Dialogue Analysis to Inform the Development of a Natural-language Tutoring System for Physics

Several cognitive scientists attribute the effectiveness of tutorial dialogue to its interactive nature. Although this view has empirical support, the notion of ―interactivity‖ is too vague to guide the development of natural-language tutorial dialogue systems. This paper describes our attempts to operationalize particular forms of interactivity: tutor abstraction and specification of student dialogue moves, and tutor prompts for specification. We describe and illustrate the process by which we specified decision rules for abstraction and specification in automated tutorial dialogues about physics. Correlational analyses suggest that particular types of interactive abstraction and specification relations predict student learning, as measured by pretest to posttest gain score—for example, tutor prompts for the student to specify the individual forces that comprise the net force. Since particular kinds of abstraction and specification relations are associated with particular decision rules, these findings are guiding our selection of rules to implement in a tutorial dialogue system for physic