Do simple word features predict dialogue acts among students working together?
Arts and Sciences
Is it possible for a computer to tell when students working together online are engaging with each other in their conversation? The Computer-Mediated Problem-Solving (COMPS) project is working to address that question. The goal is to have the computer figuratively look over the shoulders of students at work, judging whether they are more-or-less on task. This student project works with dialogue acts that typify students working productively, e.g. sharing ideas, negotiating, directing the problem solving task. The experiment uses transcripts of 1200 turns of synchronous dialogue, students type-chatting together solving exercises in a computer programming class. These transcripts have been manually pre-tagged showing which turns exhibit which dialogue behaviors. We then tabulate common words and phrases which are statistically associated with these behaviors. As a simple example, a common word like "but" might be associated with students disagreeing with each other, which would be a type of negotiating dialogue act, which would be expected from students engaging with each other in a problem-solving activity. The research question is then: are these word features predictive of the same dialogue acts in other contexts? Does the association discovered in the training data help to identify the same behaviors in conversations between different students solving different problems? We test the hypothesis using other synchronous dialogues of different students solving the same and different computer programming problems. We then test further, using threaded discussion board postings where students asynchronously discuss different topics.
Nowacki, Jaeda, "Do simple word features predict dialogue acts among students working together?" (2021). Symposium on Undergraduate Research and Creative Expression (SOURCE). 935.