Steps towards Multi-Party Dialogue Management

Abstract: Steps towards Multi-Party Dialogue ManagementAbstract The aim of this work is to enable the construction of artificial conversational agents capable of working in a multi-party setting. By an artificial conversational agent we mean a software agent (possibly with embodiment, personality etc.), capable of engaging in conversation with other conversational agents (human or artificial). By a multi-party setting we mean that more than two agents are engaged in the communication or dialogue. Components designed for two-party dialogue management including sequencing and turn management cannot, in general, handle multi-party dialogue phenomena, such as keeping track of the obligations regarding turn management and the rights and obligations to address an issue.We present an SCXML implementation of Issue-Based Dialogue Management, scaled up to handle more than two participants. The QUD-based dialogue manager makes a distinction between distributive questions (to be answered individually by each addressee) and collective questions (to be answered by the addressees collectively), in order to provide the right kind of grounding behaviour and in order to manage correctly the obligations put on the participants of the dialogue. The dialogue manager also features an SCXML implementation of the SSJ turn management model.We find that it is possible to scale up the Issue-Based Dialogue Management model from two-party to multi-party dialogues, given that distributive and collective readings of questions can be identified.KEY WORDS: SCXML, multi-party, multi-party dialogue, dialogue management, turn management, QUD, issue-based dialogue management.The thesis is written in English.