There isn't anything special about what NELL does for sarcastic,
distortionary, or decietful statements. At present, we don't go very far
in terms of being able to read individual sentences. We rely on an
assumption that true or at least commonly-believed statements occur more
frequently than ones that are sarcastic, deceitful, or whatever else.
Another thing we aren't doing as much of as we'd like to is inferring
things from what is already in NELL's KB. We could, for instance, use
this to find conflicting knowledge, and then try to focus on reading more
relevant information in order to resolve the conflict. If we can assume
that true statements are more logically consistent with each other than
false ones are, then something like this could help weed out false ones.