Textual Style Transfer on English Detoxification Dataset
0.98STAcondBERT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| condBERTEvaluation Setup=Baselines2022.06 | 0.98 | 0.77 | 0.82 | 0.62 | |
| mT5 base*Evaluation Setup=Multilingual Setup, Training Language=All English and Russian data available (not equalized)2022.06 | 0.893 | 0.787 | 0.942 | 0.657 | |
| mBART 5000Evaluation Setup=Multilingual Setup, Training Language=English and Russian (equalized), Training Iterations=50002022.06 | 0.887 | 0.889 | 0.866 | 0.64 | |
| mT5 base ENGEvaluation Setup=Cross-lingual Setup, Training Language=English2022.06 | 0.86 | 0.834 | 0.833 | 0.587 | |
| mBART 5000 ENGEvaluation Setup=Cross-lingual Setup, Training Language=English, Training Iterations=50002022.06 | 0.857 | 0.84 | 0.873 | 0.616 | |
| mT5 small ENGEvaluation Setup=Cross-lingual Setup, Training Language=English2022.06 | 0.844 | 0.858 | 0.826 | 0.591 | |
| mBART 3000 ENGEvaluation Setup=Cross-lingual Setup, Training Language=English, Training Iterations=30002022.06 | 0.842 | 0.856 | 0.876 | 0.617 | |
| mT5 baseEvaluation Setup=Multilingual Setup, Training Language=English and Russian (equalized)2022.06 | 0.833 | 0.826 | 0.83 | 0.556 | |
| mT5 smallEvaluation Setup=Multilingual Setup, Training Language=English and Russian (equalized)2022.06 | 0.826 | 0.841 | 0.763 | 0.513 | |
| DeleteEvaluation Setup=Baselines2022.06 | 0.81 | 0.93 | 0.64 | 0.46 | |
| mBART 5000 RUSEvaluation Setup=Cross-lingual Setup, Training Language=Russian, Training Iterations=50002022.06 | 0.806 | 0.484 | 0.864 | 0.242 | |
| mT5 small RUSEvaluation Setup=Cross-lingual Setup, Training Language=Russian2022.06 | 0.776 | 0.521 | 0.535 | 0.169 | |
| mBART 5000 (FSMT)Evaluation Setup=Backtranslation Setup, Engine=FSMT2022.06 | 0.744 | 0.746 | 0.893 | 0.415 | |
| mBART 5000 (Google)Evaluation Setup=Backtranslation Setup, Engine=Google2022.06 | 0.678 | 0.762 | 0.568 | 0.284 | |
| mBART 3000 RUSEvaluation Setup=Cross-lingual Setup, Training Language=Russian, Training Iterations=30002022.06 | 0.547 | 0.778 | 0.888 | 0.299 | |
| mT5 base RUSEvaluation Setup=Cross-lingual Setup, Training Language=Russian2022.06 | 0.365 | 0.906 | 0.696 | 0.171 |