On Learning Text Style Transfer with Direct Rewards

Yixin Liu, Graham Neubig, John Wieting

In most cases, the lack of parallel corpora makes it impossible to directly train supervised models for the text style transfer task. In this paper, we explore training algorithms…