Adaptation to affective factors: architectural impacts for natural language generation and dialogue
John A. Bateman, Cécile L. Paris · 2005
We are currently developing automatic natural language dialogue capabilities for assistence systems where the perceived ‘naturalness’ and ease of use for the systems’ users is a central consideration. We are, in addition, aiming at generic solutions that will enable a broad range of users to be supported. Our applications focus on issues that arise in the spatial domain—that is, our assistence systems are concerned with shared user-system tasks that are to be carried out in both shared and non-shared spatial configurations. The potential for miscommunication in these kinds of situations is extremely high since the perception and representation of space by our systems and by their human users is generally quite different; nevertheless, the interaction between user and system must be made to work even in the face of mutual non-comprehension. This potentially frustrating situation must be explicitly recognised and catered for in order to keep the system and user in an effective working relationship, thus raising issues of producing language with appropriate affect and of recognising the user’s affective state. In this paper, we present how this capability is being added into our dialogue architecture as an integral part of the mechanisms for dialogue support in general. We relate this to existing mechanisms for affective language generation/interpretation, user-adaptability and the importance of maintaining a tight inter-relationship between language and situation.