In a blog post published yesterday, the researchers from the company's USA and Asia labs claimed that their system has already achieved human parity when tried on a commonly used test set of news stories known as newstest2017. Humans perform the same thing when manually translating to verify that their translation is correct, and machines seem to benefit from using the same technique. "We just didn't realize we'd be able to hit it so soon".
With several 2,000 sentence tests run and confirmation from 3rd party bilingual language consultants Microsoft Researchers had enough data to confidently attest to the accuracy of its new machine translation.
The researchers, based in Microsoft's Asia and USA labs, claim that their AI system has achieved human parity on the newstest2017 which is a commonly used test set of news stories developed by industry and academic partners.
Technical fellow in charge of Microsoft's speech, natural language and machine translation efforts, Xuedong Huang, called it a "major milestone" in one of the most challenging natural language processing tasks. Every time they sent a sentence through the system to be translated from Chinese to English, the research team also translated it back from English to Chinese.
"Given the best-case situation as far as data and availability of resources goes, we wanted to find out if we could actually match the performance of a professional human translator", said Menezes.
To gain as many real-world test metrics, Microsoft Research has begun seeding the new translation system here.
Arul Menezes, partner research manager of Microsoft'smachine translation team.
According to the researchers, the new methods they used on the system include dual learning for fact-checking translations, deliberate networks for refining translations, and joint training for English-to-Chinese and Chinese-to-English translation systems boost. It translates new English sentences into Chinese in order to obtain new sentence pairs which are then used to augment the training dataset used for the opposite direction - Chinese to English. There is also an agreement regularization to enable the generation of translations through reading, either from left-to-right or vice versa.
Microsoft says it will continue to work on its translation system and the hope is the new AI will be rolled out to accurately translate other languages.
Federer too perfect for Chung
Anderson, last year's US Open runner-up and the seventh seed, edged past Pablo Carreno Busta 4-6 6-3 7-6 (8-6). It is his best beginning to the season since 2006 when he started 16-0 and won 33 of his first 34 matches.
How to MI vs. Montana
The Grizzlies scored just two points in the first 11 1/2 minutes of the second half as the Wolverines gradually pulled away. The Grizzlies (26-8) were limited to 32.1 percent field-goal shooting, including 3 of 15 attempts from 3-point range.