Baidu’s Translation Tech Rivals Google’s in Speed and Sophistication | Social Media
Even the strongest translation technology often suffers from a lag, due to the innate differences in sentence structure across languages. Last week, Chinese search engine Baidu unveiled a tool aiming to eliminate the lag and hasten conversations across cultural barriers.
The tool runs on STACL artificial intelligence technology or Simultaneous Translation with Anticipation and Controllable Latency. Once trained, the tool uses artificial technology to predict where a sentence is going, and fill in typical gaps with learned knowledge. With similar languages (like Spanish and Italian, for example), this process has historically been easier. But dissimilar languages, like English and Chinese, lags were inevitable—and impossible to shorten.
For brands and companies creating content that spans cultures, the need to quickly translate will be crucial; this application of AI will eventually be able to facilitate that in a number of ways. At the same time, principal designer Liang Huang foresees a world where human translators are seen as unnecessary- and he wants no part of it. “We hope this AI technology will reduce the burden on human interpreters. But we are not trying to replace human interpreters […] especially for high stakes occasions which require consistent and more precise interpretation.”
Still, STACL technology offers a precise and prompt translation that thus far has been out of bounds of many rival product, including those developed by Google. In a demo on their blog, Baidu used a sentence from English and translated it into Chinese to demonstrate how the system works: “President Bush meets with Russian president Putin in Moscow.” Sentence structure in each language has made prior translators slow, as the tool “reads” to the end of a sentence before starting the translation process.
“In Chinese, the verb ‘meets’ is at the end of the sentence,” CNBC explains. Other tools, including ones in development at Baidu rival Google, can’t translate in a timely fashion because they have to wait for the active verb. “But when translated into English, it becomes the third word in the sentence, as is appropriate in that language. Thanks to the training with the data sets, Baidu’s tool is able to predict the word that comes in the English language even before it is spoken.”
There’s still time before the product can fully take hold in countries other than China; thus far, the tool only has the training to translate Chinese to English, English to Chinese, and Chinese to German. Further, the tool is only as accurate as the training as it has completed, and there’s a great deal more to do in a number of other languages. But the initial promise that Baidu’s unique AI has shown, provides hope for more seamless conversations across the world in a number of different domains.
Learn the latest trends, insights and best practices from the brightest minds in media and technology. Sign up for SMW Insider to watch full-length sessions from official Social Media Week conferences live and on-demand.