Abstract:

This paper describes a child’s mental model of an Embodied Conversational Agent (ECA) acting as a learning Companion. Embodied Conversational Agents are screen-based characters that interact with humans through the medium of conversation. Acting as a learning Companion, an ECA can support children or offer guidance toward improving the learning process. When a child uses an ECA they necessarily form some sort of mental model of the ECA and this will affect the manner in which they learn and ultimately determine the effectiveness of their learning. This paper considers what a child’s natural mental model of an ECA is, and how we can best develop an ECA interaction strategy to fit that mental model.

ECAs have previously been encountered by users acting in the role of assistants, sales advisers, guides or tutors. They are required to create a comfortable experience for the user and they need to cooperate, provide appropriate feedback, and encourage analytic or rational behaviour rather than dependent conduct. However, little is know about the mental model that children have of these synthetic characters, how they perceive them or what their expectations are regarding what they are willing to share with these synthetic entities.

There is no existing method to effectively measure someone’s mental model so we designed and implemented our own methodology based on the Wizard of Oz protocol, focus groups and the repertory grid. We began by introducing groups of children to a prototype Companion emulated using the Wizard of Oz Protocol. Next we ran a focus group session to analyse the words children use to conceptualize the Companion. After this we created a repertory grid with each child to find out which concepts they associated with the Companion. At the end of each session we held 10 minute interviews with groups of children in order to capture some of the more relevant perceptions toward learning Companions.

This allowed us to confirm the results of previous studies which suggest that when interacting with ECAs people try to invent a relationship in which Companions are expected to combine human and agent behaviour. In addition, results suggest that personalization has an important role in the engagement that children develop with learning Companions. For example, children were affected when the Companion used personal details such as their name or mentioned some detail from a previous encounter. Another important finding is the wiling suspension of disbelief that children exercised when communicating with the Companion. In all likelihood the children understood the Companions were not sentient, as they had encountered similar synthetic agents in video-games or television, but none-the-less they were happy to play along and share personal details or opinions. It’s hoped that these results can help us form a better model of what a learning Companion is capable of and subsequently implement improved interaction strategies in the future.

Keywords:

Virtual Learning Companions, Embodied Conversational Agents, Interaction Design.