Embodied conversation agents (ECAs) are virtual humans that are able to converse with people by producing and responding to verbal and non-verbal communication, including varied facial expressions, hand gestures and other body language. Since face-to-face communication is a highly efficient means of interaction, ECAs are a promising technology with which users can enter into dialogues with computers. Reeves and Nass have demonstrated, for example, that even though people are on the whole aware that computers do not have human minds, they are willing to interpret artefacts as “social actors” and ascribe human characteristics to them.
In this project, we investigated novel methods of simulating both the physical and cognitive capabilities of embodied conversation agents. The intention was to improve on previous studies of ECAs, which have primarily originated in the computer science community. The functionality of the application and the embedded agents was a shortcoming in these studies, as few actual embodied conversation agents have significant natural language processing capabilities, and their behaviour is therefore typically canned or highly constrained. Our investigation, however, allowed us to evaluate the persuasive potential of a highly functional ECA.
Date: Aug 2005 – July 2008
Funding: Unilever Corporate Research £101,198
Collaborators: Marco de Boni (Unilever Corporate Research)