BERT

Perturbation CheckLists for Evaluating NLG Evaluation Metrics
The Heads Hypothesis: A Unifying Statistical Approach Towards Understanding Multi-Headed Attention in BERT
The heads hypothesis: A unifying statistical approach towards understanding multi-headed attention in BERT
Improving Dialog Evaluation with a Multi-reference Adversarial Dataset and Large Scale Pretraining
On the weak link between importance and prunability of attention heads
Towards Interpreting BERT for Reading Comprehension Based QA
On Incorporating Structural Information to improve Dialogue Response Generation