@inproceedings{holmstrom-etal-2026-systematic,
title = "A Systematic Comparison of Multilingual Interpretability Methods Reveals Anisotropy-Driven Failures",
author = {Holmstr{\"o}m, Oskar and
Bollmann, Marcel and
Kuhlmann, Marco},
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.999/",
doi = "10.18653/2026.findings-emnlp.999",
pages = "21115--21164",
ISBN = "979-8-89176-462-0",
abstract = "Multilingual language models develop shared cross-lingual representations, and various interpretability methods claim to quantify this sharing. These methods have been developed largely in isolation, and when they disagree, it is unclear whether the disagreement reflects a property of the model or an artifact of the measurement. We compare four sharing metrics{---}CKA, ANC, GMM dominance per token, and ILO{---}across 21 base models from five families (125M{--}14B parameters) and correlate each with cross-lingual transfer on five downstream tasks. We find that the metrics differ in their quantification of cross-lingual sharing in these models and suggest that the disagreement traces to *anisotropy*, the tendency of representations to cluster in a narrow cone of the embedding space. Only ILO{'}s correlation with cross-lingual transfer (Spearman{'}s {\ensuremath{\rho}} = 0.90) survives controls for model size, family, and per-task variation. We therefore recommend ILO as the primary sharing metric, to be reported alongside anisotropy diagnostics."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="holmstrom-etal-2026-systematic">
<titleInfo>
<title>A Systematic Comparison of Multilingual Interpretability Methods Reveals Anisotropy-Driven Failures</title>
</titleInfo>
<name type="personal">
<namePart type="given">Oskar</namePart>
<namePart type="family">Holmström</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marcel</namePart>
<namePart type="family">Bollmann</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Kuhlmann</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>Multilingual language models develop shared cross-lingual representations, and various interpretability methods claim to quantify this sharing. These methods have been developed largely in isolation, and when they disagree, it is unclear whether the disagreement reflects a property of the model or an artifact of the measurement. We compare four sharing metrics—CKA, ANC, GMM dominance per token, and ILO—across 21 base models from five families (125M–14B parameters) and correlate each with cross-lingual transfer on five downstream tasks. We find that the metrics differ in their quantification of cross-lingual sharing in these models and suggest that the disagreement traces to *anisotropy*, the tendency of representations to cluster in a narrow cone of the embedding space. Only ILO’s correlation with cross-lingual transfer (Spearman’s \ensuremathρ = 0.90) survives controls for model size, family, and per-task variation. We therefore recommend ILO as the primary sharing metric, to be reported alongside anisotropy diagnostics.</abstract>
<identifier type="citekey">holmstrom-etal-2026-systematic</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.999</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.999/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>21115</start>
<end>21164</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T A Systematic Comparison of Multilingual Interpretability Methods Reveals Anisotropy-Driven Failures
%A Holmström, Oskar
%A Bollmann, Marcel
%A Kuhlmann, Marco
%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 holmstrom-etal-2026-systematic
%X Multilingual language models develop shared cross-lingual representations, and various interpretability methods claim to quantify this sharing. These methods have been developed largely in isolation, and when they disagree, it is unclear whether the disagreement reflects a property of the model or an artifact of the measurement. We compare four sharing metrics—CKA, ANC, GMM dominance per token, and ILO—across 21 base models from five families (125M–14B parameters) and correlate each with cross-lingual transfer on five downstream tasks. We find that the metrics differ in their quantification of cross-lingual sharing in these models and suggest that the disagreement traces to *anisotropy*, the tendency of representations to cluster in a narrow cone of the embedding space. Only ILO’s correlation with cross-lingual transfer (Spearman’s \ensuremathρ = 0.90) survives controls for model size, family, and per-task variation. We therefore recommend ILO as the primary sharing metric, to be reported alongside anisotropy diagnostics.
%R 10.18653/2026.findings-emnlp.999
%U https://aclanthology.org/2026.findings-emnlp.999/
%U https://doi.org/10.18653/2026.findings-emnlp.999
%P 21115-21164
Markdown (Informal)
[A Systematic Comparison of Multilingual Interpretability Methods Reveals Anisotropy-Driven Failures](https://aclanthology.org/2026.findings-emnlp.999/) (Holmström et al., Findings 2026)
ACL