We have found a bug with character normalization (which messes up the parsing of the punctuation). We'll fix this today.