Work has taken me all around the world the past several years, and everywhere I have been English is spoken. In Germany, India, Malaysia, Denmark, Singapore, Hong Kong, Norway, Italy, Hungary, and even in France the commercial language is English. This is especially true in technical fields. Further, I perceived an accelerating velocity to the learning of English; soon Babel would be abolished.
Walter Russell Mead, in a must-share post to new college students says
The purpose of taking a language today has less to do with learning to talk to foreigners than it used to; foreigners seem to be learning English faster than we are learning their languages
Not so fast says Nicolas Olster in a new book The Last Lingua Franca: English Until the Return of Babel reviewed this week in the Economist.
Today’s But if Not is the story of how computer translation software is closing the communications gap.
How’d Google do it? Among other techniques, they scanned millions of books, from 52 languages; and where a book had been translated they put their algorithm to work. Stop and think about that. They don’t just have a database of direct word translations. They have mapped nuance. Where a book uses a turn of phrase that does not directly translate, the human translator has often replaced the phrase with an appropriate local language phrase – and now Google has it. Whoa! You can see the promise here.
It’s not as if this is the first time anyone has ever learned a language by reading a translation – Thomas Jefferson taught himself Italian during the long ship voyage to France using a side-by-side translation book of Dante’s Inferno. But it is the first time it’s ever been done writ large across many languages. And the more that we commercially translate user manuals, web pages, stories, and magazines the better Google’s translation engine will become.
By why stop there? Google has also figured out how to translate the spoken language. As best as I understand how they did it, Google started with English. They accumulated a database of phonems and accents with the now defunct GOOG411 service, a free 1-800 voice-powered directory assistance service. The GOOG411 caller, with regional accents in full flare, would say the name of a business (“Joe’s Auto Pahhts”). The voice system would confirm “Joe’s Auto Parts?” and whammo, they had the New England post-vowelic R. Since all audio can be digitally mapped, you can see how they could accumulate an enormous and powerful database from the millions of free directory assistance calls. Again, they weren’t limited by a linear set of rules.
I have already written about the amazing voice-to-text feature on the DROIDX. Surely they are using the voice engine derived from GOOG411. With voice now in text its a small leap to uses its translation engine to turn it into another language. But why put an intermediate step in between?
Just like we have books translated into many languages, we have audio books translated into many languages. Turn Google’s algorithms loose on these and you get the Google Translator Phone.
Further still, these technologies can only continue to get better and better; by continuously retraining the database engines against an ever increasing set of translated material.
I have work responsibility to coordinate global offerings and support local web development for subsidiary offices. Not long ago this task involved many hours of coordination. Now I use Chrome to translate and read the web pages of my international peers.
For the global 6.3 billion non-native English speakers, learning English is SO 2010.