Encoding schemes like UTF-8 don’t affect compressed size much. What matters is the quantity of information.

You could probably do this with minimal overhead by organizing the words by language assuming a codepage for each set.