arXiv cs.CL· Yerim Hong, Sohyun Son, Heeseung Chung, Mujeen Sung·· 12 小时前AI 评分34
EXCISE：定位并剔除医疗问答中的干扰句
EXCISE: Locating and Excising Distracting Sentences in Medical Question Answering
AI 导读
研究者提出推理时方法 EXCISE，在不知道哪句是干扰句的情况下，屏蔽对答案影响最大的句子，但仅当模型隐藏状态将其标记为与病例不符时才执行。在自建的 MedQA-TD 基准上，EXCISE 在三个模型上均取得 16.7-22.8 个准确率点的提升，代价是在无干扰同题上损失 1.5-2.1 点，优于提示、改写和 steering 基线。
正文
Abstract:Language models are increasingly considered for clinical decision support, where their answers should be grounded in clinical evidence and remain robust to irrelevant information. Prior work shows that a single irrelevant sentence can lower the accuracy of medical models, but not where inside the model its effect can be removed, or how to do so without knowing which sentence it is. We find that blocking attention to a known distractor restores the answer, whereas head masking can suppress correct answers. We also show that the model's hidden states mark the distractor as the sentence that least fits the case even when the model follows it. To remove a distractor's effect without knowing which sentence it is, we propose EXCISE, an inference-time method that blocks the sentence most influencing the answer, but only if the model's hidden states mark it as inconsistent with the case. On MedQA-TD (MedQA with Targeted Distractors), a benchmark we construct in which each distractor targets a specific wrong option, EXCISE gains 16.7-22.8 accuracy points at a 1.5-2.1-point cost on the same questions without distractors, outperforming prompting, rewriting, and steering baselines for all three models. Its gains extend to MedDistractQA and BiasMedQA, as well as chain-of-thought generation.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.35558 [cs.CL] |
| (or arXiv:2610.35558v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.35558 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yerim Hong [view email]
[v1]
Mon, 26 Oct 2026 14:59:54 UTC (738 KB)
来源：arXiv cs.CL · arxiv.org