I fear I may be completely off on this, but it looked like a very intriguing problem you inquired about. My appologies up front for any confusion I may inadvertantly cause.

Would something like Text::English be of help? Or Text::Metaphone or Text::Soundex for early approximations?

This sounded a lot like what I found when I read rob_au's posting Natural Language Index Stemming and looked up the topic of "stemming," which sounds very close to what you are describing (although I have never worked with it, thus my preface above). Super Search gave several postings when I entered that term ("stemming"), so perhaps that might be helpful as well, and while I haven't tried the Google search against the site, I would guess it may also give some interesting results as well. I don't know for sure, though, although I would be very interested to hear of your results.