Skype Translator is on the way

In May, Satya Nadella and Skype Corporate Vice President Gurdeep Singh Pall unveiled Skype Translator, Microsoft’s breakthrough in real-time speech translation at Re/code’s inaugural Code Conference. Since then, the engineering team has been hard at work to get the technology behind Skype Translator ready for a preview release. Starting today, we are rolling out a Skype Translator preview program sign-up page.

There is more here, the pointer is from Lotta Moberg.


I told my wife I shouldn't spend years learning Russian to speak to her parents because future. There is no Great Stagnation.

Maybe I should make a bet...

Even if all translation software fails miserably, no one — and I mean no one — should learn a new language simply to communicate with his wife's parents.

Actually, not needing to talk / listen to my mom-in-law sounds like a win-win to me.

Awesome, just that.

Probably was a spinoff from an NSA product.

A computer science/linguist researcher recently said that computers are now better at picking out linguistic patterns than humans. He is convinced that within ten years, computer translation will be indistinguishable from the top human translations. Of course, the computer translations will be free, and we'll see this on Skype like interpretation as well.

There is no Great Stagnation.

(But we always knew that.)

Languages are the next cursive in education. Look for them to disappear from schools in a generation.

Languages or foreign languages?

Foreign languages.

The larger trend is already there, isn't it? Many PhD programs in Science / Engineering used to have a foreign language requirement. Most got rid of it in the 1990's.

Has there been any real progress in the quality of automatic language translation?

The last time I tried Google Translate it was marginally usable in what I'd assume are best-case scenarios: English-to-Other Western European Languages.

Deep neural nets have made great progress in statistical pattern recognition and no progress at all in language understanding. This is an impressive example of what you can do without any understanding of what is being said.

Until the translator understands what is being said, there is a ceiling on the quality of the translation.

Sounds circular. Whenever quality of translation improves we will say, aha, now the AI is understanding what's being said?

My language pair is Japanese/English and I'd say there have been significant improvements in what is considered the hardest major language. Google Translate's improvements from 2005 then around 2009 allow Indian translators, who often work for rates around 1/4 that of Western translators, to figure out more difficult grammar which wasn't possible before 2011 or 2012.

And it will keep improving until essentially perfect.

Well, it's almost a Babel fish. (I actually forgot that the fish didn't just translate for you, it actually connected directly into your brain so that you automatically understood all languages. Just looked it up.)

Anyway, yeah, so add translators and interpreters to the list of jobs which technology is eating. And before someone says "this will never replace a skilled interpreter!" - yes, maybe if you're an American executive meeting with your Chinese supply chain partners, and you need absolute trust in the idiomatic translations, you'll spring for the human. But for a lot of lower-end stuff the computer will be adequate.

Then you've got the cool new applications that it brings about, like of course Skype with people in other languages, or simultaneous auto-translation of local events that normally couldn't afford human interpreters.

Do people actually pay for skilled human translators for stuff any more lower end than "American executive meeting with Chinese supply chain partner"?

Great! Soon we'll be able to receive unwanted telephone calls from around the whole world. :-(

Eventually digital live voice translation will be feasible, but this first attempt is going to be ridiculously unusable. Take the errors of the best digital transcriber, compound them with the errors of the best digital translator, and that's your best case scenario. And I'm not exactly betting on best case scenario here.

Whenever people scoff at this or other developments, saying it will never work or work well enough, I think of Kurzweil's response to why only 1% of the human genome had been mapped after years of work:

"Great, we're almost finished."

And he was right.

I pulled my kids out of mandarin in Singapore's school system. The enormous amount of time and effort required to learn the language simply isn't justified by the progress of instant translation technology.

In Canada they spend $1 billion a year on mandatory bilingualism to appease Quebec. Interesting times ahead.

For all the hype about the human genome mapping project, has it actually led to any significant applications? I'm not being sarcastic, just curious. 23andme perhaps?

I wonder what will be next - class action suits against Skype for negligent misinterpretation whereby people lost money or incurred injuries due to mistranslated calls with customer service personnel? Judging from experience of Google Translate and a few other services you either make moronically easy sentences that cannot convey anything but the most basic concepts or you get "lost in translation". Now, if you don't at least understand the target language a little bit, you can risk money, life and limb that way!

Comments for this post are closed