Hallucination Detection on HalluScore
97.77F1-ScoreCrossHallu
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CrossHalluBackbone=Phi4-mini, Train dataset=HalluScore, Transfer setting group=Same-language baseline (BS)2026.07 | 97.77 | 57.19 | 97.04 | |
| CrossHalluBackbone=Phi4-mini, Train dataset=TruthfulQA AR, Transfer setting group=Cross-domain (CD)2026.07 | 97.77 | 72.59 | 98.43 | |
| CrossHalluBackbone=Aya, Train dataset=HalluScore, Transfer setting group=Same-language baseline (BS)2026.07 | 97 | 80.37 | 97.96 | |
| CrossHalluBackbone=Aya, Train dataset=TruthfulQA AR, Transfer setting group=Cross-domain (CD)2026.07 | 96.69 | 86.9 | 98.66 | |
| CrossHalluBackbone=Mistral, Train dataset=HalluScore, Transfer setting group=Same-language baseline (BS)2026.07 | 96.18 | 69.28 | 95.56 | |
| CrossHalluBackbone=Silma, Train dataset=HalluScore, Transfer setting group=Same-language baseline (BS)2026.07 | 95.96 | 61.84 | 94.8 | |
| CrossHalluBackbone=Aya, Train dataset=TruthfulQA EN, Transfer setting group=Cross-lingual + cross-domain (CL-CD)2026.07 | 94.79 | 76.98 | 98.07 | |
| CrossHalluBackbone=Silma, Train dataset=TruthfulQA AR, Transfer setting group=Cross-domain (CD)2026.07 | 91.94 | 64.8 | 95.64 | |
| CrossHalluBackbone=Silma, Train dataset=TruthfulQA EN, Transfer setting group=Cross-lingual + cross-domain (CL-CD)2026.07 | 90.96 | 62.57 | 94.66 | |
| CrossHalluBackbone=Allam, Train dataset=TruthfulQA AR, Transfer setting group=Cross-domain (CD)2026.07 | 88.95 | 55.21 | 84.22 | |
| CrossHalluBackbone=Allam, Train dataset=HalluScore, Transfer setting group=Same-language baseline (BS)2026.07 | 87.89 | 60.4 | 86.41 | |
| CrossHalluBackbone=Mistral, Train dataset=TruthfulQA AR, Transfer setting group=Cross-domain (CD)2026.07 | 86.46 | 71.8 | 95.1 | |
| CrossHalluBackbone=Allam, Train dataset=TruthfulQA EN, Transfer setting group=Cross-lingual + cross-domain (CL-CD)2026.07 | 86.29 | 56.39 | 83.54 | |
| CrossHalluBackbone=Phi4-mini, Train dataset=TruthfulQA EN, Transfer setting group=Cross-lingual + cross-domain (CL-CD)2026.07 | 84.83 | 63.79 | 97.93 | |
| CrossHalluBackbone=Qwen2.5, Train dataset=HalluScore, Transfer setting group=Same-language baseline (BS)2026.07 | 84.57 | 72.44 | 87.17 | |
| CrossHalluBackbone=Mistral, Train dataset=TruthfulQA EN, Transfer setting group=Cross-lingual + cross-domain (CL-CD)2026.07 | 77.74 | 66.17 | 95.39 | |
| CrossHalluBackbone=Qwen2.5, Train dataset=TruthfulQA EN, Transfer setting group=Cross-lingual + cross-domain (CL-CD)2026.07 | 72.18 | 69.03 | 85.27 | |
| CrossHalluBackbone=Qwen2.5, Train dataset=TruthfulQA AR, Transfer setting group=Cross-domain (CD)2026.07 | 30.17 | 68.94 | 85.24 |