Watsonsim

[Sean Gallagher]

Hello all!

  I'm Sean Gallagher, I'm an RA at UNC Charlotte, and I develop Watsonsim (status), a class project developing a question answering system, hopefully not too unlike those mentioned here.

I heard about this group when Pasky mentioned it to me on our status blog, and it's interesting to hear about your developments, especially since I have been studying YodaQA for inspiration.

If you're interested in our project, here are a few highlights: It targets Jeopardy questions, it uses online and offline sources (including Bing search results), and it doesn't use UIMA. Right now, it gets about 31.2% accuracy and 52.4% recall on runs of 5K questions. Our drawbacks are mostly that it takes around a minute to answer a question, and you have to sacrifice two black goats to navigate the setup successfully. In fact, we are having a very tough time sharing the requisite data at all on account of the bandwidth.  I'd like to improve the recall and we have some concepts floating around the lab about how to do that - most involving improving our LAT handling.

Thanks and I look forward to reading more about your projects!