One year ago, Microsoft announced that neural networks are now powering all the speech translation on Microsoft Translator. Microsoft also made it available to all developers who use the Microsoft Translator speech API inside their apps and services. Since then, Microsoft is trying to expand the use of neural machine translation (NMT) to improve translations in all of the Microsoft Translator products and the Microsoft Translator Text and Speech APIs.
NMT offers huge advances in translation quality over the existing statistical machine translation technology. So Microsoft today announced the expansion of NMT to 10 new languages bringing the total number of NMT-powered languages to 21. The new languages are listed below.
Microsoft mentioned that all Chinese simplified and Hindi translations to and from English are now using the new NMT systems by default. Microsoft is also introducing something called Hybrid neural translation that takes advantage of the quality improvements of NMT even when one of the translating languages is not NMT-supported language.
In machine translation technology, because of the lack of available training data between any two languages, English is used as a “pivot language” to translate from one language to the other. If you were to translate between two languages, for instance from French to Chinese, in the background the machine translation tool would typically first translate the French into English, and then translate the English into Chinese. Microsoft Translator is no different, whether it is for the traditional statistical or the newer neural machine translation system.
Finally, Microsoft is also releasing neural network translation as an on premises service, learn more about it here.