@inproceedings{ohi-etal-2024-likelihood,
title = "Likelihood-based Mitigation of Evaluation Bias in Large Language Models",
author = "Ohi, Masanari and
Kaneko, Masahiro and
Koike, Ryuto and
Loem, Mengsay and
Okazaki, Naoaki",
editor = "Ku, Lun-Wei and
Martins, Andre and
Srikumar, Vivek",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2024",
month = aug,
year = "2024",
address = "Bangkok, Thailand",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.findings-acl.193/",
doi = "10.18653/v1/2024.findings-acl.193",
pages = "3237--3245",
abstract = "Large Language Models (LLMs) are widely used to evaluate natural language generation tasks as automated metrics.However, the likelihood, a measure of LLM`s plausibility for a sentence, can vary due to superficial differences in sentences, such as word order and sentence structure.It is therefore possible that there might be a likelihood bias if LLMs are used for evaluation: they might overrate sentences with higher likelihoods while underrating those with lower likelihoods.In this paper, we investigate the presence and impact of likelihood bias in LLM-based evaluators.We also propose a method to mitigate the likelihood bias.Our method utilizes highly biased instances as few-shot examples for in-context learning.Our experiments in evaluating the data-to-text and grammatical error correction tasks reveal that several LLMs we test display a likelihood bias.Furthermore, our proposed method successfully mitigates this bias, also improving evaluation performance (in terms of correlation of models with human scores) significantly."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ohi-etal-2024-likelihood">
<titleInfo>
<title>Likelihood-based Mitigation of Evaluation Bias in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Masanari</namePart>
<namePart type="family">Ohi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Masahiro</namePart>
<namePart type="family">Kaneko</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ryuto</namePart>
<namePart type="family">Koike</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mengsay</namePart>
<namePart type="family">Loem</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naoaki</namePart>
<namePart type="family">Okazaki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-08</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2024</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lun-Wei</namePart>
<namePart type="family">Ku</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andre</namePart>
<namePart type="family">Martins</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vivek</namePart>
<namePart type="family">Srikumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Bangkok, Thailand</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Large Language Models (LLMs) are widely used to evaluate natural language generation tasks as automated metrics.However, the likelihood, a measure of LLM‘s plausibility for a sentence, can vary due to superficial differences in sentences, such as word order and sentence structure.It is therefore possible that there might be a likelihood bias if LLMs are used for evaluation: they might overrate sentences with higher likelihoods while underrating those with lower likelihoods.In this paper, we investigate the presence and impact of likelihood bias in LLM-based evaluators.We also propose a method to mitigate the likelihood bias.Our method utilizes highly biased instances as few-shot examples for in-context learning.Our experiments in evaluating the data-to-text and grammatical error correction tasks reveal that several LLMs we test display a likelihood bias.Furthermore, our proposed method successfully mitigates this bias, also improving evaluation performance (in terms of correlation of models with human scores) significantly.</abstract>
<identifier type="citekey">ohi-etal-2024-likelihood</identifier>
<identifier type="doi">10.18653/v1/2024.findings-acl.193</identifier>
<location>
<url>https://aclanthology.org/2024.findings-acl.193/</url>
</location>
<part>
<date>2024-08</date>
<extent unit="page">
<start>3237</start>
<end>3245</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Likelihood-based Mitigation of Evaluation Bias in Large Language Models
%A Ohi, Masanari
%A Kaneko, Masahiro
%A Koike, Ryuto
%A Loem, Mengsay
%A Okazaki, Naoaki
%Y Ku, Lun-Wei
%Y Martins, Andre
%Y Srikumar, Vivek
%S Findings of the Association for Computational Linguistics: ACL 2024
%D 2024
%8 August
%I Association for Computational Linguistics
%C Bangkok, Thailand
%F ohi-etal-2024-likelihood
%X Large Language Models (LLMs) are widely used to evaluate natural language generation tasks as automated metrics.However, the likelihood, a measure of LLM‘s plausibility for a sentence, can vary due to superficial differences in sentences, such as word order and sentence structure.It is therefore possible that there might be a likelihood bias if LLMs are used for evaluation: they might overrate sentences with higher likelihoods while underrating those with lower likelihoods.In this paper, we investigate the presence and impact of likelihood bias in LLM-based evaluators.We also propose a method to mitigate the likelihood bias.Our method utilizes highly biased instances as few-shot examples for in-context learning.Our experiments in evaluating the data-to-text and grammatical error correction tasks reveal that several LLMs we test display a likelihood bias.Furthermore, our proposed method successfully mitigates this bias, also improving evaluation performance (in terms of correlation of models with human scores) significantly.
%R 10.18653/v1/2024.findings-acl.193
%U https://aclanthology.org/2024.findings-acl.193/
%U https://doi.org/10.18653/v1/2024.findings-acl.193
%P 3237-3245
Markdown (Informal)
[Likelihood-based Mitigation of Evaluation Bias in Large Language Models](https://aclanthology.org/2024.findings-acl.193/) (Ohi et al., Findings 2024)
ACL