@inproceedings{duan-etal-2026-large,
title = "Do Large Language Models Acquire Phrase-Based Processing? Evidence from Eye Movements and Model-Brain Alignment After Fine-Tuning",
author = "Duan, Xufeng and
Ma, Zhengwu and
Yao, Zhaoqian and
Li, Jixing and
Cai, Zhenguang",
editor = "Voigt, Rob and
Warstadt, Alex and
Feldman, Naomi and
Linzen, Tal",
booktitle = "Proceedings of the Society for Computation in Linguistics 2026",
month = jul,
year = "2026",
address = "San Diego, CA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.scil-main.44/",
pages = "464--476",
ISBN = "979-8-89176-412-5",
abstract = "Autoregressive large language models (LLMs) process text token-by-token, yet the human language system operates over multi-word units. We ask whether aggregating LLM representations at the phrase level yields a closer correspondence to human reading behavior and language cortex than the default word-level representations, and whether phrase-segmentation fine-tuning amplifies this correspondence. Using Meta-Llama-3.1-8B (base and fine-tuned), we provide three converging lines of evidence. First, phrase-level attention features predict regressive eye-saccade patterns more closely than word-level features; a partial correlation analysis with a shuffled-boundary control indicates that this is not solely an aggregation artifact and that linguistic chunk boundaries explain unique variance beyond word-level attention. Second, fMRI encoding analyses show that fine-tuning selectively improves phrase encoding in left superior temporal gyrus and inferior frontal gyrus, with no improvement for word representations. Third, representational similarity analysis confirms a phrase-specific gain in model-brain geometric alignment. These results identify phrase-level representation as a critical granularity for LLM{--}human correspondence and suggest that targeted training can model human-like compositional processing, linking computational representations to hierarchical theories of language."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="duan-etal-2026-large">
<titleInfo>
<title>Do Large Language Models Acquire Phrase-Based Processing? Evidence from Eye Movements and Model-Brain Alignment After Fine-Tuning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xufeng</namePart>
<namePart type="family">Duan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhengwu</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhaoqian</namePart>
<namePart type="family">Yao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jixing</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhenguang</namePart>
<namePart type="family">Cai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Society for Computation in Linguistics 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Rob</namePart>
<namePart type="family">Voigt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alex</namePart>
<namePart type="family">Warstadt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naomi</namePart>
<namePart type="family">Feldman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tal</namePart>
<namePart type="family">Linzen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, CA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-412-5</identifier>
</relatedItem>
<abstract>Autoregressive large language models (LLMs) process text token-by-token, yet the human language system operates over multi-word units. We ask whether aggregating LLM representations at the phrase level yields a closer correspondence to human reading behavior and language cortex than the default word-level representations, and whether phrase-segmentation fine-tuning amplifies this correspondence. Using Meta-Llama-3.1-8B (base and fine-tuned), we provide three converging lines of evidence. First, phrase-level attention features predict regressive eye-saccade patterns more closely than word-level features; a partial correlation analysis with a shuffled-boundary control indicates that this is not solely an aggregation artifact and that linguistic chunk boundaries explain unique variance beyond word-level attention. Second, fMRI encoding analyses show that fine-tuning selectively improves phrase encoding in left superior temporal gyrus and inferior frontal gyrus, with no improvement for word representations. Third, representational similarity analysis confirms a phrase-specific gain in model-brain geometric alignment. These results identify phrase-level representation as a critical granularity for LLM–human correspondence and suggest that targeted training can model human-like compositional processing, linking computational representations to hierarchical theories of language.</abstract>
<identifier type="citekey">duan-etal-2026-large</identifier>
<location>
<url>https://aclanthology.org/2026.scil-main.44/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>464</start>
<end>476</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Do Large Language Models Acquire Phrase-Based Processing? Evidence from Eye Movements and Model-Brain Alignment After Fine-Tuning
%A Duan, Xufeng
%A Ma, Zhengwu
%A Yao, Zhaoqian
%A Li, Jixing
%A Cai, Zhenguang
%Y Voigt, Rob
%Y Warstadt, Alex
%Y Feldman, Naomi
%Y Linzen, Tal
%S Proceedings of the Society for Computation in Linguistics 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, CA
%@ 979-8-89176-412-5
%F duan-etal-2026-large
%X Autoregressive large language models (LLMs) process text token-by-token, yet the human language system operates over multi-word units. We ask whether aggregating LLM representations at the phrase level yields a closer correspondence to human reading behavior and language cortex than the default word-level representations, and whether phrase-segmentation fine-tuning amplifies this correspondence. Using Meta-Llama-3.1-8B (base and fine-tuned), we provide three converging lines of evidence. First, phrase-level attention features predict regressive eye-saccade patterns more closely than word-level features; a partial correlation analysis with a shuffled-boundary control indicates that this is not solely an aggregation artifact and that linguistic chunk boundaries explain unique variance beyond word-level attention. Second, fMRI encoding analyses show that fine-tuning selectively improves phrase encoding in left superior temporal gyrus and inferior frontal gyrus, with no improvement for word representations. Third, representational similarity analysis confirms a phrase-specific gain in model-brain geometric alignment. These results identify phrase-level representation as a critical granularity for LLM–human correspondence and suggest that targeted training can model human-like compositional processing, linking computational representations to hierarchical theories of language.
%U https://aclanthology.org/2026.scil-main.44/
%P 464-476
Markdown (Informal)
[Do Large Language Models Acquire Phrase-Based Processing? Evidence from Eye Movements and Model-Brain Alignment After Fine-Tuning](https://aclanthology.org/2026.scil-main.44/) (Duan et al., SCiL 2026)
ACL