When training a language model on a massive text corpus (Common Crawl, Wikipedia dumps), you may want to bin English and non‑English documents separately. A fgselectiveallnonenglishbin routine would: