@inproceedings{huang-etal-2026-disentangling,
title = "Disentangling In-Language Alignment and Cross-Lingual Safety Borrowing in Low-Resource Languages",
author = "Huang, Zeguang and
Spenpa, Zhenma and
Chu, Mio Yip Sien and
Cao, Xi and
Qun, Nuo and
Tashi, Nyima",
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.778/",
doi = "10.18653/2026.findings-emnlp.778",
pages = "16447--16459",
ISBN = "979-8-89176-462-0",
abstract = "We study two modes of low-resource language safety: in-language alignment, where models refuse harmful instructions in the target language, and cross-lingual safety borrowing, where they rely on a better-aligned proximate high-resource language, including via response-language fallback. We evaluate 11 commercial APIs on English, Mandarin, Cantonese, Tibetan, and Russian, plus 3 Tibetan-specialized open-source models, using native-speaker-verified datasets. Results reveal two patterns: Tibetan, for which response-language fallback provides limited aggregate protection in our setting, shows a safety cliff with ASR of 12.69{--}60.96{\%}, far above English; Cantonese, which shares script and vocabulary with Mandarin, keeps aggregate ASR in single digits, but Cantonese-reply ASR exceeds aggregate ASR for all 11 models (sign test, p {\ensuremath{<}} 0.001), exposing in-language safety gaps concealed by aggregate metrics. We release Cantonese-AdvBench and Tibetan-AdvBench."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2026-disentangling">
<titleInfo>
<title>Disentangling In-Language Alignment and Cross-Lingual Safety Borrowing in Low-Resource Languages</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zeguang</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhenma</namePart>
<namePart type="family">Spenpa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mio</namePart>
<namePart type="given">Yip</namePart>
<namePart type="given">Sien</namePart>
<namePart type="family">Chu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xi</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nuo</namePart>
<namePart type="family">Qun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nyima</namePart>
<namePart type="family">Tashi</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>We study two modes of low-resource language safety: in-language alignment, where models refuse harmful instructions in the target language, and cross-lingual safety borrowing, where they rely on a better-aligned proximate high-resource language, including via response-language fallback. We evaluate 11 commercial APIs on English, Mandarin, Cantonese, Tibetan, and Russian, plus 3 Tibetan-specialized open-source models, using native-speaker-verified datasets. Results reveal two patterns: Tibetan, for which response-language fallback provides limited aggregate protection in our setting, shows a safety cliff with ASR of 12.69–60.96%, far above English; Cantonese, which shares script and vocabulary with Mandarin, keeps aggregate ASR in single digits, but Cantonese-reply ASR exceeds aggregate ASR for all 11 models (sign test, p \ensuremath< 0.001), exposing in-language safety gaps concealed by aggregate metrics. We release Cantonese-AdvBench and Tibetan-AdvBench.</abstract>
<identifier type="citekey">huang-etal-2026-disentangling</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.778</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.778/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>16447</start>
<end>16459</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Disentangling In-Language Alignment and Cross-Lingual Safety Borrowing in Low-Resource Languages
%A Huang, Zeguang
%A Spenpa, Zhenma
%A Chu, Mio Yip Sien
%A Cao, Xi
%A Qun, Nuo
%A Tashi, Nyima
%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 huang-etal-2026-disentangling
%X We study two modes of low-resource language safety: in-language alignment, where models refuse harmful instructions in the target language, and cross-lingual safety borrowing, where they rely on a better-aligned proximate high-resource language, including via response-language fallback. We evaluate 11 commercial APIs on English, Mandarin, Cantonese, Tibetan, and Russian, plus 3 Tibetan-specialized open-source models, using native-speaker-verified datasets. Results reveal two patterns: Tibetan, for which response-language fallback provides limited aggregate protection in our setting, shows a safety cliff with ASR of 12.69–60.96%, far above English; Cantonese, which shares script and vocabulary with Mandarin, keeps aggregate ASR in single digits, but Cantonese-reply ASR exceeds aggregate ASR for all 11 models (sign test, p \ensuremath< 0.001), exposing in-language safety gaps concealed by aggregate metrics. We release Cantonese-AdvBench and Tibetan-AdvBench.
%R 10.18653/2026.findings-emnlp.778
%U https://aclanthology.org/2026.findings-emnlp.778/
%U https://doi.org/10.18653/2026.findings-emnlp.778
%P 16447-16459
Markdown (Informal)
[Disentangling In-Language Alignment and Cross-Lingual Safety Borrowing in Low-Resource Languages](https://aclanthology.org/2026.findings-emnlp.778/) (Huang et al., Findings 2026)
ACL