1 paper
Henry Li Xinyuan, Ray Lee, Jerry Chen +1
While the way intermediate representations are generated in encoder-decoder sequence-to-sequence models typically allow them to preserve the semantics of the input sentence, input…