<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chandler-etal-2025-deloitte">
    <titleInfo>
        <title>Deloitte (Drocks) at SemEval-2025 Task 3: Fine-Grained Multi-lingual Hallucination Detection Using Internal LLM Weights</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Alex</namePart>
        <namePart type="family">Chandler</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Harika</namePart>
        <namePart type="family">Abburi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sanmitra</namePart>
        <namePart type="family">Bhattacharya</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Edward</namePart>
        <namePart type="family">Bowen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Nirmala</namePart>
        <namePart type="family">Pudota</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2025-07</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Sara</namePart>
            <namePart type="family">Rosenthal</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Aiala</namePart>
            <namePart type="family">Rosá</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Debanjan</namePart>
            <namePart type="family">Ghosh</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Marcos</namePart>
            <namePart type="family">Zampieri</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Vienna, Austria</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-273-2</identifier>
    </relatedItem>
    <abstract>Large Language Models (LLMs) have greatly advanced the field of Natural Language Generation (NLG). Despite their remarkable capabilities, their tendency to hallucinate—producing inaccurate or misleading information-remains a barrier to wider adoption. Current hallucination detection methods mainly employ coarse-grained binary classification at the sentence or document level, overlooking the need for precise identification of the specific text spans containing hallucinations. In this paper, we proposed a methodology that generates supplementary context and processes text using an LLM to extract internal weights (features) from various layers. These extracted features serve as input for a neural network classifier designed to perform token-level binary detection of hallucinations. Subsequently, we map the resulting token-level predictions to character-level predictions, enabling the identification of spans of hallucinated text, which we refer to as hallucination spans. Our model achieved a top-ten ranking in 13 of the 14 languages and secured first place for the French language in the SemEval: Multilingual Shared-task on Hallucinations and Related Observable Overgeneration Mistakes (Mu-SHROOM), utilizing the Mu-SHROOM dataset provided by the task organizers.</abstract>
    <identifier type="citekey">chandler-etal-2025-deloitte</identifier>
    <location>
        <url>https://aclanthology.org/2025.semeval-1.144/</url>
    </location>
    <part>
        <date>2025-07</date>
        <extent unit="page">
            <start>1089</start>
            <end>1097</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Deloitte (Drocks) at SemEval-2025 Task 3: Fine-Grained Multi-lingual Hallucination Detection Using Internal LLM Weights
%A Chandler, Alex
%A Abburi, Harika
%A Bhattacharya, Sanmitra
%A Bowen, Edward
%A Pudota, Nirmala
%Y Rosenthal, Sara
%Y Rosá, Aiala
%Y Ghosh, Debanjan
%Y Zampieri, Marcos
%S Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-273-2
%F chandler-etal-2025-deloitte
%X Large Language Models (LLMs) have greatly advanced the field of Natural Language Generation (NLG). Despite their remarkable capabilities, their tendency to hallucinate—producing inaccurate or misleading information-remains a barrier to wider adoption. Current hallucination detection methods mainly employ coarse-grained binary classification at the sentence or document level, overlooking the need for precise identification of the specific text spans containing hallucinations. In this paper, we proposed a methodology that generates supplementary context and processes text using an LLM to extract internal weights (features) from various layers. These extracted features serve as input for a neural network classifier designed to perform token-level binary detection of hallucinations. Subsequently, we map the resulting token-level predictions to character-level predictions, enabling the identification of spans of hallucinated text, which we refer to as hallucination spans. Our model achieved a top-ten ranking in 13 of the 14 languages and secured first place for the French language in the SemEval: Multilingual Shared-task on Hallucinations and Related Observable Overgeneration Mistakes (Mu-SHROOM), utilizing the Mu-SHROOM dataset provided by the task organizers.
%U https://aclanthology.org/2025.semeval-1.144/
%P 1089-1097

Markdown (Informal)

[Deloitte (Drocks) at SemEval-2025 Task 3: Fine-Grained Multi-lingual Hallucination Detection Using Internal LLM Weights](https://aclanthology.org/2025.semeval-1.144/) (Chandler et al., SemEval 2025)

• Deloitte (Drocks) at SemEval-2025 Task 3: Fine-Grained Multi-lingual Hallucination Detection Using Internal LLM Weights (Chandler et al., SemEval 2025)

ACL

• Alex Chandler, Harika Abburi, Sanmitra Bhattacharya, Edward Bowen, and Nirmala Pudota. 2025. Deloitte (Drocks) at SemEval-2025 Task 3: Fine-Grained Multi-lingual Hallucination Detection Using Internal LLM Weights. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 1089–1097, Vienna, Austria. Association for Computational Linguistics.