Pamela W. Jordan, Maxim Makatchev, and Umarani Pappuswamy
The Why2-Atlas tutoring system (VanLehn et al. 2002) presents students with qualitative physics questions and encourages them via natural language dialogue to explain their answers. We describe changes we are making to the current version of Why2-Atlas to better utilize a proof-based representation of student essays for motivating dialogue. The abductive proof of the student’s explanation for one physics problem serves as a very localized model of likely student beliefs. We will discuss how we can use such a localized student model to select dialogue topics (as the system currently does), to improve the accuracy of the model via clarification requests and to acknowledge and take advantage of the content of the student’s essay during the dialogue.