Intersectional Bias in Japanese Large Language Models from a Contextualized Perspective

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Yanaka, Hitomi, He, Xinqi, Lu, Jie, Han, Namgi, Oh, Sunjin, Kumon, Ryoma, Matsuoka, Yuma, Watabe, Katsuhiko, Itatsu, Yuko
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912503714283520
author Yanaka, Hitomi
He, Xinqi
Lu, Jie
Han, Namgi
Oh, Sunjin
Kumon, Ryoma
Matsuoka, Yuma
Watabe, Katsuhiko
Itatsu, Yuko
author_facet Yanaka, Hitomi
He, Xinqi
Lu, Jie
Han, Namgi
Oh, Sunjin
Kumon, Ryoma
Matsuoka, Yuma
Watabe, Katsuhiko
Itatsu, Yuko
contents An increasing number of studies have examined the social bias of rapidly developed large language models (LLMs). Although most of these studies have focused on bias occurring in a single social attribute, research in social science has shown that social bias often occurs in the form of intersectionality -- the constitutive and contextualized perspective on bias aroused by social attributes. In this study, we construct the Japanese benchmark inter-JBBQ, designed to evaluate the intersectional bias in LLMs on the question-answering setting. Using inter-JBBQ to analyze GPT-4o and Swallow, we find that biased output varies according to its contexts even with the equal combination of social attributes.
format Preprint
id arxiv_https___arxiv_org_abs_2506_12327
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Intersectional Bias in Japanese Large Language Models from a Contextualized Perspective
Yanaka, Hitomi
He, Xinqi
Lu, Jie
Han, Namgi
Oh, Sunjin
Kumon, Ryoma
Matsuoka, Yuma
Watabe, Katsuhiko
Itatsu, Yuko
Computation and Language
Artificial Intelligence
An increasing number of studies have examined the social bias of rapidly developed large language models (LLMs). Although most of these studies have focused on bias occurring in a single social attribute, research in social science has shown that social bias often occurs in the form of intersectionality -- the constitutive and contextualized perspective on bias aroused by social attributes. In this study, we construct the Japanese benchmark inter-JBBQ, designed to evaluate the intersectional bias in LLMs on the question-answering setting. Using inter-JBBQ to analyze GPT-4o and Swallow, we find that biased output varies according to its contexts even with the equal combination of social attributes.
title Intersectional Bias in Japanese Large Language Models from a Contextualized Perspective
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.12327