<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="utami-sasano-2026-still">
    <titleInfo>
        <title>Can We Still Trace L1 Signals? Investigating the Resilience of Native Language Signals in the LLM Era</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Nabelanita</namePart>
        <namePart type="family">Utami</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ryohei</namePart>
        <namePart type="family">Sasano</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>The evolution of writing assistance tools from machine translation to large language models (LLMs) has changed how researchers write. Before the rise of LLM-based tools, non-native authors often rely more on word look-ups and individual sentence translations. Due to native language (L1) interference of the authors, the resulting work, although high in fluency, are often to have a certain “characteristics”. Some examples include phrases, idioms, or certain grammar points from the author’s native language. With recent advancements, LLM-based tools can translate, proofread grammar, and even paraphrase process to make them more fluent-sounding. This is extremely helpful for authors, but it raises an interesting question: does the rise of LLM assistance in writings by non-native authors erase L1 characteristics? This study investigates this occurrence by analyzing native language identification (NLI) trends in papers, specifically abstracts, across three eras: pre-neural network (NN), pre-LLM, and post-LLM. We found consistent decline in NLI performance over time. Interestingly, the post-LLM era reveals anomalies, specifically Chinese and French show unexpected resistance or divergent trends. Additionally, we performed an experiment by using recent LLM models to paraphrase abstracts and then feed the rewritten texts into our classifiers to confirm LLM effect on writing styles. The result show that LLM use in papers reduce significantly, and the decline of NLI performance in this rewritten experiment is consistent with the NLI performance trends across periods.</abstract>
    <identifier type="citekey">utami-sasano-2026-still</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2042/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>46549</start>
            <end>46567</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Can We Still Trace L1 Signals? Investigating the Resilience of Native Language Signals in the LLM Era
%A Utami, Nabelanita
%A Sasano, Ryohei
%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 utami-sasano-2026-still
%X The evolution of writing assistance tools from machine translation to large language models (LLMs) has changed how researchers write. Before the rise of LLM-based tools, non-native authors often rely more on word look-ups and individual sentence translations. Due to native language (L1) interference of the authors, the resulting work, although high in fluency, are often to have a certain “characteristics”. Some examples include phrases, idioms, or certain grammar points from the author’s native language. With recent advancements, LLM-based tools can translate, proofread grammar, and even paraphrase process to make them more fluent-sounding. This is extremely helpful for authors, but it raises an interesting question: does the rise of LLM assistance in writings by non-native authors erase L1 characteristics? This study investigates this occurrence by analyzing native language identification (NLI) trends in papers, specifically abstracts, across three eras: pre-neural network (NN), pre-LLM, and post-LLM. We found consistent decline in NLI performance over time. Interestingly, the post-LLM era reveals anomalies, specifically Chinese and French show unexpected resistance or divergent trends. Additionally, we performed an experiment by using recent LLM models to paraphrase abstracts and then feed the rewritten texts into our classifiers to confirm LLM effect on writing styles. The result show that LLM use in papers reduce significantly, and the decline of NLI performance in this rewritten experiment is consistent with the NLI performance trends across periods.
%U https://aclanthology.org/2026.emnlp-main.2042/
%P 46549-46567

Markdown (Informal)

[Can We Still Trace L1 Signals? Investigating the Resilience of Native Language Signals in the LLM Era](https://aclanthology.org/2026.emnlp-main.2042/) (Utami & Sasano, EMNLP 2026)

• Can We Still Trace L1 Signals? Investigating the Resilience of Native Language Signals in the LLM Era (Utami & Sasano, EMNLP 2026)

ACL

• Nabelanita Utami and Ryohei Sasano. 2026. Can We Still Trace L1 Signals? Investigating the Resilience of Native Language Signals in the LLM Era. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 46549–46567, Budapest, Hungary. Association for Computational Linguistics.