Disfluency Handling for Robot Teammates

Abstract

Findings from the Psycholinguistic literature have shown that disfluencies may serve a coordination function for both the speaker and the listener. Given that disfluencies are common in spontaneous speech, particularly in high workload tasks, it is important that robots are able to utilize the information contained in disfluencies to improve interaction. My work focuses on this goal of improving coordination in human-robot teams through the detection and utilization of disfluent speech. I use a multi-disciplinary approach that involves conducting empirical studies to investigate how disfluencies influence grounding and coordination in teams, and using the results of these studies to inform the design of computational mechanisms that enable robots to detect and utilize disfluencies in situated interaction. In this paper, I present the results of an empirical investigation and show how the findings are being used to develop computational mechanisms for disfluency handling. Future work involving the integration of these mechanisms in a cognitive robotic architecture is also discussed.