By Peter Griffin 29/11/2016


If you’ve experienced the anxiety and embarrassment of trying to communicate with people whose languages you don’t understand, you’ll be very interested in what the newest generation of translation technology is allowing.

Earlier this month I visited Microsoft Research in Redmond, Seattle and got to see the newest version of Microsoft Translator, which allows real-time translation of group chats. I had a three-way conversation in English, French and German, each of us talking into our smartphones and seeing the appropriate translation on our screens.

It wasn’t perfect – our words were occasionally misinterpreted. But it was powerful enough to give us a coherent transcript of a conversation conducted in three languages by three people with distinctive and differing accents and vocal cadence.

Microsoft's Olivier Fortana with smart devices running the new Microsoft Translator
Microsoft’s Olivier Fortana with smart devices running the new Microsoft Translator

The technology is the result of recent advances in speech recognition, artificial intelligence and machine learning.

“We know what the Holy Grail is, its the universal translator from Star Trek, its the Babel fish,” said Olivier Fortana, director of product strategy for Microsoft Translator.

Listen to an interview and demo with Microsoft’s Olivier Fortana of the group translate function of Microsoft Translator (CREDIT Christian Sachsinger)

“We didn’t want to do the hardware, it doesn’t make sense, everyone has a smart device. So what we did was say let’s build this universal translator using the smart devices people have. We are talking about real-time, multi-language translation for in-person conversations.”

If you are in any doubt as to the usefulness of that, Fortana has a compelling example – the hundreds of thousands of refugees currently settling into German society and trying to get to grips with the language.

“From day one they could go into a regular German school. They could follow what the teacher is saying as he is talking and as their German gets better, they slowly and slowly use it less.”

Group chat

Microsoft Translator is already available as an app on Android, iOS, Windows devices and via web browsers that allows translation of two-way conversations in a number of languages and even lets you take photos of street signs to get a translation on your smartphone. It’s incredibly useful technology, the engine of which drives translation across Microsoft services like Skype and Bing.

The next iteration of the app, likely to be available before the end of the year, is the multi-lingual group chat translation that I saw a demo of in Redmond. It will support text translation of 60 languages and translation of nine spoken languages.

Microsoft Translator records the spoken words of its users for the purposes of improving its translating accuracy, said Fortana – who shares a curiously similar name to that of Microsoft’s speech-activated personal assistant Cortana, which is embedded in Windows 10.

“Obviously we are not talking about replacing translators at the UN,” he said.

“The quality is not here yet. People ask me, how can you guarantee that it is good? We can’t, it is machine translation, if you want to guarantee it, you put in a human. It is a scenario based approach. What is the scenario and what is the alternative? We are talking about a free system that opens new things that were totally closed before.”

Neural translation

Microsoft Translator is now powered by neural networks, which are computers modeled on the human brain and nervous systems. They allow more accurate translation because they mimic human speech patterns, producing more human-sounding translations.

The more conversations Translator processes with neural networks, the more the system will learn and accurately reproduce what humans intended to say. Neural networks are increasingly replacing the Statistical Machine Translation that powers translation services. Google Translate is also utilising neural networks.

While the technology powers Microsoft’s own services, it is also licensed as an API to developers to run text and speech translation in their own services.

You can try out neural network-based language translations based on Microsoft’s technology here.

translator

Peter Griffin visited Redmond as a guest of Microsoft.

Audio courtesy of Christian Sachsinger from Munich public service radio station Bayerischer Rundfunk

TOMORROW: Microsoft and the future of food – how the internet of things and smart sensors are being harnessed to improve food production.