Yeah, language detection on short texts is quite complex. In my practice, N-grams doesn’t work well for them.