Towards robust agent-based dialogue systems

Summary form only given. There has been several decades of work in AI in defining conversational agents that based on agency, that model dialog in terms of reasoning about goals, plans, and intentions. While theoretically powerful, these projects have not resulted in dialogue systems with the level of robustness attained using statistical information extraction approaches. Because it uses deeper understanding of language and can involve reasoning processes, however, the agent-based approach has the potential for handling much richer applications than the statistical approaches. For the past decade we have been working to develop models that find a good balance between generality and robustness. I would describe these efforts and propose some ideas on further bridging the gap between the two technologies