Straight to the Gradient: Learning to Use Novel Tokens for Neural Text Generation.

Xiang Lin, Simeng Han, Shafiq R. Joty

Browse the full ICML paper archive.