This corpus was compiled as part of the AI-TraLow project, funded by MCIN/AEI. It aggregates approximately 97.8 million sentence pairs across 8 language pairs covering non-EU and migrant languages (Arabic, Chinese, Hindi, Korean, Japanese) and, among the EU official languages, the ones with lower resource availability (Estonian, Maltese, Irish). All these languages come paired with Spanish.

The dataset is intended primarily for training machine translation (MT) systems targeting less resourced European languages and migrant / minority languages in Spain and Europe. It consolidates 30 source datasets drawn from OPUS, ELG, the NLLB collection, the NTEU project, and other sources into a single unified large dataset (see Source Data section for the full list).

The data spans a wide range of domains including news, legal texts, medicine, software localisation, encyclopaedic content (Wikipedia), among others.

For each language pair approximately 15 million sentence pairs are provided, with the only exception of the following pairs, for which fewer resources were available:

Language Pair	Sentence Pairs
ES-MT	4,907,188
ES-GA	2,866,017

Supported Tasks and Leaderboards

Machine Translation: parallel sentence pairs for the language pairs ES-AR, ES-ZH, ES-HI, ES-KO, ES-JA, ES-ET, ES-MT, ES-GA.

Languages

The dataset covers 8 language pairs, all with Spanish (es) as one side:

Pair	Language	Category	Size (sentence pairs)
ES-AR	Arabic (`ar`)	Non-EU / migrant	14,991,369
ES-ZH	Chinese Simplified (`zh`)	Non-EU / migrant	14,996,924
ES-HI	Hindi (`hi`)	Non-EU / migrant	14,994,669
ES-KO	Korean (`ko`)	Non-EU / migrant	14,993,592
ES-JA	Japanese (`ja`)	Non-EU / migrant	14,996,750
ES-ET	Estonian (`et`)	EU official low-resource	15,027,899
ES-MT	Maltese (`mt`)	EU official low-resource	4,907,188
ES-GA	Irish (`ga`)	EU official low-resource	2,866,017
Total			97,774,408

Dataset Structure

The dataset is provided in parquet format, with a separate file for each language pair. For larger language pairs, several parquet files containing partial datasets may be provided.

Each row contains a parallel sentence pair with the following structure:

{
  "id": 0,
  "lang_pair": "es-ar",
  "source_lang": "es",
  "target_lang": "ar",
  "source": "...",
  "target": "..."
}

Data Fields

id: Unique numeric identifier for the sentence pair
lang_pair: Language pair ISO codes (e.g., "es-ar")
source_lang: ISO code of the source language
target_lang: ISO code of the target language
source: Original sentence in the source language
target: Translated sentence in the target language

Data Splits

The dataset contains a single split: train.

Dataset Creation

Curation Rationale

This dataset was created as part of the AI-TraLow project, which aims to advance machine translation for low-resource and extremely low-resource languages using decoder-only LLMs. By developing strategies to mitigate data scarcity, the project extends the reach of LLMs to underrepresented languages, supporting the preservation of cultural heritage, the integration of vulnerable migrant communities, and the empowerment of underrepresented language communities. This dataset is intended as a training resource for machine translation systems, comprising approximately 97.8 million sentence pairs across 8 language pairs, all paired with Spanish. It covers two complementary groups of underrepresented languages: non-EU and migrant languages (Arabic, Chinese, Hindi, Korean, Japanese), and EU official languages with comparatively lower resource availability (Estonian, Maltese, Irish). This selection was designed to address both the translation needs of migrant communities in Spain and the persistent resource gaps affecting smaller EU languages.

Source Data

Initial Data Collection and Normalization

The corpus is a combination of the following original datasets collected from various public sources:

Source corpus	URL	Language pair(s)
NLLB		ES-AR, ES-ZH, ES-HI, ES-KO, ES-JA, ES-ET
UNPC		ES-AR, ES-ZH
MultiCCAligned		ES-ZH, ES-HI, ES-KO, ES-JA
WikiMatrix		ES-ZH
News-Commentary		ES-ZH
MultiParaCrawl		ES-ZH, ES-GA, ES-MT
Tatoeba		ES-ZH
NTEU (ELG)		ES-ET, ES-GA, ES-MT
DGT		ES-ET, ES-GA, ES-MT
EMEA		ES-MT
EUBookshop		ES-GA, ES-MT
KDE4		ES-GA
GNOME		ES-GA
XLENT		ES-GA

Data Processing

Data processing was handled through our Translation Data Processing Pipeline ParaCLEAN, which includes the following processing steps:

Input handling: Read corpora in plain text (.txt), tab-separated (.tsv) or translation memory (.tmx) formats.
Compute embeddings: Compute multilingual sentence embeddings (default: LaBSE).
Language identification: Depending on the specific language pair, language probability is calculated using GlotLID or Lingua.py.
Filtering: Filter by user-defined embedding scores and language probability thresholds.
Deduplication: Remove duplicate sentence pairs and fuzzy matches across corpora.
Normalisation: Standardise punctuation, spacing, and casing.

Who are the source language producers?

OPUS NLP and its constituent corpora
NLLB Team
ELG - European Language Grid
NTEU Project
Tatoeba users community
OpenSubtitles.org
Wikimedia Foundation
United Nations - producer of UNPC and MultiUN
Facebook AI / Meta - producer of MultiCCAligned, WikiMatrix, and XLENT
WMT (Conference on Machine Translation) - producer of News-Commentary
ParaCrawl Project - producer of MultiParaCrawl
Directorate-General for Translation, European Commission - producer of DGT
Joint Research Centre, European Commission - producer of JRC-Acquis
European Medicines Agency - producer of EMEA
Publications Office of the European Union - producer of EUBookshop
KDE Community - producer of KDE4
GNOME Project - producer of GNOME

Annotations

Annotation process

The dataset does not contain any manual annotations beyond the parallel alignments, which were either preserved from source datasets or validated through automated alignment scoring.

Who are the annotators?

[N/A]

Personal and Sensitive Information

Given that this dataset is derived from pre-existing datasets that contain crawled data, and that no specific anonymisation process has been applied, personal and sensitive information may be present in the data. This needs to be considered when using the data for training models.

Considerations for Using the Data

Social Impact of Dataset

This dataset contributes parallel resources for eight language pairs combining Spanish with languages spoken by migrant and/or other foreign communities in Spain (Arabic, Chinese, Hindi, Korean, Japanese) as well as lower-resourced EU official languages (Estonian, Maltese, Irish). Improved MT systems for these language pairs can facilitate communication with public services, support access to information, and help preserve and document these languages digitally, while also contributing to a better balance in the representation of EU official languages, many of which remain underserved by existing MT resources despite their official status.

Discussion of Biases

No specific bias mitigation strategies were applied to this dataset beyond deduplication and minimal quality filtering. Inherent biases may exist within the data, reflecting the biases present in the source datasets, which include web-crawled content, subtitles, news articles, and other user-generated or institutionally produced text.

Other Known Limitations

The data sources vary considerably in domain (legal, institutional, subtitles, web-crawled), which may result in uneven register and stylistic inconsistency across the dataset. Additionally, some language pairs rely heavily on automatically aligned or mined corpora (e.g., NLLB, MultiCCAligned), which can introduce alignment noise or translation quality issues, particularly for the lowest-resource pairs where fewer human-curated sources were available. The number of sentence pairs also varies substantially across language pairs, which may lead to imbalanced representation if the dataset is used without further sampling or weighting.

Additional Information

Dataset Curators

Machine Translation Group, AI Institute, Barcelona Supercomputing Center (ai_institute_mt@bsc.es).

Funding

This work is supported by MLLM4TRA (PID2024-158157OB-C32) funded by MCIN/AEI/10.13039/501100011033/FEDER, UE.

Acknowledgements

[N/A]

Licensing Information

This work is licensed under a Attribution-ShareAlike 4.0 International license. This dataset includes data obtained through web crawling. The provider makes no representations or warranties regarding the ownership or licensing status of the underlying content. Users are responsible for assessing and complying with all applicable intellectual property, privacy, data protection, and other legal requirements.

Citation Information

[N/A]

Contributions

[N/A]

Downloads last month: 7

Point of Contact:

ai_institute_mt@bsc.es

Total file size:

18 GB