更好的检索，更差的鲁棒性：多跳RAG如何放大上游ASR错误

摘要

基于语音的应用在接入任何检索模块之前，会先通过自动语音识别（ASR）将口语查询转为文本，因此ASR错误会作为固定的上游约束进入处理流程。我们通过实验检验了标准检索增强生成（RAG）的两种扩展——实体图谱链接与迭代式改写——是会吸收还是放大这些错误。利用神经语音合成生成的四种英语口音，我们在三个多跳问答基准（HotpotQA、2WikiMultiHopQA和MuSiQue）上评估了四种RAG配置，并以纯文本基准（clean-text oracle）作为对照。尽管结构更丰富的配置在ASR输入下通常保持更高的绝对F1值，但两种扩展都会放大错误：在所有三个基准上，其组合方案下，从纯文本到最高词错误率（WER）口音之间的F1差距，比朴素稠密检索方案大36%–67%。主要的失败模式是一个或多个查询实体被破坏，在2WikiMultiHopQA上，这一原因占所有四种方法性能下降案例的87%–96%。两种轻量级的表层形式缓解措施仅能缩小剩余差距的一小部分，这表明下游检索结构会放大剩余的实体错误。我们在https://github.com/ZhenghuaBao/spoken-multihop-rag 发布了代码和数据。

English

Speech-based applications pass spoken queries through automatic speech recognition (ASR) before any retrieval module, so ASR errors enter the pipeline as a fixed upstream constraint. We empirically test whether two extensions to standard retrieval-augmented generation (RAG), entity-graph linking and iterative reformulation, absorb or amplify these errors. Using four English accents synthesized through neural TTS, we evaluate four RAG configurations on three multi-hop QA benchmarks (HotpotQA, 2WikiMultiHopQA and MuSiQue) against a clean-text oracle. Although the structurally richer configurations generally retain higher absolute F1 under ASR input, both extensions amplify the error: the F1 gap from clean text to the highest-WER accent is 36-67% larger under their combination than under naive dense retrieval, on all three benchmarks. The dominant failure mode is corruption of one or more query entities, accounting for 87-96% of degradation cases on 2WikiMultiHopQA across all four methods. Two lightweight surface-form mitigations leave most of the gap intact, indicating that downstream retrieval structure amplifies remaining entity errors. We release code and data at https://github.com/ZhenghuaBao/spoken-multihop-rag .