@inproceedings{korre-etal-2023-harmful,
title = "Harmful Language Datasets: An Assessment of Robustness",
author = "Korre, Katerina and
Pavlopoulos, John and
Sorensen, Jeffrey and
Laugier, L{\'e}o and
Androutsopoulos, Ion and
Dixon, Lucas and
Barr{\'o}n-cede{\~n}o, Alberto",
editor = "Chung, Yi-ling and
R{\{}{\textbackslash}{''}ottger{\}}, Paul and
Nozza, Debora and
Talat, Zeerak and
Mostafazadeh Davani, Aida",
booktitle = "The 7th Workshop on Online Abuse and Harms (WOAH)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.woah-1.24/",
doi = "10.18653/v1/2023.woah-1.24",
pages = "221--230",
abstract = "The automated detection of harmful language has been of great importance for the online world, especially with the growing importance of social media and, consequently, polarisation. There are many open challenges to high quality detection of harmful text, from dataset creation to generalisable application, thus calling for more systematic studies. In this paper, we explore re-annotation as a means of examining the robustness of already existing labelled datasets, showing that, despite using alternative definitions, the inter-annotator agreement remains very inconsistent, highlighting the intrinsically subjective and variable nature of the task. In addition, we build automatic toxicity detectors using the existing datasets, with their original labels, and we evaluate them on our multi-definition and multi-source datasets. Surprisingly, while other studies show that hate speech detection models perform better on data that are derived from the same distribution as the training set, our analysis demonstrates this is not necessarily true."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="korre-etal-2023-harmful">
<titleInfo>
<title>Harmful Language Datasets: An Assessment of Robustness</title>
</titleInfo>
<name type="personal">
<namePart type="given">Katerina</namePart>
<namePart type="family">Korre</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">John</namePart>
<namePart type="family">Pavlopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jeffrey</namePart>
<namePart type="family">Sorensen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Léo</namePart>
<namePart type="family">Laugier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ion</namePart>
<namePart type="family">Androutsopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lucas</namePart>
<namePart type="family">Dixon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alberto</namePart>
<namePart type="family">Barrón-cedeño</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>The 7th Workshop on Online Abuse and Harms (WOAH)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yi-ling</namePart>
<namePart type="family">Chung</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paul</namePart>
<namePart type="family">R{\textbackslash”ottger}</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Debora</namePart>
<namePart type="family">Nozza</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>
<name type="personal">
<namePart type="given">Aida</namePart>
<namePart type="family">Mostafazadeh Davani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Toronto, Canada</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>The automated detection of harmful language has been of great importance for the online world, especially with the growing importance of social media and, consequently, polarisation. There are many open challenges to high quality detection of harmful text, from dataset creation to generalisable application, thus calling for more systematic studies. In this paper, we explore re-annotation as a means of examining the robustness of already existing labelled datasets, showing that, despite using alternative definitions, the inter-annotator agreement remains very inconsistent, highlighting the intrinsically subjective and variable nature of the task. In addition, we build automatic toxicity detectors using the existing datasets, with their original labels, and we evaluate them on our multi-definition and multi-source datasets. Surprisingly, while other studies show that hate speech detection models perform better on data that are derived from the same distribution as the training set, our analysis demonstrates this is not necessarily true.</abstract>
<identifier type="citekey">korre-etal-2023-harmful</identifier>
<identifier type="doi">10.18653/v1/2023.woah-1.24</identifier>
<location>
<url>https://aclanthology.org/2023.woah-1.24/</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>221</start>
<end>230</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Harmful Language Datasets: An Assessment of Robustness
%A Korre, Katerina
%A Pavlopoulos, John
%A Sorensen, Jeffrey
%A Laugier, Léo
%A Androutsopoulos, Ion
%A Dixon, Lucas
%A Barrón-cedeño, Alberto
%Y Chung, Yi-ling
%Y R{\textbackslash”ottger}, Paul
%Y Nozza, Debora
%Y Talat, Zeerak
%Y Mostafazadeh Davani, Aida
%S The 7th Workshop on Online Abuse and Harms (WOAH)
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F korre-etal-2023-harmful
%X The automated detection of harmful language has been of great importance for the online world, especially with the growing importance of social media and, consequently, polarisation. There are many open challenges to high quality detection of harmful text, from dataset creation to generalisable application, thus calling for more systematic studies. In this paper, we explore re-annotation as a means of examining the robustness of already existing labelled datasets, showing that, despite using alternative definitions, the inter-annotator agreement remains very inconsistent, highlighting the intrinsically subjective and variable nature of the task. In addition, we build automatic toxicity detectors using the existing datasets, with their original labels, and we evaluate them on our multi-definition and multi-source datasets. Surprisingly, while other studies show that hate speech detection models perform better on data that are derived from the same distribution as the training set, our analysis demonstrates this is not necessarily true.
%R 10.18653/v1/2023.woah-1.24
%U https://aclanthology.org/2023.woah-1.24/
%U https://doi.org/10.18653/v1/2023.woah-1.24
%P 221-230
Markdown (Informal)
[Harmful Language Datasets: An Assessment of Robustness](https://aclanthology.org/2023.woah-1.24/) (Korre et al., WOAH 2023)
ACL
- Katerina Korre, John Pavlopoulos, Jeffrey Sorensen, Léo Laugier, Ion Androutsopoulos, Lucas Dixon, and Alberto Barrón-cedeño. 2023. Harmful Language Datasets: An Assessment of Robustness. In The 7th Workshop on Online Abuse and Harms (WOAH), pages 221–230, Toronto, Canada. Association for Computational Linguistics.