Retrieval-Augmented Linguistic Calibration
About
Linguistic cues such as "I believe" and "probably" offer an intuitive interface for communicating confidence, yet a generalisable, principled calibration framework for linguistic confidence expressions remains underexplored. In particular, co-occurring linguistic cues, contextual variation, and subjective audience interpretation pose unique challenges. We therefore model linguistic confidence as a distribution over plausible perceived probability values that a statement is correct, capturing interpretation variability that scalar representations discard. Within this distributional framework, we introduce faithfulness as a complementary evaluation dimension and present Faithfulness Divergence (FD), an information-theoretic metric quantifying the surprise induced in audience beliefs upon truth revelation. Building on these foundations, we present Retrieval-Augmented Linguistic Calibration (RALC), a lightweight post-hoc pipeline that propagates calibrated confidence signals back into natural language via retrieval-augmented rewriting. Across three QA benchmarks and five LLM families, RALC improves in-domain faithfulness and calibration up to 66% and 58%, respectively, outperforming black-box and grey-box calibration baselines.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Calibration | MMLU | -- | 58 | |
| Question Answering | TruthfulQA | AUROC0.664 | 14 | |
| Linguistic-space calibration | SQuAD 2.0 | Mean Faithfulness Divergence Reduction66.4 | 9 | |
| Linguistic-space calibration | TruthfulQA | Faithfulness Divergence Reduction70.4 | 9 | |
| Question Answering | MMLU | AUROC (MMLU QA)0.636 | 5 | |
| Question Answering | SQuAD 2.0 | AUROC58.8 | 5 |