The problem has already be reported (with Japanese, Chinese, and Persian):
We’ll add a new “calculation method” (by words or characters) setting so that you can adapt the calculation accordingly to the specific language and update the mentioned thread as soon as it’s implemented.