Good luck, report back. My canonical 1st step w/stuff that looks like NLP, is to flip thru the Jurafsky/Martin and Norvig Russell AI texts and identify paths of least resistance, like naive Bayesian or LSI, that have perl/python/ruby/C packages that i can thrash around in with my data .
some other NLP packages: http://www.ellogon.org/ http://web.media.mit.edu/~hugo/montylingua/ -- http://mail.python.org/mailman/listinfo/python-list
