<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="crosbie-shutova-2025-induction">
    <titleInfo>
        <title>Induction Heads as an Essential Mechanism for Pattern Matching in In-context Learning</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Joy</namePart>
        <namePart type="family">Crosbie</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ekaterina</namePart>
        <namePart type="family">Shutova</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2025-04</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: NAACL 2025</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Luis</namePart>
            <namePart type="family">Chiruzzo</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Alan</namePart>
            <namePart type="family">Ritter</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Lu</namePart>
            <namePart type="family">Wang</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Albuquerque, New Mexico</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-195-7</identifier>
    </relatedItem>
    <abstract>Large language models (LLMs) have shown a remarkable ability to learn and perform complex tasks through in-context learning (ICL). However, a comprehensive understanding of its internal mechanisms is still lacking. This paper explores the role of induction heads in a few-shot ICL setting. We analyse two state-of-the-art models, Llama-3-8B and InternLM2-20B on abstract pattern recognition and NLP tasks. Our results show that even a minimal ablation of induction heads leads to ICL performance decreases of up to ~32% for abstract pattern recognition tasks, bringing the performance close to random. For NLP tasks, this ablation substantially decreases the model’s ability to benefit from examples, bringing few-shot ICL performance close to that of zero-shot prompts. We further use attention knockout to disable specific induction patterns, and present fine-grained evidence for the role that the induction mechanism plays in ICL.</abstract>
    <identifier type="citekey">crosbie-shutova-2025-induction</identifier>
    <identifier type="doi">10.18653/v1/2025.findings-naacl.283</identifier>
    <location>
        <url>https://aclanthology.org/2025.findings-naacl.283/</url>
    </location>
    <part>
        <date>2025-04</date>
        <extent unit="page">
            <start>5034</start>
            <end>5096</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Induction Heads as an Essential Mechanism for Pattern Matching in In-context Learning
%A Crosbie, Joy
%A Shutova, Ekaterina
%Y Chiruzzo, Luis
%Y Ritter, Alan
%Y Wang, Lu
%S Findings of the Association for Computational Linguistics: NAACL 2025
%D 2025
%8 April
%I Association for Computational Linguistics
%C Albuquerque, New Mexico
%@ 979-8-89176-195-7
%F crosbie-shutova-2025-induction
%X Large language models (LLMs) have shown a remarkable ability to learn and perform complex tasks through in-context learning (ICL). However, a comprehensive understanding of its internal mechanisms is still lacking. This paper explores the role of induction heads in a few-shot ICL setting. We analyse two state-of-the-art models, Llama-3-8B and InternLM2-20B on abstract pattern recognition and NLP tasks. Our results show that even a minimal ablation of induction heads leads to ICL performance decreases of up to ~32% for abstract pattern recognition tasks, bringing the performance close to random. For NLP tasks, this ablation substantially decreases the model’s ability to benefit from examples, bringing few-shot ICL performance close to that of zero-shot prompts. We further use attention knockout to disable specific induction patterns, and present fine-grained evidence for the role that the induction mechanism plays in ICL.
%R 10.18653/v1/2025.findings-naacl.283
%U https://aclanthology.org/2025.findings-naacl.283/
%U https://doi.org/10.18653/v1/2025.findings-naacl.283
%P 5034-5096

Markdown (Informal)

[Induction Heads as an Essential Mechanism for Pattern Matching in In-context Learning](https://aclanthology.org/2025.findings-naacl.283/) (Crosbie & Shutova, Findings 2025)

• Induction Heads as an Essential Mechanism for Pattern Matching in In-context Learning (Crosbie & Shutova, Findings 2025)

ACL

• Joy Crosbie and Ekaterina Shutova. 2025. Induction Heads as an Essential Mechanism for Pattern Matching in In-context Learning. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 5034–5096, Albuquerque, New Mexico. Association for Computational Linguistics.