Exploiting Conversational Topic Awareness for Interactive Robots

This paper proposes a computational approach to spot conversational topics for socially interactive robots. A speech recognition system called Julius has been used to transcribe the audible speech of an interlocutor to text. A rule-based topic-aware chatbot system coupled with a real-time web-based information extraction system have been employed to generate appropriate speech outcome for a humanoid robot named ROBIN. The conversational topic rules have been implemented on top of the existing chatbot system. Wikipedia contents have been extracted, allowing the robot to fetch factual information on-the-fly. Bodily movements of the robot have been generated based on the emotional state of the speech delivered by a human interaction partner. Additionally, the robot's vision system has been utilized so as to trigger event-specific behaviour out of the robot. The robot has been used in various interaction scenarios to establish topically-relevant affective interaction with humans. The experimental results show promising improvement over the earlier interaction system as far as the incorporation of a topic-aware conversation system is concerned.