This is the code-repository corresponding the following paper:
Evaluating Explanations: How much do explanations from the teacher aid students?
Danish Pruthi, Rachit Bansal, Bhuwan Dhingra, Livio Baldini Soares, Michael Collins, Zachary C. Lipton, Graham Neubig, William W. Cohen
Transactions of the Association for Computational Linguistics (TACL, 2021)
Update: If you are looking to evaluate explanations using our metric, you can use this repository instead as this: (1) computes our proposed simulatability metric and also generates explanations that are optimized according to our metric, and (2) is a modern reimplementation of our work, and (iii) is overall better structured and maintained.