LingLunch : Stephen Jones

Stephen Jones (Universiteit Groningen)
Grammar-based cognitive modelling of language

Human language processing is a long-established topic for cognitive modelling. Models including that of Lewis & Vasishth (2005) have replicated experimental time-courses for processing complex sentence structures, but this has been at the expense of a robust syntactic representation. One contributory factor to the poor fit between grammar and cognitive models has been the inability of many linguistic theories to account for the grammatical constraints that operate during the incremental growth of an utterance. This talk reports on the elaboration of an incremental theory of grammar from an established formalism, Lexical Functional Grammar (Bresnan & Kaplan, 1982), and the development of a model in the ACT-R cognitive architecture that can parse a set of complex sentences in English and Korean. In the model, a single production set describes the combinatorial operations used to build structure in either language, and language-specific grammatical constraints are held in the lexicon. The model also includes an initial treatment of the relationship between prosody and syntax, using the single production set to replicate observed effects from both languages. Although the model is able to generate structural representations of sentences that match human interpretations, architectural constraints within ACT-R prevented the replication of time-courses. The talk concludes with a discussion of these issues and identifying the directions for future research.