MASS Masked Sequence to Sequence Pre-training for Language Generation

MASS:MaskedSequencetoSequencePre-trainingforLanguageGenerationKaitaoSong1XuTan2TaoQin2JianfengLu1Tie-YanLiu2Abstractwhilepre-traininghasplentyofdata(Girshicketal.,2014;Szegedyetal.,2015;Ouyangetal....