RT-SFT: Text Style Transfer from Non-Parallel Corpora by Roundtrip Translation
This study proposes a method to transfer text style using only monolingual data and roundtrip translation, bypassing the need for parallel corpora. It synthesizes 'style-neutral' texts by translating monolingual in-style text back and forth, creating a pseudo-parallel dataset. An LLM is then finetuned on this synthetic data to perform style transfer, outperforming existing methods.