The system was developed using deep learning, dual learning, joint training, deliberation networks and agreement regularization and can be tested out at here. Microsoft develop the machine translation system using a sample of around 2,000 sentences and its accuracy was verified by bilingual language consultants. The task of machine translation of written sentences was much more complex than simple pattern recognition because different words can be used to convey the same meaning and previous attempts to translate web pages or articles could only give you a general idea of what was being said rather than provide a deep level of understanding.
People have been trying to reach this milestone for decades and even Microsoft's own employees admitted that they believed they wouldn't reach this milestone as soon as they have. When the rate of AI advancement surprises even the experts of the field, you have to wonder what other milestones could be closer than we imagined!
Click here to read the full article.