Knowledge Representation for Generating Locating Gestures in Route Directions

When humans give route directions, they use gestures to indicate the location of landmarks. The form of these gestures reflects that speakers take one of several perspectives when producing them. They may locate the landmark with respect to the speaker, with respect to the person following the route, or with respect to other landmarks. A corpus study shows that which perspective is chosen is partly determined by the function of the discourse segment these gestures occur in. Since locating gestures are so prevalent in direction-giving, in this paper we address the kinds of dialogue information and knowledge representation that is needed to generate them automatically.