@inproceedings{basak-he-2026-understanding,
title = "Understanding Subjectivity Transformation in Academic Meta-Review Generation",
author = "Basak, Sanju and
He, Suining",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1922/",
doi = "10.18653/2026.findings-emnlp.1922",
pages = "40975--40990",
ISBN = "979-8-89176-462-0",
abstract = "Academic meta-reviews must synthesize several reviewer opinions into a single, consistent judgment, balancing verifiable factual claims against subjective evaluations. Yet how human meta-reviewers transform reviewer subjectivity during summarization and whether large language models (LLMs) reproduce this behavior remains under-explored. To address this, we characterize subjectivity transformation in academic meta-review writing using a sentence-level subjectivity scorer, evaluating four open LLMs across six prompt configurations. Our analysis reveals that meta-review writing performs a regression to the mean of subjectivity: when a meta-review sentence is matched to its source reviewer{'}s sentence, the change in subjectivity negatively correlates with the source{'}s original subjectivity. In particular, actual, individual reviewer statements are made more interpretive, while highly subjective language tends to be moderated toward a balanced tone. All four open LLMs reproduce such a transformation closely. Conditioning the prompt on reviewer subjectivity yields only a small calibration gain where the generated meta-review{'}s overall subjectivity moves slightly closer to the human reference, yet leaves the transformation itself unchanged, implying that there exists a robust latent behavior of LLM summarization rather than a prompt-controllable one."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="basak-he-2026-understanding">
<titleInfo>
<title>Understanding Subjectivity Transformation in Academic Meta-Review Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sanju</namePart>
<namePart type="family">Basak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Suining</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Academic meta-reviews must synthesize several reviewer opinions into a single, consistent judgment, balancing verifiable factual claims against subjective evaluations. Yet how human meta-reviewers transform reviewer subjectivity during summarization and whether large language models (LLMs) reproduce this behavior remains under-explored. To address this, we characterize subjectivity transformation in academic meta-review writing using a sentence-level subjectivity scorer, evaluating four open LLMs across six prompt configurations. Our analysis reveals that meta-review writing performs a regression to the mean of subjectivity: when a meta-review sentence is matched to its source reviewer’s sentence, the change in subjectivity negatively correlates with the source’s original subjectivity. In particular, actual, individual reviewer statements are made more interpretive, while highly subjective language tends to be moderated toward a balanced tone. All four open LLMs reproduce such a transformation closely. Conditioning the prompt on reviewer subjectivity yields only a small calibration gain where the generated meta-review’s overall subjectivity moves slightly closer to the human reference, yet leaves the transformation itself unchanged, implying that there exists a robust latent behavior of LLM summarization rather than a prompt-controllable one.</abstract>
<identifier type="citekey">basak-he-2026-understanding</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1922</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1922/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>40975</start>
<end>40990</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Understanding Subjectivity Transformation in Academic Meta-Review Generation
%A Basak, Sanju
%A He, Suining
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F basak-he-2026-understanding
%X Academic meta-reviews must synthesize several reviewer opinions into a single, consistent judgment, balancing verifiable factual claims against subjective evaluations. Yet how human meta-reviewers transform reviewer subjectivity during summarization and whether large language models (LLMs) reproduce this behavior remains under-explored. To address this, we characterize subjectivity transformation in academic meta-review writing using a sentence-level subjectivity scorer, evaluating four open LLMs across six prompt configurations. Our analysis reveals that meta-review writing performs a regression to the mean of subjectivity: when a meta-review sentence is matched to its source reviewer’s sentence, the change in subjectivity negatively correlates with the source’s original subjectivity. In particular, actual, individual reviewer statements are made more interpretive, while highly subjective language tends to be moderated toward a balanced tone. All four open LLMs reproduce such a transformation closely. Conditioning the prompt on reviewer subjectivity yields only a small calibration gain where the generated meta-review’s overall subjectivity moves slightly closer to the human reference, yet leaves the transformation itself unchanged, implying that there exists a robust latent behavior of LLM summarization rather than a prompt-controllable one.
%R 10.18653/2026.findings-emnlp.1922
%U https://aclanthology.org/2026.findings-emnlp.1922/
%U https://doi.org/10.18653/2026.findings-emnlp.1922
%P 40975-40990
Markdown (Informal)
[Understanding Subjectivity Transformation in Academic Meta-Review Generation](https://aclanthology.org/2026.findings-emnlp.1922/) (Basak & He, Findings 2026)
ACL