BERT: Pre-training of Deep Bidirectional Transformers

Bidirectional encoder representations from transformers enable fine-tuning for many NLP tasks. BERT set new benchmarks across eleven natural language processing tasks.