Jacqueline C. Kowtko

Using prosodic information to constrain language models for spoken dialogue

We present work intended to improve speech recognition performance for computer dialogue by taking into account the way that dialogue context and intonational tune interact to limit the possibilities for what an utterance might be. We report on the extra constraint achieved in a bigram language model, expressed in terms of entropy, by using separate submodels for different sorts of dialogue acts, and trying to predict which submodel to apply by analysis of the intonation of the sentence being recognised.

ON THE INTONATION OF MONO- AND DI-SYLLABIC WORDS WITHIN THE DISCOURSE FRAMEWORK OF CONVERSATIONAL GAMES

Recent studies on the analysis of intonational function examine a range of materials from cue phrases in monologue (Litman and Hirschberg, 1990) and dialogue (Hirschberg and Litman, 1987; Hockey, 1991) to longer utterances in both monologue and dialogue (McLemore, 1991). Results match specific intonational tunes to certain discourse functions which are more or less well defined. Although these results make a convincing case that intonation does signal a change in discourse structure, the specification of discourse function remains vague. A suitable taxonomy is needed to fine-tune the relationship between intonation and discourse function. A recent analysis of dialogue (Kowtko et al., 1991) provides a framework of conversational games which allows more fine-grained examination of prosodic function. The current paper introduces an intonational analysis of mono-and di-syllabic words based upon such a framework and compares results in progress with previous work on intonation.