Improving error diagnosis using intermediate representations

The paper describes techniques to improve error diagnosis by enabling the student to explain his or her underlying thinking to the system. It is argued that attempts to infer this information using such techniques as plan recognition and natural language interfaces are too complex and error prone. Instead, techniques of explicit creation and annotation of underlying plans and goals are proposed, including the use of explicit instantiation of general rules. A tutor for propositional logic, EPIC, has been developed to illustrate the use of these techniques, and observations made of students using the tutor. The results indicate the power of the technique to identify plans, particularly erroneous ones, during the process of problem solving, when the student may have entered few steps of his or her proposed solution proof. Furthermore, the provision of an interface to permit the articulation of underlying plans was found to be more supportive as a learning environment than merely permiting the entry of the solution steps.