CSAIL at TREC 2007 Question Answering

MIT CSAIL’s entries for the TREC 2007 question answering track built on our systems of previous years, updating them for the new corpora. Our greatest efforts went into the system that handles the ‘other’ questions, looking for new descriptive information about the topic. We noticed in our experiments with Nuggeteer (Marton and Radul, 2006) that some of the parameters made a big difference in the results, and decided to restructure our scoring to be able to tune its parameters. This represents the first such use of the Nuggeteer software that we are aware of, and yielded excellent results.