Generating Effective Responses for a Plan-based Dialogue System in Human-robot Interaction

A high-level dialogue system involves symbolic reasoning and deep understanding of user input, and its object is to handle complex dialogue phenomena such as inference and negotiation. However, it is difficult for high-level dialogue techniques to be extended toward a practical dialogue system that must generate effective responses in a real environment because the high-level dialogue techniques are based on only a narrow band of knowledge and a complicated model. This paper proposes a new dialogue model based on plan-recognition model that is one of high-level techniques. We simplify the plan-recognition model, which includes well-defined recipes and inference rules, to be applied to a practical dialogue system. In addition, we use a multi-level inference schema that employs a discourse stack and a confirmation strategy for effectively recognizing the user's deep intention. As a result, the proposed dialogue system is successfully embedded into a multi-modal interaction system for humanrobot interaction.