Sunday, February 20, 2011

Star Trek's translator- nearly ready

mswift@mercurynews.com

Company profile: Google On "Star Trek," even the aliens with blue skin and antennas sprouting from their foreheads spoke with Capt. Kirk in perfect English, thanks to the universal translator, long a staple of science fiction.

After more than two decades of research and investment by corporations, universities and the government, machine-based speech recognition and language translation, while still in its infancy, has made significant headway. For the first time, the technology has reached the point where it is good enough for real-world commercial products.Ortsbo offers a more basic form of machine translation, based on text, that is attracting millions of users in its first few months of operation. The new service, a unit of Toronto-based Intertainment Media, offers real-time text translation in more than 50 languages for Instant Message networks including Facebook, Yahoo Messenger, MSN and Tencent QQ, the most popular IM network in China.

Then there is Google. Leveraging its vast computing power, the Mountain View Internet giant recently rolled out updates to its Google Translate apps for the iPhone and Android smartphones that offer limited speech translation between languages. While it's not quite Star Trek's universal translator, an experimental feature of the Android app even translates real-time spoken conversation between English and Spanish.
"If you had told me five years ago that we would get to where we've gotten," said Franz Och, who leads
Google's machine translation group, "I wouldn't have believed it."

Google, which also offers a software extension to its Chrome browser that translates documents, has a powerful business incentive for translation. Only a small share of Internet content is in some of the world's most commonly spoken languages, like Arabic or Hindi. If Google can give speakers of those languages access to the large volume of online content written in English, for example, it can gain a huge audience, and the revenue that comes with it.

Google's cloud network of distributed servers, which may be the world's largest computer network, can host millions of simultaneous translations. Ortsbo, which seven months after its launch already has a traffic volume it didn't expect until 2013, is racing to build a global cloud computing network to handle the flood of chatter from China, Taiwan, South Korea and Brazil, as well as from the U.S.

"You're seeing the real emerging markets come to life, and the networks that go with them," said David Lucatch, president of Toronto-based Ortsbo, which has more than 5.1 million users, a 27 percent jump since January. Lucatch says Ortsbo's online traffic is growing faster than Facebook in its first year because he's making social networks bigger, by allowing members who lacked a common language to chat.

Ortsbo -- www.ortsbo.com -- runs on 12 instant message networks, including Facebook Chat, Twitter, AIM, MSN, Google Talk, Yahoo Messenger and large Chinese networks like Tencent's QQ. The service is far from perfect: It can take the app more than 10 minutes to load onto your PC or Mac, and the machine-powered text translations regularly make the literal translation errors to which computers are prone, struggling with such colloquial expressions as "the best of two worlds."

Lucatch says that the startup is working to address those engineering problems, but that the traffic shows the powerful demand to bridge the language barrier, even if it isn't perfect yet. Ortsbo allows instant message conversations in 53 languages and 2,756 language pairings, meaning you can type in English and have your Facebook friend in Hong Kong see your IM in Chinese characters. Or, Google Talk users in Copenhagen and Katmandu could chat between Danish and Nepali, while Twitter followers in Reykjavik and Jakarta could send direct messages back and forth in Icelandic and Indonesian.

"We saw the opportunity to connect people in a new way without asking them to give up the things they like and already value, like Facebook and Google and Yahoo," Lucatch said.
The Google Translate app, meanwhile, allows users to speak or type a phrase into their phone and have it be translated into any one of more than 50 languages, some audible and some in text only. The iPhone app can "speak" translations in a synthesized voice in 23 different languages. The Android version has an experimental "conversation mode" for English and Spanish, a feature that combines speech recognition and language translation. Two people who spoke just Spanish or English could hold a phone between them, allow it to record a phrase in one language, send it to Google's cloud network for translation, and "speak" it back in the other language -- all in a few seconds. Microsoft also has a translation app that runs on its Windows Phone 7 platform.

Efforts at computer translation reached a turning point in the 1990s as scientists gradually abandoned trying to program computers with the rules of grammar and diction in favor of a statistical approach. Computer scientists like David Chiang of the Information Sciences Institute at USC say they have made great progress in having computers look at millions of sentences and their translations, using powerful statistical models to allow computers to infer the rules of grammar -- an application of artificial intelligence that mimics the human mind.

Still, computer scientists note that informal speech and text -- laden with jargon, localized expressions, inconsistent grammar, and the "ums" and "aahs" that pepper everyday conversation -- is among the most difficult problems in machine translation and speech recognition. Indeed, when Twitter announced a "Twitter Translation Center" on Monday, it said it would rely on humans to translate the site into less commonly spoken local languages.

"We've found that the quality is higher when humans translate," said Carolyn Penner, a Twitter spokeswoman. Even Ortsbo plans to use humans to translate common expressions like "your neck of the woods," that can be nonsensical if a machine translates them literally.

Chiang called the Google app "impressive," but said it also shows the limitations of machine translation.
"If I'm traveling to a foreign country, and I want to be able to ask, 'Where is the bathroom?' I'm pretty confident this app can do this," Chiang said. "But if I want to negotiate a business deal, I think we probably have some way to go."

Och joined Google in 2004 after a personal recruiting call from co-founder Larry Page, who assured the scientist that linguistic translation was central to Google's mission "to organize the world's information." But despite the advances Google and others have made in machine translation, Och believes there will be a role for human translators for the foreseeable future.

Machine translation "opens up possibilities that we've never thought of before," he said. "There will be more need for (human) translators, just because it's encouraging people to do things across the language barrier."

Contact Mike Swift at 408-271-3648. Follow him at Twitter.com/swiftstories.





No comments:

Post a Comment