Third International Conference on Spoken Language Processing (ICSLP 94)
In this paper, we present a constructive model of man-machine dialogue for problem solving. Our spoken dialogue system aims to solve scheduling problems cooperatively by interacting user and system. An important feature of this model is that it separates the knowledge about dialogue structure (conversational space) and about problem (problem solving space) in order to make correspondence to utterances and actions. Conversational space is designed for identifying user's speech act and constructing dialogue segment. Problem solving space represents the structure of the problem and the most appropriate direction for solving problem. For incorporating probabilistic reasoning, each space is represented by network. This time, we adopt this model to keyword based plan recognition and topic prediction of next utterance. In small experiment, this system hit the topics of next user's utterance at 47% at the first candidate, and at 92% within the third candidates.
Bibliographic reference. Araki, Masahiro / Watanabe, Taro / Quimbo, Felix Merida / Doshita, Shuji (1994): "A cooperative man-machine dialogue model for problem solving", In ICSLP-1994, 883-886.