Text Detoxification on CLEF-Tatar
100STAhuman-baseline
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| human-baseline2026.06 | 100 | 93.6 | 87.8 | 82.5 | |
| Tatoxa2026.06 | 98.2 | 85.9 | 81.1 | 69.5 | |
| mT0+Gemini Pro v2.52026.06 | 95.1 | 81.1 | 80.5 | 64 | |
| Gemini Pro v2.52026.06 | 92.8 | 83 | 80.5 | 63.6 | |
| Claude Opus 4.62026.06 | 90.4 | 77 | 78 | 56.2 | |
| GPT-5 Chat2026.06 | 90 | 73.4 | 80.2 | 53.9 | |
| mT0+vocab deletion2026.06 | 84.3 | 87.3 | 82 | 61.6 | |
| DeepSeek V3.22026.06 | 80.4 | 89.5 | 82 | 60.4 | |
| mT0prompt=Russian prompt2026.06 | 79 | 85.4 | 81.6 | 56.4 | |
| mT0prompt=Tatar prompt2026.06 | 78.6 | 86.5 | 82.1 | 57.1 | |
| mT0prompt=English prompt2026.06 | 77.8 | 87.9 | 82.7 | 58 | |
| Vocab deletion2026.06 | 77.7 | 89.6 | 82.5 | 57.9 |