@inproceedings{kara-isitt-etal-2026-unseen,
title = "Unseen Harm: Measuring Cross-Script Safety Inconsistency with ``Missed-in-{U}rdu'' Scores in {LLM} Hate Speech Detection.",
author = "Kara-Isitt, Fawzia and
Khosla, Sonal and
Swift, Stephen",
editor = "Calabrese, Agostina and
Davidson, Thomas and
de Kock, Christine and
Khurana, Urja and
Marchiori Manerba, Marta and
Piot, Paloma and
Talat, Zeerak",
booktitle = "Proceedings of the 10th Workshop on Online Abuse and Harms ({WOAH})",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.woah-1.6/",
pages = "49--62",
ISBN = "979-8-89176-487-3",
abstract = "This paper investigates the near-total absence of Urdu from mainstream LLM safety evaluation and nine years of WOAH proceedings. A complete enumeration of all 205 papers across nine ALW/WOAH editions confirms zero dedicated Urdu papers across the entire period. To assess whether this absence has measurable consequences, five large language models (GPT-4o, Claude Sonnet, Gemini 2.5 Flash, Llama-3.1, and Qwen-2.5) were evaluated across six datasets spanning Nastaliq Urdu, Roman Urdu, English, and code-switched Urdu-English, producing approximately 28,000 classifications. Label instability between original-script and English-translation classification ranged from 15.9{\%} to 31.6{\%}, and a missed-in-Urdu rate content flagged as harmful in English but passed as normal in Urdu ranged from 2.4{\%} to 9.9{\%}. Pairwise significance testing confirms that frontier closed models form a statistically indistinguishable cluster significantly outperforming smaller open-weight models on both metrics. The paper also identifies the absence of a dedicated Nastaliq Urdu with English code-switched hate speech dataset as a structural gap, and raises questions about annotation bias in existing Urdu corpora. The findings demonstrate that Urdu speakers are exposed to measurably weaker content moderation safety assurance, a gap that this paper documents, quantifies, and calls on the research community to address, considering that it is in the top 10 most spoken and used language in the world."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kara-isitt-etal-2026-unseen">
<titleInfo>
<title>Unseen Harm: Measuring Cross-Script Safety Inconsistency with “Missed-in-Urdu” Scores in LLM Hate Speech Detection.</title>
</titleInfo>
<name type="personal">
<namePart type="given">Fawzia</namePart>
<namePart type="family">Kara-Isitt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sonal</namePart>
<namePart type="family">Khosla</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stephen</namePart>
<namePart type="family">Swift</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 10th Workshop on Online Abuse and Harms (WOAH)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Agostina</namePart>
<namePart type="family">Calabrese</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="family">Davidson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christine</namePart>
<namePart type="family">de Kock</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Urja</namePart>
<namePart type="family">Khurana</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marta</namePart>
<namePart type="family">Marchiori Manerba</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paloma</namePart>
<namePart type="family">Piot</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zeerak</namePart>
<namePart type="family">Talat</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">979-8-89176-487-3</identifier>
</relatedItem>
<abstract>This paper investigates the near-total absence of Urdu from mainstream LLM safety evaluation and nine years of WOAH proceedings. A complete enumeration of all 205 papers across nine ALW/WOAH editions confirms zero dedicated Urdu papers across the entire period. To assess whether this absence has measurable consequences, five large language models (GPT-4o, Claude Sonnet, Gemini 2.5 Flash, Llama-3.1, and Qwen-2.5) were evaluated across six datasets spanning Nastaliq Urdu, Roman Urdu, English, and code-switched Urdu-English, producing approximately 28,000 classifications. Label instability between original-script and English-translation classification ranged from 15.9% to 31.6%, and a missed-in-Urdu rate content flagged as harmful in English but passed as normal in Urdu ranged from 2.4% to 9.9%. Pairwise significance testing confirms that frontier closed models form a statistically indistinguishable cluster significantly outperforming smaller open-weight models on both metrics. The paper also identifies the absence of a dedicated Nastaliq Urdu with English code-switched hate speech dataset as a structural gap, and raises questions about annotation bias in existing Urdu corpora. The findings demonstrate that Urdu speakers are exposed to measurably weaker content moderation safety assurance, a gap that this paper documents, quantifies, and calls on the research community to address, considering that it is in the top 10 most spoken and used language in the world.</abstract>
<identifier type="citekey">kara-isitt-etal-2026-unseen</identifier>
<location>
<url>https://aclanthology.org/2026.woah-1.6/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>49</start>
<end>62</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Unseen Harm: Measuring Cross-Script Safety Inconsistency with “Missed-in-Urdu” Scores in LLM Hate Speech Detection.
%A Kara-Isitt, Fawzia
%A Khosla, Sonal
%A Swift, Stephen
%Y Calabrese, Agostina
%Y Davidson, Thomas
%Y de Kock, Christine
%Y Khurana, Urja
%Y Marchiori Manerba, Marta
%Y Piot, Paloma
%Y Talat, Zeerak
%S Proceedings of the 10th Workshop on Online Abuse and Harms (WOAH)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-487-3
%F kara-isitt-etal-2026-unseen
%X This paper investigates the near-total absence of Urdu from mainstream LLM safety evaluation and nine years of WOAH proceedings. A complete enumeration of all 205 papers across nine ALW/WOAH editions confirms zero dedicated Urdu papers across the entire period. To assess whether this absence has measurable consequences, five large language models (GPT-4o, Claude Sonnet, Gemini 2.5 Flash, Llama-3.1, and Qwen-2.5) were evaluated across six datasets spanning Nastaliq Urdu, Roman Urdu, English, and code-switched Urdu-English, producing approximately 28,000 classifications. Label instability between original-script and English-translation classification ranged from 15.9% to 31.6%, and a missed-in-Urdu rate content flagged as harmful in English but passed as normal in Urdu ranged from 2.4% to 9.9%. Pairwise significance testing confirms that frontier closed models form a statistically indistinguishable cluster significantly outperforming smaller open-weight models on both metrics. The paper also identifies the absence of a dedicated Nastaliq Urdu with English code-switched hate speech dataset as a structural gap, and raises questions about annotation bias in existing Urdu corpora. The findings demonstrate that Urdu speakers are exposed to measurably weaker content moderation safety assurance, a gap that this paper documents, quantifies, and calls on the research community to address, considering that it is in the top 10 most spoken and used language in the world.
%U https://aclanthology.org/2026.woah-1.6/
%P 49-62
Markdown (Informal)
[Unseen Harm: Measuring Cross-Script Safety Inconsistency with "Missed-in-Urdu" Scores in LLM Hate Speech Detection.](https://aclanthology.org/2026.woah-1.6/) (Kara-Isitt et al., WOAH 2026)
ACL