Lecture: Brian Riordan

The next CL lunch talk will be given by Brian Riordan.

When: Friday, Jan. 30, 12:00-1:00 p.m.
Where: Memorial Hall 401

Brian Riordan will present, "Redundancy in perceptual and linguistic experience: Comparing feature-based and distributional theories of semantic representation."

Abstract:
In recent years a variety of machine learning techniques have been developed to learn semantic representations directly from linguistic structure. Variations of these approaches have been used in cognitive science to model a variety of semantic phenomena, from semantic priming to word association. The success of these approaches suggests that much information relevant to semantic representations is encoded in the linguistic environment. However, these models have been criticized as inadequate cognitive theories of human semantic learning and representation because their representations are purely symbolic and are not grounded in perception and action. Models based on human-generated features have been argued to give access to such perceptual aspects of lexical semantic representation. I explore the hypothesis that the amount of perceptual information that can be learned from purely distributional statistics has been underappreciated. In this work I demonstrate that, in a semantic clustering task, several distributional models perform as well as human-generated feature-based representations. Furthermore, when trained on child-directed speech, the same distributional models perform as well as sensorimotor-based feature representations of children's lexical semantic knowledge. These results provide support for the hypothesis that, to a large extent, information relevant for extracting semantic categories is redundantly encoded in perceptual and linguistic experience.