Xuedong Huang, a Microsoft Technical Fellow in AI and Research, leads Microsoft’s speech and language efforts. (Microsoft Photo)
Microsoft says it has built the first machine that can translate parts of news stories from Chinese to English as well as humans can.
Researchers have toiled to find a way to level up machine translation for decades, but advances in artificial intelligence helped power Microsoft’s latest achievements. Deep neural networks — algorithms that can recognize patterns and learn — paired with new AI technologies from Microsoft that help translation machines do the equivalent of editing, fact-checking and revising their work helped get Microsoft to the milestone of human parity.
Microsoft’s Allison Linn detailed the program, which includes researchers working out of labs in the U.S. and Asia. The machines took on a popular database of general news stories developed last year by industry and academic partners. Microsoft researchers detailed their findings in a research paper, claiming their technology matches the quality of professional human translations.
While the human parity milestone is a huge accomplishment, Microsoft acknowledges the technology still has a long way to go. News articles in this database are generalized, so they don’t tackle in-depth concepts with their own jargon. And Chinese to English is something of a low-hanging fruit for this kind of research in that there is a lot of data to pull from.
“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,” Arul Menezes, a partner research manager of Microsoft’s machine translation team who helped lead the project, said in the blog post.
Computer vision expert Harry Shum leads Microsoft’s artificial intelligence initiatives as the executive vice president of the company’s AI and Research group. (GeekWire Photo / Todd Bishop)
Microsoft has put a lot of time and energy into advancing its AI and machine learning technologies focused on language. Last year, Microsoft said its speech recognition technology achieved an industry milestone matching the error rate of multiple human transcribers in a recognized accuracy test. Another Microsoft AI program bested humans earlier this year on a reading comprehension data set developed at Stanford.
Microsoft researchers will continue to hone the company’s translation and understanding abilities to become as good as humans in more adverse circumstances. The company will also continue integrating these breakthroughs into its suite of productivity …read more