Identifying Communicative Functions in Dialogue Systems

摘要

Improvements of naturalness in voice human-machine interaction systems can be achieved also by modelling specific aspects of spoken language use like the so-called disfluency phenomena many of which are, despite their name, very important in communication as they strongly contribute to convey speaker's communicative intentions. Filled pause, for example, are planning strategy used by speakers to signal their intention of "holding the floor" in a conversation. In previous work on Bari variety of Italian, it was established that duration and F0 shape can be considered as important acoustic cues for identifying a special category of filled pauses, which are produced in Italian by prolonging the ending vowel of a word. In this paper, a classification task has been carried out in order to assess the reliability of the above mentioned statistically determined paramenters. Results confirm that duration and F0 shape are reliable acoustic cues for identifying "word final lengthening" filled pauses in 3 varieties of Italian, namely those of Bari, Naples and Pisa. Results also confirm that such parameters can be successfully used in prosodic boundary detection in Bari Italian spontaneous speech, which conveys speaker's intentionality like, for example, discourse organisation.