Imsidag-community/nllb_en_kab

NLLB English - Kabyle Dataset

This dataset contains parallel sentences in English and Kabyle, cleaned and filtered using the GlotLid model. The dataset is derived from the OPUS-NLLB corpus and has been processed to ensure high-quality sentence pairs.

Dataset Structure

• nllb_en_kab.parquet: A Parquet file containing the cleaned English-Kabyle sentence pairs.

Dataset Statistics

• Total Sentence Pairs: 2,484,297
• English Sentences: 2,484,297
• Kabyle Sentences: 2,484,297

Data Preprocessing

1. Source: The dataset is derived from the OPUS-NLLB corpus.
2. Language Detection: The GlotLid v3 model was used to filter out non-Kabyle sentences. Only sentences with a confidence score of 0.95 or higher were kept.
3. Normalization: A custom fixer/normalizer was used to correct non-standard Kabyle Latin characters. Such as the following replacements were made:
   • ţţ → tt
   • țț → tt
   • ε → ɛ
   • ϵ → ɛ
   • γ → ɣ
   • Γ → Ɣ
   • Σ → Ɛ
   • Ԑ → Ɛ
   • ğ → ǧ
   • ş → ṣ

Usage

You can load this dataset using the datasets library:

from datasets import load_dataset

dataset = load_dataset("Imsidag-community/nllb_en_kab", split="train")
print(dataset)

License

This dataset is licensed under the CC-BY-NC 4.0 License.

Citation

If you use this dataset in your research, please cite it as follows:

@dataset{
    nllb_en_kab,
    title = {NLLB English-Kabyle Dataset},
    authors = {Athmane MOKRAOUI},
    year = {2025},
    url = {https://huggingface.co/datasets/Imsidag-community/nllb_en_kab}
}