FOR IMMEDIATE RELEASE CONTACT: Sept. 1, 1993 Steve Eisenberg Pittsburgh Supercomputing Center 412-268-6132
Kitano, a computer scientist specializing in artificial intelligence (AI), is bypassing traditional AI methods to tackle the translation challenge. And his efforts, which involved the Connection Machine CM-2 at the Pittsburgh Supercomputing Center, earned him the most prestigious award in artificial intelligence for researchers under 35. He received the Computers and Thought Award Aug. 31 at the 13th International Joint Conference on Artificial Intelligence in Chambery, France.
This year's award marks the first time a researcher has won for work on a massively parallel computer, such as the CM-2 from Cambridge, Mass.-based Thinking Machines Corp. "The nomination letters highlight his outstanding contributions to the use of massively parallel software and hardware in artificial intelligence -- for example, in speech-to-speech translation," says Conference Chair Wolfgang Wahlster of the German Research Center for AI, who organized the worldwide search for the award winner.
Until recently, the 100s to 1,000s of processors in a massively parallel computer were largely absent from AI research because traditional AI models had relied on single-processor computers, such as high-end workstations, and were not appropriate for massively parallel machines. So Kitano, 32, wrote a new program that took advantage of the CM-2's 32,000 processors. The result was one of the world’s first speech-to-speech translation systems relying on massively parallel computing -- the type of computer firepower, he says, that will dominate AI's future.
Kitano received his doctorate in 1991 from Japan's Kyoto University, and he splits his research time between Carnegie Mellon University, where he is a visiting scientist at the Center for Machine Translation (CMT), and the Sony Computer Science Laboratory in Japan. CMT is a research branch of Carnegie Mellon's School of Computer Science and is devoted to basic and applied research in all aspects of natural language processing.
The traditional AI approach to speech-to-speech translation, which requires a few seconds to a few minutes to translate one sentence, is extremely rigid. All knowledge is explicitly represented, and rules strictly run the show. "There are several reasons this fails in the real world," says Kitano, who once was a professional simultaneous interpreter. "It is almost impossible to obtain a complete set of knowledge for a given problem."
The solution, Kitano maintains, is building on the memory-based computer system first proposed in the early 1980s. "The common thread is to view memory as the foundation of intelligence," he says. When the memory-based approach is used for translation with the CM-2, an English sentence arrives at the computer, which then rifles through its 32,000 processors -- each containing a few sentences -- and comes up with the appropriate match, along with its Japanese translation.
For instance, if the English sentence were, "Kitano won the Computers and Thought Award," the machine would locate two similar English sentences, such as "Yukawa won the Nobel Prize" or "Sena won the Grand Prix," and their Japanese translations. And because the "Nobel Prize" is closer in meaning to the "Computers and Thought Award," the machine would pick the first sentence, make the appropriate changes and produce the Japanese translation. So far, Kitano has achieved a 75 percent accuracy rate when testing his program on 1,600 sentences with the CM-2.
The Computers and Thought Award was established in 1971 and is awarded every two years. Nominations for the $2,000 award are invited from everyone in the international artificial intelligence community. The Pittsburgh Supercomputing Center, a joint project of Carnegie Mellon University and the University of Pittsburgh together with Westinghouse Electric Corp., was established in 1986 by a grant from the National Science Foundation with support from the Commonwealth of Pennsylvania.
go back to contents page