Thanks to the partnership between the Donald Bren School of Information and Computer Sciences (ICS) and the Allen Institute for Artificial Intelligence (AI2), Assistant Professor of Computer Science Sameer Singh and two ICS undergraduate students, Jens Tuyls and Junlin Wang, were part of a team awarded the Best Demo Paper at the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP 2019). The paper, “AllenNLP Interpret: A Framework for Explaining Predictions of NLP Models,” is a collaboration with AI2 researchers Eric Wallace, Sanjay Subramanian and Matt Gardner, and everyone was on hand at the conference in Hong Kong to receive the award.

Machine learning models for NLP have become complex, making them very difficult to understand. According to the team, this can “leave researchers and practitioners scratching their heads asking, why did my model make this prediction?”

The toolkit they’ve developed makes it easy for researchers to apply standard interpretation techniques so they can better understand what is going on inside their models. AllenNLP Interpret offers a suite of such techniques applicable to most models, APIs for developing new interpretation methods, and reusable front-end components for visualizing the interpretation results. In announcing the Best Demo Paper award, the selection commitee acknowledged that “since the need to better understand notoriously opaque neural network systems is huge, the system is likely to stimulate more research.”

The project website demonstrates the toolkit’s flexibility and utility by implementing live demos for five interpretation methods on a variety of models and tasks.

— Shani Murray