EXTENDED LEARNING MECHANISMS IN A PRODUCTION SYSTEM THEORY OF COGNITION

Abstract

Aim of the Thesis: The objective of this thesis is to model the cognitive processes involved in learning Smalltalk. The research aim arises because no computational theory of cognition suitable for guiding the development of an intelligent instructional system for teaching Small talk can be found. A suitable theory must provide an adequate model of students learning Smalltalk by accounting parsimoniously for the learning phenomena exhibited by these students, namely, the schematization of memory for declarative facts, the reliance on prior problem solving episodes, and increasing expertise in terms of programming skill. The thesis proposes a theory, COGNITlO, that integrates three learning mechanisms within a single computational framework suitable for guiding the design of an intelligent tutoring system for Small talk. Scope of the Thesis: The thesis extends ACT*, a production system theory of cognition. ACT*, which accounts for learning through knowledge compilation, has been used successfully in modeling the skill acquisition aspect of learning. In this thesis, the ACT* architecture is enhanced by integrating two other learning mechanisms, namely, schema formation and episode storage/retrieval. The main research work described in this thesis involves extending and integrating existing cognition theories as well as providing evidence for the psychological validity of COGNITIO. COGNITIO's psychological validity is illustrated by a computer model that simulates a student learning the knowledge and skills involved in message construction and execution - the most fundamental aspect of Smalltalk programming. General principles and specific implications for the design of intelligent tutoring systems are also derived from COGNITIO.