<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yudhistira-etal-2026-rupiah">
    <titleInfo>
        <title>~rupiah at WMT26: Training-Free Context-Aware Subtitle Translation with QE Selection</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Pieter</namePart>
        <namePart type="family">Yudhistira</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sayyidah</namePart>
        <namePart type="family">Azzahra</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Dzaki</namePart>
        <namePart type="family">Malik</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Tirana</namePart>
        <namePart type="given">Noor</namePart>
        <namePart type="family">Fatyanosa</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>Proceedings of the Eleventh Conference on Machine Translation</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Barry</namePart>
            <namePart type="family">Haddow</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Tom</namePart>
            <namePart type="family">Kocmi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Philipp</namePart>
            <namePart type="family">Koehn</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Christof</namePart>
            <namePart type="family">Monz</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">979-8-89176-474-3</identifier>
    </relatedItem>
    <abstract>We present the training-free subtitle translation system developed by Team ~rupiah for the WMT26 Video Subtitle Translation shared task. The system translates Chinese subtitles into English and Indonesian using Hy-MT2-7B while preserving the source cue indices and timestamps. To support discourse and terminology consistency, each cue is conditioned on episode metadata, preceding source cues, translation history, and an automatically constructed episode glossary. At inference time, the system generates candidates at six temperatures, removes exact duplicates, filters candidates containing Han characters, and selects a hypothesis using reference-free CometKiwi quality estimation. A deterministic lexical-recovery step then replaces glossary terms that remain untranslated in the selected hypothesis. In the preliminary official evaluation over 100 videos per direction, ~rupiah ranks fifth of eight systems for Chinese-to-English and third of four for Chinese-to-Indonesian. Rubric-level results indicate that the largest performance gap relative to the top-ranked system is in accuracy and fidelity, whereas the gaps in fluency and subtitle conventions are smaller. Reference-based scores over 89 videos per direction provide additional diagnostics of the complete pipeline. These aggregate evaluations do not isolate the contribution of individual components. Source code available and can be accessed here: https://github.com/Pieter414/-rupiah-training-free-video-translation-subtitle</abstract>
    <identifier type="citekey">yudhistira-etal-2026-rupiah</identifier>
    <location>
        <url>https://aclanthology.org/2026.wmt-main.206/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>2623</start>
            <end>2630</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T ~rupiah at WMT26: Training-Free Context-Aware Subtitle Translation with QE Selection
%A Yudhistira, Pieter
%A Azzahra, Sayyidah
%A Malik, Dzaki
%A Fatyanosa, Tirana Noor
%Y Haddow, Barry
%Y Kocmi, Tom
%Y Koehn, Philipp
%Y Monz, Christof
%S Proceedings of the Eleventh Conference on Machine Translation
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-474-3
%F yudhistira-etal-2026-rupiah
%X We present the training-free subtitle translation system developed by Team ~rupiah for the WMT26 Video Subtitle Translation shared task. The system translates Chinese subtitles into English and Indonesian using Hy-MT2-7B while preserving the source cue indices and timestamps. To support discourse and terminology consistency, each cue is conditioned on episode metadata, preceding source cues, translation history, and an automatically constructed episode glossary. At inference time, the system generates candidates at six temperatures, removes exact duplicates, filters candidates containing Han characters, and selects a hypothesis using reference-free CometKiwi quality estimation. A deterministic lexical-recovery step then replaces glossary terms that remain untranslated in the selected hypothesis. In the preliminary official evaluation over 100 videos per direction, ~rupiah ranks fifth of eight systems for Chinese-to-English and third of four for Chinese-to-Indonesian. Rubric-level results indicate that the largest performance gap relative to the top-ranked system is in accuracy and fidelity, whereas the gaps in fluency and subtitle conventions are smaller. Reference-based scores over 89 videos per direction provide additional diagnostics of the complete pipeline. These aggregate evaluations do not isolate the contribution of individual components. Source code available and can be accessed here: https://github.com/Pieter414/-rupiah-training-free-video-translation-subtitle
%U https://aclanthology.org/2026.wmt-main.206/
%P 2623-2630

Markdown (Informal)

[~rupiah at WMT26: Training-Free Context-Aware Subtitle Translation with QE Selection](https://aclanthology.org/2026.wmt-main.206/) (Yudhistira et al., WMT 2026)

• ~rupiah at WMT26: Training-Free Context-Aware Subtitle Translation with QE Selection (Yudhistira et al., WMT 2026)

ACL

• Pieter Yudhistira, Sayyidah Azzahra, Dzaki Malik, and Tirana Noor Fatyanosa. 2026. ~rupiah at WMT26: Training-Free Context-Aware Subtitle Translation with QE Selection. In Proceedings of the Eleventh Conference on Machine Translation, pages 2623–2630, Budapest, Hungary. Association for Computational Linguistics.