<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bakos-etal-2026-pre">
    <titleInfo>
        <title>A Pre-Generation Epistemic Routing Axis in Language Models</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Steve</namePart>
        <namePart type="family">Bakos</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Heidar</namePart>
        <namePart type="family">Davoudi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Aijun</namePart>
        <namePart type="family">An</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>When answering factual questions, language models may respond correctly, hallucinate, or abstain. We investigate whether this epistemic disposition is encoded in hidden states before any answer token is generated. Using a balanced Wikipedia QA corpus with pre/post article-date splits and rollout-consensus labels, we analyze prompt-only MLP activations from four instruction-tuned 7–9B models. PCA reveals a dominant pre-generation epistemic routing axis that orders outcomes as ABSTAIN, HALLUCINATION, and CORRECT. We then apply activation steering as a bounded causal stress test. Dense residual-write edits can shift many hallucinations toward abstention or correctness, while sparse MLP-intermediate edits localize the steering signal. These results frame the axis as a steering audit: it identifies where fixed edits improve factual behavior and where correct loss or broken outputs call for conditional controllers.</abstract>
    <identifier type="citekey">bakos-etal-2026-pre</identifier>
    <identifier type="doi">10.18653/2026.findings-emnlp.1324</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1324/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>28066</start>
            <end>28091</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T A Pre-Generation Epistemic Routing Axis in Language Models
%A Bakos, Steve
%A Davoudi, Heidar
%A An, Aijun
%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 bakos-etal-2026-pre
%X When answering factual questions, language models may respond correctly, hallucinate, or abstain. We investigate whether this epistemic disposition is encoded in hidden states before any answer token is generated. Using a balanced Wikipedia QA corpus with pre/post article-date splits and rollout-consensus labels, we analyze prompt-only MLP activations from four instruction-tuned 7–9B models. PCA reveals a dominant pre-generation epistemic routing axis that orders outcomes as ABSTAIN, HALLUCINATION, and CORRECT. We then apply activation steering as a bounded causal stress test. Dense residual-write edits can shift many hallucinations toward abstention or correctness, while sparse MLP-intermediate edits localize the steering signal. These results frame the axis as a steering audit: it identifies where fixed edits improve factual behavior and where correct loss or broken outputs call for conditional controllers.
%R 10.18653/2026.findings-emnlp.1324
%U https://aclanthology.org/2026.findings-emnlp.1324/
%U https://doi.org/10.18653/2026.findings-emnlp.1324
%P 28066-28091

Markdown (Informal)

[A Pre-Generation Epistemic Routing Axis in Language Models](https://aclanthology.org/2026.findings-emnlp.1324/) (Bakos et al., Findings 2026)

• A Pre-Generation Epistemic Routing Axis in Language Models (Bakos et al., Findings 2026)

ACL

• Steve Bakos, Heidar Davoudi, and Aijun An. 2026. A Pre-Generation Epistemic Routing Axis in Language Models. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 28066–28091, Budapest, Hungary. Association for Computational Linguistics.