Katz Explains Contributions To Watson Jeopardy! Challenge
February 14, 2011
When the theme music for Jeopardy! fades into the background tonight and the race for the buzzer begins, Watson, IBM’s question answering computing system, will be relying on several ideas and algorithms, including ones originally developed at CSAIL by Principal Research Scientist Boris Katz.
When IBM set out to create a computing system that could compete on and win Jeopardy!, the company approached researchers and scientists for input on how to make the machine faster, more knowledgeable and more competitive. Katz, head of the InfoLab Group at CSAIL, presented several concepts that will allow Watson to actually gain an understanding of a question, so that an accurate conclusion can be reached.
Watson incorporates ideas pioneered by Katz’s program START, the world’s first Web-based natural language question answering system. Katz’s object-property-value data model makes it possible to view the vast amount of semi-structured information available on the Web as a uniform database, thus enabling this data to be effectively retrieved in response to natural language questions. This model also recognizes many disparate ways of phrasing object-property-value questions.
Katz’s model of syntactic decomposition helps Watson decipher complex, multi-pronged questions by allowing the system to understand that it needs to tackle several sub-questions. The system then uses an algorithm that helps it decide which sub-questions to answer and in what order, and compiles the gathered information into a cohesive, and hopefully correct, answer.
“It is very important for a computer playing Jeopardy! to be able to break down the question and understand the pieces,” said Katz. The object-property-value data model combined with syntactic decomposition makes it possible to answer complex questions using information fused from many data sources.
Katz is eagerly anticipating Watson’s debut on Jeopardy! tonight, as well as how the technology developed to build Watson will be used once the game show challenge is complete.
“It will be wonderful to show the world (on Jeopardy!) how far in the field we have progressed,” explained Katz. “I also think the core technology developed in natural language question answering systems could be used in many interesting domains that require analyzing large amounts of text, such as medical and legal applications.”