February 19th, 2016

How Microsoft Uses Neural Networks To Translate Languages Across The World

Mobile App Testing Learn tips, techniques and trends for launching great mobile apps Get It Now
Like what you're reading? Subscribe to the weekly newsletter. Subscribe For Free
written by

Welcher Weg ist der Bibliothek?

The concept of neural networks—systems of computers that can think and understand like human beings—has been around for decades. But it has only been in the last couple of years that the practical application of actual neural networks has been achieved by computer scientists.

Language is the area where we are seeing the first jumps from theoretical computer science to actual human application of neural networks. Microsoft Research has been working on using neural networks for language translation for years.

Peter Lee, corporate vice president of Microsoft Research, told me in an interview in 2013 how Microsoft was focusing on language and translation as its first commercial implementation of neural networks:

One is that there has been, there is right now for us, a resurgence of hope and optimism in being able to solve some of the longest standing problems in core artificial intelligence. To get machines that see and hear and understand reason at levels that understand or match human capabilities.

“I think we are seeing that first in dealing with language. I think language is coming first because it is a little bit of a simpler problem but one that has commercial implications. So, that is moving really fast.

The Microsoft Translate engine—powered by deep neural networks—are now baked into the Bing search engine, Skype and dedicated apps for both iOS and Android. This week, Microsoft introduced new features for the Android and iOS translate apps that should make it much easier for people to get real-world translations from wherever they are.

5 Issues Facing the Future of the Apps Economy Get ahead of the future of the apps economy by knowing the current trends in app development Get It Now

Taking Neural Network Translation Offline For Android


The Microsoft Translate app for Android will now work offline if a user downloads language specific packets within the app.

Microsoft touts that the offline translation is powered by deep neural networks and is the first to be able to provide high quality translations when a user is not connected to the Internet.

“We are pleased to introduce today the world’s first Deep Neural Network-powered offline engine, available in the Microsoft Translator app for Android. By downloading free offline packs, users can get near online-quality translations, even when they are not connected to the Internet,” Microsoft wrote.

The offline packs are first available in Chinese Simplified, French, German, Italian, Polish, Portuguese, Russian, Spanish and Vietnamese.

The packets are fairly heavy. We downloaded the Italian and French packets (ARC will be traveling in Europe over the next two weeks) and found that they are about 250 MB each … a substantial size of memory for one app to consume on a smartphone. We’ll find out in the next couple of weeks if the offline packets are worth it.

Translating Pictures: Optical Character Recognition For iOS

Optical character recognition—OCR—is how machines read real images and understand them. One of the most famous uses of OCR has been how the Google Books initiative has scanned and digitized thousands of books and scripts to put them online. OCR has been mixed with regressive neural networks by researchers at the University of California, Berkeley to scan, understand and digitize ancient Tibetan documents.

Microsoft will use OCR in the Microsoft Translate app for iOS to allow people to take pictures of signs in other languages and have the translation overlay the picture on a person’s smartphone.

Microsoft stated:

This feature lets you translate text from your camera roll or saved pictures. Rather than typing the text or speaking aloud, you can translate pictures from your phone with the translation appearing in an overlay above the existing text in the picture.

Translation from your camera roll is great for translating signs, menus, and fliers; and translation of saved images is perfect for images from emails, the Internet, and social media.

Google Translate Now Covers 103 Languages

Google Translate has been available since 2006 and also use neural networks and machine learning to perform translation through the Internet. Google announced this week that it has added 13 new languages to Google Translate that brings the number of languages covered by Google to 103. Google said that the 13 new languages, which cover some out-of-the-way dialects such as Scots Gaelic, Samoan and Luxembourgish (what they speak in Luxembourg, apparently) cover 120 million people across the globe.

“As we scan the Web for billions of already translated texts, we use machine learning to identify statistical patterns at enormous scale, so our machines can ‘learn’ the language,” wrote Sveta Kelman, senior program manager for Google Translate.

  • valakos

    google translate is not very accurate, needs to improve