Algorithmic debugging to support cognitive diagnosis in tutoring systems

Cognitive modelling in intelligent tutoring systems aims at identifying a learner's skills and knowledge from his answers to tutor questions and other observed behaviour. In this paper, we propose an innovative variant of Shapiro's algorithmic debugging technique whose application can be used to pin-point learners' erroneous behaviour in terms of an irreducible disagreement to the execution trace of an expert model. Our variant has two major benefits: in contrast to traditional approaches, it does not rely on an explicit encoding on mal-rules, and second, it induces a natural teacher-learner dialogue with no need for the prior scripting of individial turns or higher-level dialogue planning