@inproceedings{kwon-etal-2026-computational,
title = "Computational surprisal is not enough: Discourse prominence modulates human eye-movement patterns but not {GPT} expectations in {K}orean",
author = "Kwon, Nayoung and
Shin, Gyu-Ho and
Mun, Seongmin",
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.1443/",
doi = "10.18653/2026.findings-emnlp.1443",
pages = "30783--30796",
ISBN = "979-8-89176-462-0",
abstract = "Large Language Model (LLM) surprisal is increasingly used as a proxy for human sentence processing, yet whether model expectations reflect the cognitive factors governing real-time human behavior remains uncertain. We examine Korean subject{--}verb honorific agreement in subject- and object-control constructions, where human eye-tracking data reveal a prominence-driven asymmetry. Evaluating three Korean-capable models, we find that surprisal was sensitive to honorific-feature mismatches, yielding attraction-like effects, but was not modulated by distractor prominence. Although surprisal selectively improved reading-time predictions, this reflected item-level predictive overlap rather than prominence sensitivity. GPT models thus capture surface regularities but do not reproduce the construction-sensitive, prominence-driven asymmetry of human dependency resolution, suggesting that LLMs diverge from humans precisely where processing depends on working-memory constraints rather than distributional expectation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kwon-etal-2026-computational">
<titleInfo>
<title>Computational surprisal is not enough: Discourse prominence modulates human eye-movement patterns but not GPT expectations in Korean</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nayoung</namePart>
<namePart type="family">Kwon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gyu-Ho</namePart>
<namePart type="family">Shin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Seongmin</namePart>
<namePart type="family">Mun</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>Large Language Model (LLM) surprisal is increasingly used as a proxy for human sentence processing, yet whether model expectations reflect the cognitive factors governing real-time human behavior remains uncertain. We examine Korean subject–verb honorific agreement in subject- and object-control constructions, where human eye-tracking data reveal a prominence-driven asymmetry. Evaluating three Korean-capable models, we find that surprisal was sensitive to honorific-feature mismatches, yielding attraction-like effects, but was not modulated by distractor prominence. Although surprisal selectively improved reading-time predictions, this reflected item-level predictive overlap rather than prominence sensitivity. GPT models thus capture surface regularities but do not reproduce the construction-sensitive, prominence-driven asymmetry of human dependency resolution, suggesting that LLMs diverge from humans precisely where processing depends on working-memory constraints rather than distributional expectation.</abstract>
<identifier type="citekey">kwon-etal-2026-computational</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1443</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1443/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>30783</start>
<end>30796</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Computational surprisal is not enough: Discourse prominence modulates human eye-movement patterns but not GPT expectations in Korean
%A Kwon, Nayoung
%A Shin, Gyu-Ho
%A Mun, Seongmin
%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 kwon-etal-2026-computational
%X Large Language Model (LLM) surprisal is increasingly used as a proxy for human sentence processing, yet whether model expectations reflect the cognitive factors governing real-time human behavior remains uncertain. We examine Korean subject–verb honorific agreement in subject- and object-control constructions, where human eye-tracking data reveal a prominence-driven asymmetry. Evaluating three Korean-capable models, we find that surprisal was sensitive to honorific-feature mismatches, yielding attraction-like effects, but was not modulated by distractor prominence. Although surprisal selectively improved reading-time predictions, this reflected item-level predictive overlap rather than prominence sensitivity. GPT models thus capture surface regularities but do not reproduce the construction-sensitive, prominence-driven asymmetry of human dependency resolution, suggesting that LLMs diverge from humans precisely where processing depends on working-memory constraints rather than distributional expectation.
%R 10.18653/2026.findings-emnlp.1443
%U https://aclanthology.org/2026.findings-emnlp.1443/
%U https://doi.org/10.18653/2026.findings-emnlp.1443
%P 30783-30796
Markdown (Informal)
[Computational surprisal is not enough: Discourse prominence modulates human eye-movement patterns but not GPT expectations in Korean](https://aclanthology.org/2026.findings-emnlp.1443/) (Kwon et al., Findings 2026)
ACL