Integrating Joint Intention Theory, Belief Reasoning, and Communicative Action for Generating Team-Oriented Dialogue

摘要

The goal of this research is to develop an architecture that can guide an agent during collaborative teamwork. The architecture should generate communication and dialogue during the performance of collaborative multi-agent tasks as a byproduct of the agent's rationally pursuing its intentions. This paper describes how a joint intention interpreter that is integrated with a reasoner over beliefs and communicative acts can form the core of a dialogue engine. As an interpreter of joint actions, the architecture enables agent programmers to describe a domain declaratively, specifying an agent's individual and joint intentions, plans and actions at a high level. The interpreter attempts to fulfill the agent's individual and joint intentions, subject to its beliefs and mutual beliefs. As a consequence, the system engages in dialogue through the planning and execution of communicative acts necessary to attain the collaborative task at hand. The dialogue engine is general enough to be applicable to both agent-agent and agent-human teams. The system has been implemented in a combination of Java and Prolog, and can be shown to obey the predictions of joint intention.