Hallucination Detection on PHOENIX 2014T
94AUCMETA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| METAModel Type=Gloss-free (GF)2025.10 | 94 | 99.2 | 87.2 | |
| ConfidenceModel Type=Gloss-free (GF)2025.10 | 92.6 | 98.9 | 86 | |
| GroundingModel Type=Gloss-free (GF)2025.10 | 91.8 | 98.6 | 93.8 | |
| METAModel Type=Gloss-based (GB)2025.10 | 89.9 | 97.8 | 86.4 | |
| PerplexityModel Type=Gloss-free (GF)2025.10 | 88.8 | 97.9 | 71.3 | |
| Token entropyModel Type=Gloss-free (GF)2025.10 | 87.1 | 97.5 | 73.1 | |
| GroundingModel Type=Gloss-based (GB)2025.10 | 82.7 | 95.4 | 89.9 | |
| Token entropyModel Type=Gloss-based (GB)2025.10 | 63.9 | 90 | 58.1 | |
| ConfidenceModel Type=Gloss-based (GB)2025.10 | 61.8 | 88.7 | 64.5 | |
| PerplexityModel Type=Gloss-based (GB)2025.10 | 59.5 | 89.2 | 56.1 |