A method for the coupling of belief systems through human-robot language interaction

This paper describes a method of multi-modal language processing that reflects experiences shared by the user and the robot. Through incremental online optimization in the process of interaction, the robot's system of beliefs, which is represented by a stochastic model, is formed coupling with that of the user. The belief system of the robot consists of belief modules, a confidence that each belief is shared by the user (local confidence), and a confidence that all the belief modules and the local confidence are identical to those of the user (global confidence). Based on the system of beliefs, the robot can interpret even fragmental and ambiguous utterances, and can act and generate utterances appropriate for a given situation.