ABSTRACT
Now that we have described most of the other major compo-nents of the system, we have laid the necessary foundationsfor this chapter on the dialogue generation that is the core of our work. From the very beginning, the focus of our funding from the Office of Naval Research was on the generation of tutoring language. We contemplated trying to base this project on an intelligent tutoring system to which we could add language generation without building a whole system from scratch. We seriously considered trying to develop our research on top of the hypothetico-deductive problem-solving tutor under development at Rush Medical College and at IIT (Chen, 1999; Luksa, 1994; Michael, Haque, Rovick, & Evens, 1989). We finally decided, instead, to add dynamic planning, student modeling, a domain knowledge base, and natural language understanding and generation to CIRCSIM, so that we could build on the experience of Joel Michael and Allen Rovick. This was also the domain in which they felt most comfortable carrying out the extensive series of human tutoring sessions that we planned to use as a basis of our study of sublanguage and of tutoring strategies. In the event, we built a totally new system. As Moser and Moore (1995) pointed out, if a tutoring system is going to benefit from a natural
language interface, it must be provided with the properties that make human tutoring language so effective and much more research is needed to establish just what these properties are.
