A Framework for Building Conversational Agents Based on a Multi-Expert Model

This paper presents a novel framework for building symbol-level control modules of animated agents and robots having a spoken dialogue interface. It features distributed modules called experts each of which is specialized to perform certain kinds of tasks. A common interface that all experts must support is specified, and any kind of expert can be incorporated if it has the interface. Several modules running in parallel coordinate the experts by accessing them through the interface, so that the whole system can achieve flexible control, such as interruption handling and parallel task execution.