“An evaluation of animation in a pedagogical agent” by Williams, Belynne and Bodenheimer

This work presents an empirical evaluation of a pedagogical agent animated through a manifold learning technique. We describe how the dimensionality reduction and embedding attributes of manifold learning techniques allow us to parameterize video data to create a model-free, videorealistic, talking pedagogical agent. The agent is used within a learning-by-teaching system focused on elementary school science education. The system was evaluated on students aged 9- 11 (U.S. fifth grade) in local public schools. A single condition was tested, whether the agent was animated or not. The specific task used to evaluate the results was a debugging task concerning the nitrogen cycle of plants. Students using the animated agent performed statistically significantly better than students without the animated agent. Additionally, survey questions indicate that the students prefer the system with animation.

1. Bargh, J. A., and Schul, Y. 1980. On the cognitive benefits of teaching. Journal of Educational Psychology72, 593–604.
2. Davis, J., Leelawong, K., Belynne, K., Bodenheimer, B., Biswas, G., Vye, N., and Bransford, J. 2003. Intelligent user interface design for teachable agent systems. In IUI 2003, 26–33.
3. Tenenbaum, J. B., de Silva, V., and Langford, J. C. 2000. A global geometric framework for nonlinear dimensionality reduction. Science 290, 2319–2323.