<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yong-etal-2026-controllable">
    <titleInfo>
        <title>Controllable Affective Generation via Latent Vector Steering</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Xixian</namePart>
        <namePart type="family">Yong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Siyuan</namePart>
        <namePart type="family">Chang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yingying</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xian</namePart>
        <namePart type="family">Wu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xiao</namePart>
        <namePart type="family">Zhou</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>Large Language Models (LLMs) often produce emotionally flattened responses after alignment, limiting their effectiveness in affect-sensitive applications. In this paper, we propose EmoVec, a lightweight framework for controllable affective generation via latent vector steering. EmoVec extracts emotion-specific directions from paired neutral and emotion-conditioned responses using contrastive activation addition, and further purifies them through task-specific debiasing and principal subspace removal. During inference, these vectors are injected into the final residual stream with static or scenario-adaptive scaling, enabling continuous control over emotional intensity without updating model weights. Experiments across three LLMs and eight emotions show that EmoVec consistently improves emotional salience while largely preserving semantic content, fluency, and coherence. Ablation studies and human evaluation further confirm the effectiveness of vector purification and adaptive scaling, demonstrating EmoVec as a practical inference-time method for affective control in deployed LLMs.</abstract>
    <identifier type="citekey">yong-etal-2026-controllable</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1707/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>36494</start>
            <end>36510</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Controllable Affective Generation via Latent Vector Steering
%A Yong, Xixian
%A Chang, Siyuan
%A Zhang, Yingying
%A Wu, Xian
%A Zhou, Xiao
%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 yong-etal-2026-controllable
%X Large Language Models (LLMs) often produce emotionally flattened responses after alignment, limiting their effectiveness in affect-sensitive applications. In this paper, we propose EmoVec, a lightweight framework for controllable affective generation via latent vector steering. EmoVec extracts emotion-specific directions from paired neutral and emotion-conditioned responses using contrastive activation addition, and further purifies them through task-specific debiasing and principal subspace removal. During inference, these vectors are injected into the final residual stream with static or scenario-adaptive scaling, enabling continuous control over emotional intensity without updating model weights. Experiments across three LLMs and eight emotions show that EmoVec consistently improves emotional salience while largely preserving semantic content, fluency, and coherence. Ablation studies and human evaluation further confirm the effectiveness of vector purification and adaptive scaling, demonstrating EmoVec as a practical inference-time method for affective control in deployed LLMs.
%U https://aclanthology.org/2026.findings-emnlp.1707/
%P 36494-36510

Markdown (Informal)

[Controllable Affective Generation via Latent Vector Steering](https://aclanthology.org/2026.findings-emnlp.1707/) (Yong et al., Findings 2026)

• Controllable Affective Generation via Latent Vector Steering (Yong et al., Findings 2026)

ACL

• Xixian Yong, Siyuan Chang, Yingying Zhang, Xian Wu, and Xiao Zhou. 2026. Controllable Affective Generation via Latent Vector Steering. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 36494–36510, Budapest, Hungary. Association for Computational Linguistics.