Hallucination Detection on WMT DE-EN 18 (test)
91.72AUROCLaBSE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LaBSEdetector_type=External2022.12 | 91.72 | 26.91 | |
| Wass-Combodetector_type=Model-based, aggregation=mean scores across five random seeds2022.12 | 87.17 | 47.56 | |
| CometKiwidetector_type=External2022.12 | 86.96 | 53.61 | |
| Wass-to-Datadetector_type=Model-based, aggregation=mean scores across five random seeds2022.12 | 84.2 | 48.15 | |
| Seq-Logprobdetector_type=Model-based2022.12 | 83.4 | 59.02 | |
| Wass-to-Unifdetector_type=Model-based2022.12 | 80.37 | 72.22 | |
| Attn-ign-SRCdetector_type=Model-based2022.12 | 79.36 | 72.83 |