Grounding Information in Route Explanation Dialogues

Abstract

We have studied the role of several factors influencing the grounding of information in a French corpus of direction-giving dialogues. We focused on how speakers agree on specific locations that are parts of a route explanation. Spatial knowledge is thus seen here as a means of a precise semantic analysis of a certain type of dialog act (feedback responses). We show here the different kinds of feedbacks that can be accounted for by our data, looking at the type of lexical cues in utterances, the type of dialog act targeted by the feedback and the span of the conversation that falls under the scope of such a feedback. Dialog acts also depend on the type of spatial descriptions that are used (motion description, introduction of landmarks or description of landmarks). We then propose an interpretation of this as different mechanisms for speakers' agreement.