OiQ/daa-pairs

A randomized Moroccan Darija phrase-pair dataset with three script variants per phrase: Arabic, Arabizi (Latin), and Mixte (mixed script). This dataset is a post-processed version of the generation output. Segment order has been randomized per source to prevent models from relying on positional/sequential patterns during training. Column Description video_id Source YouTube video ID (empty for articles) article_id