Current syntactic theory features a basic opposition with respect to the status of multiple specifiers. While the 'classical' antisymmetry theory of Kayne (1994) rules out multiple specifiers of a single projection in principle, multiple specifiers are an integral component of 'classical' Minimalism (Chomsky 1995, ch. 4). This paper attempts to resolve this opposition by arguing that multiple specifiers are freely available throughout the syntactic derivation, but ruled out at PF, where the LCA (or its analogue) is presumably relevant.

From an historical standpoint, the original motivation for multiple specifiers in Chomsky (1995) was to provide the escape hatch previously supplied by AGR projections, in contexts where a single projection (e.g. vP) must simultaneously host the landing site for a category moved to check a strong feature and the trace of another category moved to check a feature in a higher projection. The paradigm case of this type is object shift contexts where vP hosts the trace of the moved subject in one specifier and the shifted object in another. The point of departure for this treatment was Ura's (1996) analysis of raising contexts where non-subjects appear to be able to raise over subjects; Ura proposed that this possibility followed from the availability of multiple specfiers in the projection from which extraction occurs. Needless to say, both the object shift configuration proposed by Chomsky and the raising configurations studied by Ura utilize exactly two specifiers, only one of which is visible at PF: the other is always occupied by a trace.

Proposed instances of multiple specifiers filled by two or more overt categories fall into three patterns. The first are Japanese/Korean-style multiple nominative constructions, also studied (although not in detail) by Ura (1). The second are object shift contexts where the subject is claimed not to leave vP, as in Icelandic transitive expletive constructions. The third are South Slavic-style multiple wh movement questions. We argue, in each case citing substantial previous literature, that each of these multiple movement (or multiple move and merge) contexts in fact involve specifiers of distinct projections.

In the case of multiple nominative constructions, arguments that multiple nominative-marked subjects reside in the specifiers of distinct projections are given by Tateishi (1995). We add two arguments here. First, note that MNCs allow stranding of a quantifier associated with the 'outer subject' between it and the 'inner subject', as in (2). Since a stranded quantifier is a diagnostic for the presence of a trace associated with the quantified NP (Miyagawa 1989), this suggests that another specifier, and thus another projection, intervenes between the 'inner' and 'outer' subject. While this argument indicates that MNCs must sometimes involve specifiers of distinct projections, a second argument shows that multiple specifiers are in general not available in MNCs. Kuno (1972) shows that while 'possessor raising' type MNCs are possible in the case of the possessor of a subject (3), they are never available in the case of a possessor of an object (4). This fact is most straightforwardly captured as a (relativized) minimality/shortest move effect: extraction of the possessor of a subject involves no movement over another category; extraction of the possessor of an object involves movement over a subject. But if multiple specifiers (of, say TP) were freely available in MNCs, the latter movement should be licit, since the specifier hosting the base (inner) subject and the specifier hosting the extracted possessor would be equidistant. Alternatively, under a theory built on 'tucking in' of the category moved to a multiple specifier configuration, the pattern in (4) (where the extracted possessor lands to the right of the subject) should be possible. But (4) is completely ungrammatical.

In the case of Icelandic transitive expletives with object shift, we apply recent work of Jonas (in progress) showing that the specifiers hosting the shifted object and the subject are distinct. In the case of multiple wh movement questions, we draw on Boskovic's (1999) arguments that such patterns in South Slavic involve movements of different types to distinct projections. We conclude that multiple specifiers as a derivational concept and a version of the LCA applying at spellout are fully compatible.

Data

(1) Tanaka ga ie ga kazi ni natta. (Japanese)
Tanaka NOM house NOM fire DAT become
'It is Tanaka whose house caught on fire.'

(2) Zyuumin ga kinoo hutari ie ga kazi ni natta.
resident NOM yesterday 2 house NOM fire DAT become
'Two residents' houses caught on fire yesterday.'

(3) *Musume ga Tanaka ga [t nikki] o yonda.
daughter ACC Tanaka NOM diary ACC read
'It is his daughter that Tanaka read the diary (of).'

(4) *Tanaka ga musume ga [t nikki] o yonda.
Tanaka NOM daughter NOM diary ACC read

Boskovic, Z. 1999. Sometimes in SpecCP, sometimes in-situ. In Step-by step: Essays on minimalism in honor of Howard Lasnik, ed. Roger martin, David Michaels, and Juan Uriagereka. MIT Press.

Miyagawa, S. 1989. Structure and Case marking in Japanese. Academic Press.

Ura, Hiroyuki. 1996. Multiple feature-checking: A theory of grammatical function splitting. MIT dissertation.