W1: MODELLING HUMAN TEACHING TACTICS AND STRATEGIES

Ben du Boulay, University of Sussex, Brighton, UK.
Art Graesser, University of Memphis, Memphis, USA.
Jim Greer, University of Saskatchewan, Saskatoon, Canada.
Susanne Lajoie, McGill University, Montreal, Canada.
Mark Lepper, Stanford University, Palo Alto, USA.
Rose Luckin, University of Sussex, Brighton, UK.
Johanna Moore, University of Edinburgh, Edinburgh, UK.
Natalie Person, Rhodes College, Memphis, USA.

The purpose of this workshop is to explore the issues concerned with capturing human teaching tactics and strategies as well as attempts to model and evaluate those tactics and strategies in systems.

The former topic covers studies both of expert as well as "ordinary" teachers. The latter includes issues of modelling motivation, timing, conversation, learning as well as simply knowledge traversal.

We see this workshop as a follow-on from the panel discussion at AI-ED'99 that stimulated a debate about the whole issue of how and whether, and with what effect, human teaching tactics can/should be modelled. The description of that panel was as follows:

"According to Bloom, one-on-one tutoring is the most successful form of instruction. Bloom was referring to human tutoring but the AI-ED community has replicated this finding with computer tutors in intelligent tutoring systems where computers generate adaptive forms of tutoring for individual learners. Recently, the AI-ED community has been exploring issues of human tutoring in terms of how experts coach novices, when do tutors tutor, how do they tutor in terms of the types of things they say to the learner, and when do they fade their assistance? One issue this panel will address is should computer tutors mimic human tutors or are there special advantages or disadvantages of computers that should be drawn on or avoided? "Even if the computer could accurately diagnose the student's affective state and even if the computer could respond to that state (in combination with its diagnosis of the learner's cognitive state) exactly as a human tutor would, there remains one final potential difficulty: the plausibility, or perhaps the acceptability, problem. The issue here is whether the same actions and the same statements that human tutors use will have the same effect if delivered instead by a computer, even a computer with a virtually human voice.'' (Lepper et al., 1993)

Human-to-human tutoring incorporates mechanisms that are associated with normal conversational dialogue, but rarely incorporate most ideal tutoring strategies. Some of the normal conversational mechanisms can be simulated on computer, whereas others are too difficult to incorporate in current computational technologies. It would be prudent for an ITS to incorporate both ideal pedagogical strategies and some conversational mechanisms that are within the immediate grasp of modern technologies. But this solution is periodically confronted with trade-offs and conflicts between ideal strategies and natural conversation. These issues will be addressed by this panel."