Fetching the paper…

MASS: Masked Sequence to Sequence Pre-training for Language Generation · Around