Perceptual Science Series
Modeling believable human-computer interaction with an embodied conversational agent (ECA): face-to-face communication of uncertainty
Dr. Insuk Oh
Monday, October 09, 2006, 01:00pm - 02:00pm
School of Communication, Information, and Library Studies and the Department of Computer Science
Given that an agent must act under uncertainty and humans communicate uncertainty during conversation, endowing an embodied conversational agent (ECA) with the ability to express and recognize it becomes one of the key challenges. The primary goal of the research was to develop an integrated framework for a systematic way of studying human behaviors for ECAs to achieve believable interaction. The research consists of four distinct stages: the conversation experiment to elicit spontaneous uncertainty expressions (Stage I), the signal identification experiment to find reliable uncertainty signals by separate communication channels (Stage II), the first coding stage to analyze human behaviors (Stage III-Part 1), the second coding stage to add human behaviors to an ECA (Stage III-Part 2), and the recognition experiment to identify reliable uncertainty signals with the ECA (Stage IV). The study results provide support for the existence of reliable audiovisual cues of uncertainty in human-human interaction. When identified visual cues were added to our ECA, RUTH (Rutgers University Talking Head), people could reliably recognize uncertainty signals from RUTH, too. We concluded that our research provided empirical evidences of uncertainty expression and recognition during face-to-face communication and, further, a methodologically solid framework of adding multimodal human behaviors to ECAs for believable interaction between a human and an ECA.