Google is Using a Neural Machine System to Translate Languages

By S. Rina / 1475160360
(Photo : GettyImages/JeffJMitchell) Google's neural machine translation (NMT) system may reduce translation errors by 55 to 85 percent.

Google has announced that it will use a neural machine translation (NMT) system to make its translation service more efficient. The technology giant said that the new system might reduce translation errors by 55 to 85 percent. Google is currently celebrating the 10th-year anniversary of its Google Translate service.

The new feature has been embedded into the web and mobile versions of Google Translate. However, currently, it is only used for translating Chinese into English. Google is also using its deep neural networks for other of its applications such as Google Allo.

Google Brain Team elaborated that the company has used machine intelligence to boost speech recognition and image recognition systems. However, the use of machine intelligence for translation proved to be "challenging." The new Google Neural Machine Translation system (GNMT) uses the latest training techniques for improving machine translation quality.

Neural machine improves the translation output by considering the entire sentence to be one unit. This is in contrast to phrase-based machine translation systems which consider words and phrases independently for translation purpose. The system is reported to have worked particularly well for English to German and English to French translations, where it managed to reduce errors by up to 60 percent.

However, the system is not foolproof yet. The system still makes errors which may have been avoided by a human translator. The system is particularly prone to dropping words or mistranslating rate terms. It is also not fully capable of understanding context. The team said that "there is still a lot of work we can do to serve our users better."