Relevant publications from NALA: What Would a Teacher Do? Predictng Future Talk Moves.
Technology is a big part of today’s classrooms; both remote learning and the use of smart devices and computers in classrooms provide us the potential to harness advances in AI and NLP for enhancing teaching and learning.
We are interested in using NLP to understand and improve interactions between teachers and students. This involves identifying effective teaching patterns from classroom discourse in order to 1) provide feedback to teachers, 2) develop NLP models that can learn intervention strategies to improve engagement. This specific problem domain provides rich and varied challenges from the perspective of NLP research, especially in dialogue understanding and generation. Classroom discussions involve multi-party conversations between students and teachers, thus differing from work around task-oriented or open-domain dialogue systems. Additionally, the structure of content in this domain is unique, due to the variety of discourse structures employed by teachers – such as repetition, rhetorical and probing questions. We investigate how to develop models that can effectively understand such conversations.
Another important goal of our research is to progress towards the development of an AI system that can serve as a collaborative partner to students. We envision such a system as being able to guide students towards having productive discussions that promote learning. For this purpose, our system should be able to generate dialogue that is content-specific, e.g., ask a question about a problem that students are discussing. We should also be able to generate dialogue that engages students, e.g., ask students if they agree with another student. To this end, we are working broadly on problems in both natural language understanding and generation as applied to discourse in an educational setting. We are collaborating with domain experts such as teachers, linguists and learning scientists to develop robust models that can be deployed in classrooms, with careful considerations of societal and ethical issues around their use.
NALA is part of the NSF National AI Institute for Student-AI Teaming. More information about the AI Institute can be found here.