>Currently released model supports most of the languages except East Asian (Chinese Simplified, Traditional, Japanese, Korean, Thai). You should expect good results if a language uses space as a main token delimitter.

Any plans to change this?

What does MS fall back on for these languages?