While I was away and the whole world was on vacations, there has been some news in the domain of artificial intelligence and neural networks. And this news concern my field of predilection – machine translation.

At the beginning of the year I described and tested Google Translate revolutionary features of instant translation combined with the visual recognition. Last week Google reported to have added a whole bunch of new languages available to this service, passing from 7 to 27 languages. Amazingly enough, according to the WIRED Magazine UK, this “leap in functionality, has been down to the use of neural machine networks, that can process even more vast amounts of data in a fraction of the time”[i]. Google Translate software engineer Ottavio Good explains in the official blog post how “Google Translate Squeezes Deep Learning onto a Phone”[ii] providing a step by step explanation of the whole process – from image recognition sorting out letters and bringing them into words to integration with the translation service. The deep learning technology is used to train neural networks to recognize letters and be able to separate them to other objects that may be seen in the image – background pictures, landscape, etc. Ottavio explains the challenge of making neural networks work efficiently on the limited resources of a mobile phone having slow Internet connection and little of computational power. Google has approached the challenge by “developing a very small neural net”[iii] and limiting severely “the density of information it handles”[iv].

However deep learning is not limited to natural language processing, although it has allowed great progress in the area.

MIT Technology Review reported from the Deep Learning Summit in Boston earlier this year describing how deep learning technology is used by eBay, Baidu and pharmaceutical companies[v]. Deep learning capacity to train neural network to recognize objects and concepts and cognate the world is being used extensively in modelling fails and breaches in software, “predicting market shifts like a sudden plunge in a currency’s value[vi] or computerized rug discovery.

The technology is changing the industries bringing in more possibilities and changing job market.

image source: pixabay.com



[i] Google app adds 20 languages to text translation service by Liat Clark for WIRED, July 29, 2015, online http://www.wired.co.uk/news/archive/2015-07/29/google-translate-has-27-languages, accessed on August 2, 2015

[ii] How Google Translate squeezes deep learning onto a phone by Ottavio Good, July 29, 2015, online http://googleresearch.blogspot.co.uk/2015/07/how-google-translate-squeezes-deep.html, accessed August 3, 2015

[iii] How Google Translate squeezes deep learning onto a phone by Ottavio Good, July 29, 2015, online http://googleresearch.blogspot.co.uk/2015/07/how-google-translate-squeezes-deep.html, accessed August 3, 2015

[iv] How Google Translate squeezes deep learning onto a phone by Ottavio Good, July 29, 2015, online http://googleresearch.blogspot.co.uk/2015/07/how-google-translate-squeezes-deep.html, accessed August 3, 2015

[v] Deep Learning Catches On in New Industries from Fashion to Finance by Will Knight for MIT Technology Review, May 30, 2015, online http://www.technologyreview.com/news/537806/deep-learning-catches-on-in-new-industries-from-fashion-to-finance/, accessed on August 3, 2015

[vi] Deep Learning Catches On in New Industries from Fashion to Finance by Will Knight for MIT Technology Review, May 30, 2015, online http://www.technologyreview.com/news/537806/deep-learning-catches-on-in-new-industries-from-fashion-to-finance/, accessed on August 3, 2015