The fact that WordNet only contains open-class words might be a benefit, actually. If it's not in WordNet, the OP probably doesn't want it in the concordance.

...except that that idealizes WN's coverage of English. You'd probably do well, though with the added heuristic: "and it's a short, uncapitalized word". (i.e. "Short, uncapitalized words that aren't in WN should probably be ignored.")

...and it ignores the problem of homonyms. (e.g. 'in' has 7 senses in WordNet)

Nonetheless, felt the need to "defend" WN (since someone's quoting from the FAQ I wrote).

Better solutions have been mentioned elsewhere in the thread. But, I'll also add that my approach to the problem would probably be more Information Retrieval oriented. I'd use KStem, a stemming algorithm whose output is (in the usual case) an actual word. It tries to correct some of the problems with Porter. That paper is pretty academic (but good). There's also downloadable Java code.

Then I'd just use tf-idf style weights to pick out the interesting words.

Seems easier than parsing, even if it's shallow parsing.