<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rajput-etal-2026-evaluation">
    <titleInfo>
        <title>Evaluation Drift in LLM Personality Induction: Are We Moving the Goalpost?</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Prateek</namePart>
        <namePart type="given">Kumar</namePart>
        <namePart type="family">Rajput</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yewei</namePart>
        <namePart type="family">Song</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Iyiola</namePart>
        <namePart type="given">Emmanuel</namePart>
        <namePart type="family">Olatunji</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jacques</namePart>
        <namePart type="family">Klein</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Tegawendé</namePart>
        <namePart type="family">Bissyande</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-05</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the Fifteenth Language Resources and Evaluation Conference</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Stelios</namePart>
            <namePart type="family">Piperidis</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Núria</namePart>
            <namePart type="family">Bel</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Henk</namePart>
            <namePart type="family">van den Heuvel</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Nancy</namePart>
            <namePart type="family">Ide</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Simon</namePart>
            <namePart type="family">Krek</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Antonio</namePart>
            <namePart type="family">Toral</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>ELRA Language Resource Association</publisher>
            <place>
                <placeTerm type="text">Palma de Mallorca, Spain</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Can large language models reliably express a human-like personality, or are they merely mimicking surface cues without a stable underlying profile? We study this question on the long-form Essays Dataset, preferred over short, mood-driven text to target stable traits. Using a questionnaire-based (self-evaluation) test: IPIP-NEO, we ask: (i) does post-training (SFT, DPO, ORPO) stabilize questionnaire scores under prompt rephrasings, and (ii) can it induce target Big Five profiles from unguided essays? Across five models, fine-tuning consistently reduces variance in questionnaire responses, mitigating the fragility seen in pre-trained models. Yet accuracy on the full five-dimensional profile remains near chance even when single-trait scores improve, indicating that unguided essays lack the cues needed for faithful personality expression. We argue for scenario-grounded datasets or interactive elicitation that accumulates test-aligned evidence over time.</abstract>
    <identifier type="citekey">rajput-etal-2026-evaluation</identifier>
    <identifier type="doi">10.63317/4zdw5tmzx58h</identifier>
    <location>
        <url>https://aclanthology.org/2026.lrec-1.881/</url>
    </location>
    <part>
        <date>2026-05</date>
        <extent unit="page">
            <start>11272</start>
            <end>11285</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Evaluation Drift in LLM Personality Induction: Are We Moving the Goalpost?
%A Rajput, Prateek Kumar
%A Song, Yewei
%A Olatunji, Iyiola Emmanuel
%A Klein, Jacques
%A Bissyande, Tegawendé
%Y Piperidis, Stelios
%Y Bel, Núria
%Y van den Heuvel, Henk
%Y Ide, Nancy
%Y Krek, Simon
%Y Toral, Antonio
%S Proceedings of the Fifteenth Language Resources and Evaluation Conference
%D 2026
%8 May
%I ELRA Language Resource Association
%C Palma de Mallorca, Spain
%F rajput-etal-2026-evaluation
%X Can large language models reliably express a human-like personality, or are they merely mimicking surface cues without a stable underlying profile? We study this question on the long-form Essays Dataset, preferred over short, mood-driven text to target stable traits. Using a questionnaire-based (self-evaluation) test: IPIP-NEO, we ask: (i) does post-training (SFT, DPO, ORPO) stabilize questionnaire scores under prompt rephrasings, and (ii) can it induce target Big Five profiles from unguided essays? Across five models, fine-tuning consistently reduces variance in questionnaire responses, mitigating the fragility seen in pre-trained models. Yet accuracy on the full five-dimensional profile remains near chance even when single-trait scores improve, indicating that unguided essays lack the cues needed for faithful personality expression. We argue for scenario-grounded datasets or interactive elicitation that accumulates test-aligned evidence over time.
%R 10.63317/4zdw5tmzx58h
%U https://aclanthology.org/2026.lrec-1.881/
%U https://doi.org/10.63317/4zdw5tmzx58h
%P 11272-11285

Markdown (Informal)

[Evaluation Drift in LLM Personality Induction: Are We Moving the Goalpost?](https://aclanthology.org/2026.lrec-1.881/) (Rajput et al., LREC 2026)

• Evaluation Drift in LLM Personality Induction: Are We Moving the Goalpost? (Rajput et al., LREC 2026)

ACL

• Prateek Kumar Rajput, Yewei Song, Iyiola Emmanuel Olatunji, Jacques Klein, and Tegawendé Bissyande. 2026. Evaluation Drift in LLM Personality Induction: Are We Moving the Goalpost?. In Proceedings of the Fifteenth Language Resources and Evaluation Conference, pages 11272–11285, Palma de Mallorca, Spain. ELRA Language Resource Association.