@inproceedings{elchafei-etal-2026-contradictory,
title = "Contradictory Evidence in {LLM}s: Tracing Internal Conflict Behavior with White-Box Signals",
author = "Elchafei, Passant and
Monshizadehkamkar, Ali and
Masoudian, Shahed and
Swain, Monorama and
Hou, Yufang and
Schedl, Markus",
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.1838/",
pages = "39240--39252",
ISBN = "979-8-89176-462-0",
abstract = "When LLMs are given contradictory passages, the generated response may appear decisive while masking how internal signals behave under conflicting evidence. Does it favor one passage, blend both, or ignore both? We study model behavior under inter-context contradiction by introducing a white-box tracing framework that extracts internal model signals based on hidden-state shifts, counterfactual attribution via passage removal, and token-level uncertainty during generation under two single-passage prompts and one dual-passage prompt with conflicting evidence. We evaluate four open-weight LLMs from the LLaMA (8B and 70B), OLMo, and Qwen on WikiContradict, a human-annotated QA benchmark with contradictory passages. Each dual-passage answer is labeled as committing to passage A or passage B, blending both, or fabricating an answer unsupported by either passage. Comparing these labels with the extracted signals shows that larger models reduce fabrication, but hidden-state and probability-level signals can disagree with the final answer. Moreover, fabricated answers often remain high-confidence, showing that confidence alone is unreliable for identifying ungrounded behavior under contradictory evidence. We further validate on ConflictBank, a larger misinformation-conflict QA dataset, showing consistent disagreement patterns across models. Finally, we test whether pre-generation white-box signals are informative beyond interpretation by training classifiers to predict output behavior. Classifiers using white-box signals outperform probability baselines, showing these signals contain diagnostic information about model behavior under contradictory evidence beyond log probabilities."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="elchafei-etal-2026-contradictory">
<titleInfo>
<title>Contradictory Evidence in LLMs: Tracing Internal Conflict Behavior with White-Box Signals</title>
</titleInfo>
<name type="personal">
<namePart type="given">Passant</namePart>
<namePart type="family">Elchafei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ali</namePart>
<namePart type="family">Monshizadehkamkar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shahed</namePart>
<namePart type="family">Masoudian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Monorama</namePart>
<namePart type="family">Swain</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yufang</namePart>
<namePart type="family">Hou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Markus</namePart>
<namePart type="family">Schedl</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>When LLMs are given contradictory passages, the generated response may appear decisive while masking how internal signals behave under conflicting evidence. Does it favor one passage, blend both, or ignore both? We study model behavior under inter-context contradiction by introducing a white-box tracing framework that extracts internal model signals based on hidden-state shifts, counterfactual attribution via passage removal, and token-level uncertainty during generation under two single-passage prompts and one dual-passage prompt with conflicting evidence. We evaluate four open-weight LLMs from the LLaMA (8B and 70B), OLMo, and Qwen on WikiContradict, a human-annotated QA benchmark with contradictory passages. Each dual-passage answer is labeled as committing to passage A or passage B, blending both, or fabricating an answer unsupported by either passage. Comparing these labels with the extracted signals shows that larger models reduce fabrication, but hidden-state and probability-level signals can disagree with the final answer. Moreover, fabricated answers often remain high-confidence, showing that confidence alone is unreliable for identifying ungrounded behavior under contradictory evidence. We further validate on ConflictBank, a larger misinformation-conflict QA dataset, showing consistent disagreement patterns across models. Finally, we test whether pre-generation white-box signals are informative beyond interpretation by training classifiers to predict output behavior. Classifiers using white-box signals outperform probability baselines, showing these signals contain diagnostic information about model behavior under contradictory evidence beyond log probabilities.</abstract>
<identifier type="citekey">elchafei-etal-2026-contradictory</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1838/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>39240</start>
<end>39252</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Contradictory Evidence in LLMs: Tracing Internal Conflict Behavior with White-Box Signals
%A Elchafei, Passant
%A Monshizadehkamkar, Ali
%A Masoudian, Shahed
%A Swain, Monorama
%A Hou, Yufang
%A Schedl, Markus
%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 elchafei-etal-2026-contradictory
%X When LLMs are given contradictory passages, the generated response may appear decisive while masking how internal signals behave under conflicting evidence. Does it favor one passage, blend both, or ignore both? We study model behavior under inter-context contradiction by introducing a white-box tracing framework that extracts internal model signals based on hidden-state shifts, counterfactual attribution via passage removal, and token-level uncertainty during generation under two single-passage prompts and one dual-passage prompt with conflicting evidence. We evaluate four open-weight LLMs from the LLaMA (8B and 70B), OLMo, and Qwen on WikiContradict, a human-annotated QA benchmark with contradictory passages. Each dual-passage answer is labeled as committing to passage A or passage B, blending both, or fabricating an answer unsupported by either passage. Comparing these labels with the extracted signals shows that larger models reduce fabrication, but hidden-state and probability-level signals can disagree with the final answer. Moreover, fabricated answers often remain high-confidence, showing that confidence alone is unreliable for identifying ungrounded behavior under contradictory evidence. We further validate on ConflictBank, a larger misinformation-conflict QA dataset, showing consistent disagreement patterns across models. Finally, we test whether pre-generation white-box signals are informative beyond interpretation by training classifiers to predict output behavior. Classifiers using white-box signals outperform probability baselines, showing these signals contain diagnostic information about model behavior under contradictory evidence beyond log probabilities.
%U https://aclanthology.org/2026.findings-emnlp.1838/
%P 39240-39252
Markdown (Informal)
[Contradictory Evidence in LLMs: Tracing Internal Conflict Behavior with White-Box Signals](https://aclanthology.org/2026.findings-emnlp.1838/) (Elchafei et al., Findings 2026)
ACL