<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="min-etal-2026-probing">
    <titleInfo>
        <title>Probing Reasoning Length Control in Language Models via Neurons</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Dongwook</namePart>
        <namePart type="family">Min</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jeonhwi</namePart>
        <namePart type="family">Kang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Cheoneum</namePart>
        <namePart type="family">Park</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Seung-Hoon</namePart>
        <namePart type="family">Na</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Yue</namePart>
            <namePart type="family">Zhang</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Ruifeng</namePart>
            <namePart type="family">Xu</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Ayu</namePart>
            <namePart type="family">Purwarianti</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Tong</namePart>
            <namePart type="family">Rong</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jun</namePart>
            <namePart type="family">Suzuki</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Hengqin, China</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-444-6</identifier>
    </relatedItem>
    <abstract>Reasoning-oriented language models can generate thousands of tokens before answering, yet the internal factors that determine their reasoning length remain poorly understood. This paper studies whether reasoning length can be controlled through sparse neuron interventions, complementing prompt-based, budget-based, and controller-based approaches. The proposed method discovers reasoning-length-associated neurons from paired long and concise correct trajectories on a training split, ranks MLP neurons by long-minus-concise activation contrast, and tests the selected neurons on disjoint evaluation splits by scaling their activations during generation. The evaluation covers math, science, and commonsense reasoning tasks, including GSM8K, MATH500, ARC-Challenge, CommonsenseQA, and QASC, under a strict split-separated protocol that never discovers neurons on held-out evaluation examples. In several, but not all, task/model/budget regimes, selected neurons modulate generated length differently from layer-matched random controls, with a modest single-run specificity result on CommonsenseQA accompanied by a small accuracy increase. These results provide task-, model-, and prompt-protocol-specific evidence that sparse MLP neurons can, in some split-controlled regimes, provide neuron-level control signals for generated reasoning length.</abstract>
    <identifier type="citekey">min-etal-2026-probing</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-aacl.244/</url>
    </location>
    <part>
        <date>2026-11</date>
        <extent unit="page">
            <start>4791</start>
            <end>4808</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Probing Reasoning Length Control in Language Models via Neurons
%A Min, Dongwook
%A Kang, Jeonhwi
%A Park, Cheoneum
%A Na, Seung-Hoon
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F min-etal-2026-probing
%X Reasoning-oriented language models can generate thousands of tokens before answering, yet the internal factors that determine their reasoning length remain poorly understood. This paper studies whether reasoning length can be controlled through sparse neuron interventions, complementing prompt-based, budget-based, and controller-based approaches. The proposed method discovers reasoning-length-associated neurons from paired long and concise correct trajectories on a training split, ranks MLP neurons by long-minus-concise activation contrast, and tests the selected neurons on disjoint evaluation splits by scaling their activations during generation. The evaluation covers math, science, and commonsense reasoning tasks, including GSM8K, MATH500, ARC-Challenge, CommonsenseQA, and QASC, under a strict split-separated protocol that never discovers neurons on held-out evaluation examples. In several, but not all, task/model/budget regimes, selected neurons modulate generated length differently from layer-matched random controls, with a modest single-run specificity result on CommonsenseQA accompanied by a small accuracy increase. These results provide task-, model-, and prompt-protocol-specific evidence that sparse MLP neurons can, in some split-controlled regimes, provide neuron-level control signals for generated reasoning length.
%U https://aclanthology.org/2026.findings-aacl.244/
%P 4791-4808

Markdown (Informal)

[Probing Reasoning Length Control in Language Models via Neurons](https://aclanthology.org/2026.findings-aacl.244/) (Min et al., Findings 2026)

• Probing Reasoning Length Control in Language Models via Neurons (Min et al., Findings 2026)

ACL

• Dongwook Min, Jeonhwi Kang, Cheoneum Park, and Seung-Hoon Na. 2026. Probing Reasoning Length Control in Language Models via Neurons. In Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026, pages 4791–4808, Hengqin, China. Association for Computational Linguistics.