Improving the engineering of model tracing based intelligent program diagnosis

Abstract

The model tracing based approach to intelligent program diagnosis and tutoring emphasises the fact that the student's design decisions are traced as the student develops the program. A system based on this model monitors the user's actions as he/she moves along the solution path, automatically analyses partial solutions for semantic errors and misconceptions, and offers guidance whenever he/she deviates from the correct solution path. In this way, the system always checks to see if the student is following the design path of an ideal model. Buggy paths are pruned as soon as they are discovered. Through this approach to automatic diagnosis and tutoring, a model tracing system can firstly diagnose very specific errors and misconceptions and provide clear advice and explanations within a proper and immediate context; secondly, explicitly guide the user in the process of organising different programming concepts and statements; and thirdly, simplify the engineering of automatic diagnosis by preventing multiple bugs and errors. However, this approach is very directive and interventionist. The user is highly constrained in the solutions that can be developed, as he/she must conform to the task decomposition and coding sequence enforced by model tracing systems. To bypass these pitfalls associated with the traditional model tracing approach, solutions and guidelines are suggested for improving the engineering of model tracing based diagnosis.