A Comparison of Architectures and Pretraining Methods for Contextualized Multilingual Word Embeddings
2019·Arxiv