arXiv cs.CL· Ruoran Xu, Yilun Zhao, Ziyu Chen, Xiaoqiang kang, Arman Cohan·· 11 小时前AI 评分38
在线策略蒸馏的“证据绕过”问题：答案更准，证据响应更弱
Better Answers, Weaker Evidence Control: Evidence Bypass in On-Policy Distillation
AI 导读
研究发现在线策略蒸馏（OPD）在提升答案准确率的同时会削弱模型对支撑证据的响应，并将这一现象命名为“证据绕过”。在 QASPER 上用 Qwen3-4B 学生模型测试，标准 OPD 使原文档准确率提升 3.4 个百分点、反事实准确率下降 5.3 个百分点、绕过率上升 15.7 个百分点，PeerQA 上呈相同趋势。
正文 · 原文
Abstract:On-policy distillation (OPD) can improve answer accuracy while weakening responsiveness to supporting evidence. We investigate this separation in scientific-document question answering through matched document interventions that change the supported answer while preserving the question and unrelated context. We identify evidence bypass: a student retains an originally correct answer after the document changes to support a different one. On QASPER with a Qwen3-4B student, standard OPD raises original-document accuracy by 3.4 percentage points, lowers counterfactual accuracy by 5.3 points, and increases bypass by 15.7 points; the same directional pattern appears on PeerQA. Controlled perturbations localize the shift to answer-critical evidence, and bypass grows with the student's pre-distillation closed-book preference for the original answer. A factorized study of teacher targets, privileged information, divergence, and supervision density finds that explicitly supplying the teacher with annotated answer-critical spans yields the largest joint improvement in evidence response and claim support. These findings motivate Evidence-Paired OPD (EP-OPD), which combines on-policy supervision on matched documents supporting different answers to the same question with an answer-level margin that favors each document's supported answer over its paired alternative. Across both datasets and two student sizes, EP-OPD recovers 54-66% of standard OPD's counterfactual-accuracy loss while preserving 80-85% of its original-document accuracy gain. Our results establish evidence responsiveness as a distinct training and evaluation target alongside answer accuracy. Our code is publicly available at: this https URL.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.01967 [cs.CL] |
| (or arXiv:2611.01967v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.01967 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ruoran Xu [view email]
[v1]
Sun, 1 Nov 2026 15:05:28 UTC (2,107 KB)
来源：arXiv cs.CL · arxiv.org