B 2 : A Tutoring Shell for Bayesian Networks that Supports

This paper describes B2, an extension to BANTER. BANTER is a tutoring shell that allows medical students to practice their decision-making skills by considering a number of medical cases that diier from each other in a controlled manner. The underlying inference engine uses a Bayesian network. B2 provides a natural language interface to BANTER that is consistent with principles for cooperative interaction. B2 represents three signiicant advances over the original tool: First, B2 adds knowledge about the structure of the medical domain, allowing it to generalize its responses when appropriate. Second, B2 adds a discourse model, enabling it to reason about the content and structure of the interaction , e.g. to avoid giving irrelevant information. Third, B2 integrates natural language and graphical user interface gestures, allowing students to refer to sequences of actions as well as to elements of the ongoing verbal communication. Together, these improvements make B2 more useful to students, because they allow students to better direct the system's explanation to their needs.