Personalisation and automation in a virtual conversation skills tutor for children with autism

Existing research involving children with autism suggests that autonomous (self directed) virtual humans can be used successfully to improve language skills, and that authorable (researcher controlled) virtual humans can be used to improve social skills. This research combines these ideas and investigates the use of autonomous virtual humans for teaching and facilitating practice of basic social skills in the areas of greeting, conversation skills, listening and turn taking. The Social Tutor software features three virtual human ‘characters’ who guide the learner through educational tasks and model social scenarios. Participants used the software for 10–15 min per day, 3–5 days per week for 3 weeks, with data collected before software use commenced, immediately after use ended, then again 2 and 4 months after this. The software evaluation revealed that the social tutor was generally well-received by participants and caregivers, with data showing a clear difference in theoretical knowledge of social skills between the experimental (social content) group and the control (placebo content) group. This paper focuses on the automated assessment, dynamic lesson sequencing, feedback, and reinforcement systems of the social tutor software, the impact these systems had on participant performance during the software evaluation, and recommendations for future development. Thus, while data reflecting overall performance of the social tutor system is provided for context, the main data presented is limited to that specifically relating to the performance and user perceptions of the aforementioned systems.