how about it just determines you didn't mean that, and correct it before the message is even sent? Machine learning should be able to do this rather easily.