It depends on language. In some languages it is not trivial to detect word boundaries, no matter what encoding you are using.