Google Translate is the go-to destination when we get confronted by language barriers, needing help in understanding foreign words. The service supports 103 languages at the moment and will expand in the future. Although more than 100 languages is a great feat, the main problem with Google Translate is that it usually provides a simplified translation, without regards to context, colloquialisms, slang, and different syntax rules.

The result is often a translation able to give you a general idea of what the foreign language text is all about, but not much more. Of course, there’re errors in translation because of the huge language library supported. All those issues make Google Translate a great tool for instant translation on a daily basis, with advanced, more accurate translation still being a huge obstacle Google Translate has to pass in order to become a full-fledged translation service.

google-translate-neural-machine

This could soon become a reality, with the introduction of Neural Machine Translation. Instead of using Phrase-Based Machine Translation based on a translation of one word at a time, thus making regular errors while translating long sentences filled with interconnections and different contexts, Neural Machine Translation system translates entire sentences at a time, having a deeper understanding of the meaning and connections between words.

Further, the new translation system uses a broader context in order to come up with the most relevant translation, it even rearranges words in the sentence and tries to use proper grammar  so translation can sound more “human.” And since it’s basically an AI-based translation system, Neural Machine Translation gets better by learning as people use it more and more. In theory, the system could become a perfect machine-translation service if used enough.

For now, the new Neural Machine Translation method is used with eight languages. As Barak Turovsky, product lead of Google Translate stated: “Today we’re putting Neural Machine Translation into action with a total of eight languages to and from English and French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish.” These eight languages present 35 percent of all translations made in Google Translate. The big plan is bringing the new system to all 103 supported languages. Google also offered its Machine Learning APIs to developers so they can check them out, maybe make them even better.