A general model of dialogue interpretation for concept tutoring systems

Abstract

The paper describes how high-level knowledge about visual images can be communicated to students through system-active and systempassive tutorial interactions. Past work in visual concept tutoring has concentrated on the theoretical principles of how humans acquire expertise in visual recognition. The few implementations there have been are domain-specific. However, a key point that has been neglected is the question of how general knowledge can be used to regulate tutorial dialogues about abnormal image features, while enforcing consistency of such dialogues. We introduce a new model of interpretation, which combines meta-level knowledge with domain-specific teaching effects in a computerbased environment which allows the student to explore example images or, alternatively, to interact with an Intelligent Tutoring System (ITS). We argue that principled knowledge can be acquired by free exploration of the image database browser and that experiential knowledge can be acquired through consistent dialogues guided by the ITS interface.