Incremental dialogue act understanding

This paper presents a machine learning-based approach to the incremental understanding of dialogue utterances, with a focus on the recognition of their communicative functions. A token-based approach combining the use of local classifiers, which exploit local utterance features, and global classifiers which use the outputs of local classifiers applied to previous and subsequent tokens, is shown to result in excellent dialogue act recognition scores for unsegmented spoken dialogue. This can be seen as a significant step forward towards the development of fully incremental, on-line methods for computing the meaning of utterances in spoken dialogue.