@inproceedings{kim-etal-2026-chronos,
title = "Chronos: Chronology-Grounded Continued Pre-training for Language Model Adaptation to Materials Science",
author = "Kim, Sangyun and
Kim, Junho and
Lee, SangKeun",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2341/",
pages = "53173--53193",
ISBN = "979-8-89176-459-0",
abstract = "Continued pre-training on domain-specific corpora has become the prevailing approach for adapting pre-trained language models (PLMs) to materials science. However, existing strategies typically treat scientific literature as independent samples, thereby overlooking the cumulative nature of scientific knowledge where novel discoveries systematically build upon prior research. To address this limitation, we propose Chronology-grounded continued pre-training for materials science (Chronos), which explicitly considers the chronological knowledge in terms of data construction and training schemes. Specifically, our framework traces the relevant prior articles by using citation-based information and constructs a chronological curriculum to progressively learn the materials knowledge. The comprehensive experiments demonstrate that Chronos outperforms state-of-the-art (SOTA) baselines on various materials science benchmarks and backbone models, achieving improvements of 4.7{\%} on the Macro-F1 score. In-depth analysis shows that considering the cumulative nature of knowledge enhances both the quality of the corpus and the effectiveness of training schemes."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kim-etal-2026-chronos">
<titleInfo>
<title>Chronos: Chronology-Grounded Continued Pre-training for Language Model Adaptation to Materials Science</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sangyun</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junho</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">SangKeun</namePart>
<namePart type="family">Lee</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Continued pre-training on domain-specific corpora has become the prevailing approach for adapting pre-trained language models (PLMs) to materials science. However, existing strategies typically treat scientific literature as independent samples, thereby overlooking the cumulative nature of scientific knowledge where novel discoveries systematically build upon prior research. To address this limitation, we propose Chronology-grounded continued pre-training for materials science (Chronos), which explicitly considers the chronological knowledge in terms of data construction and training schemes. Specifically, our framework traces the relevant prior articles by using citation-based information and constructs a chronological curriculum to progressively learn the materials knowledge. The comprehensive experiments demonstrate that Chronos outperforms state-of-the-art (SOTA) baselines on various materials science benchmarks and backbone models, achieving improvements of 4.7% on the Macro-F1 score. In-depth analysis shows that considering the cumulative nature of knowledge enhances both the quality of the corpus and the effectiveness of training schemes.</abstract>
<identifier type="citekey">kim-etal-2026-chronos</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2341/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>53173</start>
<end>53193</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Chronos: Chronology-Grounded Continued Pre-training for Language Model Adaptation to Materials Science
%A Kim, Sangyun
%A Kim, Junho
%A Lee, SangKeun
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F kim-etal-2026-chronos
%X Continued pre-training on domain-specific corpora has become the prevailing approach for adapting pre-trained language models (PLMs) to materials science. However, existing strategies typically treat scientific literature as independent samples, thereby overlooking the cumulative nature of scientific knowledge where novel discoveries systematically build upon prior research. To address this limitation, we propose Chronology-grounded continued pre-training for materials science (Chronos), which explicitly considers the chronological knowledge in terms of data construction and training schemes. Specifically, our framework traces the relevant prior articles by using citation-based information and constructs a chronological curriculum to progressively learn the materials knowledge. The comprehensive experiments demonstrate that Chronos outperforms state-of-the-art (SOTA) baselines on various materials science benchmarks and backbone models, achieving improvements of 4.7% on the Macro-F1 score. In-depth analysis shows that considering the cumulative nature of knowledge enhances both the quality of the corpus and the effectiveness of training schemes.
%U https://aclanthology.org/2026.emnlp-main.2341/
%P 53173-53193
Markdown (Informal)
[Chronos: Chronology-Grounded Continued Pre-training for Language Model Adaptation to Materials Science](https://aclanthology.org/2026.emnlp-main.2341/) (Kim et al., EMNLP 2026)
ACL