An intelligent computer tutor to guide self-explanation while learning from examples

Many studies in cognitive science show that self-explanation—the process of clarifying and making more complete to oneself the solution of an example—improves learning, and that guiding self-explanation extends these benefits. This thesis presents an intelligent computer tutor that aims to improve learning from examples by supporting self-explanation. The tutor, known as the SE (self-explanation) Coach, is innovative in two ways. First, it represents the first attempt to develop a computer tutor that supports example studying instead of problem solving. Second, it explicitly guides a domain-general, meta-cognitive skill: self-explanation. The SE-Coach is part of the Andes tutoring system for college physics and is meant to be used in conjunction with the problem solving tasks that Andes supports. In order to maximize the system capability to trigger the same beneficial cognitive processes, every element of the SE-Coach embeds existing hypotheses about the features that make self-explanation effective for learning. Designing the SE-Coach involved finding solutions for three main challenges: (1) To design an interface that effectively monitors and supports self-explanation. (2) To devise a student model that allows the assessment of example understanding from reading and self-examination actions. (3) To effectively elicit further self-explanation that improves student's example understanding. In this work we present our solutions to these challenges: (1) An interface including principled, interactive tools to explore examples and build self-explanations under the SECoach's supervision. (2) A probabilistic student model based on a Bayesian network, which integrates a model of correct self-explanation and information on the student's knowledge and studying actions to generate a probabilistic assessment of the student's example understanding. (3) Tutorial interventions that rely on the student model to detect deficits in the student's example understanding and elicit self-explanations that overcome them. In this thesis we also present the results of a formal study with 56 college students to evaluate the effectiveness of the SE-Coach. We discuss some hypotheses to explain the obtained results, based on the analysis of the data collected during the experiment.