Reacting to Agreement and Error in Spoken Dialogue Systems Using Degrees of Groundedness

Abstract

Computational models of grounding are extended to include representations of degrees of groundedness. These representations are then used for decision-making in dialogue management for spoken dialogue systems. Several domains will be explored with this model, and an implementation will be tested and evaluated. Background The general public is becoming increasingly aware of spoken dialogue systems: as telephone-based Interactive Voice Response systems, for in-car navigation, or for personal computer control, for example. Researchers study other domains such as tutoring, collaborative planning, and interacting with virtual humans. However, both commercial and research systems must become more robust before they will be widely accepted. This research proposes to extend ideas from artificial intelligence and psychology of language to build dialogue systems that model degrees of groundedness to track and react to the extent to which the system and human user are in agreement. Most models of discourse, be they from philosophy, linguistics, artificial intelligence, or psychology, use some version of a notion of common ground, the beliefs that the participants of that discourse have in common (Clark and Marshall, 1981). (Clark and Schaefer, 1989) demonstrates the importance of studying how material is added to that common ground, and defines grounding as a collaborative activity between participants who work towards establishing a set of mutually-held beliefs. Behavior such as confirmations ("OK, you said seven o'clock"), backchannelling ("uh-huh"), and corrections ("I said ten, not seven") can be explained by models of grounding. Participants work towards the grounding criterion, at which point both parties are satisfied that they agree on the