Evaluating description and reference strategies in a cooperative human-robot dialogue system

We present a human-robot dialogue system that enables a robot to work together with a human user to build wooden construction toys. We then describe a study which assessed the responses of naïve users to output that varied along two dimensions: the method of describing an assembly plan (preorder or post-order), and the method of referring to objects in the world (basic and full). Varying both of these factors produced significant results: subjects using the system that employed a preorder description strategy asked for instructions to be repeated significantly less often than those who experienced the post-order strategy, while the subjects who heard references generated by the full reference strategy judged the robot's instructions to be significantly more understandable than did those who heard the output of the basic strategy.