@inproceedings{nnamdi-etal-2026-epistemic,
title = "Epistemic Erosion: An Error Class Beyond Hallucination in {LLM} Text Generation",
author = "Nnamdi, Micky C. and
Wang, Shaunna and
Nnamdi, Chinaza and
Li, Jieru and
Marteau, Benoit Louis and
Tamo, J. Ben and
Yuan, Yining and
Wang, May Dongmei",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.223/",
pages = "4998--5026",
ISBN = "979-8-89176-439-2",
abstract = "Expert writers calibrate their claims with qualifiers such as ``likely benign{``} in a radiology report, ``may constitute{``} in a court ruling, and ``suggest{``} in a scientific paper. We ask whether this calibration survives when Large Language Models (LLMs) summarize and rewrite professional text. We define epistemic erosion as any change in a source{'}s expressed certainty, including qualified claims that are dropped or introduced without basis. We measure it with the Epistemic Faithfulness Score (EFS), computed by aligning source and output claims one-to-one with rule-based matching and assigning each distortion to one of six types, namely escalation, attenuation, omission, fabrication, scope shift, and severity warp. On clinical notes, mean EFS for 11 open-weight LLMs ranges from 0.415 to 0.493; three of these models are also evaluated on legal and scientific text. Four hosted models (GPT-4o, GPT-5.4, Phi-4, Mistral-Large-3) reach at most 0.483 in any domain. The most frequent detected distortion is omission (66{\%}), in which a qualified claim is dropped entirely; claims that survive tend to shift toward greater certainty. Model scale and medical adaptation bring no consistent gain, seven prompting strategies bring at most modest gains, and ROUGE-L and BERTScore correlate only weakly with EFS ($|r| \approx 0.1$)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="nnamdi-etal-2026-epistemic">
<titleInfo>
<title>Epistemic Erosion: An Error Class Beyond Hallucination in LLM Text Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Micky</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Nnamdi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaunna</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chinaza</namePart>
<namePart type="family">Nnamdi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jieru</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Benoit</namePart>
<namePart type="given">Louis</namePart>
<namePart type="family">Marteau</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">J</namePart>
<namePart type="given">Ben</namePart>
<namePart type="family">Tamo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yining</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">May</namePart>
<namePart type="given">Dongmei</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>Expert writers calibrate their claims with qualifiers such as “likely benign“ in a radiology report, “may constitute“ in a court ruling, and “suggest“ in a scientific paper. We ask whether this calibration survives when Large Language Models (LLMs) summarize and rewrite professional text. We define epistemic erosion as any change in a source’s expressed certainty, including qualified claims that are dropped or introduced without basis. We measure it with the Epistemic Faithfulness Score (EFS), computed by aligning source and output claims one-to-one with rule-based matching and assigning each distortion to one of six types, namely escalation, attenuation, omission, fabrication, scope shift, and severity warp. On clinical notes, mean EFS for 11 open-weight LLMs ranges from 0.415 to 0.493; three of these models are also evaluated on legal and scientific text. Four hosted models (GPT-4o, GPT-5.4, Phi-4, Mistral-Large-3) reach at most 0.483 in any domain. The most frequent detected distortion is omission (66%), in which a qualified claim is dropped entirely; claims that survive tend to shift toward greater certainty. Model scale and medical adaptation bring no consistent gain, seven prompting strategies bring at most modest gains, and ROUGE-L and BERTScore correlate only weakly with EFS (|r| \approx 0.1).</abstract>
<identifier type="citekey">nnamdi-etal-2026-epistemic</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.223/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>4998</start>
<end>5026</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Epistemic Erosion: An Error Class Beyond Hallucination in LLM Text Generation
%A Nnamdi, Micky C.
%A Wang, Shaunna
%A Nnamdi, Chinaza
%A Li, Jieru
%A Marteau, Benoit Louis
%A Tamo, J. Ben
%A Yuan, Yining
%A Wang, May Dongmei
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F nnamdi-etal-2026-epistemic
%X Expert writers calibrate their claims with qualifiers such as “likely benign“ in a radiology report, “may constitute“ in a court ruling, and “suggest“ in a scientific paper. We ask whether this calibration survives when Large Language Models (LLMs) summarize and rewrite professional text. We define epistemic erosion as any change in a source’s expressed certainty, including qualified claims that are dropped or introduced without basis. We measure it with the Epistemic Faithfulness Score (EFS), computed by aligning source and output claims one-to-one with rule-based matching and assigning each distortion to one of six types, namely escalation, attenuation, omission, fabrication, scope shift, and severity warp. On clinical notes, mean EFS for 11 open-weight LLMs ranges from 0.415 to 0.493; three of these models are also evaluated on legal and scientific text. Four hosted models (GPT-4o, GPT-5.4, Phi-4, Mistral-Large-3) reach at most 0.483 in any domain. The most frequent detected distortion is omission (66%), in which a qualified claim is dropped entirely; claims that survive tend to shift toward greater certainty. Model scale and medical adaptation bring no consistent gain, seven prompting strategies bring at most modest gains, and ROUGE-L and BERTScore correlate only weakly with EFS (|r| \approx 0.1).
%U https://aclanthology.org/2026.aacl-long.223/
%P 4998-5026
Markdown (Informal)
[Epistemic Erosion: An Error Class Beyond Hallucination in LLM Text Generation](https://aclanthology.org/2026.aacl-long.223/) (Nnamdi et al., AACL-IJCNLP 2026)
ACL
- Micky C. Nnamdi, Shaunna Wang, Chinaza Nnamdi, Jieru Li, Benoit Louis Marteau, J. Ben Tamo, Yining Yuan, and May Dongmei Wang. 2026. Epistemic Erosion: An Error Class Beyond Hallucination in LLM Text Generation. In Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4998–5026, Hengqin, China. Association for Computational Linguistics.