Linguistic-space calibration on SQuAD 2.0
66.4Mean Faithfulness Divergence ReductionRALC
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RALCSignal=Semantic Uncertainty, Calibration Training Domain=TruthfulQA2026.05 | 66.4 | 56.9 | |
| RALCSignal=Semantic Uncertainty, Calibration Training Domain=SQuAD 2.02026.05 | 66 | 58.7 | |
| RALCSignal=Linguistic Confidence, Calibration Training Domain=TruthfulQA2026.05 | 65.8 | 47.2 | |
| RALCSignal=Token Probability, Calibration Training Domain=SQuAD 2.02026.05 | 64.9 | 50.6 | |
| RALCSignal=Token Probability, Calibration Training Domain=TruthfulQA2026.05 | 64.2 | 48.6 | |
| RALCSignal=Linguistic Confidence, Calibration Training Domain=SQuAD 2.02026.05 | 62 | 43.4 | |
| RALCSignal=Token Probability, Calibration Training Domain=MMLU2026.05 | 52.6 | 38.5 | |
| RALCSignal=Semantic Uncertainty, Calibration Training Domain=MMLU2026.05 | 36.1 | 30.8 | |
| RALCSignal=Linguistic Confidence, Calibration Training Domain=MMLU2026.05 | 34.6 | 20.7 |