EVALUATION OF ECA GESTURE STRATEGIES FOR ROBUST HUMAN-MACHINE INTERACTION

Abstract

Embodied Conversational Agents (ECAs) offer us the possibility to design pleasant and efficient human-machine interaction. In this paper we present an evaluation scheme to compare dialogue-based speaker authentication and information retrieval systems with and without ECAs on the interface. We used gestures and other visual cues to improve fluency and robustness of interaction with these systems. Our tests results suggest that when an ECA is present users perceive fewer system errors, their frustration levels are lower, turn-changing goes more smoothly, the interaction experience is more enjoyable, and system capabilities are generally perceived more positively than when no ECA is present. However, the ECA seems to intensify the users' privacy concerns.