Microsoft's AI Program Can Match Human Chinese Translation Skills

Microsoft may be getting closer to putting translators out of work. A team of company researchers has adult a auto organisation that they merits tin can translate news articles from Chinese to English with the same quality as a human person.

On Midweek, the researchers called the achievement a major milestone. Many experts had believed designing a computer to match the translation abilities of a human could never be achieved, Microsoft said in a web log mail service.

"Hitting human parity in a auto translation task is a dream that all of u.s.a. take had," said Microsoft technical boyfriend Xuedong Huang in the post. "We just didn't realize nosotros'd be able to striking information technology so soon."

But are the translations really that skillful? I'm a Chinese speaker and so I tried the arrangement, which Microsoft has put online. It'due south really impressive and seems to excel at writing translations that sound natural in the English language —something which Google Translate and Microsoft Bing tin struggle to do.

However, the arrangement hedges a bit. It'll give you not one, but two unlike translations, both of which can vary in accurateness and sometimes include mistakes. (In other words, human proofreaders volition nevertheless be needed.)

What makes Microsoft'south solution special? Information technology taps into an AI-based applied science known as neural networks, or computing systems that have been modeled afterwards biological brains. In recent years, neural networks take attracted growing interest from estimator scientists for their chapters to learn like humans tin: with practice.

In Microsoft'south example, the company's researchers trained the AI system to focus on writing fluent and natural-sounding translations by doing the process repeatedly.

"One method they used is dual learning," Microsoft's web log mail service said. "Call back of this every bit a fashion of fact-checking the arrangement's work: Every fourth dimension they sent a judgement through the organization to be translated from Chinese to English language, the research team also translated it back from English language to Chinese."

This back and along taught the AI what mistakes it had made and how to fix them. On meridian of that, the researchers also trained the computing system to translate a judgement of text over and over once more until information technology was polished.

To show the AI system could friction match the capabilities of a human, Microsoft ran it through a exam that judged the results against man-produced translations. This arroyo involved hiring exterior bilingual experts to score the translations from 0 to 100, and then compare the average.

This testing method is maybe a picayune questionable; it isn't comparing one private translation against another, but looking at the overall performance. Nevertheless, the test showed the AI organization scoring slightly meliorate —from 68.6 to 69— than the human-produced translations, the highest of which came in at 68.5.

Simply despite the achievement, the researchers said in Microsoft's blog post that motorcar translation is not a solved problem. Their ain paper on the technology mentions randomly sampling 500 sentences translated by the AI system and occasionally finding various errors, including incorrect words, bad grammar, and problems with names.

Still, the piece of work represents progress. "What nosotros can predict is that definitely nosotros will do better and better," said Microsoft principal enquiry director Tie-Yan Liu in Wednesday'south web log postal service.

Although the research only covers Chinese and English language, the technologies can be useful in other languages or perhaps even beyond translation, Microsoft added.

Source: https://sea.pcmag.com/news/20085/microsofts-ai-program-can-match-human-chinese-translation-skills

Posted by: austinthresper1974.blogspot.com

Share this post