@inproceedings{hirave-oet-2026-dogs,
title = "Who Let the Dogs Out? Auditing Dogwhistle Coverage in Hate Speech Benchmarks Across Target Groups",
author = "Hirave, Reva and
Oet, Ryan",
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-1.34/",
pages = "392--410",
ISBN = "979-8-89176-487-3",
abstract = "Automated hate speech detection relies on benchmark datasets for both training and evaluation, making benchmark composition a fairness concern that precedes classifier behavior. We audit dogwhistle coverage in two widely-used benchmarks (HateXplain and Measuring Hate Speech) using an existing glossary of 340 dogwhistle entries mapped onto a four-level coding sophistication taxonomy, ranging from explicit slurs (L1) to opaque persona signals (L4). Across 59.6k posts in the benchmark union, annotation failure is the dominant failure mode: at the stereotype-based level (L2), every racial and religious group with stable match counts is labeled non-hateful at rates exceeding 60{\%}. Collection failure, where entire dogwhistle categories are absent from the union, becomes co-dominant at higher coding levels, with twenty group x level cells showing categorical gaps that re-annotation cannot fix. These failures are unevenly distributed: race:black vs. race:middle{\_}eastern fails the 4/5 rule on annotation DI at L2 (DI=0.661), and lgbtq:Trans/NB posts one of the highest annotator failure rates (81.2{\%}, second only to origin:undocumented) and the lowest presence rate of any comparably-sized group at every level. We distinguish collection failure from annotation failure as mechanisms with different remediation implications, and release our mapped glossary and pipeline publicly."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hirave-oet-2026-dogs">
<titleInfo>
<title>Who Let the Dogs Out? Auditing Dogwhistle Coverage in Hate Speech Benchmarks Across Target Groups</title>
</titleInfo>
<name type="personal">
<namePart type="given">Reva</namePart>
<namePart type="family">Hirave</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ryan</namePart>
<namePart type="family">Oet</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>Automated hate speech detection relies on benchmark datasets for both training and evaluation, making benchmark composition a fairness concern that precedes classifier behavior. We audit dogwhistle coverage in two widely-used benchmarks (HateXplain and Measuring Hate Speech) using an existing glossary of 340 dogwhistle entries mapped onto a four-level coding sophistication taxonomy, ranging from explicit slurs (L1) to opaque persona signals (L4). Across 59.6k posts in the benchmark union, annotation failure is the dominant failure mode: at the stereotype-based level (L2), every racial and religious group with stable match counts is labeled non-hateful at rates exceeding 60%. Collection failure, where entire dogwhistle categories are absent from the union, becomes co-dominant at higher coding levels, with twenty group x level cells showing categorical gaps that re-annotation cannot fix. These failures are unevenly distributed: race:black vs. race:middle_eastern fails the 4/5 rule on annotation DI at L2 (DI=0.661), and lgbtq:Trans/NB posts one of the highest annotator failure rates (81.2%, second only to origin:undocumented) and the lowest presence rate of any comparably-sized group at every level. We distinguish collection failure from annotation failure as mechanisms with different remediation implications, and release our mapped glossary and pipeline publicly.</abstract>
<identifier type="citekey">hirave-oet-2026-dogs</identifier>
<location>
<url>https://aclanthology.org/2026.woah-1.34/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>392</start>
<end>410</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Who Let the Dogs Out? Auditing Dogwhistle Coverage in Hate Speech Benchmarks Across Target Groups
%A Hirave, Reva
%A Oet, Ryan
%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 hirave-oet-2026-dogs
%X Automated hate speech detection relies on benchmark datasets for both training and evaluation, making benchmark composition a fairness concern that precedes classifier behavior. We audit dogwhistle coverage in two widely-used benchmarks (HateXplain and Measuring Hate Speech) using an existing glossary of 340 dogwhistle entries mapped onto a four-level coding sophistication taxonomy, ranging from explicit slurs (L1) to opaque persona signals (L4). Across 59.6k posts in the benchmark union, annotation failure is the dominant failure mode: at the stereotype-based level (L2), every racial and religious group with stable match counts is labeled non-hateful at rates exceeding 60%. Collection failure, where entire dogwhistle categories are absent from the union, becomes co-dominant at higher coding levels, with twenty group x level cells showing categorical gaps that re-annotation cannot fix. These failures are unevenly distributed: race:black vs. race:middle_eastern fails the 4/5 rule on annotation DI at L2 (DI=0.661), and lgbtq:Trans/NB posts one of the highest annotator failure rates (81.2%, second only to origin:undocumented) and the lowest presence rate of any comparably-sized group at every level. We distinguish collection failure from annotation failure as mechanisms with different remediation implications, and release our mapped glossary and pipeline publicly.
%U https://aclanthology.org/2026.woah-1.34/
%P 392-410
Markdown (Informal)
[Who Let the Dogs Out? Auditing Dogwhistle Coverage in Hate Speech Benchmarks Across Target Groups](https://aclanthology.org/2026.woah-1.34/) (Hirave & Oet, WOAH 2026)
ACL