Language Analysis and Generation in Algebra Tutorial Dialogues for Language-Based Intelligent Tutoring Systems

摘要

The North Carolina A&T State University algebra tutoring dialogue project collects and analyzes algebra tutoring dialogues with the aim of describing tutoring strategies and language with enough rigor that they may be evaluated and incorporated in machine tutoring. The tutoring is computer- mediated chat-style, with the tutor and student collaboratively solving a problem and editing equations. Transcripts of these dialogues are annotated for tutorial intentions. language, and behaviors. Comparison of tutoring by more and less expert tutors allows us to focus on the strategies that are most effective. The most prominent we observe is a metacognitive strategy where the tutor tries to elicit from the student a categorization of the problem under discussion and to name a solution method. We combine these strategies into a structured set of tutorial goals, with example sentences for each. In subsequent tutoring the computer then suggests these goals and sentences to the - human tutor, who can pick use them in the dialogue. Measuring how much the tutors follow the computer-assistance, as well as student learning gains, allows to evaluate the validity of our structural analysis of tutoring.