@inproceedings{zundanovic-etal-2026-croatian,
title = "The {C}roatian Dataset Seed Submission to the {WMT}26 Open Language Data Initiative Shared Task",
author = "Zundanovi{\'c}, Dragana and
Leventi{\'c}, Hrvoje and
Bo{\v{z}}i{\'c} Lenard, Dragana and
Romi{\'c}, Kre{\v{s}}imir",
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.201/",
pages = "2571--2583",
ISBN = "979-8-89176-474-3",
abstract = "We present a new English{--}Croatian parallel corpus for the WMT26 Open Language Data Initiative (OLDI) Seed shared task, consisting of 6,193 sentence pairs. To ensure high quality, the resource was produced via machine translation followed by a two-stage verification process (post-editing and independent revision) conducted by professional translators. We validate the corpus by fine-tuning NLLB-200, TranslateGemma, and Gemma-4-31B. The dataset improves every dedicated translation model in both directions and the strongest open LLMs in the Croatian{\textrightarrow}English direction. Through a training-data ablation, we demonstrate that the benefit of human verification increases with model capacity and is visible only to neural metrics, not to surfaceoverlap metrics such as chrF++. Finally, we analyze the quality ceiling for the strongest models, identifying a trade-off between naturalness and surface overlap in the English{\textrightarrow}Croatian direction."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zundanovic-etal-2026-croatian">
<titleInfo>
<title>The Croatian Dataset Seed Submission to the WMT26 Open Language Data Initiative Shared Task</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dragana</namePart>
<namePart type="family">Zundanović</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hrvoje</namePart>
<namePart type="family">Leventić</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dragana</namePart>
<namePart type="family">Božić Lenard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Krešimir</namePart>
<namePart type="family">Romić</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 new English–Croatian parallel corpus for the WMT26 Open Language Data Initiative (OLDI) Seed shared task, consisting of 6,193 sentence pairs. To ensure high quality, the resource was produced via machine translation followed by a two-stage verification process (post-editing and independent revision) conducted by professional translators. We validate the corpus by fine-tuning NLLB-200, TranslateGemma, and Gemma-4-31B. The dataset improves every dedicated translation model in both directions and the strongest open LLMs in the Croatian→English direction. Through a training-data ablation, we demonstrate that the benefit of human verification increases with model capacity and is visible only to neural metrics, not to surfaceoverlap metrics such as chrF++. Finally, we analyze the quality ceiling for the strongest models, identifying a trade-off between naturalness and surface overlap in the English→Croatian direction.</abstract>
<identifier type="citekey">zundanovic-etal-2026-croatian</identifier>
<location>
<url>https://aclanthology.org/2026.wmt-main.201/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2571</start>
<end>2583</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T The Croatian Dataset Seed Submission to the WMT26 Open Language Data Initiative Shared Task
%A Zundanović, Dragana
%A Leventić, Hrvoje
%A Božić Lenard, Dragana
%A Romić, Krešimir
%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 zundanovic-etal-2026-croatian
%X We present a new English–Croatian parallel corpus for the WMT26 Open Language Data Initiative (OLDI) Seed shared task, consisting of 6,193 sentence pairs. To ensure high quality, the resource was produced via machine translation followed by a two-stage verification process (post-editing and independent revision) conducted by professional translators. We validate the corpus by fine-tuning NLLB-200, TranslateGemma, and Gemma-4-31B. The dataset improves every dedicated translation model in both directions and the strongest open LLMs in the Croatian→English direction. Through a training-data ablation, we demonstrate that the benefit of human verification increases with model capacity and is visible only to neural metrics, not to surfaceoverlap metrics such as chrF++. Finally, we analyze the quality ceiling for the strongest models, identifying a trade-off between naturalness and surface overlap in the English→Croatian direction.
%U https://aclanthology.org/2026.wmt-main.201/
%P 2571-2583
Markdown (Informal)
[The Croatian Dataset Seed Submission to the WMT26 Open Language Data Initiative Shared Task](https://aclanthology.org/2026.wmt-main.201/) (Zundanović et al., WMT 2026)
ACL