Dataset Card for EU Acts: Ukrainian → EU Languages (le-llm/EU-acts-uk-to-langs)

Dataset Description

Dataset Summary

EU-acts-uk-to-langs is a sentence- and segment-aligned parallel corpus of EU legal acts, pairing Ukrainian (uk) source segments with their counterparts in the 24 official EU languages. The corpus is derived from TMX memories built from: (a) translations of EU acts published at the official portal of the Parliament of Ukraine, and (b) EU acts available on EUR-Lex. The resource originates from ELRC-SHARE “EU acts in Ukrainian” and is redistributed under CC BY 4.0.

Languages

• Source: Ukrainian (uk)
• Targets: the official EU languages listed below (24 total including en).

Dataset Structure

Data Fields

Each record is one aligned segment pair:

• source_lang: always uk
• target_lang: one of the 24 EU language codes listed below
• text: template: "Український текст:\n\n{{uk_text}}\n\nПортугальський текст:\n\n{{pt_text}}

Language code -> Ukrainian display name

LANG_NAME_UK = { "bg": "Болгарський", "cs": "Чеський", "da": "Данський", "de": "Німецький", "el": "Грецький", "en": "Англійський", "es": "Іспанський", "et": "Естонський", "fi": "Фінський", "fr": "Французький", "ga": "Ірландський", "hr": "Хорватський", "hu": "Угорський", "it": "Італійський", "lt": "Литовський", "lv": "Латвійський", "mt": "Мальтійський", "nl": "Нідерландський", "pl": "Польський", "pt": "Португальський", "ro": "Румунський", "sk": "Словацький", "sl": "Словенський", "sv": "Шведський", # Add any others you see in your TMX folder: "cs_CZ": "Чеський", "de_DE": "Німецький", "en_GB": "Англійський", "en_US": "Англійський", "el_GR": "Грецький", "pt_PT": "Португальський", "pt_BR": "Португальський", }

Splits

• train: full corpus (3,056,791 aligned pairs across all target languages).

Language coverage and pair counts

Code	Language	Segment count (pairs)
bg	Bulgarian	133,031
cs	Czech	130,003
da	Danish	129,504
de	German	129,623
el	Greek	131,428
en	English	129,941
es	Spanish	131,406
et	Estonian	130,220
fi	Finnish	129,311
fr	French	130,778
ga	Irish	94,399
hr	Croatian	133,291
hu	Hungarian	130,703
it	Italian	131,001
lt	Lithuanian	130,075
lv	Latvian	131,147
mt	Maltese	83,812
nl	Dutch	130,337
pl	Polish	131,572
pt	Portuguese	130,936
ro	Romanian	132,471
sk	Slovak	131,336
sl	Slovenian	131,131
sv	Swedish	129,335

Dataset Creation

Source Data

• Original TMX compilation: ELRC-SHARE (3,056,791 TUs).

Intended Uses

• Training and evaluation for machine translation from Ukrainian to EU languages.
• Domain adaptation for legal/administrative translation.
• Research on multilingual legal translation and terminology.

How to Use

from datasets import load_dataset

ds = load_dataset("le-llm/EU-acts-uk-to-langs", split="train")

# Filter to a specific target language, e.g., Polish
pl = ds.filter(lambda x: x["target_lang"] == "pl")
print(pl[0])

Citation

• ELRC-SHARE (CC-BY-SA-4.0).

• BibTeX

@dataset{EU-acts-langs-to-uk,
  title   = {EU Acts: Ukrainian → EU Languages },
  author  = {lapa-llm},
  year    = {2025},
  url     = {https://huggingface.co/datasets/lapa-llm/EU-acts-uk-to-langs}
}

License

CC-BY-SA-4.0

This dataset is part of the "Lapa" - Ukrainian LLM initiative to advance natural language processing for the Ukrainian language.