@inproceedings{shafiabadi-wisniewski-2026-data,
title = "Data Shapes What, Scale Shapes How: Lexical Learning Dynamics in {LLM}s",
author = "Shafiabadi, Nazanin and
Wisniewski, Guillaume",
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.1812/",
pages = "38703--38716",
ISBN = "979-8-89176-462-0",
abstract = "We investigate how model scale shapes lexical acquisition in LLMs, introducing trajectory comparison as a methodological tool and three new convergence metrics. Tracking surprisal dynamics across four Pythia model sizes and 153 training checkpoints, we report two main findings. First, scale drives a global rise in inter-word surprisal coherence: synonym curves become more correlated at larger scales, but baseline correction reveals that the advantage of semantic relatedness over random pairs is invariant with scale, indicating that what scales is broad inter-word coherence rather than semantic abstraction specifically. Second, learning dynamics are shaped primarily by training data rather than by model capacity: surprisal trajectories are highly similar across model sizes, and our new metrics confirm that larger models converge faster but not along qualitatively different paths. Together, these results call for baseline-corrected analyses and longitudinal evaluation when assessing lexical competence in LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shafiabadi-wisniewski-2026-data">
<titleInfo>
<title>Data Shapes What, Scale Shapes How: Lexical Learning Dynamics in LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nazanin</namePart>
<namePart type="family">Shafiabadi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guillaume</namePart>
<namePart type="family">Wisniewski</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>We investigate how model scale shapes lexical acquisition in LLMs, introducing trajectory comparison as a methodological tool and three new convergence metrics. Tracking surprisal dynamics across four Pythia model sizes and 153 training checkpoints, we report two main findings. First, scale drives a global rise in inter-word surprisal coherence: synonym curves become more correlated at larger scales, but baseline correction reveals that the advantage of semantic relatedness over random pairs is invariant with scale, indicating that what scales is broad inter-word coherence rather than semantic abstraction specifically. Second, learning dynamics are shaped primarily by training data rather than by model capacity: surprisal trajectories are highly similar across model sizes, and our new metrics confirm that larger models converge faster but not along qualitatively different paths. Together, these results call for baseline-corrected analyses and longitudinal evaluation when assessing lexical competence in LLMs.</abstract>
<identifier type="citekey">shafiabadi-wisniewski-2026-data</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1812/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>38703</start>
<end>38716</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Data Shapes What, Scale Shapes How: Lexical Learning Dynamics in LLMs
%A Shafiabadi, Nazanin
%A Wisniewski, Guillaume
%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 shafiabadi-wisniewski-2026-data
%X We investigate how model scale shapes lexical acquisition in LLMs, introducing trajectory comparison as a methodological tool and three new convergence metrics. Tracking surprisal dynamics across four Pythia model sizes and 153 training checkpoints, we report two main findings. First, scale drives a global rise in inter-word surprisal coherence: synonym curves become more correlated at larger scales, but baseline correction reveals that the advantage of semantic relatedness over random pairs is invariant with scale, indicating that what scales is broad inter-word coherence rather than semantic abstraction specifically. Second, learning dynamics are shaped primarily by training data rather than by model capacity: surprisal trajectories are highly similar across model sizes, and our new metrics confirm that larger models converge faster but not along qualitatively different paths. Together, these results call for baseline-corrected analyses and longitudinal evaluation when assessing lexical competence in LLMs.
%U https://aclanthology.org/2026.findings-emnlp.1812/
%P 38703-38716
Markdown (Informal)
[Data Shapes What, Scale Shapes How: Lexical Learning Dynamics in LLMs](https://aclanthology.org/2026.findings-emnlp.1812/) (Shafiabadi & Wisniewski, Findings 2026)
ACL