Paraphrase Quality Assessment on MAGE shared subset (evaluation 300 AI-written samples)
16.25PPLDirect Paraphrasing
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Direct Paraphrasing2026.05 | 16.25 | 4.96 | 85 | |
| Adv. Paraphrase2026.05 | 16.44 | 4.94 | 85 | |
| MAGE BERTScore Linearalpha=0.52026.05 | 17.61 | 4.9 | 84 | |
| MAGE BERTScorethreshold=12026.05 | 19.54 | 4.99 | 99 | |
| Original2026.05 | 20.34 | — | — | |
| DEPO2026.05 | 20.6 | 4.94 | 86 | |
| MAGE BERTScore Linearalpha=0.32026.05 | 20.69 | 4.93 | 83 | |
| MAGE BERTScorethreshold=02026.05 | 20.86 | 4.89 | 80 | |
| MAGE BERTScore Linearalpha=0.12026.05 | 23.72 | 4.86 | 80 | |
| AuthorMist2026.05 | 26.31 | 4.64 | 81 | |
| SilverSpeak2026.05 | 35.12 | 4.62 | 34 | |
| StealthRL2026.05 | 61.6 | 2.94 | 61 |