Incorporating Social Practices in Dialogue Systems

Abstract

AbstractCurrent dialogue management systems do not take social concepts such as norms, conventions, roles etc. into account when managing dialogues. Neither do they keep track of the personal (mental) state such as goals, needs, etc. While the data-driven approaches work quite well in some cases, they are usually domain/user dependent and not transparent. On the other hand, the rule-based methods can only work on the predefined scenarios and are not flexible in that sense. In addition, these approaches are limited to modeling only the dialogue system and do not include the human participant as part of the overall dialogue. This makes the current dialogue systems not well suited for complex and natural dialogues. In this paper, we present a dialogue management system framework that incorporates the notion of social practices as a first step to extend the type of dialogues that can be supported. The use of social practices is meant to give structure to the dialogue without restricting it to a fixed protocol. We demonstrate the use of the proposed system on a scenario between the doctor and patient roles where the doctor is a medical student and the patient is simulated by the dialogue management system. KeywordsDialogue systemsSocial AIConversational AISocial practicesChatbots