@inproceedings{ziwei-2026-hy,
title = "{H}y-{MT}2-1.8{B} for {WMT}26: Staged Multilingual Adaptation for {C}hinese{--}{S}outheast {A}sian Translation",
author = "Ziwei, Ma",
editor = "Haddow, Barry and
Kocmi, Tom and
Koehn, Philipp and
Monz, Christof",
booktitle = "Proceedings of the Eleventh Conference on Machine Translation",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wmt-main.184/",
pages = "2397--2404",
ISBN = "979-8-89176-474-3",
abstract = "We present a Hy-MT2-1.8B system for the WMT26 Chinese{--}Southeast Asian translation task, covering fourteen translation directions between Chinese and seven Southeast Asian languages. The system uses a staged adaptation pipeline consisting of monolingual continued pretraining followed by bilingual supervised fine-tuning. We compare parameter-efficient LoRA SFT with full-parameter SFT from the same continued-pretraining checkpoint under matched data and evaluation conditions. On 26,474 validation examples, full-parameter SFT achieves higher macro-average sacreBLEU and COMET scores than LoRA SFT, although the gains vary across language directions. The implementation separates data construction, model preparation, quality evaluation, and serving so that the submitted system can be reproduced and evaluated under the shared-task conditions."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ziwei-2026-hy">
<titleInfo>
<title>Hy-MT2-1.8B for WMT26: Staged Multilingual Adaptation for Chinese–Southeast Asian Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ma</namePart>
<namePart type="family">Ziwei</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 a Hy-MT2-1.8B system for the WMT26 Chinese–Southeast Asian translation task, covering fourteen translation directions between Chinese and seven Southeast Asian languages. The system uses a staged adaptation pipeline consisting of monolingual continued pretraining followed by bilingual supervised fine-tuning. We compare parameter-efficient LoRA SFT with full-parameter SFT from the same continued-pretraining checkpoint under matched data and evaluation conditions. On 26,474 validation examples, full-parameter SFT achieves higher macro-average sacreBLEU and COMET scores than LoRA SFT, although the gains vary across language directions. The implementation separates data construction, model preparation, quality evaluation, and serving so that the submitted system can be reproduced and evaluated under the shared-task conditions.</abstract>
<identifier type="citekey">ziwei-2026-hy</identifier>
<location>
<url>https://aclanthology.org/2026.wmt-main.184/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2397</start>
<end>2404</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Hy-MT2-1.8B for WMT26: Staged Multilingual Adaptation for Chinese–Southeast Asian Translation
%A Ziwei, Ma
%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 ziwei-2026-hy
%X We present a Hy-MT2-1.8B system for the WMT26 Chinese–Southeast Asian translation task, covering fourteen translation directions between Chinese and seven Southeast Asian languages. The system uses a staged adaptation pipeline consisting of monolingual continued pretraining followed by bilingual supervised fine-tuning. We compare parameter-efficient LoRA SFT with full-parameter SFT from the same continued-pretraining checkpoint under matched data and evaluation conditions. On 26,474 validation examples, full-parameter SFT achieves higher macro-average sacreBLEU and COMET scores than LoRA SFT, although the gains vary across language directions. The implementation separates data construction, model preparation, quality evaluation, and serving so that the submitted system can be reproduced and evaluated under the shared-task conditions.
%U https://aclanthology.org/2026.wmt-main.184/
%P 2397-2404
Markdown (Informal)
[Hy-MT2-1.8B for WMT26: Staged Multilingual Adaptation for Chinese–Southeast Asian Translation](https://aclanthology.org/2026.wmt-main.184/) (Ziwei, WMT 2026)
ACL