It's a test of Watson's technology, which -- very simply put -- is designed to take in huge amounts of information, process and learn from them in the same way the human brain does. Watson has already spent years picking up the weird phrases, nuances and quirks of English. But this is the first time IBM's tried to teach Watson a language that doesn't use the Latin alphabet.
"This is a major step for us," said John Gordon, IBM's vice president for the Watson Group. IBM has fed Watson texts translated from Japanese before; he said it's been hard to capture the "richness and depth" of the language, a process made even more difficult because Japanese has three different alphabets and thousands of characters.
To accomplish that task, Watson will have to work to improve itself using essentially the same process that students learning foreign languages have for years, though on a much larger scale. It will get a bank of 250,000 words, and turn them into 10,000 diagrammed sentences to identify the subject, object and verb. (Which in Japanese, this like is.) Then native speakers -- yes! humans! -- will read its first attempts at translations, correct them and feed the right sentences back into the system. Over time, Watson should be able to learn the language.
"You give it homework problems, grade its homework, and it figures out what's right," Gordon said.
IBM lays the process out in this infographic below:
Softbank was a logical partner for IBM here, Gordon said, because it has roughly 1300 subsidiaries that specialize in everything from telecommunications to gaming to robotics. It has also signaled its interest in big data, partnering last year with General Electric to sell that company's big data solution.
The breadth of Softbank's business opens plenty of applications for Watson beyond answering quiz questions. The companies have already dropped some of Watson's technology into Softbank's humanoid Pepper robot. That robot, which is designed to be empathetic and hold conversations, is due to go on sale in the coming months in Japan. (There hasn't yet been any decision about U.S. sales.) Watson, in theory, could help Pepper better understand the context of questions and make it a better companion down the line.
It also provides a good opportunity for IBM, which has never officially broken out its sales figures for Watson. A report last year from the Wall Street Journal said that Watson had totaled less than $100 million in revenue as of October 2013; IBM chief executive Virginia Rometty has said publicly that Watson should generate $10 billion in annual revenue within ten years.
IBM spokeswoman Lia Davis refuted the report's characterization of Watson's success, noting that the company’s Business Analytics department – which includes Watson – reported revenue growth of 7 percent over the past year to $17 billion this year. That department also includes IBM cloud services and its mobile partnership with Apple.
IBM is pretty confident that this type of "cognitive computing" -- in which computers learn like people -- will define the next era of computing. There's a lot of information out there and having a machine to digest and process it all saves humans a lot of time and papercuts. To date, IBM has focused the Watson project on a few core industries such as law or health care to help professionals stay abreast of new literature to best inform their daily practice.
Gordon said that the company has opened up its program to developers of all stripes to explore other ways that businesses can build on top of the Watson platform themselves.
That could lead to future partnerships, and expanded revenue for IBM down the line. Gordon said that he's enjoyed watching the applications for Watson grow, specifically mentioning a veterinary marketing firm and a group that's used Watson to help people evaluate which charities will use their donations most effectively.
The Japanese experiment will almost certainly lead to other partnerships -- and other attempts to teach Watson new alphabets and languages down the line, Gordon said.