@inproceedings{ravichandran-2026-mechanistic,
title = "Mechanistic Evidence Routing in Transformers: Margin-Gated Intervention for Retrieval-Conflict Hallucination",
author = "Ravichandran, Sridurgesh",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.100/",
pages = "2193--2211",
ISBN = "979-8-89176-439-2",
abstract = "When a language model ignores retrieved evidence and answers from memory instead, this is not a knowledge failure, the correct information is present. It is a routing failure; the model commits to its parametric answer despite the evidence signal. We show this distinction is mechanistically meaningful, localizable, and correctable at inference time without training data or weight modification.\par A single prefill-margin computation detects routing failures (AUC $\geq 0.93$ across five model dataset pairs) and gates a format-appropriate logit-bias correction. On MedQA multi-option conflicts, this achieves 77.5{\%} restoration at zero harm, exceeding the strongest prompting baseline by 67.5 pp and matching domain-matched ITI without training data. On open-ended PopQA (3,141 items, three models), balanced alias promotion achieves 80-88{\%} restoration at near-zero harm, Pareto-dominating prompt-only baselines in five of six cross-model conditions. The central mechanistic finding is that effective late-layer intervention sites are structurally distinct from the causal commitment locus (a five-head cluster, joint flip rate 88.6{\%}, $p < 10^{-9}$), suggesting that steering can succeed through residual-stream write access without exactly identifying the full decision circuit."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ravichandran-2026-mechanistic">
<titleInfo>
<title>Mechanistic Evidence Routing in Transformers: Margin-Gated Intervention for Retrieval-Conflict Hallucination</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sridurgesh</namePart>
<namePart type="family">Ravichandran</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>When a language model ignores retrieved evidence and answers from memory instead, this is not a knowledge failure, the correct information is present. It is a routing failure; the model commits to its parametric answer despite the evidence signal. We show this distinction is mechanistically meaningful, localizable, and correctable at inference time without training data or weight modification.\par A single prefill-margin computation detects routing failures (AUC \geq 0.93 across five model dataset pairs) and gates a format-appropriate logit-bias correction. On MedQA multi-option conflicts, this achieves 77.5% restoration at zero harm, exceeding the strongest prompting baseline by 67.5 pp and matching domain-matched ITI without training data. On open-ended PopQA (3,141 items, three models), balanced alias promotion achieves 80-88% restoration at near-zero harm, Pareto-dominating prompt-only baselines in five of six cross-model conditions. The central mechanistic finding is that effective late-layer intervention sites are structurally distinct from the causal commitment locus (a five-head cluster, joint flip rate 88.6%, p < 10⁻9), suggesting that steering can succeed through residual-stream write access without exactly identifying the full decision circuit.</abstract>
<identifier type="citekey">ravichandran-2026-mechanistic</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.100/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>2193</start>
<end>2211</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mechanistic Evidence Routing in Transformers: Margin-Gated Intervention for Retrieval-Conflict Hallucination
%A Ravichandran, Sridurgesh
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F ravichandran-2026-mechanistic
%X When a language model ignores retrieved evidence and answers from memory instead, this is not a knowledge failure, the correct information is present. It is a routing failure; the model commits to its parametric answer despite the evidence signal. We show this distinction is mechanistically meaningful, localizable, and correctable at inference time without training data or weight modification.\par A single prefill-margin computation detects routing failures (AUC \geq 0.93 across five model dataset pairs) and gates a format-appropriate logit-bias correction. On MedQA multi-option conflicts, this achieves 77.5% restoration at zero harm, exceeding the strongest prompting baseline by 67.5 pp and matching domain-matched ITI without training data. On open-ended PopQA (3,141 items, three models), balanced alias promotion achieves 80-88% restoration at near-zero harm, Pareto-dominating prompt-only baselines in five of six cross-model conditions. The central mechanistic finding is that effective late-layer intervention sites are structurally distinct from the causal commitment locus (a five-head cluster, joint flip rate 88.6%, p < 10⁻9), suggesting that steering can succeed through residual-stream write access without exactly identifying the full decision circuit.
%U https://aclanthology.org/2026.aacl-long.100/
%P 2193-2211
Markdown (Informal)
[Mechanistic Evidence Routing in Transformers: Margin-Gated Intervention for Retrieval-Conflict Hallucination](https://aclanthology.org/2026.aacl-long.100/) (Ravichandran, AACL-IJCNLP 2026)
ACL