<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="khan-etal-2026-translation">
    <titleInfo>
        <title>Translation Circuit Discovery and Targeted Optimization for Low-Resource Multilingual Translation</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Atai</namePart>
        <namePart type="given">Waris</namePart>
        <namePart type="family">Khan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Abdullah</namePart>
        <namePart type="family">Mazhar</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Raghu</namePart>
        <namePart type="family">Kumar</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Md</namePart>
        <namePart type="given">Shad</namePart>
        <namePart type="family">Akhtar</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>Recent multilingual large language models achieve strong translation performance, yet the internal mechanisms responsible for low-resource multilingual translation remain insufficiently understood. In this paper, we study translation circuits in multilingual decoder-only LLMs for Hindi→Urdu and Hindi→Sindhi (Devanagari). We propose TITAN, a mechanistic framework that combines causal intervention, sparse feature decomposition, and targeted preference optimization to identify and optimize translation crucial transformer components. Unlike conventional full-parameter fine-tuning, TITAN restricts optimization to sparse causally important circuits, enabling scalable and parameter-efficient adaptation while updating fewer than 5% of model parameters. Experimental results show that TITAN achieves BLEU improvements of +33.75% on Hindi→Urdu and +45.86% on Hindi→Sindhi over the strongest multilingual baselines. Human expert evaluation further confirms substantial gains in adequacy and fluency across both language pairs. Additional analyses reveal that multilingual translation is implemented through sparse structured circuits operating over a shared semantic subspace rather than through an English centric pivot representation. Overall, our findings demonstrate that mechanistic circuit discovery can directly improve low-resource multilingual translation while providing interpretable insights into multilingual transformer behavior.</abstract>
    <identifier type="citekey">khan-etal-2026-translation</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1348/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>28626</start>
            <end>28643</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Translation Circuit Discovery and Targeted Optimization for Low-Resource Multilingual Translation
%A Khan, Atai Waris
%A Mazhar, Abdullah
%A Kumar, Raghu
%A Akhtar, Md Shad
%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 khan-etal-2026-translation
%X Recent multilingual large language models achieve strong translation performance, yet the internal mechanisms responsible for low-resource multilingual translation remain insufficiently understood. In this paper, we study translation circuits in multilingual decoder-only LLMs for Hindi→Urdu and Hindi→Sindhi (Devanagari). We propose TITAN, a mechanistic framework that combines causal intervention, sparse feature decomposition, and targeted preference optimization to identify and optimize translation crucial transformer components. Unlike conventional full-parameter fine-tuning, TITAN restricts optimization to sparse causally important circuits, enabling scalable and parameter-efficient adaptation while updating fewer than 5% of model parameters. Experimental results show that TITAN achieves BLEU improvements of +33.75% on Hindi→Urdu and +45.86% on Hindi→Sindhi over the strongest multilingual baselines. Human expert evaluation further confirms substantial gains in adequacy and fluency across both language pairs. Additional analyses reveal that multilingual translation is implemented through sparse structured circuits operating over a shared semantic subspace rather than through an English centric pivot representation. Overall, our findings demonstrate that mechanistic circuit discovery can directly improve low-resource multilingual translation while providing interpretable insights into multilingual transformer behavior.
%U https://aclanthology.org/2026.findings-emnlp.1348/
%P 28626-28643

Markdown (Informal)

[Translation Circuit Discovery and Targeted Optimization for Low-Resource Multilingual Translation](https://aclanthology.org/2026.findings-emnlp.1348/) (Khan et al., Findings 2026)

• Translation Circuit Discovery and Targeted Optimization for Low-Resource Multilingual Translation (Khan et al., Findings 2026)

ACL

• Atai Waris Khan, Abdullah Mazhar, Raghu Kumar, and Md Shad Akhtar. 2026. Translation Circuit Discovery and Targeted Optimization for Low-Resource Multilingual Translation. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 28626–28643, Budapest, Hungary. Association for Computational Linguistics.