Coupling of Task and Partner Model: Investigating the Intra-Individual Variability in Gaze during Human–Robot Explanatory Dialogue

摘要

In a successful dialogue in general and a successful explanation in specific, partners need to account for both, the task model (what is relevant for the task) and the partner model (what one can contribute). The phenomenon of coupling between task and the partner model becomes especially interesting in the context of Human–Robot Interaction where humans have to deal with unknown capabilities of the robot, which can momentarily be perceived when the robot is unable to contribute to the task. Following research on the path over manner prominence in an action [31, 32, 33], a robot explained actions to a human by emphasizing two aspects – the path ("where" component) and the manner ("how" component). On critical trials, the robot occasionally omitted one of these components where participants sought missing information for the path or the manner. Participants’ information-seeking and gaze behaviour were analysed. Analysis confirms the initial predictions for, a) task model (path over manner prominence), i.e., earlier information-seeking for path-missing than manner-missing trials, and b) partner model, i.e., while information-seeking is predominantly tied to the attention on the robot’s face, when robot fails to provide resolution, attention shifts more often towards its torso – a behavior likely to indicate an exploration of the robot’s capabilities. An individual-level analysis further confirms that the intra-individual variation in the task model is partly influenced by the perceived capability of the robot.