Researchers at Microsoft have created an artificial intelligence that can translate from English to Chinese and vice versa, as accurately as a person can. Microsoft Logo on a building. (Image via Microsoft)


When learning English for the first time, most ESL (English as a Second Language) teachers urge students against using google translate or any translation tool to translate from English to their first language. The reason is that the translation is not always accurate; the tool used to translate is most likely not aware of parameters such as context, sarcasm or metaphor. Luckily companies like Google and Microsoft are continually looking to improve their services and the tools that support those services.


Thanks to Microsoft, the days of confusing translation are numbered. Researchers at the company amazed with their discovery. Many before them have tried for numerous years and are still trying to solve the issue of accuracy of machine-translated information. Obviously, many might doubt the invention, but the researchers responded to the possible doubt by releasing the work of the machine at a research conference last year.


The machine created by Microsoft has been tested with a few articles of various origins, more than 1500 sentences, and the results were compared to the translation made by professional translators and further assessed by experts of languages. The verdict was clear: Microsoft’s machine can effectively rival human translation. Microsoft’s new “baby” has the same error rate (5.9%) as a professional transcriptionist, which is why the company is boasting about reaching human parity in speech recognition. On the company’s blog, one of the researchers, Xuedong Huang, explained that none of them were expecting to reach human parity so early.


To understand why Microsoft just made history with its new machine, it is important to understand first how translation by humans works and then how the machine matches that level of understanding. When professionals are asked to translate information, each of them might provide a different version of translation without being wrong; and that is because in languages there are ideas that can be expressed through different words. The translation appeared to require a level of appreciation or judgment that an AI didn’t have. The closest AIs have come to understanding human words is recognizing voices and a few words. But even then, most AIs on the market cannot compute more than a sentence at a time. So, to have a machine translate a full page will require new skills.


To incorporate those new skills in AIs, Microsoft’s researchers had to turn to deep neural networks. It is a technique to teach Artificial Intelligences a wider range of topics; therefore, improving the machines’ ability to identify contexts. In addition, researchers taught the AI how to verify facts, check the translation for errors, check for improvements in the translation of English to Chinese or vice versa, and the ability to read from left to right or right to left. Ming Zhou, one of the members of the research team in China, believes that with all those additions, Microsoft’s new AI might help in all areas of life let alone translation of any other language.


However, before Microsoft could shout “victory,” the research team still has a few adjustments to make on the machine as well as testing it on live news stories. For the time being, they invite anyone interested to go on the company’s website and try their new translation tool. However, it isn't 100% complete yet. And read more about the API here.


Have a story tip? Message me at: cabe(at)element14(dot)com