And very interesting, making some considerations.
Grouping words one by one makes nonsense, but we have a text with the same stathistical properties of the original text. Grouping word two by two of three by three is the same, but we don't lost meaning (well, better, we don't lost too much meaning).

Using letters instead of words as units, I think (this is only a conjecture) we could "mimic" the sound of the original language, realizing a new kind of your Acme::Translator :). A kind that could be difficult to obtain even making Acme::Translator having Sex with Transcramble - Random text generator (brrr....).

Even more interesting experiments could be done using music instead of texts. Using something from the MIDI suite we could import a melody and use it to generate other similar (in the sense described by your program) melodies.

Well, just a thought, but could drive to interesting projects (I've always been interested in generative arts).