A visual anthropomorphic agent with learning capability of cooperative answering strategy through speech dialog

As the opportunity of using computer systems spreads into everyday life, the importance of friendly human interfaces is increasing. As a form of next generation human interfaces, an anthropomorphic interface agent which mimics a face-to-face communication holds promise, and some early developments have started. Its multimodality including facial expression and speech dialog fits to human perception and can enhance the friendliness of the interface. We present an anthropomorphic interface agent called VSA (Visual Software Agent), which has a moving realistic facial image and a speech dialog function. Unlike other anthropomorphic interface systems, our VSA system has connection to a WWW browser (Netscape Navigator), so that it can serve as a new interface to a vast WWW information space and effectively use multimedia data written in a standardized HTML format. As immediate applications of the VSA, it is suitable for guidance systems which are used by various people with little knowledge of computers at, for example, department stores, company reception desks, university campuses, etc. We have implemented learning capability into our anthropomorphic agent. We use reinforcement learning, in particular, profit sharing method; but our research is unique in that the learning mechanism is implemented to acquire knowledge from speech dialogs. We show our implemented learning mechanism with reference to a task of campus guidance.