Token-level and sequence-level loss smoothing for RNN language models.

Maha Elbayad, Laurent Besacier, Jakob Verbeek

Browse the full ACL paper archive.