Sean M. Burke's premise was that typos on dvorak keyboards tended to look more like wierd words rather than random mis-typings than their counterparts on qwerty keyboards. In discussion of this he wrote a plausibility function I think you may find very useful.

My suggestion is that you take a very large piece of text (grab something off gutenberg) remove all the spaces and look at the occuring 3-letter groupings (trigraphs). Then examine the text you're facing. With some experimentation you should be able to determine an approximate cut-off level whereby your text is probably a nonsensical phrase vs a non-phrase.

Studies of language for the purposes of cryptography have shown that of all possible trigraphs only some small percentage are actually common in typical language. You should have no problem distinguishing rubbish from language this way. :)

Hope it helps.

jarich