An Investigation of Language Model Interpretability via Sentence Editing

Samuel Stevens, Yu Su

Pre-trained language models (PLMs) like BERT are being used for almost all language-related tasks, but interpreting their behavior still remains a significant challenge and many im…