<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="fang-etal-2026-emotions">
    <titleInfo>
        <title>Some Emotions Run Deeper: Layer-wise Probing and Causal Intervention in Large Language Models</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Tian</namePart>
        <namePart type="family">Fang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Gaël</namePart>
        <namePart type="family">Guibon</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Davide</namePart>
        <namePart type="family">Buscaldi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>Emotion is expressed in text along a wide spectrum, from surface lexical cues to inferences entangled with content. Most layer-wise analyses of emotion in LLMs use a single corpus, leaving open whether the depth at which emotion becomes accessible is a property of the model or also of the text source. We investigate this across three datasets spanning different degrees of explicitness and contextualization in emotion expression (Twitter posts, Reddit comments, and autobiographical narratives) and eight 1B–9B open-weight LLMs from the Llama, Qwen, and Granite families. We combine layer-wise probing with offline feature scaling and online forward interventions, transfer analyses, and an early-exit classifier. We find that (i) the best probing layer shifts systematically across corpora, from input-adjacent layers to over half model depth, and this ordering persists after matching label-by-length-bin distributions; (ii)across the evaluated settings, forward-pass interventions on probe-selected bands reduce test accuracy by 5–6 points more than same-width random bands (q &lt; 0.01); (iii) selected bands transfer across datasets and emotion categories, suggesting partially shared affective information rather than strictly per-emotion substrates; and (iv) probe-selected early-exit representations outperform full-depth exits by 6.9 percentage points on average.</abstract>
    <identifier type="citekey">fang-etal-2026-emotions</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.421/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>8858</start>
            <end>8876</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Some Emotions Run Deeper: Layer-wise Probing and Causal Intervention in Large Language Models
%A Fang, Tian
%A Guibon, Gaël
%A Buscaldi, Davide
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F fang-etal-2026-emotions
%X Emotion is expressed in text along a wide spectrum, from surface lexical cues to inferences entangled with content. Most layer-wise analyses of emotion in LLMs use a single corpus, leaving open whether the depth at which emotion becomes accessible is a property of the model or also of the text source. We investigate this across three datasets spanning different degrees of explicitness and contextualization in emotion expression (Twitter posts, Reddit comments, and autobiographical narratives) and eight 1B–9B open-weight LLMs from the Llama, Qwen, and Granite families. We combine layer-wise probing with offline feature scaling and online forward interventions, transfer analyses, and an early-exit classifier. We find that (i) the best probing layer shifts systematically across corpora, from input-adjacent layers to over half model depth, and this ordering persists after matching label-by-length-bin distributions; (ii)across the evaluated settings, forward-pass interventions on probe-selected bands reduce test accuracy by 5–6 points more than same-width random bands (q < 0.01); (iii) selected bands transfer across datasets and emotion categories, suggesting partially shared affective information rather than strictly per-emotion substrates; and (iv) probe-selected early-exit representations outperform full-depth exits by 6.9 percentage points on average.
%U https://aclanthology.org/2026.findings-emnlp.421/
%P 8858-8876

Markdown (Informal)

[Some Emotions Run Deeper: Layer-wise Probing and Causal Intervention in Large Language Models](https://aclanthology.org/2026.findings-emnlp.421/) (Fang et al., Findings 2026)

• Some Emotions Run Deeper: Layer-wise Probing and Causal Intervention in Large Language Models (Fang et al., Findings 2026)

ACL

• Tian Fang, Gaël Guibon, and Davide Buscaldi. 2026. Some Emotions Run Deeper: Layer-wise Probing and Causal Intervention in Large Language Models. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 8858–8876, Budapest, Hungary. Association for Computational Linguistics.