Mark G. Core, Johanna D. Moore and Claus Zinn
A promising approach to constructing more effective computer tutors is implementing tutorial strategies that extend over multiple turns. This means that computer tutors must deal with (1) failure, (2) interruptions, (3) the need to revise their tactics, and (4) basic dialogue phenomena such as acknowledgment. To deal with these issues, we need to combine ITS technology with advances from robotics and computational linguistics. We can use reactive planning techniques from robotics to allow us to modify tutorial plans, adapting them to student input. Computational linguistics will give us guidance in handling communication management as well as building a reusable architecture for tutorial dialogue systems. A modular and reusable architecture is critical given the diffi- culty in constructing tutorial dialogue systems and the many domains to which we would like to apply them. In this paper, we propose such an architecture and discuss how a reactive planner in the context of this architecture can implement multi-turn tutorial strategies.