Learning Evidence Sufficiency Boundaries for Selective Answering in Grounded Multi-Hop QA

学习用于接地多跳问答中选择性回答的证据充分性边界

Grounded question answering systems should answer only when the supplied evidence supports the answer. In multi-hop QA, this requirement is difficult because partial evidence can make an unsupported answer appear plausible. 接地问答系统应当仅在所提供的证据支持答案时才进行回答。在多跳问答中，这一要求很难实现，因为部分证据可能会使一个不受支持的答案看起来显得合理。

We study selective answering through evidence sufficiency boundaries: for the same question, a model should abstain under unsupported or partially supported context, answer when the context first becomes sufficient, and keep the answer stable when redundant evidence is added. 我们通过“证据充分性边界”来研究选择性回答：对于同一个问题，模型在上下文不受支持或仅部分支持时应选择弃权（拒绝回答），在上下文首次变得充分时应给出答案，并在添加冗余证据时保持答案的稳定性。

We introduce Evidence Sufficiency Boundary Training, a generation-native training framework that constructs ordered evidence chains and supervises the abstain-to-answer transition directly. The method combines level supervision, a boundary flip margin, post-boundary stability, and answer recall protection. 我们引入了“证据充分性边界训练”（Evidence Sufficiency Boundary Training），这是一种原生于生成的训练框架，它构建有序的证据链，并直接监督从“弃权”到“回答”的转换过程。该方法结合了层级监督、边界翻转边际、边界后稳定性以及答案召回保护。

We build evidence chains from HotpotQA, 2WikiMultiHopQA, and MuSiQue, then evaluate models with chain metrics, raw QA utility, and unsupported-answer rates on external non-answerable sets. 我们利用 HotpotQA、2WikiMultiHopQA 和 MuSiQue 构建了证据链，随后通过链式指标、原始问答效用以及在外部不可回答数据集上的错误回答率对模型进行了评估。

With Qwen2.5-3B-Instruct and LoRA adaptation, Evidence Sufficiency Boundary Training gives the strongest boundary localization among the tested systems, with flip accuracy of 0.807 compared with 0.781 for a token-level abstention baseline. It also achieves the lowest overall unsupported-answer rate on external non-answerable evaluation, 0.095 compared with 0.101 for the same baseline, while retaining competitive raw QA F1. 在使用 Qwen2.5-3B-Instruct 和 LoRA 微调的情况下，证据充分性边界训练在所有测试系统中表现出最强的边界定位能力，其翻转准确率达到 0.807，而基于 Token 级弃权的基准模型仅为 0.781。它在外部不可回答评估中也实现了最低的总体错误回答率（0.095，基准模型为 0.101），同时保持了具有竞争力的原始问答 F1 分数。

The results show that grounded selective answering improves when training marks the evidence level where refusal should give way to answering. 研究结果表明，当训练过程明确标记出“拒绝回答”应转变为“给出答案”的证据层级时，接地选择性问答的效果会得到提升。