Grammatical number in numeral-noun constructions

A common assumption in current generative literature on cardinal numerals is that, with few exceptions, the structural relation between a numeral and the noun that it combines with is uniform, cross-linguistically as well as language-internally. Thus, the debate whether numerals are heads that select a projection of the noun as their complements (Ionin and Matushansky 2006), or specifiers of an (extended) projection of the noun (Corver and Zwarts 2006), has usually been carried under the assumption that there is just one analysis that applies to (almost) all numerals, with the possible exception of the often-noticed differences between ‘adjectival’ and ‘nominal’ numerals. In a recent paper, Ionin and Matushansky (2006)(henceforth IM) argue convincingly that the case-related properties of (some) numerals in Russian, Finnish and Inari Sami provide strong evidence for viewing them as nominal heads that recursively take another nominal projection as a complement; this analysis is then generalized to all other numerals and other languages. The first goal of this paper is to show that this reasoning is not valid; using data from a variety of languages, I argue that numerals combine with nouns in at least two distinct ways, and that IM’s head-complement analysis is only compatible with some numeral-noun constructions. The second goal is to show that the distribution of NUMBER features plays a central role in restricting the choice of structure for numeral-noun constructions. In Hebrew, cardinal numerals may appear in two distinct forms, which I refer to as free and bound. Free numerals, as in (1a), are used when there is no overt noun, or with indefinite nouns (with some exceptions to be discussed below); while bound numerals, which resemble heads of construct state nominals, are used mostly with definite nouns, as in (1b):