An intelligent tutoring system

Abstract

Students entering a new field must learn to speak the specialized language of that field. Previous research using automated measures of word overlap has found that students who modify their language to align more closely to a tutor's language show larger overall learning gains. We present an alternative approach that assesses syntactic as well as lexical alignment in a corpus of human-computer tutorial dialogue. We found distinctive patterns differentiating high and low achieving students. Our high achievers were most likely to mimic their own earlier statements and rarely made mistakes when mimicking the tutor. Low achievers were less likely to reuse their own successful sentence structures, and were more likely to make mistakes when trying to mimic the tutor. We argue that certain types of mimicking should be encouraged in tutorial dialogue systems, an important future research direction.