@inproceedings{aldayel-etal-2024-covert,
title = "Covert Bias: The Severity of Social Views{'} Unalignment in Language Models Towards Implicit and Explicit Opinion",
author = "Aldayel, Abeer and
Alokaili, Areej and
Alahmadi, Rehab",
editor = "Dementieva, Daryna and
Ignat, Oana and
Jin, Zhijing and
Mihalcea, Rada and
Piatti, Giorgio and
Tetreault, Joel and
Wilson, Steven and
Zhao, Jieyu",
booktitle = "Proceedings of the Third Workshop on NLP for Positive Impact",
month = nov,
year = "2024",
address = "Miami, Florida, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.nlp4pi-1.6",
pages = "68--77",
abstract = "While various approaches have recently been studied for bias identification, little is known about how implicit language that does not explicitly convey a viewpoint affects bias amplification in large language models. To examine the severity of bias toward a view, we evaluated the performance of two downstream tasks where the implicit and explicit knowledge of social groups were used. First, we present a stress test evaluation by using a biased model in edge cases of excessive bias scenarios. Then, we evaluate how LLMs calibrate linguistically in response to both implicit and explicit opinions when they are aligned with conflicting viewpoints. Our findings reveal a discrepancy in LLM performance in identifying implicit and explicit opinions, with a general tendency of bias toward explicit opinions of opposing stances. Moreover, the bias-aligned models generate more cautious responses using uncertainty phrases compared to the unaligned (zero-shot) base models. The direct, incautious responses of the unaligned models suggest a need for further refinement of decisiveness by incorporating uncertainty markers to enhance their reliability, especially on socially nuanced topics with high subjectivity.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="aldayel-etal-2024-covert">
<titleInfo>
<title>Covert Bias: The Severity of Social Views’ Unalignment in Language Models Towards Implicit and Explicit Opinion</title>
</titleInfo>
<name type="personal">
<namePart type="given">Abeer</namePart>
<namePart type="family">Aldayel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Areej</namePart>
<namePart type="family">Alokaili</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rehab</namePart>
<namePart type="family">Alahmadi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Third Workshop on NLP for Positive Impact</title>
</titleInfo>
<name type="personal">
<namePart type="given">Daryna</namePart>
<namePart type="family">Dementieva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Oana</namePart>
<namePart type="family">Ignat</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhijing</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rada</namePart>
<namePart type="family">Mihalcea</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Giorgio</namePart>
<namePart type="family">Piatti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joel</namePart>
<namePart type="family">Tetreault</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Steven</namePart>
<namePart type="family">Wilson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jieyu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Miami, Florida, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>While various approaches have recently been studied for bias identification, little is known about how implicit language that does not explicitly convey a viewpoint affects bias amplification in large language models. To examine the severity of bias toward a view, we evaluated the performance of two downstream tasks where the implicit and explicit knowledge of social groups were used. First, we present a stress test evaluation by using a biased model in edge cases of excessive bias scenarios. Then, we evaluate how LLMs calibrate linguistically in response to both implicit and explicit opinions when they are aligned with conflicting viewpoints. Our findings reveal a discrepancy in LLM performance in identifying implicit and explicit opinions, with a general tendency of bias toward explicit opinions of opposing stances. Moreover, the bias-aligned models generate more cautious responses using uncertainty phrases compared to the unaligned (zero-shot) base models. The direct, incautious responses of the unaligned models suggest a need for further refinement of decisiveness by incorporating uncertainty markers to enhance their reliability, especially on socially nuanced topics with high subjectivity.</abstract>
<identifier type="citekey">aldayel-etal-2024-covert</identifier>
<location>
<url>https://aclanthology.org/2024.nlp4pi-1.6</url>
</location>
<part>
<date>2024-11</date>
<extent unit="page">
<start>68</start>
<end>77</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Covert Bias: The Severity of Social Views’ Unalignment in Language Models Towards Implicit and Explicit Opinion
%A Aldayel, Abeer
%A Alokaili, Areej
%A Alahmadi, Rehab
%Y Dementieva, Daryna
%Y Ignat, Oana
%Y Jin, Zhijing
%Y Mihalcea, Rada
%Y Piatti, Giorgio
%Y Tetreault, Joel
%Y Wilson, Steven
%Y Zhao, Jieyu
%S Proceedings of the Third Workshop on NLP for Positive Impact
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, Florida, USA
%F aldayel-etal-2024-covert
%X While various approaches have recently been studied for bias identification, little is known about how implicit language that does not explicitly convey a viewpoint affects bias amplification in large language models. To examine the severity of bias toward a view, we evaluated the performance of two downstream tasks where the implicit and explicit knowledge of social groups were used. First, we present a stress test evaluation by using a biased model in edge cases of excessive bias scenarios. Then, we evaluate how LLMs calibrate linguistically in response to both implicit and explicit opinions when they are aligned with conflicting viewpoints. Our findings reveal a discrepancy in LLM performance in identifying implicit and explicit opinions, with a general tendency of bias toward explicit opinions of opposing stances. Moreover, the bias-aligned models generate more cautious responses using uncertainty phrases compared to the unaligned (zero-shot) base models. The direct, incautious responses of the unaligned models suggest a need for further refinement of decisiveness by incorporating uncertainty markers to enhance their reliability, especially on socially nuanced topics with high subjectivity.
%U https://aclanthology.org/2024.nlp4pi-1.6
%P 68-77
Markdown (Informal)
[Covert Bias: The Severity of Social Views’ Unalignment in Language Models Towards Implicit and Explicit Opinion](https://aclanthology.org/2024.nlp4pi-1.6) (Aldayel et al., NLP4PI 2024)
ACL