An Empirical Investigation of Typicality and Uniqueness Effects on Article Choice in Attributive-Possession NPs*

0. Introduction Previous analyses of the use of English definite descriptions have generally taken one of two approaches, loosely characterizable as familiarity and uniqueness. That is, felicitous use of the definite article has been argued to require that the referent of the NP be either known to the hearer within the current context of utterance (e.g. Heim 1983, Prince 1992) or uniquely identifiable to the hearer (e.g. Gundel et al. 1993, Birner & Ward 1994, inter alia). What is common to all previous analyses is that the explanatory principle is presented as categorical; i.e. a referent is familiar or not, unique or not. There is generally no provision for gradience within the theory. In what follows we will investigate how familiarityand uniqueness-based accounts of definiteness fare in accounting for the class of EMBEDDED DEFINITES known as ATTRIBUTIVE-POSSESSION constructions (McKercher 2001) and how the gradient notion of typicality impacts article choice in these constructions. Attributive-possession NPs (henceforth APNPs), which consist of embedded NPs headed by with (e.g. the man with the cane), can be used to refer to one of the figures in (1):