Isn't that the case for almost all services that do any amount of predictive typing?
Is there a reason to think that sending keystrokes to a Chinese prediction engine is inherently less secure than sending them to a prediction engine that works in any other language?