Google Translate now makes use of machine learning for Chinese to English translation
- Fariha Khan
- September 28, 2016
- 479
Adding machine learning to a number of its products, Google has recently introduced a new Google Neural Machine Translation (GNMT) system in Google Translate. Already being used as many as 18 million times a day, this system dramatically reduces errors up to 85%! Considering how tough it is to pair Chinese to English, the GNMT is doing a great job and Google deserves applause for this development.
Read also: Google takes Photo Sharing Experience to the Next Level
Machine learning vs. phrase-based production system
Compared to the earlier phrase-based production system, the Google Neural Machine Translation works in a wonderful way by reducing the errors to a considerable extent. The system was tested by Google by carrying out comparisons of tested sentences from news websites and using multi-lingual human raters.
The downsides of machine translation
Despite the fact that this system by Google is a great development, on the whole, machine translation still errs the way human translators don’t. Some of these mistakes may include dropping words, mistranslating rare terms or precise names. Translating sentences literally rather than contextually is another common error made by these systems.
In the past, neural networks were not adequately fast for real world positioning in products used by real people. But, with the help of machine learning toolkit TensorFlow and Tensor Processing Units (TPUs) hardware it became possible for Google to use the new system. Both these developments offer sufficient computational power to make use of the GNMT models, meeting Google Translate’s latency requirements.
Google’s GNMT will soon be rolled out to other language pairs
In near future, Google is planning to incorporate the other ten thousand language pairs that Google Translate supports using its Google Neural Machine translation GNMT, helping people around the world make the most of Translate.