[Evaluation Benchmark] New curated 400-sentence Gold Standard for Tarifit (rif_Latn)
Hi everyone,
To support the ongoing efforts in the No Language Left Behind (NLLB) initiative, I have released two curated evaluation benchmarks for Tarifit (Rif-Berber):
Tarifit-Catalan Public Services Dataset: https://huggingface.co/datasets/jamalinu/tarifit-catalan-public-services
Tarifit-Spanish Public Services Dataset: https://huggingface.co/datasets/jamalinu/tarifit-spanish-public-services
As a linguist and native speaker, I noticed that current massively multilingual models still face challenges with the morphological complexity and script consistency of the Rif variant, especially in technical domains. These datasets consist of 400 high-quality parallel sentences specifically focused on Public Services (Healthcare, Education, Admin), serving as a Gold Standard for measuring BLEU/ChrF++ scores more accurately in this low-resource context.
I hope this helps the community improve translation quality and digital accessibility for the Amazigh languages. Feedback is more than welcome!
Best,
Jamal