Learnability in the Context of Lexicalized Parameters: A Compositional Approach to Language Acquisition

Abstract

Language acquisition is inextricably linked to the structure of language itself. Consequently, syntactic acquisition must address the structure of syntactic representation. Research on cross-linguistic syntactic variation and acquisition has repeatedly demonstrated the various difficulties in modeling a Principles and Parameters-style grammar and learning mechanism. Perhaps, such difficulty stems from the inadequacy of a Principles and Parameters-style representation of language itself. This paper combines relevant aspects of Minimalist syntax and a learning-by-parsing approach, in the vein of Fodor and Teller (2000), in order to motivate a reanalysis of learnability using lexicalized parameter values. The learner’s goal is no longer to set grammar-wide parameter values, but rather to learn how individual lexical items are merged. In this paper, a new psychologically-plausible learnability model is mapped as the foundation for future research. Furthermore, by abstracting away from mental primitives, this paper also notes how tree-adjoining grammars may be used as computational proxies for lexicalized parameters, providing empirical ways to further study this proposed learning algorithm.