@inproceedings{hasegawa-etal-2025-knowledge,
title = "Knowledge Editing Induces Underconfidence in Language Models",
author = "Hasegawa, Ryo and
Sakai, Yusuke and
Kamigaito, Hidetaka and
Watanabe, Taro",
editor = "Frermann, Lea and
Stevenson, Mark",
booktitle = "Proceedings of the 14th Joint Conference on Lexical and Computational Semantics (*SEM 2025)",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.starsem-1.27/",
pages = "338--347",
ISBN = "979-8-89176-340-1",
abstract = "As language models continue to scale, the demand for knowledge editing, a retraining-free knowledge update method, has increased. However, since knowledge editing directly alters token prediction probabilities acquired during pretraining, the probabilities may diverge from the empirical distribution. In this study, we analyze the impact of knowledge editing to compare the alignment between token prediction probabilities and task accuracy by calculating confidence calibration before and after knowledge editing. Our results reveal that, for tasks requiring semantic understanding, the range of increase in token prediction probabilities tends to be smaller than that of accuracy improvement, suggesting that knowledge editing methods lead to less confidence in prediction."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hasegawa-etal-2025-knowledge">
<titleInfo>
<title>Knowledge Editing Induces Underconfidence in Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ryo</namePart>
<namePart type="family">Hasegawa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yusuke</namePart>
<namePart type="family">Sakai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hidetaka</namePart>
<namePart type="family">Kamigaito</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Taro</namePart>
<namePart type="family">Watanabe</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 14th Joint Conference on Lexical and Computational Semantics (*SEM 2025)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lea</namePart>
<namePart type="family">Frermann</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mark</namePart>
<namePart type="family">Stevenson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-340-1</identifier>
</relatedItem>
<abstract>As language models continue to scale, the demand for knowledge editing, a retraining-free knowledge update method, has increased. However, since knowledge editing directly alters token prediction probabilities acquired during pretraining, the probabilities may diverge from the empirical distribution. In this study, we analyze the impact of knowledge editing to compare the alignment between token prediction probabilities and task accuracy by calculating confidence calibration before and after knowledge editing. Our results reveal that, for tasks requiring semantic understanding, the range of increase in token prediction probabilities tends to be smaller than that of accuracy improvement, suggesting that knowledge editing methods lead to less confidence in prediction.</abstract>
<identifier type="citekey">hasegawa-etal-2025-knowledge</identifier>
<location>
<url>https://aclanthology.org/2025.starsem-1.27/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>338</start>
<end>347</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Knowledge Editing Induces Underconfidence in Language Models
%A Hasegawa, Ryo
%A Sakai, Yusuke
%A Kamigaito, Hidetaka
%A Watanabe, Taro
%Y Frermann, Lea
%Y Stevenson, Mark
%S Proceedings of the 14th Joint Conference on Lexical and Computational Semantics (*SEM 2025)
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-340-1
%F hasegawa-etal-2025-knowledge
%X As language models continue to scale, the demand for knowledge editing, a retraining-free knowledge update method, has increased. However, since knowledge editing directly alters token prediction probabilities acquired during pretraining, the probabilities may diverge from the empirical distribution. In this study, we analyze the impact of knowledge editing to compare the alignment between token prediction probabilities and task accuracy by calculating confidence calibration before and after knowledge editing. Our results reveal that, for tasks requiring semantic understanding, the range of increase in token prediction probabilities tends to be smaller than that of accuracy improvement, suggesting that knowledge editing methods lead to less confidence in prediction.
%U https://aclanthology.org/2025.starsem-1.27/
%P 338-347
Markdown (Informal)
[Knowledge Editing Induces Underconfidence in Language Models](https://aclanthology.org/2025.starsem-1.27/) (Hasegawa et al., *SEM 2025)
ACL
- Ryo Hasegawa, Yusuke Sakai, Hidetaka Kamigaito, and Taro Watanabe. 2025. Knowledge Editing Induces Underconfidence in Language Models. In Proceedings of the 14th Joint Conference on Lexical and Computational Semantics (*SEM 2025), pages 338–347, Suzhou, China. Association for Computational Linguistics.