The challenges in adapting traditional techniques for modeling student behavior in ill-defined domains

Designing cognitive tutors and modeling behavior for ill-defined domains require innovative methods and techniques. We combine a top-down, theoretical approach with a bottom-up, empirical approach to develop a student model for the selection of aspect in French verbs. In performing this task, we design a new representation applicable to feature-driven ill-defined problem spaces and utilize tutoring scaffolds in order to elucidate the student thought process. We then evaluate and refine our model based on empirical data collected through student think-alouds. We plan to use our preliminary results to design and evaluate a fully-developed cognitive tutor, and hope to generalize our tutor development process to other illdefined domains.