Discourse Conditions for Spatial Perspective Taking

Abstract

This paper addresses which reference systems speakers employ in verbal human-robot interaction, under which conditions they employ them, and to what extent speakers stick to one system once they have started to use it. The method is to compare human-robot dialogues that differ only with respect to a single variable, namely the robot's linguistic output. The conversation analytic and corpus analyses of our data show that speakers take path-based descriptions from the partner's perspective to be useful because they are fail-safe, yet they involve constant attention of the speaker; object-naming strategies from the partner's perspective are taken to be most efficient; external relations, such as 'go north', may be used because the artificial communication partner, the robot, seems to be particularly good at it, even though the speaker herself may not be; and egocentric descriptions are easy for the speaker but are believed to be too difficult for the robot. 1.