@inproceedings{su-etal-2026-attention,
title = "When Attention Escapes: Diagnosing and Mitigating Hallucination in Retrieval-Augmented Generation",
author = "Su, Wencheng and
Wang, Qinglong and
Ba, Zhongjie and
Chu, Zhixuan and
Zhong, Zhenyu and
Yan, Liu and
Ren, Kui",
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.411/",
pages = "8678--8696",
ISBN = "979-8-89176-462-0",
abstract = "Retrieval-Augmented Generation (RAG) mitigates hallucinations by grounding generation in retrieved evidence, yet failures persist when models underutilize the retrieved context. While prior work attributes this to conflicts between external context and internal parametric knowledge encoded in Feed-Forward Networks (FFNs), our empirical analyses reveal that FFN-derived signals are unstable and insufficient for reliable diagnosis across diverse contexts. We therefore propose the Attention Escape Score (AES), a training-free diagnostic framework that provides a mechanistic lens on RAG hallucinations via dynamic attention routing failures. We models the failures as weakened Semantic Anchoring, increased Contextual Distraction, and heightened Generation Inertia, moving beyond static component-level analyses. Extensive experiments show that AES significantly outperforms state-of-the-art baselines with robust cross-model generalization. Furthermore, we introduce Pruning Escape Routes (PER), a lightweight inference-time intervention that suppresses hallucinations by attenuating identified escape pathways, achieving improved faithfulness with negligible degradation to complex reasoning and general capabilities."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="su-etal-2026-attention">
<titleInfo>
<title>When Attention Escapes: Diagnosing and Mitigating Hallucination in Retrieval-Augmented Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wencheng</namePart>
<namePart type="family">Su</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qinglong</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhongjie</namePart>
<namePart type="family">Ba</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhixuan</namePart>
<namePart type="family">Chu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhenyu</namePart>
<namePart type="family">Zhong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liu</namePart>
<namePart type="family">Yan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kui</namePart>
<namePart type="family">Ren</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 Generation (RAG) mitigates hallucinations by grounding generation in retrieved evidence, yet failures persist when models underutilize the retrieved context. While prior work attributes this to conflicts between external context and internal parametric knowledge encoded in Feed-Forward Networks (FFNs), our empirical analyses reveal that FFN-derived signals are unstable and insufficient for reliable diagnosis across diverse contexts. We therefore propose the Attention Escape Score (AES), a training-free diagnostic framework that provides a mechanistic lens on RAG hallucinations via dynamic attention routing failures. We models the failures as weakened Semantic Anchoring, increased Contextual Distraction, and heightened Generation Inertia, moving beyond static component-level analyses. Extensive experiments show that AES significantly outperforms state-of-the-art baselines with robust cross-model generalization. Furthermore, we introduce Pruning Escape Routes (PER), a lightweight inference-time intervention that suppresses hallucinations by attenuating identified escape pathways, achieving improved faithfulness with negligible degradation to complex reasoning and general capabilities.</abstract>
<identifier type="citekey">su-etal-2026-attention</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.411/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>8678</start>
<end>8696</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T When Attention Escapes: Diagnosing and Mitigating Hallucination in Retrieval-Augmented Generation
%A Su, Wencheng
%A Wang, Qinglong
%A Ba, Zhongjie
%A Chu, Zhixuan
%A Zhong, Zhenyu
%A Yan, Liu
%A Ren, Kui
%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 su-etal-2026-attention
%X Retrieval-Augmented Generation (RAG) mitigates hallucinations by grounding generation in retrieved evidence, yet failures persist when models underutilize the retrieved context. While prior work attributes this to conflicts between external context and internal parametric knowledge encoded in Feed-Forward Networks (FFNs), our empirical analyses reveal that FFN-derived signals are unstable and insufficient for reliable diagnosis across diverse contexts. We therefore propose the Attention Escape Score (AES), a training-free diagnostic framework that provides a mechanistic lens on RAG hallucinations via dynamic attention routing failures. We models the failures as weakened Semantic Anchoring, increased Contextual Distraction, and heightened Generation Inertia, moving beyond static component-level analyses. Extensive experiments show that AES significantly outperforms state-of-the-art baselines with robust cross-model generalization. Furthermore, we introduce Pruning Escape Routes (PER), a lightweight inference-time intervention that suppresses hallucinations by attenuating identified escape pathways, achieving improved faithfulness with negligible degradation to complex reasoning and general capabilities.
%U https://aclanthology.org/2026.findings-emnlp.411/
%P 8678-8696
Markdown (Informal)
[When Attention Escapes: Diagnosing and Mitigating Hallucination in Retrieval-Augmented Generation](https://aclanthology.org/2026.findings-emnlp.411/) (Su et al., Findings 2026)
ACL