Learning to paraphrase: using paraphrase detection of spoken utterances to predict learner expertise

Abstract

Predicting student knowledge from text has become an increasingly common for automatic grading and assessment. Much of this work, however, hinges on natural language processing techniques that tend to neglect the relative locations of individual words by using a bag of words model. Instead of using this technique we segment text into a bag of utterances. We then use those utterances to examine the extent to which individuals of different levels of expertise paraphrase one another. Results indicate clear distinctions among paraphrase frequencies of the different levels of expertise. Furthermore, results suggest that experts and novices have many of the same intuitions, but that the expert’s knowledge is more accurately applied.