<?xml version="1.0" encoding="utf-8"?>
<TEI xmlns="http://www.tei-c.org/ns/1.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:hal="http://hal.archives-ouvertes.fr/" xmlns:gml="http://www.opengis.net/gml/3.3/" xmlns:gmlce="http://www.opengis.net/gml/3.3/ce" version="1.1" xsi:schemaLocation="http://www.tei-c.org/ns/1.0 http://api.archives-ouvertes.fr/documents/aofr-sword.xsd">
  <teiHeader>
    <fileDesc>
      <titleStmt>
        <title>HAL TEI export of hal-05317590</title>
      </titleStmt>
      <publicationStmt>
        <distributor>CCSD</distributor>
        <availability status="restricted">
          <licence target="https://creativecommons.org/publicdomain/zero/1.0/">CC0 1.0 - Universal</licence>
        </availability>
        <date when="2026-05-22T21:58:11+02:00"/>
      </publicationStmt>
      <sourceDesc>
        <p part="N">HAL API Platform</p>
      </sourceDesc>
    </fileDesc>
  </teiHeader>
  <text>
    <body>
      <listBibl>
        <biblFull>
          <titleStmt>
            <title xml:lang="en">MOSAIC : Multiple Observers Spotting AI Content</title>
            <title xml:lang="fr">MOSAIC : Mélange d'experts pour la détection de textes artificiels</title>
            <author role="aut">
              <persName>
                <forename type="first">Matthieu</forename>
                <surname>Dubois</surname>
              </persName>
              <email type="md5">060aa661cb75a2dff6f6121ae2dfcdc1</email>
              <email type="domain">isir.upmc.fr</email>
              <idno type="idhal" notation="string">matthieu-dubois</idno>
              <idno type="idhal" notation="numeric">1632348</idno>
              <idno type="halauthorid" notation="string">330217-1632348</idno>
              <idno type="ORCID">https://orcid.org/0009-0003-9777-7472</idno>
              <affiliation ref="#struct-1085231"/>
              <affiliation ref="#struct-1148235"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Pablo</forename>
                <surname>Piantanida</surname>
              </persName>
              <email type="md5">4d99eb094c1c3424e1f0788dcb8f365d</email>
              <email type="domain">centralesupelec.fr</email>
              <idno type="idhal" notation="string">pablo-piantanida</idno>
              <idno type="idhal" notation="numeric">736967</idno>
              <idno type="halauthorid" notation="string">28498-736967</idno>
              <idno type="ORCID">https://orcid.org/0000-0002-8717-2117</idno>
              <idno type="IDREF">https://www.idref.fr/116306300</idno>
              <affiliation ref="#struct-1148235"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">François</forename>
                <surname>Yvon</surname>
              </persName>
              <email type="md5">c62025aca0c6cf8d65b4fe9600c991af</email>
              <email type="domain">cnrs.fr</email>
              <idno type="idhal" notation="string">francois-yvon</idno>
              <idno type="idhal" notation="numeric">5347</idno>
              <idno type="halauthorid" notation="string">2529017-5347</idno>
              <idno type="ORCID">https://orcid.org/0000-0002-7972-7442</idno>
              <idno type="IDREF">https://www.idref.fr/057593531</idno>
              <idno type="GOOGLE SCHOLAR">https://scholar.google.fr/citations?user=JjfDvawAAAAJ</idno>
              <idno type="ARXIV">https://arxiv.org/a/yvon_f_1</idno>
              <affiliation ref="#struct-1085231"/>
            </author>
            <editor role="depositor">
              <persName>
                <forename>François</forename>
                <surname>Yvon</surname>
              </persName>
              <email type="md5">c62025aca0c6cf8d65b4fe9600c991af</email>
              <email type="domain">cnrs.fr</email>
            </editor>
          </titleStmt>
          <editionStmt>
            <edition n="v1" type="current">
              <date type="whenSubmitted">2025-10-16 12:07:26</date>
              <date type="whenModified">2026-04-23 09:02:01</date>
              <date type="whenReleased">2025-10-20 08:01:50</date>
              <date type="whenProduced">2025-07-02</date>
              <date type="whenEndEmbargoed">2025-10-16</date>
              <ref type="file" target="https://hal.science/hal-05317590v1/document">
                <date notBefore="2025-10-16"/>
              </ref>
              <ref type="file" subtype="greenPublisher" n="1" target="https://hal.science/hal-05317590v1/file/2025.jeptalnrecital-taln.31.pdf" id="file-5317590-4554884">
                <date notBefore="2025-10-16"/>
              </ref>
            </edition>
            <respStmt>
              <resp>contributor</resp>
              <name key="164853">
                <persName>
                  <forename>François</forename>
                  <surname>Yvon</surname>
                </persName>
                <email type="md5">c62025aca0c6cf8d65b4fe9600c991af</email>
                <email type="domain">cnrs.fr</email>
              </name>
            </respStmt>
          </editionStmt>
          <publicationStmt>
            <distributor>CCSD</distributor>
            <idno type="halId">hal-05317590</idno>
            <idno type="halUri">https://hal.science/hal-05317590</idno>
            <idno type="halBibtex">dubois:hal-05317590</idno>
            <idno type="halRefHtml">&lt;i&gt;32ème Conférence sur le Traitement Automatique des Langues Naturelles (TALN)&lt;/i&gt;, ATALA, Jul 2025, Marseille, France</idno>
            <idno type="halRef">32ème Conférence sur le Traitement Automatique des Langues Naturelles (TALN), ATALA, Jul 2025, Marseille, France</idno>
            <availability status="restricted">
              <licence target="https://creativecommons.org/licenses/by/4.0/">CC BY 4.0 - Attribution<ref corresp="#file-5317590-4554884"/></licence>
            </availability>
          </publicationStmt>
          <seriesStmt>
            <idno type="stamp" n="CNRS">CNRS - Centre national de la recherche scientifique</idno>
            <idno type="stamp" n="ISIR" corresp="SORBONNE-UNIVERSITE">Institut des Systèmes Intelligents et de Robotique</idno>
            <idno type="stamp" n="CENTRALESUPELEC">Ecole CentraleSupélec</idno>
            <idno type="stamp" n="GENCI">Publications ayant eu recours aux supercalculateurs du GENCI</idno>
            <idno type="stamp" n="UNIV-PARIS-SACLAY">Université Paris-Saclay</idno>
            <idno type="stamp" n="SORBONNE-UNIVERSITE">Sorbonne Université</idno>
            <idno type="stamp" n="SORBONNE-UNIV" corresp="SORBONNE-UNIVERSITE">Sorbonne Université 01/01/2018</idno>
            <idno type="stamp" n="SU-SCIENCES" corresp="SORBONNE-UNIVERSITE">Faculté des Sciences de Sorbonne Université</idno>
            <idno type="stamp" n="TALN-RECITAL">Conférences TALN RECITAL</idno>
            <idno type="stamp" n="UNIVERSITE-PARIS-SACLAY" corresp="UNIV-PARIS-SACLAY">Université Paris-Saclay</idno>
            <idno type="stamp" n="SU-TI">Sorbonne Université - Texte Intégral</idno>
            <idno type="stamp" n="HUB-IA">Hub Intelligence Artificielle de CentraleSupélec</idno>
            <idno type="stamp" n="ALLIANCE-SU"> Alliance Sorbonne Université</idno>
            <idno type="stamp" n="ISIR_MLIA" corresp="ISIR">Machine Learning and Information Access</idno>
            <idno type="stamp" n="ILLS">International Laboratory on Learning Systems</idno>
            <idno type="stamp" n="SUPRA_INGENIERIE">Ingénierie</idno>
            <idno type="stamp" n="CORIA-TALN2025" corresp="TALN-RECITAL">Conférence CORIA TALN 2025</idno>
          </seriesStmt>
          <notesStmt>
            <note type="audience" n="2">International</note>
            <note type="invited" n="0">No</note>
            <note type="popular" n="0">No</note>
            <note type="peer" n="1">Yes</note>
            <note type="proceedings" n="1">Yes</note>
          </notesStmt>
          <sourceDesc>
            <biblStruct>
              <analytic>
                <title xml:lang="en">MOSAIC : Multiple Observers Spotting AI Content</title>
                <title xml:lang="fr">MOSAIC : Mélange d'experts pour la détection de textes artificiels</title>
                <author role="aut">
                  <persName>
                    <forename type="first">Matthieu</forename>
                    <surname>Dubois</surname>
                  </persName>
                  <email type="md5">060aa661cb75a2dff6f6121ae2dfcdc1</email>
                  <email type="domain">isir.upmc.fr</email>
                  <idno type="idhal" notation="string">matthieu-dubois</idno>
                  <idno type="idhal" notation="numeric">1632348</idno>
                  <idno type="halauthorid" notation="string">330217-1632348</idno>
                  <idno type="ORCID">https://orcid.org/0009-0003-9777-7472</idno>
                  <affiliation ref="#struct-1085231"/>
                  <affiliation ref="#struct-1148235"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Pablo</forename>
                    <surname>Piantanida</surname>
                  </persName>
                  <email type="md5">4d99eb094c1c3424e1f0788dcb8f365d</email>
                  <email type="domain">centralesupelec.fr</email>
                  <idno type="idhal" notation="string">pablo-piantanida</idno>
                  <idno type="idhal" notation="numeric">736967</idno>
                  <idno type="halauthorid" notation="string">28498-736967</idno>
                  <idno type="ORCID">https://orcid.org/0000-0002-8717-2117</idno>
                  <idno type="IDREF">https://www.idref.fr/116306300</idno>
                  <affiliation ref="#struct-1148235"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">François</forename>
                    <surname>Yvon</surname>
                  </persName>
                  <email type="md5">c62025aca0c6cf8d65b4fe9600c991af</email>
                  <email type="domain">cnrs.fr</email>
                  <idno type="idhal" notation="string">francois-yvon</idno>
                  <idno type="idhal" notation="numeric">5347</idno>
                  <idno type="halauthorid" notation="string">2529017-5347</idno>
                  <idno type="ORCID">https://orcid.org/0000-0002-7972-7442</idno>
                  <idno type="IDREF">https://www.idref.fr/057593531</idno>
                  <idno type="GOOGLE SCHOLAR">https://scholar.google.fr/citations?user=JjfDvawAAAAJ</idno>
                  <idno type="ARXIV">https://arxiv.org/a/yvon_f_1</idno>
                  <affiliation ref="#struct-1085231"/>
                </author>
              </analytic>
              <monogr>
                <title level="m">Actes des 32ème Conférence sur le Traitement Automatique des Langues Naturelles (TALN)</title>
                <meeting>
                  <title>32ème Conférence sur le Traitement Automatique des Langues Naturelles (TALN)</title>
                  <date type="start">2025-07-02</date>
                  <date type="end">2025-07-05</date>
                  <settlement>Marseille</settlement>
                  <country key="FR">France</country>
                </meeting>
                <respStmt>
                  <resp>conferenceOrganizer</resp>
                  <name>ATALA</name>
                </respStmt>
                <imprint/>
              </monogr>
            </biblStruct>
          </sourceDesc>
          <profileDesc>
            <langUsage>
              <language ident="fr">French</language>
            </langUsage>
            <textClass>
              <keywords scheme="author">
                <term xml:lang="en">Information theory</term>
                <term xml:lang="en">Machine-generated text detection</term>
                <term xml:lang="en">Text Generation</term>
                <term xml:lang="fr">Théorie de Mixture of experts</term>
                <term xml:lang="fr">Détection de textes artificiels</term>
                <term xml:lang="fr">Génération de Textes</term>
                <term xml:lang="fr">Mélange d'experts</term>
                <term xml:lang="fr">Mélange d'experts Génération de Textes Détection de textes artificiels Théorie de Mixture of experts Text Generation Machine-generated text detection Information theory</term>
              </keywords>
              <classCode scheme="halDomain" n="info.info-tt">Computer Science [cs]/Document and Text Processing</classCode>
              <classCode scheme="halTypology" n="COMM">Conference papers</classCode>
              <classCode scheme="halOldTypology" n="COMM">Conference papers</classCode>
              <classCode scheme="halTreeTypology" n="COMM">Conference papers</classCode>
            </textClass>
            <abstract xml:lang="en">
              <p>The dissemination of Large Language Models (LLMs) has made it easier for all to produce harmful,toxic, faked or forged content. In response, various proposals have been made to automaticallydiscriminate artificially generated from human-written texts, framing the problem as a binary classifi-cation problem. Early approaches evaluate an input document with a detector LLM, assuming thatlow-perplexity scores reliably signal machine-made content. More recent systems instead considertwo LLMs and compare their probability distributions over the document to further discriminate whenperplexity alone cannot. However, using a fixed pair of models can induce brittleness in performance.We extend these approaches to the ensembling of several LLMs and derive a new, theoreticallygrounded approach to combine their respective strengths.</p>
            </abstract>
            <abstract xml:lang="fr">
              <p>La diffusion auprès du grand public de grands modèles de langue facilite la production de contenus nuisibles, médisants, malhonnêtes ou falsifiés. En réponse, plusieurs solutions ont été proposées pour identifier les textes ainsi produits, en traitant le problème comme une tâche de classification binaire. Les premières approches reposent sur l'analyse d'un document par un modèle détecteur, avec l'hypothèse qu'un faible score de perplexité indique que le contenu est artificiel. Des méthodes plus récentes proposent de comparer les distributions de probabilité calculées par deux modèles. Cependant, s'appuyer sur une paire fixe de modèles peut fragiliser les performances. Nous étendons ces méthodes en combinant plusieurs modèles et en développant une approche théoriquement fondée pour exploiter au mieux chacun d'entre eux.</p>
            </abstract>
            <particDesc>
              <org type="ir" xml:id="fr-esr-fdr-ir-064">
                <orgName type="acronym">GENCI</orgName>
                <idno type="wikidata">Q3093180</idno>
              </org>
            </particDesc>
          </profileDesc>
        </biblFull>
      </listBibl>
    </body>
    <back>
      <listOrg type="structures">
        <org type="researchteam" xml:id="struct-1085231" status="VALID">
          <orgName>Machine Learning and Information Access</orgName>
          <orgName type="acronym">MLIA</orgName>
          <date type="start">2022-01-01</date>
          <desc>
            <address>
              <addrLine>Tour 65-66 3ème étage4, place Jussieu75005 Paris – France</addrLine>
              <country key="FR"/>
            </address>
          </desc>
          <listRelation>
            <relation active="#struct-541937" type="direct"/>
            <relation active="#struct-413221" type="indirect"/>
            <relation name="UMR7222" active="#struct-441569" type="indirect"/>
          </listRelation>
        </org>
        <org type="laboratory" xml:id="struct-1148235" status="VALID">
          <idno type="IdRef">280573138</idno>
          <idno type="RNSR">202224169E</idno>
          <idno type="ROR">https://ror.org/02bkw1m29</idno>
          <orgName>International Laboratory on Learning Systems</orgName>
          <orgName type="acronym">ILLS</orgName>
          <date type="start">2022-01-01</date>
          <desc>
            <address>
              <addrLine>Montréal</addrLine>
              <country key="CA"/>
            </address>
          </desc>
          <listRelation>
            <relation active="#struct-134741" type="direct"/>
            <relation active="#struct-136804" type="direct"/>
            <relation active="#struct-411575" type="direct"/>
            <relation active="#struct-419361" type="direct"/>
            <relation name="IRL2020" active="#struct-441569" type="direct"/>
          </listRelation>
        </org>
        <org type="laboratory" xml:id="struct-541937" status="VALID">
          <idno type="IdRef">241179122</idno>
          <idno type="RNSR">200918463J</idno>
          <idno type="ROR">05neq8668</idno>
          <orgName>Institut des Systèmes Intelligents et de Robotique</orgName>
          <orgName type="acronym">ISIR</orgName>
          <date type="start">2018-01-01</date>
          <desc>
            <address>
              <addrLine>Sorbonne-Université, Boite courrier 173 4 Place JUSSIEU 75252 Paris cedex 05</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.isir.upmc.fr</ref>
          </desc>
          <listRelation>
            <relation active="#struct-413221" type="direct"/>
            <relation name="UMR7222" active="#struct-441569" type="direct"/>
          </listRelation>
        </org>
        <org type="regroupinstitution" xml:id="struct-413221" status="VALID">
          <idno type="IdRef">221333754</idno>
          <idno type="ROR">https://ror.org/02en5vm52</idno>
          <orgName>Sorbonne Université</orgName>
          <orgName type="acronym">SU</orgName>
          <date type="start">2018-01-01</date>
          <desc>
            <address>
              <addrLine>21 rue de l’École de médecine - 75006 Paris</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.sorbonne-universite.fr/</ref>
          </desc>
        </org>
        <org type="regroupinstitution" xml:id="struct-441569" status="VALID">
          <idno type="IdRef">02636817X</idno>
          <idno type="ISNI">0000000122597504</idno>
          <idno type="ROR">https://ror.org/02feahw73</idno>
          <orgName>Centre National de la Recherche Scientifique</orgName>
          <orgName type="acronym">CNRS</orgName>
          <date type="start">1939-10-19</date>
          <desc>
            <address>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.cnrs.fr/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-134741" status="VALID">
          <idno type="ROR">https://ror.org/01pxwe438</idno>
          <orgName>McGill University = Université McGill [Montréal, Canada]</orgName>
          <desc>
            <address>
              <addrLine>845, rue Sherbrooke O. Montréal (Québec) Canada H3A 0G4</addrLine>
              <country key="CA"/>
            </address>
            <ref type="url">http://www.mcgill.ca/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-136804" status="VALID">
          <idno type="IdRef">080094201</idno>
          <idno type="ISNI">0000000122224302</idno>
          <idno type="ROR">https://ror.org/0020snb74</idno>
          <idno type="Wikidata">Q3577916</idno>
          <orgName>École de technologie supérieure [Montréal]</orgName>
          <orgName type="acronym">ÉTS</orgName>
          <date type="start">1974-01-01</date>
          <desc>
            <address>
              <addrLine>1100, rue Notre-Dame Ouest Montréal (Qc) H3C 1K3 Canada</addrLine>
              <country key="CA"/>
            </address>
            <ref type="url">http://www.etsmtl.ca/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-411575" status="VALID">
          <idno type="IdRef">184443237</idno>
          <idno type="ROR">https://ror.org/019tcpt25</idno>
          <orgName>CentraleSupélec</orgName>
          <desc>
            <address>
              <addrLine>3, rue Joliot Curie,Plateau de Moulon,91192 GIF-SUR-YVETTE Cedex</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.centralesupelec.fr</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-419361" status="VALID">
          <idno type="IdRef">241345251</idno>
          <idno type="ROR">https://ror.org/03xjwb503</idno>
          <orgName>Université Paris-Saclay</orgName>
          <desc>
            <address>
              <addrLine>Bâtiment Bréguet, 3 Rue Joliot Curie 2e ét, 91190 Gif-sur-Yvette</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.universite-paris-saclay.fr/fr</ref>
          </desc>
        </org>
      </listOrg>
    </back>
  </text>
</TEI>