Have a look at some of the code from one of the open source "spam filter" projects. One thing they do is check content to make sure that it isn't gibberish (since big chunks of random gibberish are often attached to spam, to fool primitive spam filters). This would force our chat crashers to use real words.