Microsoft Translator Uses AI to Break Language Barriers on Smartphones

Microsoft squeezes its neural network translation technology into smartphones that lack specialized Artificial Intelligence hardware.

Microsoft Mobile Translator

Travelers who rely on Microsoft Translator's offline mode when they venture into unfamiliar territory can rest a little easier. The app can now deliver higher-quality, neural network translation services to modern iOS, Android and Amazon Fire devices even when they can't access the company's powerful artificial intelligence (AI) systems in the cloud.

Users of the newly-updated Translator app, available for Android and hitting iOS devices by April 21, can now download AI-enabled translation packs, a feature that was previously available to a just a couple of smartphones from Chinese device maker Huawei.

Microsoft partnered with Huawei in 2017 on downloadable neural machine translation packs for the translation app that used the AI coprocessors in the Huawei Mate 10, and later the Honor View 10, to provide faster and more accurate translations without an internet connection. Now that capability is making its way to iPhones and stock Android devices without specialized AI chips.

Arul Menezes, Machine Translation Partner Research manager at Microsoft, told eWEEK that his group "figured out how we can run neural machine translation without specialized hardware." Noting that on-device neural networks are CPU- and GPU-hungry, getting the technology to work on a phone has been a challenge, said Menezes.

Microsoft set out to refine its AI algorithms building off the experience the company gained with the Huawei alliance. The company also tapped the processing power of NEON, an SIMD (Single Instruction Multiple Data) extension found in Arm-based mobile chipsets, to enable recent Android smartphones, plus the iPhone 5S and above, to run Translator's new offline language packs.

In terms of quality and speed, the translations provided by the new Translator offline packs are "going to be very close to our online service," Menezes assured. Since users typically use the app to translate words or brief phrases, not lengthy text strings, most users won't be able to tell the difference between the cloud-based and local translations, he added.

Overall, both the online and offline flavors of Microsoft's neural network translation technology yield faster and more fluent translations than the statistical machine translation approach of the past, said Menezes. "[We made] tremendous progress in the past few years because of machine learning and neural networks," he said.

As an added bonus, the new language packs take up less mobile storage. The move to neural machine translation has reduced the size of Translator's packs by 50 percent.

There are also some perks for Android developers wishing to add the Translator's new neural network translation capabilities to their own apps.

Using Android's bound service technology, which allows one app to interact with another behind the scenes, developers can incorporate Translator's offline, neural machine translation into their own applications with a little additional code.

If the device is connected to the internet, Translator will fetch translations from the corresponding service on Microsoft Azure. For end users, Menezes promised a seamless experience, "whether you're online or offline."

The new local translation features for Android developers is currently in beta but is expected to be made generally available within 90 days.

Pedro Hernandez

Pedro Hernandez

Pedro Hernandez is a contributor to eWEEK and the IT Business Edge Network, the network for technology professionals. Previously, he served as a managing editor for the network of...