@inproceedings{gottapu-2026-evaluating,
title = "Evaluating Automatically Inferred Linguistic Perspective as an Auxiliary Signal for Toxicity Detection",
author = "Gottapu, Meghana",
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-1.20/",
pages = "228--233",
ISBN = "979-8-89176-487-3",
abstract = "Identity-related language occurs in both self-referential discussion and speech aboiut other groups, complicating toxicity detection. We ask whether an automatically inferred linguistic-perspective signal is useful as an auxiliary traiing objective. A regex and zero-shot NLI pipeline lables comments as SELF or OTHER; these are noisy linguistic proxies, not verified speak indetities. On Jigsaw Unintended Bias, we fine-tune a Toxic-BERT model with toxicity and proxy labels heads and compare it with a matched control whose perspective-loss weight is zero. The auxiliary objective gives a small descriptive mean ROC-AUC increase but lower mean accuracy. On decontaminated CivilComments-WILDS, ROC-AUC is 0.9346 while worst group accuracy is 0.5805. The main contribution is a leakage-controlled, matced evaluation showing that the full fine-tuning recipe improves toxic class F1 and ranking substantially, where as the automatic perspective proxy provides only limited additional benefit."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gottapu-2026-evaluating">
<titleInfo>
<title>Evaluating Automatically Inferred Linguistic Perspective as an Auxiliary Signal for Toxicity Detection</title>
</titleInfo>
<name type="personal">
<namePart type="given">Meghana</namePart>
<namePart type="family">Gottapu</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>Identity-related language occurs in both self-referential discussion and speech aboiut other groups, complicating toxicity detection. We ask whether an automatically inferred linguistic-perspective signal is useful as an auxiliary traiing objective. A regex and zero-shot NLI pipeline lables comments as SELF or OTHER; these are noisy linguistic proxies, not verified speak indetities. On Jigsaw Unintended Bias, we fine-tune a Toxic-BERT model with toxicity and proxy labels heads and compare it with a matched control whose perspective-loss weight is zero. The auxiliary objective gives a small descriptive mean ROC-AUC increase but lower mean accuracy. On decontaminated CivilComments-WILDS, ROC-AUC is 0.9346 while worst group accuracy is 0.5805. The main contribution is a leakage-controlled, matced evaluation showing that the full fine-tuning recipe improves toxic class F1 and ranking substantially, where as the automatic perspective proxy provides only limited additional benefit.</abstract>
<identifier type="citekey">gottapu-2026-evaluating</identifier>
<location>
<url>https://aclanthology.org/2026.woah-1.20/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>228</start>
<end>233</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Evaluating Automatically Inferred Linguistic Perspective as an Auxiliary Signal for Toxicity Detection
%A Gottapu, Meghana
%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 gottapu-2026-evaluating
%X Identity-related language occurs in both self-referential discussion and speech aboiut other groups, complicating toxicity detection. We ask whether an automatically inferred linguistic-perspective signal is useful as an auxiliary traiing objective. A regex and zero-shot NLI pipeline lables comments as SELF or OTHER; these are noisy linguistic proxies, not verified speak indetities. On Jigsaw Unintended Bias, we fine-tune a Toxic-BERT model with toxicity and proxy labels heads and compare it with a matched control whose perspective-loss weight is zero. The auxiliary objective gives a small descriptive mean ROC-AUC increase but lower mean accuracy. On decontaminated CivilComments-WILDS, ROC-AUC is 0.9346 while worst group accuracy is 0.5805. The main contribution is a leakage-controlled, matced evaluation showing that the full fine-tuning recipe improves toxic class F1 and ranking substantially, where as the automatic perspective proxy provides only limited additional benefit.
%U https://aclanthology.org/2026.woah-1.20/
%P 228-233
Markdown (Informal)
[Evaluating Automatically Inferred Linguistic Perspective as an Auxiliary Signal for Toxicity Detection](https://aclanthology.org/2026.woah-1.20/) (Gottapu, WOAH 2026)
ACL