@inproceedings{kang-etal-2025-format,
title = "When Format Changes Meaning: Investigating Semantic Inconsistency of Large Language Models",
author = "Kang, Cheongwoong and
Baek, Jongeun and
Kim, Yeonjea and
Choi, Jaesik",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-emnlp.143/",
doi = "10.18653/v1/2025.findings-emnlp.143",
pages = "2647--2667",
ISBN = "979-8-89176-335-7",
abstract = "Large language models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks. However, they remain vulnerable to semantic inconsistency, where minor formatting variations result in divergent predictions for semantically equivalent inputs. Our comprehensive evaluation reveals that this brittleness persists even in state-of-the-art models such as GPT-4o, posing a serious challenge to their reliability. Through a mechanistic analysis, we find that semantic-equivalent input changes induce instability in internal representations, ultimately leading to divergent predictions. This reflects a deeper structural issue, where form and meaning are intertwined in the embedding space. We further demonstrate that existing mitigation strategies, including direct fine-tuning on format variations, do not fully address semantic inconsistency, underscoring the difficulty of the problem. Our findings highlight the need for deeper mechanistic understanding to develop targeted methods that improve robustness."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kang-etal-2025-format">
<titleInfo>
<title>When Format Changes Meaning: Investigating Semantic Inconsistency of Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Cheongwoong</namePart>
<namePart type="family">Kang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jongeun</namePart>
<namePart type="family">Baek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yeonjea</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaesik</namePart>
<namePart type="family">Choi</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>Findings of the Association for Computational Linguistics: EMNLP 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</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">979-8-89176-335-7</identifier>
</relatedItem>
<abstract>Large language models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks. However, they remain vulnerable to semantic inconsistency, where minor formatting variations result in divergent predictions for semantically equivalent inputs. Our comprehensive evaluation reveals that this brittleness persists even in state-of-the-art models such as GPT-4o, posing a serious challenge to their reliability. Through a mechanistic analysis, we find that semantic-equivalent input changes induce instability in internal representations, ultimately leading to divergent predictions. This reflects a deeper structural issue, where form and meaning are intertwined in the embedding space. We further demonstrate that existing mitigation strategies, including direct fine-tuning on format variations, do not fully address semantic inconsistency, underscoring the difficulty of the problem. Our findings highlight the need for deeper mechanistic understanding to develop targeted methods that improve robustness.</abstract>
<identifier type="citekey">kang-etal-2025-format</identifier>
<identifier type="doi">10.18653/v1/2025.findings-emnlp.143</identifier>
<location>
<url>https://aclanthology.org/2025.findings-emnlp.143/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>2647</start>
<end>2667</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T When Format Changes Meaning: Investigating Semantic Inconsistency of Large Language Models
%A Kang, Cheongwoong
%A Baek, Jongeun
%A Kim, Yeonjea
%A Choi, Jaesik
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Findings of the Association for Computational Linguistics: EMNLP 2025
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-335-7
%F kang-etal-2025-format
%X Large language models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks. However, they remain vulnerable to semantic inconsistency, where minor formatting variations result in divergent predictions for semantically equivalent inputs. Our comprehensive evaluation reveals that this brittleness persists even in state-of-the-art models such as GPT-4o, posing a serious challenge to their reliability. Through a mechanistic analysis, we find that semantic-equivalent input changes induce instability in internal representations, ultimately leading to divergent predictions. This reflects a deeper structural issue, where form and meaning are intertwined in the embedding space. We further demonstrate that existing mitigation strategies, including direct fine-tuning on format variations, do not fully address semantic inconsistency, underscoring the difficulty of the problem. Our findings highlight the need for deeper mechanistic understanding to develop targeted methods that improve robustness.
%R 10.18653/v1/2025.findings-emnlp.143
%U https://aclanthology.org/2025.findings-emnlp.143/
%U https://doi.org/10.18653/v1/2025.findings-emnlp.143
%P 2647-2667
Markdown (Informal)
[When Format Changes Meaning: Investigating Semantic Inconsistency of Large Language Models](https://aclanthology.org/2025.findings-emnlp.143/) (Kang et al., Findings 2025)
ACL