Fetching the paper…

BERTRAM: Improved Word Embeddings Have Big Impact on Contextualized Model Performance · Around