@inproceedings{kwako-etal-2023-bert,
title = "Does {BERT} Exacerbate Gender or {L}1 Biases in Automated {E}nglish Speaking Assessment?",
author = "Kwako, Alexander and
Wan, Yixin and
Zhao, Jieyu and
Hansen, Mark and
Chang, Kai-Wei and
Cai, Li",
editor = {Kochmar, Ekaterina and
Burstein, Jill and
Horbach, Andrea and
Laarmann-Quante, Ronja and
Madnani, Nitin and
Tack, Ana{\"\i}s and
Yaneva, Victoria and
Yuan, Zheng and
Zesch, Torsten},
booktitle = "Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.bea-1.54",
doi = "10.18653/v1/2023.bea-1.54",
pages = "668--681",
abstract = "In English speaking assessment, pretrained large language models (LLMs) such as BERT can score constructed response items as accurately as human raters. Less research has investigated whether LLMs perpetuate or exacerbate biases, which would pose problems for the fairness and validity of the test. This study examines gender and native language (L1) biases in human and automated scores, using an off-the-shelf (OOS) BERT model. Analyses focus on a specific type of bias known as differential item functioning (DIF), which compares examinees of similar English language proficiency. Results show that there is a moderate amount of DIF, based on examinees{'} L1 background in grade band 912. DIF is higher when scored by an OOS BERT model, indicating that BERT may exacerbate this bias; however, in practical terms, the degree to which BERT exacerbates DIF is very small. Additionally, there is more DIF for longer speaking items and for older examinees, but BERT does not exacerbate these patterns of DIF.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kwako-etal-2023-bert">
<titleInfo>
<title>Does BERT Exacerbate Gender or L1 Biases in Automated English Speaking Assessment?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Alexander</namePart>
<namePart type="family">Kwako</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yixin</namePart>
<namePart type="family">Wan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jieyu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mark</namePart>
<namePart type="family">Hansen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kai-Wei</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Li</namePart>
<namePart type="family">Cai</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>Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Kochmar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jill</namePart>
<namePart type="family">Burstein</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andrea</namePart>
<namePart type="family">Horbach</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ronja</namePart>
<namePart type="family">Laarmann-Quante</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nitin</namePart>
<namePart type="family">Madnani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anaïs</namePart>
<namePart type="family">Tack</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Victoria</namePart>
<namePart type="family">Yaneva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zheng</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Torsten</namePart>
<namePart type="family">Zesch</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>In English speaking assessment, pretrained large language models (LLMs) such as BERT can score constructed response items as accurately as human raters. Less research has investigated whether LLMs perpetuate or exacerbate biases, which would pose problems for the fairness and validity of the test. This study examines gender and native language (L1) biases in human and automated scores, using an off-the-shelf (OOS) BERT model. Analyses focus on a specific type of bias known as differential item functioning (DIF), which compares examinees of similar English language proficiency. Results show that there is a moderate amount of DIF, based on examinees’ L1 background in grade band 912. DIF is higher when scored by an OOS BERT model, indicating that BERT may exacerbate this bias; however, in practical terms, the degree to which BERT exacerbates DIF is very small. Additionally, there is more DIF for longer speaking items and for older examinees, but BERT does not exacerbate these patterns of DIF.</abstract>
<identifier type="citekey">kwako-etal-2023-bert</identifier>
<identifier type="doi">10.18653/v1/2023.bea-1.54</identifier>
<location>
<url>https://aclanthology.org/2023.bea-1.54</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>668</start>
<end>681</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Does BERT Exacerbate Gender or L1 Biases in Automated English Speaking Assessment?
%A Kwako, Alexander
%A Wan, Yixin
%A Zhao, Jieyu
%A Hansen, Mark
%A Chang, Kai-Wei
%A Cai, Li
%Y Kochmar, Ekaterina
%Y Burstein, Jill
%Y Horbach, Andrea
%Y Laarmann-Quante, Ronja
%Y Madnani, Nitin
%Y Tack, Anaïs
%Y Yaneva, Victoria
%Y Yuan, Zheng
%Y Zesch, Torsten
%S Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023)
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F kwako-etal-2023-bert
%X In English speaking assessment, pretrained large language models (LLMs) such as BERT can score constructed response items as accurately as human raters. Less research has investigated whether LLMs perpetuate or exacerbate biases, which would pose problems for the fairness and validity of the test. This study examines gender and native language (L1) biases in human and automated scores, using an off-the-shelf (OOS) BERT model. Analyses focus on a specific type of bias known as differential item functioning (DIF), which compares examinees of similar English language proficiency. Results show that there is a moderate amount of DIF, based on examinees’ L1 background in grade band 912. DIF is higher when scored by an OOS BERT model, indicating that BERT may exacerbate this bias; however, in practical terms, the degree to which BERT exacerbates DIF is very small. Additionally, there is more DIF for longer speaking items and for older examinees, but BERT does not exacerbate these patterns of DIF.
%R 10.18653/v1/2023.bea-1.54
%U https://aclanthology.org/2023.bea-1.54
%U https://doi.org/10.18653/v1/2023.bea-1.54
%P 668-681
Markdown (Informal)
[Does BERT Exacerbate Gender or L1 Biases in Automated English Speaking Assessment?](https://aclanthology.org/2023.bea-1.54) (Kwako et al., BEA 2023)
ACL