MR-Align: Meta-Reasoning Informed Factuality Alignment for Large Reasoning Models

TL;DR

MR-ALIGN is a novel framework that improves factual accuracy in large reasoning models by aligning their reasoning process through meta-reasoning and transition-aware rewards, without external verifiers.

Contribution

It introduces a meta-reasoning informed alignment method that enhances factuality by reinforcing beneficial reasoning patterns during the model's thinking process.

Findings

Consistently improves accuracy across multiple datasets

Reduces misleading reasoning in large models

Enhances factuality without external verification

Abstract

Large reasoning models (LRMs) show strong capabilities in complex reasoning, yet their marginal gains on evidence-dependent factual questions are limited. We find this limitation is partially attributable to a reasoning-answer hit gap, where the model identifies the correct facts during reasoning but fails to incorporate them into the final response, thereby reducing factual fidelity. To address this issue, we propose MR-ALIGN, a Meta-Reasoning informed alignment framework that enhances factuality without relying on external verifiers. MR-ALIGN quantifies state transition probabilities along the model's thinking process and constructs a transition-aware implicit reward that reinforces beneficial reasoning patterns while suppressing defective ones at the atomic thinking segments. This re-weighting reshapes token-level signals into probability-aware segment scores, encouraging coherent reasoning trajectories that are more conducive to factual correctness. Empirical evaluations across four factual QA datasets and one long-form factuality benchmark show that MR-ALIGN consistently improves accuracy and truthfulness while reducing misleading reasoning. These results highlight that aligning the reasoning process itself, rather than merely the outputs, is pivotal for advancing factuality in LRMs.