Calibrated two-sided evidence sets improve claim support and refutation
Two-Sided Conformal Evidence Sets for Retrieval-Augmented Generation
Machine LearningComputation and Language
Summary
Sometimes, when computers generate answers using retrieved information, they may find passages that support, contradict, or don’t relate to the answer. The authors designed a method called Conformal Evidence Sets (CES) that separately reports supporting and contradicting evidence with strong guarantees on including the right kind of passages. Their method works well when the data conditions match, but can lose accuracy when tested on different distributions. They also found that correctly covering evidence and accurately judging support or refutation quality are related but distinct goals that need separate measures.
What this means in practice
- •For natural language processing engineers: Produce more trustworthy AI-generated answers by providing separate, calibrated sets of supporting and contradicting source passages during generation.
- •For legal document analysts: Identify and distinguish supporting and contradicting textual evidence for claims with calibrated confidence to aid case analysis.
Authors
Ben Jenkins, Mihaela Cardei
Abstract
Retrieval-augmented generation (RAG) can surface passages that support a claim, contradict it, or fail to address it, yet existing calibrated grounding methods are typically one-sided and model only supporting evidence. We introduce Conformal Evidence Sets (CES), a two-sided framework that returns separate support and refutation sets with finite-sample inclusion coverage of at least $1-α$ on each side under exchangeability; a Bonferroni union yields joint coverage of at least $1-2α$. On labeled candidate pools from the contrastive VitaminC benchmark, CES closely tracks nominal coverage across $α\!\in\!\{0.05,0.10,0.15,0.20,0.30\}$, with maximum deviations of $3.0$ percentage points for support and $5.4$ for refutation, and reaches essentially the same empirical coverage--size frontier as a test-tuned fixed-threshold heuristic without any test exposure. Stress tests mark the boundary of that guarantee: under FEVER distribution shift both sides under-cover, by up to $12.7$ percentage points, and support-side group-stratified recalibration substantially reduces the support gap. Scorer swaps, hard-negative retrieval, cross-dataset transfer, and a live end-to-end pilot expose a clear separation: when exchangeability holds and relevant evidence is retrieved, evidence inclusion remains calibrated even with an imperfect scorer, whereas the support/refute verdict tracks scorer quality directly. Calibrated evidence coverage and reliable verdicts are therefore distinct properties and should be reported separately.