@thilo @aral It's not too hard to determine whether a given string of non-Latin characters resembles a word in the post's language. You just need a morphological lexicon and basic understanding of NLP, which anyone who works on screen readers definitely has.