Wednesday, January 19, 2011
eWeek reports: "IBM researchers have been working for four years on its DeepQA supercomputer, which it showed off at a media event Jan. 13, just before it was scheduled to take on two former Jeopardy champions in a two-day $1 million tournament for the long-running TV game show. Dubbed Watson, after IBM founder Thomas J. Watson Sr., the supercomputer can understand natural human language–complete with puns, plays on words, and slang–and answer the complex questions that are typical in the Jeopardy trivia game. Watson analyzes “real language,” or spoken language, instead of just looking for keywords like a search engine. It understands context and correlates the question with the millions of pieces of information it has stored to find a specific answer, said David Ferrucci, the principal investigator on the Watson project. “The hard part for Watson is finding and justifying the correct answer, computing confidence that it’s right and doing it fast enough,” said Ferrucci. Researchers trained Watson with 200 million pages of text, or about 1 million books, ranging from sources like encyclopedias, movie scripts, newspapers and even children’s book abstracts. “Watson is not just storing all that information,” it is also correlating all of it to make sense out of it, according to Bernie Spang, director of strategy for the software group at IBM Research said."