The code from the paper is available at http://groups.csail.mit.edu/rbg/code/content_structure/code....

You can see a demo of the site at condensr.com. There will be substantial changes to the site including predicting the best dish at a restaurant on the site soon. An academic paper should follow shortly.

Interesting fact: I was asked to program something that did just this for my interview with Yelp.

Anyone know what software (open source, Java/Python) that is out there that can give me similar results? I've looked at LingPipe, but the closest thing it does is sentiment classification.

Check out the open source Stanford Parser (Java) http://nlp.stanford.edu/software/lex-parser.shtml

I also wrote a free open source javascript API for the Stanford Parser if you want to play around with NLP in JS. http://nlp.naturalparsing.com/browserparser/parse (it is a bit slow at the moment, someone is using it to parse all of wikipedia).

Python's NLTK has a lot of the raw tools at hand to do NLP tasks.