I'm interested in this primarily as an English teacher. I need to be able to spot the linguistic tics and errors and recognize where it likely came from.

That might well turn out to be the Red Queen's Race. It's only a guess, but I suspect that competitive models, the advances resulting from competition, and the advances and experimentation associated with catching and correcting mistakes will mean that you'll generally be playing catch up.

Frankly, I don't even have anything more useful to offer than the unrealistic suggestion that all such work be performed in class using locked down word processing appliances or in longhand. It may be that the days of assigning unsupervised schoolwork are over.