[pfeyz]

perhaps you can integrate a part-of-speech tagger from nltk or montylingua to help it pick the right kind of word.

http://www.nltk.org/ http://web.media.mit.edu/~hugo/montylingua/

[viraptor]

Please don't use montylingua. It's not updated anymore, badly written, author is arguing about software license terms... and it's generally worse than nltk in almost every way. You'll be better off leaving montylingua alone.