General EN-AR Filtered

General Arabic-English datasets, with rule-based and semantic filtering (train > 0.45 - test/dev > 0.9) as well as (>= 0.1) fasttext language detection.