ParaKab – Many Languages, One Kabyle: A Multilingual Parallel Corpus for a Low-Resource Language

Description of the Dataset

This dataset consists of three parallel corpora involving the Kabyle language, covering the following language pairs:

Kabyle-Arabic

Kabyle-French

Kabyle-English

Together, these corpora comprise approximately one million (1M) aligned sentence pairs, forming a multilingual Kabyle-centric parallel dataset. The resource is intended for research in natural language processing (NLP), particularly for low-resource languages (LRLs), machine translation, and cross-linguistic studies.

The data were compiled from multiple sources, primarily from publicly available resources distributed via the OPUS platform, as well as selected public websites. All texts underwent a rigorous pipeline including dataset selection, cleaning, normalization, alignment, verification, and deduplication, resulting in a ready-to-use dataset for research and machine learning applications.

Academic Context

This dataset was developed as part of a university research project on low-resource languages, focusing on Kabyle, a northern Tamazight (Berber) language. The project aims to contribute to the digital inclusion of low-resource languages and to support their integration into modern NLP systems.

Licensing and Data Rights

Each sentence pair in the dataset is linked to its original data source and, where applicable, its corresponding license or rights holder, ensuring transparency and compliance with source-specific terms of use.

For the publication of this corpus, a general license is provided for the dataset as a whole. Users are responsible for respecting the original licenses and usage conditions associated with each source when reusing the data.