<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yagiz-horasan-2026-beyond">
    <titleInfo>
        <title>Beyond the Token: Correcting the Tokenization Bias in XAI via Morphologically-Aligned Projection</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Muhammet</namePart>
        <namePart type="given">Anil</namePart>
        <namePart type="family">Yagiz</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Fahrettin</namePart>
        <namePart type="family">Horasan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-03</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the Second Workshop Natural Language Processing for Turkic Languages (SIGTURK 2026)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Kemal</namePart>
            <namePart type="family">Oflazer</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Abdullatif</namePart>
            <namePart type="family">Köksal</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Onur</namePart>
            <namePart type="family">Varol</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Rabat, Morocco</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-370-8</identifier>
    </relatedItem>
    <abstract>Current interpretability methods for Large Language Models (LLMs) operate on a fundamental yet flawed assumption: that subword tokens represent independent semantic units. We prove that this assumption creates a fidelity bottleneck in Morphologically Rich Languages (MRLs), where semantic meaning is densely encoded in sub-token morphemes. We term this phenomenon the Tokenization-Morphology Misalignment (TMM). To resolve TMM, we introduce MAFEX (Morpheme-Aligned Faithful Explanations), a theoretically grounded framework that redefines feature attribution as a linear projection from the computational (token) basis to the linguistic (morpheme) basis. We evaluate our method on a diverse suite of Turkish LLMs, including BERTurk, BERTurk-Sentiment, Cosmos-BERT, and Kumru-2B. On our embedded benchmark (N=20), MAFEX achieves an average F1@1 of 91.25% compared to 13.75% for standard token-level baselines (IG, SHAP, DeepLIFT), representing a +77.5% absolute improvement, establishing it as the new standard for faithful multilingual interpretability.</abstract>
    <identifier type="citekey">yagiz-horasan-2026-beyond</identifier>
    <location>
        <url>https://aclanthology.org/2026.sigturk-1.19/</url>
    </location>
    <part>
        <date>2026-03</date>
        <extent unit="page">
            <start>228</start>
            <end>235</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Beyond the Token: Correcting the Tokenization Bias in XAI via Morphologically-Aligned Projection
%A Yagiz, Muhammet Anil
%A Horasan, Fahrettin
%Y Oflazer, Kemal
%Y Köksal, Abdullatif
%Y Varol, Onur
%S Proceedings of the Second Workshop Natural Language Processing for Turkic Languages (SIGTURK 2026)
%D 2026
%8 March
%I Association for Computational Linguistics
%C Rabat, Morocco
%@ 979-8-89176-370-8
%F yagiz-horasan-2026-beyond
%X Current interpretability methods for Large Language Models (LLMs) operate on a fundamental yet flawed assumption: that subword tokens represent independent semantic units. We prove that this assumption creates a fidelity bottleneck in Morphologically Rich Languages (MRLs), where semantic meaning is densely encoded in sub-token morphemes. We term this phenomenon the Tokenization-Morphology Misalignment (TMM). To resolve TMM, we introduce MAFEX (Morpheme-Aligned Faithful Explanations), a theoretically grounded framework that redefines feature attribution as a linear projection from the computational (token) basis to the linguistic (morpheme) basis. We evaluate our method on a diverse suite of Turkish LLMs, including BERTurk, BERTurk-Sentiment, Cosmos-BERT, and Kumru-2B. On our embedded benchmark (N=20), MAFEX achieves an average F1@1 of 91.25% compared to 13.75% for standard token-level baselines (IG, SHAP, DeepLIFT), representing a +77.5% absolute improvement, establishing it as the new standard for faithful multilingual interpretability.
%U https://aclanthology.org/2026.sigturk-1.19/
%P 228-235

Markdown (Informal)

[Beyond the Token: Correcting the Tokenization Bias in XAI via Morphologically-Aligned Projection](https://aclanthology.org/2026.sigturk-1.19/) (Yagiz & Horasan, SIGTURK 2026)

• Beyond the Token: Correcting the Tokenization Bias in XAI via Morphologically-Aligned Projection (Yagiz & Horasan, SIGTURK 2026)

ACL

• Muhammet Anil Yagiz and Fahrettin Horasan. 2026. Beyond the Token: Correcting the Tokenization Bias in XAI via Morphologically-Aligned Projection. In Proceedings of the Second Workshop Natural Language Processing for Turkic Languages (SIGTURK 2026), pages 228–235, Rabat, Morocco. Association for Computational Linguistics.