Investigating the Contribution of Distributional Semantic Information for Dialogue Act Classification

This paper presents a series of experiments in applying compositional distributional semantic models to dialogue act classifica-tion. In contrast to the widely used bag-of-words approach, we build the meaning of an utterance from its parts by composing the distributional word vectors using vec-tor addition and multiplication. We inves-tigate the contribution of word sequence, dialogue act sequence, and distributional information to the performance, and com-pare with the current state of the art ap-proaches. Our experiment suggests that that distributional information is useful for dialogue act tagging but that simple mod-els of compositionality fail to capture cru-cial information from word and utterance sequence; more advanced approaches (e.g. sequence- or grammar-driven, such as cat-egorical, word vector composition) are re-quired.