Dialogue Act Prediction Using Stochastic Context-Free Grammar Induction

Abstract

This paper presents a model-based approach to dialogue management that is guided by data-driven dialogue act prediction. The statistical prediction is based on stochastic context-free grammars that have been obtained by means of grammatical inference. The prediction performance of the method compares favourably to that of a heuristic baseline and to that of n-gram language models. The act prediction is explored both for dialogue acts without realised semantic content (consisting only of communicative functions) and for dialogue acts with realised semantic content.