Question Answering on FreshQA (train test)
28.37BLEUSeq2Sick
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Seq2SickProtection=Surface-level perturbation2026.01 | 28.37 | 29.85 | 14.8 | 30.12 | 38.7 | |
| BAEProtection=Surface-level perturbation2026.01 | 28.18 | 29.62 | 14.77 | 30.32 | 39.27 | |
| No alterationProtection=None2026.01 | 27.05 | 29.58 | 17.36 | 28.66 | 39.68 | |
| Disclaimer InjectionProtection=Disclaimer Injection2026.01 | 5.58 | 10.17 | 3.3 | 8.7 | 4.01 |