A Primer in BERTology: What we know about how BERT works
It provides a comprehensive overview for researchers and practitioners in NLP, but is incremental as it synthesizes existing work.
The paper surveys over 150 studies to summarize current knowledge about how BERT works, including its learned information, modifications, and compression approaches, without presenting new experimental results.
Transformer-based models have pushed state of the art in many areas of NLP, but our understanding of what is behind their success is still limited. This paper is the first survey of over 150 studies of the popular BERT model. We review the current state of knowledge about how BERT works, what kind of information it learns and how it is represented, common modifications to its training objectives and architecture, the overparameterization issue and approaches to compression. We then outline directions for future research.