Description

This repository collects open source parallel aligned corpuses Catalan to several languages.

Parallel corpus

We use these corpuses to train the Softcatalà neural translation system:

• English - Catalan
• German - Catalan
• French - Catalan
• Italian - Catalan
• Japanese - Catalan
• Dutch - Catalan
• Portuguese - Catalan
• Spanish - Catalan
• Occitan - Catalan
• Galician - Catalan
• Basque - Catalan

The corpus with extension xz need to be descompressed with xz.

You can do this easily by typing:

make extract-corpus

Catalan monolingual corpus

For backtranslation you may be interested in a monolingual Catalan corpus. You can create a monolingual corpus by typing:

make build-monolingual

This creates a single Catalan file with all unique strings across all language pairs.

Sources of the corpus used

We strongly recommend the following sources of aligned Catalan parallel corpuses:

• https://opus.nlpl.eu/
• https://github.com/facebookresearch/LASER/tree/master/tasks/WikiMatrix
• https://www.statmt.org/cc-aligned/

On top of these previously available corpus, we have created the following corpus:

• Europarl-catalan
• Tilde-MODEL-catalan
• Open source corpus in serval directions using Softcatalà translation tools

Do you want to help?

See here (In Catalan)

Contact

Contact Jordi Mas jmas@softcatala.org

Metadescription

The following table is necessary for this dataset to be indexed by search engines such as Google Dataset Search.

property	value
name	Open source aligned text corpus English, German, French, Italian, Japanese, Portuguese, Spanish, Occitan, Galician, Basque, etc to/from Catalan.
description	Open source aligned text corpus for building NLP applications (e.g. machine translation). Already existing corpus have been clean up and new corpus have been introduced: Europarl Catalan, Tilde Catalan and open source translation memories.
sameAs	https://github.com/Softcatala/parallel-catalan-corpus/
url	https://github.com/Softcatala/parallel-catalan-corpus/
creator	Softcatalà