At some point though you have to acknowledge that a specific use of language belongs to the medium through which you're counting word frequencies. There are also specific writing styles (including sentence/paragraph sizes, unnecessary repetitions, focusing on other metrics than readability) associated with newspapers, novels, e-mails to your boss, anything really. As long as text was written by a human who was counting on at least some remote possibility that another human might read it, this is way more legitimate use of language than just generating it with a machine.