ISCA Archive Interspeech 2004
ISCA Archive Interspeech 2004

Automatic detection of dialog acts based on multilevel information

Sophie Rosset, Lori Lamel

This paper reports on our experience in the automatic detection of dialog acts in human-human spoken dialog corpora. Two hypotheses underlie this work: first, word position is important in identifying the dialog act; and second, there is a strong grammar constraining the sequence of dialog acts. A Memory Based Learning approach has been used to detect dialog acts. Experiments are carried out with a known number of utterances per speaker turn, and with a hypothesized number of utterances determined using a language model for automatic utterance boundary detection. In order to verify our first hypothesis, the model trained on a French corpus was tested on an English corpus for a similar task and on a French corpus from a different domain. A correct dialog act detection rate of 83% is obtained for the same domain and language conditions and about 75% for the cross-language and cross-domain conditions.