One of the most elaborate Nativist accounts of Verb Argument Acquisition (VAS) is Pinker's (1984, 1989) "semantic bootstrapping". According to Pinker, early acquisition of VAS is regulated by a "canonical mapping" scheme; that is, a default mapping between thematic roles and syntactic functions such that most AGENT roles are initially assigned to the Subject, most THEME/PATIENT roles to the Direct Object, and most LOCATION/GOAL/ SOURCE roles to the Oblique Object. The proposed mapping scheme is assumed be universal, and to reflect properties of children’s innate capacity for language acquisition.

To evaluate Pinker's account, the present paper uses naturalistic longitudinal data collected on a bi-weekly basis from two Hebrew-speaking girls, Smadar and Lior, between ages 1;5-2;9. Analysis of the Hebrew data suggests that a canonical mapping scheme of the sort proposed by Pinker may not, in fact, facilitate VAS acquisition, and moreover, that it may not be sufficient to account for  VAS acquisition crosslinguistically. Instead, I argue that  VAS is initially learned for individual verbs, and that the initial choice of arguments appears to be determined by input and by pragmatic factors like new versus old information. These claims need to be fully established by future research.