QA technology, most recognizably used in
IBM's Watson computer system,
takes a question expressed in natural language (such as the text on this page,
for example), and seeks to understand it in much greater detail to return a
precise response. Koichi and Hiroshi led the effort behind Watson's ability to
attach meaning to the words, expressed as clues, on the quiz show Jeopardy! in
The need to understand a wide variety of
subjects to play Jeopardy! demanded a new approach to QA computing confidence
and speed for Watson to succeed.
“The traditional approach in QA technology
needed to set rules in order to answer questions in a particular field. Before
Watson, the existing technology was incapable of answering ambiguous questions
written in natural languages, or questions outside its programmed areas of
expertise. By taking a more-flexible approach to how Watson stored massive
amounts of data, quickly extracted and indexed information, and calculated a
statistical bias, it was able to understand the broad range of topics on
Jeopardy! -- and win," Hiroshi said as he looked back to his time as a
member of the Watson project.
“The biggest reason for Watson's victory on
Jeopardy! was that it could store and access a wealth of high quality
information put into its system. But perhaps the biggest reason that Watson
drew such significant attention was because of its promise as a technology
capable of understanding and finding value in unstructured data, which has
exponentially increased in recent years," Koichi said.
"The QA technology that the Watson
project team developed brought further sophistication to the methods for
information access. Through the combination of its highly advanced QA
technology, and improved text mining technology, intellectual enterprise
solutions that integrate structured information with unstructured data (for
example, obtained from sensors and social media) will advance many fields of
Today, Hiroshi and Koichi's research focuses
improving QA technology in a Japanese environment, applying Watson's QA system
in healthcare through collaboration with their colleagues at IBM's Thomas J. Watson Research
Center in the US, and furthering what QA technology can understand,
and how we can interact with it.
(Left) Koichi Takeda
worked on statistical analysis of background text sources on the Watson
project. His research interests include text mining, question answering, and
unstructured information management. (Right) Hiroshi Kanayama worked on mining
evidence from background text sources. His research interests include syntactic
parsing, text mining, and sentiment analysis.