Transferring Human Tutor's Style to Pedagogical Agent: A Possible Way by Leveraging Variety of Artificial Intelligence Achievements

Pedagogical agents (P A) are lifelike characters presented on a computer screen that guide users through multimedia learning environments. Evidences show that P A has effect on promoting learning. However, P A field still faces common problems to be solved, such as how to promote the trust relationship among P A and learners. Such questions relate to how to make P A behaves more like human. This paper argues that solving these problems requires at least P A to simulate humans better in appearance, speech and motion, a possible solution is to transfer a human tutor's style to P A. Traditional P A production methods rely on the authoring tool, such as Microsoft agent, this type of methods are highly depending on manual production, so it is difficult to transfer the style of human tutor to P A well. Meanwhile, the production cycle is long, the cost is high, and the adaptability is not ideal. Based on the analysis of the achievements of interdisciplinary literature, this paper points out that based on various machine learning technologies, the above problems can be solved to a great extent. The related technologies are more likely to enhance the human-like of PA, such as the establishment of teacher strategy and behavior prediction model.