REFACT: Adaptive Fact Restatement for Compact and Faithful Chain-of-Thought Reasoning

2026-07-22

AI Summary

This work addresses the vulnerability of large language models to sparse, noisy, or conflicting context during long-chain reasoning, which often leads to factually inconsistent outputs lacking adequate citation support. To mitigate this, the authors propose REFACT, a novel framework featuring an adaptive fact restatement mechanism that dynamically determines when and at what granularity to restate facts, transforming original citations into intermediate states that better support reasoning. REFACT integrates a two-stage SFT-to-RL training pipeline and introduces a utility reward that encourages traceable, well-formed, and answer-sufficient citations. Experimental results demonstrate that REFACT substantially improves question-answering accuracy and counterfactual faithfulness on LongBench, LV-Eval, and ConFiQA, while reducing token consumption and yielding more compact, dense reasoning traces.

Abstract

Large language models increasingly rely on long-form reasoning for complex tasks, yet their reasoning traces may drift away from the supplied context when evidence is sparse, noisy, or in conflict with parametric knowledge. Existing grounding methods either attach citations after generation or encourage evidence retrieval inside the trace, but they often do not ensure that cited content is sufficient for the local inference and final answer. We propose REFACT, an adaptive fact-restatement citation framework that trains models to decide when a reasoning step needs contextual grounding and at what granularity source facts should be restated. This design avoids both unsupported inference and indiscriminate fact copying by turning citations into answer-supporting intermediate states. REFACT is optimized with a two-stage SFT-to-RL pipeline in which a citation-utility reward encourages cited facts to be well-formed, source-traceable, and answer-sufficient. Experiments on LongBench, LV-Eval, and ConFiQA show that REFACT improves long-context QA and counterfactual faithfulness while substantially reducing token consumption. Further analysis shows that REFACT preserves more answer-bearing evidence with fewer restated facts, yielding reasoning traces that are denser rather than longer. All code and data are available at https://github.com/NEUIR/REFACT.

Problem

• chain-of-thought reasoning
• contextual grounding
• fact restatement
• faithfulness
• long-context QA

Innovation

• adaptive fact restatement
• chain-of-thought reasoning
• contextual grounding
• citation-utility reward
• token-efficient reasoning