Lee Becker's Assignment 14

Entailment Driven Dialogues for Intelligent Tutoring

Computerized tutoring systems are an effective tool for learning when used in conjunction with traditional teaching methods.[1] These systems provide further opportunities for assessing student progress and understanding. Currently most systems are limited to doing assessment by using a limited and directed line of reasoning. One way to encourage deeper understanding of the material is through an interactive dialog. Most dialog systems still require a high degree of tailoring to pair the appropropriate tutor and student responses.

This paper describes the My Science Tutor program, a Socratic style tutor that uses a classroom technique known as Questioning the Author (QtA) for the basis of its dialogue. Instead of giving the student the correct answer, the tutor uses QtA to try to elicit facts from the student. Dialog generation is driven by a model of student comprehension which is based on confidence levels derived from semantic analysis of students' responses and degree of correctness as determined by an entailment algorithm. This approach provides a much more scalable and less work-intensive way to generate dialog for intelligent tutoring agents.

[1] DRK-12 Proposal Improving Science Learning in Inquiry-based Programs, Boulder Language Technologies, W. Ward, et al.