Tweeted By @rctatman
Really nice comparison of different embedding methods, including contextual embeddings. It's from February, so it includes Bert but not GPT-2 or later models. https://t.co/S4QijmqfTy
— Rachael Tatman (@rctatman) July 10, 2019
(It's in Chinese but I found the Google translate to English in Chrome very readable. Yay NLP!)