Reading all of this has given me the beginnings of an idea. Although I'm not a machine learning/NLP expert the idea is obviously non-trivial to an extreme (maybe impossible, currently): What if there were a forum similar in style to reddit that had a form of auto-moderation that identified logical fallacies and other irrelevant information and then highlighted those within arguments.

For example if there was an ad hominem attack "Fuck you Ellen Pao." The auto moderator would allow the post but would highlight the text red with a small bubble identifying it as ad hominem, users could react appropriately.

Or, maybe you have a long well thought out argument that has a straw man in it. The auto-mod could highlight the straw man to point out that it exists and maybe even format the post to show the point from which the rest of the position is based on that.

Maybe this wouldn't have to be actively presented but just a tool that comments were run through so users could be warned prior to their submissions. Obviously the complications with this would be ridiculous and you could never stop training the algorithm that was processing your text. But the implications of having such a thing seem pretty incredible.