Changing concepts of machine dialogue management

Abstract

An important research question in machine dialogue management is how to go beyond hand-crafted approaches. Currently, statistical approaches are prevalent. However, recently, researchers have focused again on the representational capacities of the human language processing system. The point of departure for this paper is that this methodological shift is necessary if we aim at designing advanced conversational agents. In line with this, the paper introduces an extension of the focus tree model of attentional information in machine dialogue. It proposes an approach to automatic construction of a focus tree for a given interaction domain. In addition, the paper provides a brief insight into previous work on this model in order to report on a representational, cognitively-inspired and domain-independent approach to machine dialogue management. Finally, the paper discusses selected aspects of a prototype conversational agent designed to maintain natural language interaction in therapeutic settings between the therapist and the robot.