@inproceedings{neubecker-etal-2026-language,
title = "Language-Triggered Value Inconsistencies in Multilingual {LLM}s",
author = {Neubecker, Mauricio and
Eder, Maximilian and
W{\"u}rfl, Tim and
Hoelscher-Obermaier, Jason and
Krishnan, Jeyashree and
Mandyam Rangarajan, Ajay},
editor = "Ranaldi, Leonardo and
Pucci, Giulia and
Ranaldi, Federico and
Valentino, Marco and
Birch, Alexandra and
Zanzotto, Fabio Massimo",
booktitle = "The Proceedings of the Workshop on Open Reasoning Across Cultures Languages ({ORACLE})",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.oracle-main.9/",
pages = "134--149",
ISBN = "979-8-89176-483-5",
abstract = "Large Language Models are deployed across languages, yet whether their responses to valueladen questions remain stable across prompt languages is unclear. We evaluate sixteen instruction-tuned models, seven in the 7{--}9B band and nine larger, all open-weight except one. On translated GlobalOpinionQA items we compare answer distributions across seven languages and against country-matched survey pluralities. Model responses vary systematically by prompt language, yet in every language they stay closer to US survey responses than to the country-matched population. On items that carry both a US and a Chinese human answer, all seven smaller models match the US one more often than the Chinese one. In a crossed prompt-language and national-identity cue experiment, responses stay closer to the prompt-language condition than to the statedidentity condition, including in the larger models, even though a cue-uptake probe confirms the identity cue is registered. This demonstrates language dominance over identity cues, although it does not by itself identify training data as the cause. This inconsistency matters most when the request is harmful, since a model judged safe in one language need not be safe in another. On paired English{--} Chinese prompts from TRIDENT, a benchmark of expert-validated harmful requests, four of the five non-Chinese-origin models in the 7{--}9B band receive significantly higher harmfulness ratings in Chinese, whereas neither Chineseorigin model does. The language gap replicates on six larger models, though the origin split does not. The choice of translator also changes the measured answers, so we report metrics and comparisons that do not depend on it, and we treat raw cross-lingual distances as uninterpretable on their own."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="neubecker-etal-2026-language">
<titleInfo>
<title>Language-Triggered Value Inconsistencies in Multilingual LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mauricio</namePart>
<namePart type="family">Neubecker</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maximilian</namePart>
<namePart type="family">Eder</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tim</namePart>
<namePart type="family">Würfl</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jason</namePart>
<namePart type="family">Hoelscher-Obermaier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jeyashree</namePart>
<namePart type="family">Krishnan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ajay</namePart>
<namePart type="family">Mandyam Rangarajan</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>The Proceedings of the Workshop on Open Reasoning Across Cultures Languages (ORACLE)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Leonardo</namePart>
<namePart type="family">Ranaldi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Giulia</namePart>
<namePart type="family">Pucci</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Federico</namePart>
<namePart type="family">Ranaldi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Valentino</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alexandra</namePart>
<namePart type="family">Birch</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fabio</namePart>
<namePart type="given">Massimo</namePart>
<namePart type="family">Zanzotto</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">979-8-89176-483-5</identifier>
</relatedItem>
<abstract>Large Language Models are deployed across languages, yet whether their responses to valueladen questions remain stable across prompt languages is unclear. We evaluate sixteen instruction-tuned models, seven in the 7–9B band and nine larger, all open-weight except one. On translated GlobalOpinionQA items we compare answer distributions across seven languages and against country-matched survey pluralities. Model responses vary systematically by prompt language, yet in every language they stay closer to US survey responses than to the country-matched population. On items that carry both a US and a Chinese human answer, all seven smaller models match the US one more often than the Chinese one. In a crossed prompt-language and national-identity cue experiment, responses stay closer to the prompt-language condition than to the statedidentity condition, including in the larger models, even though a cue-uptake probe confirms the identity cue is registered. This demonstrates language dominance over identity cues, although it does not by itself identify training data as the cause. This inconsistency matters most when the request is harmful, since a model judged safe in one language need not be safe in another. On paired English– Chinese prompts from TRIDENT, a benchmark of expert-validated harmful requests, four of the five non-Chinese-origin models in the 7–9B band receive significantly higher harmfulness ratings in Chinese, whereas neither Chineseorigin model does. The language gap replicates on six larger models, though the origin split does not. The choice of translator also changes the measured answers, so we report metrics and comparisons that do not depend on it, and we treat raw cross-lingual distances as uninterpretable on their own.</abstract>
<identifier type="citekey">neubecker-etal-2026-language</identifier>
<location>
<url>https://aclanthology.org/2026.oracle-main.9/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>134</start>
<end>149</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Language-Triggered Value Inconsistencies in Multilingual LLMs
%A Neubecker, Mauricio
%A Eder, Maximilian
%A Würfl, Tim
%A Hoelscher-Obermaier, Jason
%A Krishnan, Jeyashree
%A Mandyam Rangarajan, Ajay
%Y Ranaldi, Leonardo
%Y Pucci, Giulia
%Y Ranaldi, Federico
%Y Valentino, Marco
%Y Birch, Alexandra
%Y Zanzotto, Fabio Massimo
%S The Proceedings of the Workshop on Open Reasoning Across Cultures Languages (ORACLE)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-483-5
%F neubecker-etal-2026-language
%X Large Language Models are deployed across languages, yet whether their responses to valueladen questions remain stable across prompt languages is unclear. We evaluate sixteen instruction-tuned models, seven in the 7–9B band and nine larger, all open-weight except one. On translated GlobalOpinionQA items we compare answer distributions across seven languages and against country-matched survey pluralities. Model responses vary systematically by prompt language, yet in every language they stay closer to US survey responses than to the country-matched population. On items that carry both a US and a Chinese human answer, all seven smaller models match the US one more often than the Chinese one. In a crossed prompt-language and national-identity cue experiment, responses stay closer to the prompt-language condition than to the statedidentity condition, including in the larger models, even though a cue-uptake probe confirms the identity cue is registered. This demonstrates language dominance over identity cues, although it does not by itself identify training data as the cause. This inconsistency matters most when the request is harmful, since a model judged safe in one language need not be safe in another. On paired English– Chinese prompts from TRIDENT, a benchmark of expert-validated harmful requests, four of the five non-Chinese-origin models in the 7–9B band receive significantly higher harmfulness ratings in Chinese, whereas neither Chineseorigin model does. The language gap replicates on six larger models, though the origin split does not. The choice of translator also changes the measured answers, so we report metrics and comparisons that do not depend on it, and we treat raw cross-lingual distances as uninterpretable on their own.
%U https://aclanthology.org/2026.oracle-main.9/
%P 134-149
Markdown (Informal)
[Language-Triggered Value Inconsistencies in Multilingual LLMs](https://aclanthology.org/2026.oracle-main.9/) (Neubecker et al., ORACLE 2026)
ACL
- Mauricio Neubecker, Maximilian Eder, Tim Würfl, Jason Hoelscher-Obermaier, Jeyashree Krishnan, and Ajay Mandyam Rangarajan. 2026. Language-Triggered Value Inconsistencies in Multilingual LLMs. In The Proceedings of the Workshop on Open Reasoning Across Cultures Languages (ORACLE), pages 134–149, Budapest, Hungary. Association for Computational Linguistics.