@inproceedings{li-etal-2026-vrr,
title = "{VRR}: Verification-Guided Retrieval Revision for Multi-Step {QA}",
author = "Li, Zhijie and
Wang, Lei and
Li, Tianyuan and
Han, Bangju and
Ahmat, Ahtamjan and
Dong, Rui and
Wang, Zhen",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.143/",
pages = "2962--2992",
ISBN = "979-8-89176-462-0",
abstract = "Retrieval-augmented language models improve knowledge-intensive question answering, but they remain brittle on multi-step questions, where an early retrieval error can steer subsequent reasoning toward the wrong entity or relation. Existing methods can often detect that the current answer or evidence is unreliable, yet they rarely turn this diagnosis into an explicit retrieval correction. We propose Verification-Guided Retrieval Revision (VRR), a framework that uses verification as a control signal for multi-step retrieval. VRR couples a Generator that proposes search and answers, a Judge that verifies answer correctness and evidential support while emitting structured failure diagnoses, and a Retrieval Context Manager (RCM) that converts these diagnoses into retriever-facing updates such as clue selection, query rewriting, and state-aware evidence ranking. To make this loop learnable, we train VRR in three stages: iterative cold-start training to stabilize search, judge-shaped policy optimization to improve long-horizon reasoning, and diagnosis-conditioned RCM training for retrieval adaptation. Across seven open-domain QA benchmarks with a fixed retriever and corpus, training on HotpotQA and transferring to out-of-domain datasets, VRR surpasses the strongest prior baseline by 4.2 EM on average under budget-aligned evaluation, with larger gains on multi-hop benchmarks (+5.4)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-vrr">
<titleInfo>
<title>VRR: Verification-Guided Retrieval Revision for Multi-Step QA</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zhijie</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lei</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tianyuan</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bangju</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ahtamjan</namePart>
<namePart type="family">Ahmat</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rui</namePart>
<namePart type="family">Dong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhen</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Retrieval-augmented language models improve knowledge-intensive question answering, but they remain brittle on multi-step questions, where an early retrieval error can steer subsequent reasoning toward the wrong entity or relation. Existing methods can often detect that the current answer or evidence is unreliable, yet they rarely turn this diagnosis into an explicit retrieval correction. We propose Verification-Guided Retrieval Revision (VRR), a framework that uses verification as a control signal for multi-step retrieval. VRR couples a Generator that proposes search and answers, a Judge that verifies answer correctness and evidential support while emitting structured failure diagnoses, and a Retrieval Context Manager (RCM) that converts these diagnoses into retriever-facing updates such as clue selection, query rewriting, and state-aware evidence ranking. To make this loop learnable, we train VRR in three stages: iterative cold-start training to stabilize search, judge-shaped policy optimization to improve long-horizon reasoning, and diagnosis-conditioned RCM training for retrieval adaptation. Across seven open-domain QA benchmarks with a fixed retriever and corpus, training on HotpotQA and transferring to out-of-domain datasets, VRR surpasses the strongest prior baseline by 4.2 EM on average under budget-aligned evaluation, with larger gains on multi-hop benchmarks (+5.4).</abstract>
<identifier type="citekey">li-etal-2026-vrr</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.143/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2962</start>
<end>2992</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T VRR: Verification-Guided Retrieval Revision for Multi-Step QA
%A Li, Zhijie
%A Wang, Lei
%A Li, Tianyuan
%A Han, Bangju
%A Ahmat, Ahtamjan
%A Dong, Rui
%A Wang, Zhen
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F li-etal-2026-vrr
%X Retrieval-augmented language models improve knowledge-intensive question answering, but they remain brittle on multi-step questions, where an early retrieval error can steer subsequent reasoning toward the wrong entity or relation. Existing methods can often detect that the current answer or evidence is unreliable, yet they rarely turn this diagnosis into an explicit retrieval correction. We propose Verification-Guided Retrieval Revision (VRR), a framework that uses verification as a control signal for multi-step retrieval. VRR couples a Generator that proposes search and answers, a Judge that verifies answer correctness and evidential support while emitting structured failure diagnoses, and a Retrieval Context Manager (RCM) that converts these diagnoses into retriever-facing updates such as clue selection, query rewriting, and state-aware evidence ranking. To make this loop learnable, we train VRR in three stages: iterative cold-start training to stabilize search, judge-shaped policy optimization to improve long-horizon reasoning, and diagnosis-conditioned RCM training for retrieval adaptation. Across seven open-domain QA benchmarks with a fixed retriever and corpus, training on HotpotQA and transferring to out-of-domain datasets, VRR surpasses the strongest prior baseline by 4.2 EM on average under budget-aligned evaluation, with larger gains on multi-hop benchmarks (+5.4).
%U https://aclanthology.org/2026.findings-emnlp.143/
%P 2962-2992
Markdown (Informal)
[VRR: Verification-Guided Retrieval Revision for Multi-Step QA](https://aclanthology.org/2026.findings-emnlp.143/) (Li et al., Findings 2026)
ACL
- Zhijie Li, Lei Wang, Tianyuan Li, Bangju Han, Ahtamjan Ahmat, Rui Dong, and Zhen Wang. 2026. VRR: Verification-Guided Retrieval Revision for Multi-Step QA. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 2962–2992, Budapest, Hungary. Association for Computational Linguistics.