Traditionally, intelligent tutoring systems have provided feedback on the basis of a so-called expert model. Expert model tutors incorporate production rules associated with error free and efficient task performance. These systems intervene with corrective feedback as soon as a student deviates from a solution path.

This thesis explores the effects of providing feedback on the basis of a so-called intelligent novice cognitive model. An intelligent novice tutor allows students to make errors, and provides guidance through the exercise of error detection and correction skills. The underlying cognitive model in such a tutor includes both rules associated with solution generation, and rules relating to error detection and correction. There are two pedagogical motivations for feedback based on an intelligent novice model. First, novice performance is often error prone and students may need error detection and correction skills in order to succeed in real world tasks. Second, the opportunity to reason about the causes and consequences of errors may allow students to form a better model of the behavior of domain operators.

Learning outcomes associated with the two models were experimentally evaluated. Results show that learners who receive intelligent novice feedback demonstrate better learning overall, including better retention and transfer performance than students receiving expert model based feedback.

Another focus of the research described here has been to help students form a robust and accurate encoding of declarative knowledge prior to procedural practice with an intelligent tutoring system. Examples have been widely used as a component of declarative instruction. However, research suggests that the effectiveness of examples is limited by the fact that inferences concerning the specific conditions under which operators may be applicable are only implicit in most examples, and may not be apparent to students without self-explanation. This thesis explores the effectiveness of a technique referred to in this thesis as example walkthroughs. Example walkthroughs interactively guide students through the study of examples. They present question prompts that help students make the inferences necessary to select problem solving operators that will lead to a solution. Students make these inferences by responding to multiple choice prompts. Evaluations suggest that example walkthroughs may provide a cost effective way to boost learning outcomes in intelligent tutoring systems.

136 pages

Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu