CLASP
The Centre for Linguistic Theory and Studies in Probability

The Informative Speech Act: Linguistic Disambiguation on the basis of Epistemic Probability of Interpretations

Abstract

In a series of influential papers, Lassiter and Goodman have proposed to use a variant of the rational speech act model to account for semantic learning of linguistic parameters of lexical items. In this talk, I propose a reformulation of the Lassiter-Goodman model in information-theoretic terms. This reformulation sets the stage not only for a critique of the model, but shows that semantic learning can be given a direct epistemic account for linguistic disambiguation, doing away with RSA. I will conclude the talk with a comparison of the two models.