonder [whether hei will win the race] In (64) the resumptive pronoun he occupies the subject position. Our analysis, however, can account for the fact properly. According to our analysis, a subject relative pronoun is attracted by T's strong [D] and moves to the first specifier position of TP. Since a [Top] feature that a relative pronoun has is pied-piped as a free rider, T's strong [Top] feature undergoes checking when a subject relative pronoun raises to the first specifier position of TP: (65) the man [TP who[D][Top][T' T[D][Top][VP t saw me]]] In (65) there are no places that a resumptive can occur because who does not move to any positions other than the first Spec of TP, and the VP-internal position where a subject is base-generated is not a candidate for a resumptive pronoun. Therefore, (62b) is excluded. As for (64), the subject relative pronoun moves to the second Spec of the topmost embedded clause and leaves behind the trace. Thus, a resumptive pronoun is possible as seen in (66): (66) the man [TP who [T' I [T' T[Top] [VPwonder [CP whether he 48 ENGLISH LINGUISTICS, VOLUME 14 (1997)

will win the race]]]]]18 Next, let us examine the following contrast demonstrated in (67): (67) a. This is the child who there is nobody who is willing to accept. (Kuno (1976: 423)) b. *This is the child that there is no one who the authorities persuade to accept. (Chung and McCloskey (1983: 708-709)) In the most deeply embedded sentence of (67), two elements are relativized: in (67a) the subject and the object of the verb, and in (67b) the object of persuade and the object of accept. Grammaticality of (67a) is better than that of (67b).19 How can this difference of gram- maticality be accounted for? Let us begin with the traditional CP analysis. Following this analysis, we have the following representation for (67): (68) a. ...the child [CP whoi [C' C [TP there is [DP nobody [CP whoj [C' C [TP tj is willing to accept ti]]]]]]] b. ...the child [CP OPi [C' that [TP there is [DP no one [CP whoj [C' C [TP the authorities persuade tj accept ti]]]]]]] In (68) the relative pronouns occupy the Spec of CPs. There are no structural differences between (68a) and (68b). Each trace obeys a proper head government requirement in the sense of Rizzi (1990): in (68a) the subject trace is head-governed by C agreeing with who and the object trace is head-governed by the verb. In (68b) each trace is head-governed by its verb. Both sentences equally violate subjacency: whoi and OPi skip over two boundaries, namely CP and DP. According to the traditional CP analysis, coupled with a head govern- ment requirement, both sentences in (67) should be excluded as subjacency violation and thus the grammaticality difference cannot be explained. Our analysis, however, can explain the difference of grammaticality in (67) properly. Note that a subject relative pronoun does not move from the first specifier position of TP in our analysis. We have the following representation for (67):

18 In (66) D assigns a [Top] feature to the topmost T. 19 We abstract away from the marginality of the example, which is due to movement across an extraposed relative clause. TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 49

(69) a. ...the child [TP whoi there is [DP nobody [TP(SPEC1) who [VP is willing to accept ti]]]] b. ...the child [CP OPj that [TP there is [DP no one [TP(SPEC2) whoi[T'(SPEC1) theauthorities [VP persuade ti to accept ti]]]]]] While the most deeply embedded TP of (69a) has only one specifier, that of (69b) has two specifiers. If we assume that TP with one specifier is not a barrier for further extraction and that TP with two specifiers is a barrier for further extraction, the grammaticality differ- ence can be accounted for easily. In other words, we can attribute ungrammaticality of (67b (=69b)) to the topic island as shown in (70): (70) *Which booki did Mary say that [TP to Billj [T' she will give ti tj]] (Culicover (1991: 4)) In (70) which book moves across the second specifier of TP created by topicalization in the embedded clause.20 In order to explain the contrast discussed above, Chomsky (1986) assumes that a subject relative pronoun does not move to the Spec of CP at S-structure and the empty specifier of CP is utilized as an escape hatch for further extraction: (71) ...the child [CP whoj [C' C [TP there is [DP nobody [CP tj [C' C [TP who is willing to accept tj]]]]]]] Our analysis gives a principled explanation to Chomsky's assumption: a subject relative pronoun does not move to any positions other than the first Spec of TP.

6. Summary In this paper we, adopting the Minimalist Program by Chomsky (1995), argued that topicalization is driven by a [Top] feature assigned to T and that a topic element raises to the second Spec of TP for feature checking. We showed that this analysis can account for lack of short subject topicalization in terms of the Economy condition, and that it enables us to explain the distribution of a topic element in the various constructions.

20 I tentatively assume that a projection that has two specifiers is a barrier for further extraction, leaving the principled explanation for the assumption for future research. 50 ENGLISH LINGUISTICS, VOLUME 14 (1997)

We then argued that our analysis opens up a way to reanalyze relativization, which is seemingly unrelated to topicalization, in terms of morphological movement driven by a [Top] feature. As a conse- quence of this analysis, we identify a driving force for movement of a wh-relative pronoun as a [Top] feature assigned to T and proposed that relative clauses with a wh-relative pronoun consist of TP rather than CP with a subject wh-relative pronoun in the first specifier TP and an object wh-relative pronoun in the second specifier.

REFERENCES

Baltin, Mark (1982)“A Landing site Theory of Movement Rule,”Linguistic Inquiry 13, 1-38. Boskovic, Zeljko (1996)“Selection of Infinitival Complements,”Natural Language and Linguistic Theory 14, 269-304. Chomsky, Noam (1982) Some Concepts and Consequences of the Theory of Government and Binding, MIT Press, Cambridge, MA. Chomsky, Noam (1986) Barriers, MIT Press, Cambridge, MA. Chomsky, Noam (1995) The Minimalist Program, MIT Press, Cambridge, MA. Chung, Sandra and James. McCloskey (1983)“On the Interaction of Certain Island Facts in GPSG,”Linguistic Inquiry 14, 704-713. Culicover, Peter (1991)“Topicalization, Inversion, and Complementizers in English,”ms., Ohio State University. Egashira, Hiroki (1997) Topicalization and Relativization: A Minimalist Approach, Master thesis, Tokyo Metropolitan University. Fujita, Koji (1997)“Local Economy and a Generalized Theory of Attract,” Eibungaku Hyoron (Review of English Literature) 69, 97-145, Kyoto University. Fukui, Naoki (1993)“Parameters and Optionalty,”Linguistic Inquiry 24, 329-420. Grimshaw, Jane (1993)“Minimal Projection, Heads, and Optimality,”ms., Rutgers University. Hooper, Joan and Sandra Thompson (1973)“On the Applicability of Root Transformations,”Linguistic Inquiry 4, 459-497. Imai, Kunihiko and Heizo Nakajima (1972) Bun II (Sentences II), Kenkyusya, Tokyo. Imai, Kunihiko, Heizo Nakajima, Shigeo Tonoike, Kimiya Adachi, and Hajime Fukuchi (1989) Ippo Susunda Eibunpoo (Who's Afraid of Bars and Barriers), Taisyukan, Tokyo. Jackendoff, Ray (1977) X-bar Syntax: A Study of Phrase Structure, MIT Press, Cambridge, MA. TOPICALIZATION AND RELATIVIZATION IN MINIMALIST SYNTAX 51

Kuno, Susumu (1973) The Structure of the Japanese Language, MIT Press, Cambridge, MA. Kuno, Susumu (1976)“Subject, Theme, and Speaker's Empathy-Reexamina- tion of Relativization Phenomena,”Subject and Topic, ed. by C. N. Li, 417-444, Academic Press, New York. Lasnik, Howard and Juan Uriagereka (1988) A Course in GB Syntax, MIT Press, Cambridge, MA. Nakajima, Heizo (1992)“Syntactic Differences between Main and Embedded Clauses and split-comp Hypothesis,”Metropolitan Linguistics 12, 18-38. Nakamura, Masaru (1994)“Topicalization, Neg-Preposing, and Locative Preposing,”Current Topics in English and Japanese, ed. by Masaru Nakamura, 151-177, Hituji Syobo, Tokyo. Reinhart, Tanya (1981)“A Second Comp Position,”Theory of Markedness in Generative Grammar, ed. by A. Belletti, L. Brandi, and L. Rizzi, 517-558, Scuola Normale Superore, Pisa. Rizzi, Luigi (1990) Relativized Minimality, MIT Press, Cambridge, MA. Rizzi, Luigi (1991)“Residual Verb Second and Wh-criterion,”Technical Report in Formal and Conceptual Linguistics 2, 23-50, Fuculte des Letters, Universite de Geneve. Rochemont, Michael (1989)“Topic Islands and the Subjacency Parameter,” Canadian Journal of Linguistics 34, 145-170. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik (1985) A Comprehensive Grammar of the English Language, Longman, London. Schachter, Paul (1973)“Focus and Relativization,”Language 49, 19-46.

Ura, Hiroyuki (1994)“Varieties of Raising and the Feature-based Phrase

Structure Theory,”MIT Occasional Papers in Linguistics 7, Department of

Linguistics and Philosophy, MIT.

2-7-16-203 Hashimoto Sagamihara-shi Kanagawa 229-11 e-mail:[email protected]