Deriving Multiple Head and Phrasal Movement: The Cluster Hypothesis

1 Introduction

A central claim of the Minimalist Program (Chomsky 1993, 1994, 1995) is that movement is triggered solely by the need to check features. In this squib I will argue that this analysis provides a natural account for the formation of complex (multisegmental) categories—that is, for "cluster formation." I assume the Cluster Hypothesis (CH).

(1) Cluster Hypothesis

A feature F that is attracted by K attracts a feature of the same type F.

(1) is only a descriptive generalization. In fact, (1) interacts with Attract F (see Chomsky 1995).

(2) Attract F

K attracts F if F is the closest feature that can enter into a checking relation with a sublabel of K.

(3) Closeness

β is closer to K than α if β c-commands α.

Given the definitions in (2)-(3), and given a structure like [ . . . α . . . β . . . γ . . . ] where α asymmetrically c-commands β and β asymmetrically c-commands γ, α cannot attract γ because β is closer to α. However, β can attract γ and α can attract the complex [_β β γ] (or [_β γ β]) if, as stated in (1), the feature F of β that is attracted by α may itself attract the analogous feature of γ. As shown in section 2, independent evidence for cluster formation as a result of (1) can be gained from complex X⁰ categories consisting of multiple verbs and multiple clitics. Furthermore, in section 3 I will argue that in multiple wh-and focus constructions, cluster formation applies overtly in languages such as Bulgarian and Malagasy and covertly in Japanese.

2 Head Movement and the Cluster Hypothesis

Assuming with Chomsky (1995, to appear) that (syntactic) head movement is triggered by a [+affix] feature, we can account for the distribution of several incorporation phenomena on the basis of (1). Consider first verb incorporation as discussed in Baker 1988:371. Baker notes that in Quechua, multiple verb incorporation, transformationally derived from a base like (5a), takes the form in (4b) (= (5b)) and not (4a). As pointed out by Baker, (4a) shows that it is impossible to incorporate the verbs 'make' in CP₂ and 'eat' in CP₃ directly and independently of each other into the matrix verb. Assuming that incorporation in (4) is triggered by a feature [+affix] (i.e., that 'want', 'make', and 'eat' realize a [+affix] feature), 'want' may not attract 'eat' independently of 'make' as in (4a). According to the CH, the only possible derivation is (4b) (= (5b)), where 'eat' (V₃) is attracted by 'make' (V₂) and the complex verb'eat+make' is attracted by the matrix verb'want' (V₁). This derivation yields the well-formed (4b).

(4)

1. a. *Mikhu-naya-chi-wa-n. eat-want-make-1SG.OBJECT-3SUBJECT 'It makes me feel like eating.'

2. b. Mikhu-chi-naya-wa-n. eat-make-want-1SG.OBJECT-3SUBJECT

(5)

1. a. [_CP1 I want [_CP2 PRO make [_CP3 him eat]]]

2. b.

Clitic-climbing phenomena provide further evidence for the CH. Let us consider the following examples from Spanish. I follow Jaeggli (1986) and Sportiche (1996), who assume that clitics in Spanish are associated with an overt lexical NP/DP or pro. The former case represents the "clitic-doubling" construction.

(6) Juan y Maria la            llamaron a ella/pro.

Juan and Maria CL.3SG.FEM called her 'Juan and Maria called her.'

Furthermore, I assume that (as proposed by, e.g., Suñer (1988), Fernández (1989), Franco (1991), Runner (1991), Zubizarreta (1992), Sabel (1995), and Sportiche (1996)) clitics must be analyzed as heads of a functional projection (e.g., as Agr elements) or (as proposed by Torrego (1996)) they must be analyzed as heads of a DP complement selected by the light verb (as in . . . [_vP [v] [_DP [D] [_VP DP [V] . . . ]]]).

Let us now turn to examples with multiple clitic climbing. Examples (7)-(8) contain two restructuring verbs, querer 'want' and permitir 'allow' (Aissen and Perlmutter 1983:366, 367). Four derivations are possible. The clitics may remain in the sentences in which they are base-generated (7a); the clitic from CP₃ may move to CP₂ (7b); the clitic associated with permitir 'allow' may move to CP₁ (7c); or both clitics...