Facebook is once again unleashing advanced AI networks to the planet. This time, rather than being surprised by the capability of computers to create their own language, the company is utilizing them to offer its users better communication. It means that Facebook translations are going a step forward as these are now entirely powered by artificial intelligence techniques.
At the moment, 50 percent of the Facebook community doesn’t speak English and is unable to understand each other’s languages. In order to remove such communication barriers, Facebook’s Applied Machine Learning team designed an AI-based automatic translation system that helps over 800Mn people every month view translated posts in their News Feed. The team has been training its AI to better comprehend how things like typos, intent, and slang work to offer more accurate Facebook translations.
Read more at Facebook Celebrates World Emoji Day 2017
The social media mogul is making use of a type of neural-network called a convolutional neural network (CNN) and recurrent neural network (RNN) to automatically translate content on its platform.
Earlier, Facebook was using phrase-based machine translation models, breaking down sentences into phrases or words. The shortcoming of such models was particularly obvious when translating between languages with really different sentence structures. Now, the company is using neural networks to build a better translator, which can translate full sentences as well as their context, generating much more accurate Facebook translations.
The example Facebook provided is below. The first translation is produced by a phrase-based system while the second is produced by a neural network.
According to the company,
“Completing the transition from phrase-based to neural machine translation is a milestone on our path to providing Facebook experiences to everyone in their preferred language. We will continue to push the boundaries of neural machine translation technology, with the aim of providing humanlike translations to everyone on Facebook.”
Further, the use of neural networks lets Facebook deal with unknown words and slangs in an effective manner. The Silicon Valley-based giant created a ‘lexicon’ from its training data. So, when the neural network finds a word in a sentence that it hasn’t seen before, it produces a placeholder for that word and searches it in the lexicon. This also allows abbreviations like ‘tmrw’ to be translated into their intended meaning, i.e. tomorrow.
It has also been reported that the company is also starting to explore multilingual models that can translate many different language directions.
You may also like: Facebook is to Come with Its Own Video Chat Device