Unsupervised Text Style Transfer using Language Models as Discriminators

Unsupervised text style transfer requires learning disentangled representations of attributes (e.g., negative/positive sentiment, plaintext/ciphertext orthography) and underlying content.