高橋和子
Dialogue Model Based on Data Transfer
Abstract:In this paper, we propose a dialogue model based on data transfer. We deal with the dialogues having a goal of knowledge acquisition of a designated fact, such as hotel reservation and route guide.
In this model, a dialogue is regarded as a sequence of inner states representing agents'
beliefs, and a new modal operator need-to-know is introduced to describe the timing of
utterances. For example, when an agent is conveyed some specific fact, she believes that
she needs to know another related fact. Then, the belief (called a seed) invokes the next
utterance. We show that this mechanism is simple enough to implement.
Moreover, interactive belief revision can be handled in this model. If each data is represented as a proposition, we cannot express inconsistency between a pair of data. Thus, we use a data type feature, which is the pair of a label and a value. If two features have the same label and different values, then they are inconsistent. We show that the processes of confirmation and correction to achieve the mutual belief can be treated in a unified manner.