In the past month(s) some of the news coming out from major technology companies have caught the media’s attention. At the same time my co-host for the Globally Speaking Podcast, Renato Beninatto , was really pushing us to make our most complex podcast to date on the topic as part of a series on the machines taking over:) Since then I have started doing my research in preparation for our conversations with leaders in the field. At this point I’m likely to help others talk intelligently at a cocktail party.
To date much of the conversation around Machine Translation, which has been around since the 1950s, mainly consisted of statistical and rule-based systems. Rule Based systems as the names signifies is based up linguistic rules that set how words will be translated, words from target language will replace the source language. Statistical machine translation focused on pattern recognition within translation and provided target based on huge amounts of parallel texts. Most of the effective machine engines for a time ended up being a Hybrid machine translation engine that incorporated the best of both methodologies..
Now neural networks are on the scene. To understand the effectiveness and basic outline of this technology check out a this great article, From not working to neural networking. As I understand the strength of the neural network is the depth of the data that is process. Rather then being limited to a number of rules or a corpus of strings to improve the quality. Neural machine translation operates beyond a string and by exposing it to a huge number of examples it will learn without telling it what to look for. There is a sight that even allows you to demo the Neural Machine Translation by LISA, which has been trained on a lot of data from the UN and European Parliament.
So since this has been going on for some time, why the news all of a sudden? For one Alan Packer from Facebook came out and said that the other forms (specifically statistical) of machine translation have reached their usefulness and Facebook is now focusing on the use of neural networks. Check out Rachel Metz’s article, Facebook Plans to Boost Its Translations Using Neural Networks This Year. Then this month Google has come out and said that Google Neural Machine Translation reduces errors by 60 percent, cool. NPR picking it up here, they interviewed a translator naysayer who felt the need to reinforce that professional translators will be needed.
Then during Google’s “Made by Google” Event, the implications of this break thru and artificial intelligence (AI) were discussed by Sudar Pichai during the early part of the event. The part about the implications for voice technology were interest, why do we only have one voice?
With all the excitement it is important to have other voices outside of Google and Facebook, so that we don’t all get caught up in the next wave of hype. For that check out the article, Hyperbolic? Experts Weigh In on Google Neural Translate from Florian Faes at Slater. Overall the opinions are rather favorable, so check it out.
Now you are on the road of looking like a star with your friends and new acquaintances. Glad to hear about other resources you are finding key on getting you up to speed with this great technology, if it’s something we use I’ll be sure to mention it on the podcast!
Sometimes I wish that I could go into a time machine right now and just look at my self and say, ‘Calm down. Things are gonna be fine. Things are gonna be all great. Just relax.’