NSF logo

Speech-Based Learning Analytics for Collaboration: 1432606

Principal Investigator: Cynthia D’Angelo
CoPrincipal Investigator(s): Elizabeth Shriberg, Harry Bratt
Organization: SRI International

The Education Core Research (ECR) program funds foundational research in STEM learning, STEM learning environments, workforce development, and broadening participation in STEM. One of the pressing current challenges in schools as curriculum focuses increasingly on inquiry, application, and synthesis (rather than factual recall) is how to efficiently manage, support, and assess collaboration in the classroom. While teaching with collaborative learning may be more productive, it also creates greater burdens on educators, including the need to have real-time diagnosis of problems in group work. Technology has been used in a variety of ways to provide real-time data, but this project takes a novel approach: using cutting edge computer science, specifically automatic speech processing technology, and studies how it might be used to support real-time diagnosis of collaboration among learners.

Using speech processing to recognize the words spoken, and then analyze whether collaboration is taking place, is an area of research that is far from applicability. This project will instead use speech processing technology to look for other aspects of speech that may serve as useful gross indicators of collaboration, such as turn taking, which is much easier to apply using current technological infrastructure. The proposal is a collaboration between learning scientists, computer scientists, linguists, and teachers, and will address three research questions: first, they will study (using human observation) how well indicators (such as turn-taking, or vocal tone associated with frustration) predict overall collaboration quality (i.e., collaborative learning behavior and outcomes); second, they will examine how well automated speech processing technology can identify these indicators; and finally, they will attempt to validate whether the automated system can usefully gauge the student collaboration. This research will be undertaken in the context of a well-studied Mathematics curriculum (Cornerstone Mathematics). Prior research on Cornerstone Math shows that while it is highly successful on average, student outcomes vary significantly with student discourse quality. The project proposes to study the automated system both in a controlled laboratory setting in which acoustics can be carefully controlled, and in a realistic field setting using diverse students in 5 middle school classrooms. The intellectual merit includes improving the state of the art in automated speech recognition technology to examine paralinguistic and prosodic features useful in gauging collaboration; novel research on the ways successful and unsuccessful collaboration shows certain acoustic markers; and design of new computer-supported collaborative learning technology to support teacher diagnosis of learner collaboration in real time. Broader impacts of the work relate to the eventual possibility that teachers could more easily implement collaborative learning in their classrooms, while paying attention to the teams or groups that need help the most.

Tags: , , , , ,