What Do Rationales Communicate? A Message-Intervention Study in Role-Specialized QA

What Do Rationales Communicate? A Message-Intervention Study in Role-Specialized QA

推理过程传达了什么?角色化问答中的消息干预研究

Abstract: Role-specialized QA pipelines increasingly pass rationales from a reasoner to a verifier, but it is unclear what this message actually buys: better answers, stronger support assessment, or a new failure surface.

摘要: 角色化问答(Role-specialized QA)流水线越来越多地将推理过程(rationales)从推理器(reasoner)传递给验证器(verifier),但目前尚不清楚这种消息传递究竟带来了什么:是更好的答案、更强的证据评估,还是引入了新的故障面。

We introduce a message-intervention diagnostic that fixes the evidence and candidate answer while varying only the rationale passed across the reasoner-to-verifier boundary.

我们引入了一种消息干预诊断方法,在固定证据和候选答案的同时,仅改变在推理器与验证器之间传递的推理过程。

On 400 MuSiQue, HotpotQA, and 2WikiMultiHopQA examples with DeepSeek as generator and verifier, faithful rationales add almost no answer accuracy over no rationale, while corrupted rationales strongly alter support judgments.

在以 DeepSeek 作为生成器和验证器的 400 个 MuSiQue、HotpotQA 和 2WikiMultiHopQA 示例测试中,忠实的推理过程相比于没有推理过程,几乎没有提升答案准确率,而受损(corrupted)的推理过程却会显著改变对证据的支持判断。

Under a blind verifier prompt, harmless paraphrases shift support by only 0—2.5%, whereas corrupted rationales shift support by 10—22%; an explicit rationale-checking prompt amplifies the same pattern to 34—55%.

在盲验证器提示下,无害的改写仅会使支持度偏移 0—2.5%,而受损的推理过程会导致支持度偏移 10—22%;若使用明确的推理检查提示,这一模式会被放大至 34—55%。

Final answers move less (2—30%), and only 2.9—35.3% of corrupted support flips co-occur with answer changes.

最终答案的变化较小(2—30%),且只有 2.9—35.3% 的受损支持度翻转伴随着答案的改变。

Human audits show why this matters: 16/42 valid corruptions are corruption-overtrust cases, and blind humans reject or mark unclear 9/10 audited corrupted rationales that the model accepts.

人工审计揭示了其重要性:在 42 个有效的受损案例中,有 16 个属于“受损过度信任”案例;对于模型接受的受损推理过程,盲测人类审计员拒绝或标记为“不明确”的比例高达 9/10。

Cross-model and task-boundary checks show when the channel is active, amplified, inert, or folded into the task label.

跨模型和任务边界检查显示了该通道在何时处于活跃、放大、惰性状态,或被折叠进任务标签中。

Rationale sharing should be evaluated as a verification-message mechanism, not merely as a route to higher answer accuracy.

推理过程共享应被评估为一种验证消息机制,而不仅仅是提高答案准确率的途径。