Microsoft and Google compete for neural translation crown
Wed 16 Nov 2016
As Google introduces details of its new Neural Machine Translation system, Microsoft too has announced the use of neural networks behind its Translator service.
In a blog post, the Translator team explained that all of its translation apps, including Skype Translator and the Microsoft Translator app for mobile devices, are supported by ‘state-of-the-art’ neural network technology.
Using the scaling capacity and compute power offered by Microsoft’s AI supercomputer and Cognitive Toolkit, the service has released the neural offering across a range of languages: Arabic, Chinese Mandarin, English, French, German, Italian, Brazilian Portuguese, Russian and Spanish – as well as Japanese text and additional languages yet to come. This represents over 80% of translations performed every day over Translator, according to Microsoft.
For now, Google lags behind slightly in its neural translation in terms of the supported languages offered. In the release by Google Translate product lead Barak Turovsky, the company announced that eight languages would be covered by the new Neural Machine offering – English, French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish. Google explained that these languages cover around 35% of Translate queries.
Google also noted that it plans to roll the neural capability out to its total 103 Translation-supported languages in the future.
Neural networks have seen growing interest over recent years, particularly in speech and image-processing applications. In the virtual translator sphere, neural networks are providing opportunities for significant advancements in translation quality over traditional Statistical Machine Translation (SMT), capable of capturing the implied context of a sentence rather than simply translating word for word.
‘Even though Microsoft’s use of neural networks for speech and text translation is still at an early stage, it is producing superior translations to what SMT provides. As with any new technology (we’re in the so-called ascent phase of the s-curve), we know the quality improvements neural networks provide today are only a first step towards future improvements,’ said Microsoft Translator in its statement.
Facebook is also working on neural networks for translation purposes. In July, the social network announced that it is deploying neural technology to build a ‘multilingual composer’ which automatically translates posts for users into their preferred language.