ABSTRACT This paper presents an approach to extraction of dialog acts and topics from utterances in a spoken dialog system. Two knowledge sources are used to describe the dialog history. One is a transition network of dialog acts and the other is a tree of topics which might appear in domain communications. Dialog acts and topics are extracted through bottom-up and top-down analyses. Bottom-up candidates are decided by applying a set of specially designed rules to the semantic representation of an utterance, and top-down candidates by using the current state of the dialog history. The logical ANDs between bottom-up and top-down candidates are taken to decide the dialog act and topic of an utterance. This method was examined with a corpus of fourteen dialogs including 335 utterances. Correct extraction rates were 85% for the topic and 82% for the dialog act.
Cite as: Niimi, Y., Takinaga, N., Nishimoto, T. (1998) Extraction of the dialog act and the topic from utterances in a spoken dialog system. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 1138, doi: 10.21437/ICSLP.1998-500
@inproceedings{niimi98_icslp,
author={Yasuhisa Niimi and Noboru Takinaga and Takuya Nishimoto},
title={{Extraction of the dialog act and the topic from utterances in a spoken dialog system}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 1138},
doi={10.21437/ICSLP.1998-500}
}