Yesterday, we published an English – Kabyle Parallel Corpus.
130 883 aligned sentence pairs extracted from our contributions and contributions of the community on Tatoeba database.
The corpus is aligned pair by sentence-id (en-kab).
By number of sentences, Kabyle language is ranked 5th on Tatoeba with 772 002 submitted sentences (september 14th, 2025).
The dataset will be updated from time to time via :
HF dataset : https://huggingface.co/datasets/Imsidag-community/english-kabyle-parallel
We used some "basic fixer and normalizer" to correct some words and sentences.
The fixer/normalizer is used to correct some errors and map some char sequences to the standardized kabyle alphabet published by the CLDR/Unicode.
Tool used to download and align the corpus, kabyle-nlp-toolkit : https://github.com/BoFFire/kabyle-nlp-toolkit
Expect more parallel `en-kab` corpus thanks to the project `translation-memory-tools` built by @softcatala with 
We tried to customize it locally for kabyle language and it works : we are able to build a kabyle translation memory from all the translations we submitted over the years by translating FLOSS Software and projects.
Link : https://github.com/Softcatala/translation-memory-tools
athmane mokraoui [BoF] ⏚ꝃñ⌁⁂
