@inproceedings{chen-etal-2026-heal,
title = "{HEAL}: Anchored Hyperspherical Equilibrium for Multilingual Safety Alignment",
author = "Chen, Wenwei and
Chen, Hongliang and
Zhan, Kun",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1294/",
pages = "29463--29472",
ISBN = "979-8-89176-459-0",
abstract = "Safety alignment in large language models is markedly uneven across languages: behaviors learned from English preference data fail to transfer reliably into lower-resource languages, and existing remedies based on per-language preference data or pairwise translation pivots do not scale. We propose Hyperspherical Equilibrium for Aligned Languages (HEAL), a lightweight auxiliary objective that augments any DPO-style alignment pipeline without altering its supervised pairs. HEAL applies two opposing forces to the multilingual representations of the same prompt: a shrink term draws each non-English representation toward a stop-gradient English anchor, while a dilation term repels same-prompt language representations through a pairwise hinge. At equilibrium, representations occupy an English-centered shell-like region that supports cross-lingual transfer while preserving per-language distinctness. Across multiple backbones, HEAL substantially strengthens multilingual safety with the largest gains on the weakest languages, while preserving general capability and showing favorable out-of-distribution behavior on jailbreak evaluation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-heal">
<titleInfo>
<title>HEAL: Anchored Hyperspherical Equilibrium for Multilingual Safety Alignment</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wenwei</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongliang</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kun</namePart>
<namePart type="family">Zhan</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Safety alignment in large language models is markedly uneven across languages: behaviors learned from English preference data fail to transfer reliably into lower-resource languages, and existing remedies based on per-language preference data or pairwise translation pivots do not scale. We propose Hyperspherical Equilibrium for Aligned Languages (HEAL), a lightweight auxiliary objective that augments any DPO-style alignment pipeline without altering its supervised pairs. HEAL applies two opposing forces to the multilingual representations of the same prompt: a shrink term draws each non-English representation toward a stop-gradient English anchor, while a dilation term repels same-prompt language representations through a pairwise hinge. At equilibrium, representations occupy an English-centered shell-like region that supports cross-lingual transfer while preserving per-language distinctness. Across multiple backbones, HEAL substantially strengthens multilingual safety with the largest gains on the weakest languages, while preserving general capability and showing favorable out-of-distribution behavior on jailbreak evaluation.</abstract>
<identifier type="citekey">chen-etal-2026-heal</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1294/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>29463</start>
<end>29472</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HEAL: Anchored Hyperspherical Equilibrium for Multilingual Safety Alignment
%A Chen, Wenwei
%A Chen, Hongliang
%A Zhan, Kun
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F chen-etal-2026-heal
%X Safety alignment in large language models is markedly uneven across languages: behaviors learned from English preference data fail to transfer reliably into lower-resource languages, and existing remedies based on per-language preference data or pairwise translation pivots do not scale. We propose Hyperspherical Equilibrium for Aligned Languages (HEAL), a lightweight auxiliary objective that augments any DPO-style alignment pipeline without altering its supervised pairs. HEAL applies two opposing forces to the multilingual representations of the same prompt: a shrink term draws each non-English representation toward a stop-gradient English anchor, while a dilation term repels same-prompt language representations through a pairwise hinge. At equilibrium, representations occupy an English-centered shell-like region that supports cross-lingual transfer while preserving per-language distinctness. Across multiple backbones, HEAL substantially strengthens multilingual safety with the largest gains on the weakest languages, while preserving general capability and showing favorable out-of-distribution behavior on jailbreak evaluation.
%U https://aclanthology.org/2026.emnlp-main.1294/
%P 29463-29472
Markdown (Informal)
[HEAL: Anchored Hyperspherical Equilibrium for Multilingual Safety Alignment](https://aclanthology.org/2026.emnlp-main.1294/) (Chen et al., EMNLP 2026)
ACL