The technology is still in the experimental phase

Feb 18, 2010 09:55 GMT  ·  By

Google is an odd company; it manages to be as nimble as a startup yet has all the advantages of a huge tech corp. It has plenty of experimental projects and small teams working on innovative products but can also tap into its vast technology catalog to create new products without too much hassle. One great example of this is the translation technology integrated into Google Goggles, still in prototype phase. Showcased by CEO Eric Schmidt himself at the Mobile World Congress this week, the technology does exactly what you'd expect, it translates the text captured in photos.

"[A]t the Mobile World Congress in Barcelona, Eric Schmidt and I demonstrated a prototype version of Google Goggles that showcases the potential of integrating Google's machine translation and image recognition technologies. In the video below, you'll see how we use Goggles to take a picture of a German menu and instantly translate the text into English," writes Google Scientist Hartmut Neven on the recently launched Google Translate Blog.

We must underline that this is a very early tech demo, so don't expect to have it on your phone too soon. Still, even if only in certain conditions, it's actually working, you can take a photo of something and have the text in it translated, something that seems a stretch even in Sci-Fi movies.

Google can do this because it already has all the powerful technologies needed to achieve machine translation of a photo. First off, the tool needs to detect if there is a text in the image and then recognize it. This is trivial for humans but it's a daunting task for computers. Luckily, Google has some of the best OCR (optical character recognition) technologies at its disposal thanks to projects like Google Books for which it scanned millions of books.

This is just half of the work, after the initial text has been discerned, it still needs to be translated, which is where Google's extensive translation technologies come in. Again, it has some of the best translation algorithms and huge amounts of translation data to compare to so, for Google Goggles, all it needs is to tap into the already existing infrastructure.

Even so, the translation feature in Google Goggles is very much experimental. For one, it only works for German to English translations though this could rather easily be expanded to include any language that uses the Latin alphabet. The biggest problem is with the OCR technology. It's one thing to understand the text in a scanned book, and even that is far from perfect, and it's a completely different thing to detect and understand text in a phone camera photo. Still, Google says that "soon your phone will be able to translate signs, posters and other foreign text instantly into your language." 'Soon' may mean at least a couple of years but, even so, it would be an impressive achievement.