首个细粒度推理忠实性评测，定位模型推理中的每一步错误。
GRACE: Step-Level Benchmark for Faithful Reasoning over Context
- 构建人类标注的步骤级忠实性数据集，逐步判断推理是否符合依据。
- 发现当前模型在正确答案下仍有60%以上步骤存在事实偏差。
- 适合关注模型可解释性与可信推理的研究者使用。
许多推理任务需要模型基于输入上下文进行判断，如文档问答和规则推演。链式思维（CoT）提示虽呈现透明推理过程，但单个步骤可能悄悄偏离原始证据，即使最终答案正确。现有方法仅在输出层面检测幻觉，无法定位错误发生位置或类型。我们提出GRACE，首个基于人工标注的步骤级忠实性基准，包含来自10个模型在4个数据集上的推理链条，每个步骤均标注忠实性、错误类别及自然语言解释。通过无监督聚类发现数据驱动的错误分类体系，将失败归为两类：推理错误（GRACE-Inference）和事实根基错误（GRACE-Grounding），每类含四个子类。评测集经人工标注，设计具有挑战性。实验显示当前模型仍有显著提升空间。将步骤级忠实性信号引入强化学习流程，可同时提升下游准确率与推理可靠性。
原文摘要 · Abstract (English)
Many reasoning tasks require models to reason over input context, from document-grounded question answering to rule-based deduction. Chain-of-Thought (CoT) prompting produces traces that appear transparent, yet individual steps can silently deviate from the source evidence, even when the final answer is correct. Existing methods detect hallucinations at the response level but fail to identify where in the chain a failure occurs or what type it is. We introduce GRACE, the first human-annotated step-level faithfulness benchmark with a data-driven error taxonomy for context-grounded textual reasoning. GRACE covers CoT traces from 10 models across 4 source datasets, with each step annotated for faithfulness, error category, and natural language explanation. A data-driven taxonomy, discovered bottom-up via unsupervised clustering, organizes failures into two tracks: GRACE-Inference (deductive errors) and GRACE-Grounding (factual grounding errors), with four categories each. The evaluation set is human-annotated and challenging by design. Our experiments reveal substantial headroom for current models. In addition, integrating step-level faithfulness signals into reinforcement learning pipelines improves both downstream accuracy and reasoning reliability.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。