<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="beliga-etal-2026-injecting">
    <titleInfo>
        <title>Injecting Structured Lexicographic Knowledge into LLMs for Non-Literal Expression Disambiguation: A Controlled Study on Croatian</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Slobodan</namePart>
        <namePart type="family">Beliga</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ivana</namePart>
        <namePart type="family">Filipović Petrović</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ana</namePart>
        <namePart type="family">Meštrović</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 Learning Non-Literal Expressions with Small Data @ LREC 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Markus</namePart>
            <namePart type="family">Egg</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Valia</namePart>
            <namePart type="family">Kordoni</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>ELRA Language Resources Association (ELRA)</publisher>
            <place>
                <placeTerm type="text">Palma, Mallorca (Spain)</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>In potentially idiomatic expressions (PIEs), the same surface form may receive either a literal or an idiomatic interpretation depending on context, making automatic literal–idiomatic disambiguation challenging. This is acute for Croatian, where annotated data and locally runnable generative models are limited. We present a study of Croatian PIE literal–idiomatic disambiguation examining how structured lexicographic knowledge can improve open-weight, decoder-only LLMs without fine-tuning. Using a new expert-annotated concordance dataset – CroPIEs, we compare baseline prompting to inference-time knowledge injection via retrieval-augmented generation (RAG) from a Croatian phraseological dictionary. We isolate the contribution of three knowledge types: definitional knowledge (structured meanings), contextual knowledge as curated prototypical usage examples, and their combination. Results show consistent improvements in macro-F1 for both GaMS-2B-Instruct and GaMS-9B-Instruct models. Definitional knowledge is generally more stable than examples alone, while examples can be effective but less consistent across expressions. The strongest and most reliable gains are obtained when definitions and examples are combined, indicating a synergistic effect between explicit meaning descriptions and contextual cues. Per-class analyses show that injected lexicographic evidence mitigates baseline biases between Literal and Idiomatic predictions, improving decision balance in a low-resource setting with small data of compact, expert-curated lexicographic evidence injected at inference time.</abstract>
    <identifier type="citekey">beliga-etal-2026-injecting</identifier>
    <identifier type="doi">10.63317/4s8wcqht63fc</identifier>
    <location>
        <url>https://aclanthology.org/2026.nonliteral-1.3/</url>
    </location>
    <part>
        <date>2026-05</date>
        <extent unit="page">
            <start>21</start>
            <end>30</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Injecting Structured Lexicographic Knowledge into LLMs for Non-Literal Expression Disambiguation: A Controlled Study on Croatian
%A Beliga, Slobodan
%A Filipović Petrović, Ivana
%A Meštrović, Ana
%Y Egg, Markus
%Y Kordoni, Valia
%S Proceedings of Learning Non-Literal Expressions with Small Data @ LREC 2026
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F beliga-etal-2026-injecting
%X In potentially idiomatic expressions (PIEs), the same surface form may receive either a literal or an idiomatic interpretation depending on context, making automatic literal–idiomatic disambiguation challenging. This is acute for Croatian, where annotated data and locally runnable generative models are limited. We present a study of Croatian PIE literal–idiomatic disambiguation examining how structured lexicographic knowledge can improve open-weight, decoder-only LLMs without fine-tuning. Using a new expert-annotated concordance dataset – CroPIEs, we compare baseline prompting to inference-time knowledge injection via retrieval-augmented generation (RAG) from a Croatian phraseological dictionary. We isolate the contribution of three knowledge types: definitional knowledge (structured meanings), contextual knowledge as curated prototypical usage examples, and their combination. Results show consistent improvements in macro-F1 for both GaMS-2B-Instruct and GaMS-9B-Instruct models. Definitional knowledge is generally more stable than examples alone, while examples can be effective but less consistent across expressions. The strongest and most reliable gains are obtained when definitions and examples are combined, indicating a synergistic effect between explicit meaning descriptions and contextual cues. Per-class analyses show that injected lexicographic evidence mitigates baseline biases between Literal and Idiomatic predictions, improving decision balance in a low-resource setting with small data of compact, expert-curated lexicographic evidence injected at inference time.
%R 10.63317/4s8wcqht63fc
%U https://aclanthology.org/2026.nonliteral-1.3/
%U https://doi.org/10.63317/4s8wcqht63fc
%P 21-30

Markdown (Informal)

[Injecting Structured Lexicographic Knowledge into LLMs for Non-Literal Expression Disambiguation: A Controlled Study on Croatian](https://aclanthology.org/2026.nonliteral-1.3/) (Beliga et al., NonLiteral 2026)

• Injecting Structured Lexicographic Knowledge into LLMs for Non-Literal Expression Disambiguation: A Controlled Study on Croatian (Beliga et al., NonLiteral 2026)

ACL

• Slobodan Beliga, Ivana Filipović Petrović, and Ana Meštrović. 2026. Injecting Structured Lexicographic Knowledge into LLMs for Non-Literal Expression Disambiguation: A Controlled Study on Croatian. In Proceedings of Learning Non-Literal Expressions with Small Data @ LREC 2026, pages 21–30, Palma, Mallorca (Spain). ELRA Language Resources Association (ELRA).