It's likely due to the corpus though. It's multilingual, but the dataset they trained on is representative of "the Internet" so the latin languages (English, Fr

It's likely due to the corpus though. It's multilingual, but the dataset they trained on is representative of "the Internet" so the latin languages (English, French, Spanish, Italian, German, etc...) are overly represented.