Immediate Translation Available Now with Google Translate New Features

Remember we have discussed simultaneous translation proposed by Microsoft for Skype? That was back in December and the tool was still in preview (and it still is[i]).

Well, after a short period of rumors and announcements, Google has actually launched a similar feature in Google Translate app released simultaneously for Android and iOS. There are two major feature updates actually. Updated Google Translate uses the camera for instant translations of written text (working offline!) with the recently acquired Word Lens technology[ii]. The text recognition is pretty neat, the translation is very much word-by-word as for now.  Integrating different segmentation patterns (not only words and paragraphs) and the translation of phraseological units will be very much welcome, but this feature is already a great help for travelers!

Another feature (which was already partially present in the Android app, but has been greatly enhanced) is actually instant speech translation. And this feature, in spite of all hopes, is not available offline so far.

As the feature is based on the combination of linguistic services – speech recognition and machine translation, it is unavailability without Wi-Fi connection should possibly be due to the slow and rather inefficient speech recognition performance offline. (As discussed earlier Google has introduced offline speech recognition features in Android KitKat version and has opened the corresponding API). Still it depends heavily on the smartphone’s memory and the processor’s calculation capacity, so is basically limited to separated words and most frequent commands. The translation feature is rather performant though. Google has opted for statistical machine translation that was quite an innovative approach. The volumes of data that Google possesses and efficient machine learning mechanisms allow to have efficiently trained engine and provide good translations.

Coming back to the speech recognition step of the instant translation tool, another amazing feature, after the language pair is indicated, the speech recognition looks for both languages at the same time and is capable to tell one language from the other.

It is extremely interesting to see that the speech recognition is the component that has weaker performance compared to the one of the machine translation component – I have tested it with Russian, English and French. It is rather smooth with English and much slower and erroneous with French and Russian. The user base for English speech recognition should be massively bigger, and Google, leveraging greatly from the network effects of its tools (having more users lead to improving and enhancing the service), has made it efficient.

The good news is that the “Star Trek Universal Translator” feature of Google Translate has been reviewed greatly in French media including 20minutes.fr, Android France, Le Monde Informatique, CNET France, etc. Having a bigger base of early adopters in France could allow improve speech recognition engine and have a more intelligent (trained) mechanism.

 


[i] Skype Translator http://www.skype.com/en/translator-preview/, accessed January 15, 2015
 

[ii] Chris Welsh “Google buys Word Lens, the app that translates languages with your phone's camera” for The Verge, May 16, 2014, available online http://www.theverge.com/2014/5/16/5724230/google-buys-word-lens-translation-app