The McGurk effect in symmetric neural networks

Abstract

This article brings together two topics on which Paul Boersma and I have worked together: the interaction of auditory and visual cues in speech perception, and symmetric neural networks as a decision mechanism. The McGurk effect, first described by McGurk and MacDonald in 1976, reveals how seer-listeners integrate conflicting visual and auditory information, and how the former can even override the latter. Boersma provided an Optimality-Theoretic formalization of this process; in this article, I present two neural network models to show that the McGurk effect is an emergent epiphenomenon in the perception of speech.