Here comes a Faster and Accurate AI-Based Translation System by Facebook

May 11, 2017, 1:16 p.m. By: Hasan Ali Gumani


There are billions of users on Facebook, and they speak oodles of languages. At present, the social media giant supports more than 45 languages. It simply means that if you speak English and you come across a Spanish or let’s say, you speak French and see content in German, you see a link that says “See Translation.”

On Tuesday, Facebook declared that its team of machine learning experts is done creating a neural network that translates languages nine times faster and very accurately as compared to other systems. The team that developed the new system works at FAIR – Facebook A.I. Research.

Modeled after the neural network of the human brain, the new system helps in translating a sentence from one language to another and summarizing the text.

There are multiple neural networks, but only the standard approach has been put to use so far. This approach works by looking at one word at a time and predicting the output word in the new language. However, Facebook went a step ahead and used a technique called convolutional neural network (CNN). This method looks at words in a group rather than in isolation.

Though Facebook has still not announced how it is planning to integrate the new technology but it has released the tech publicly as open-source so that other coders can leverage its benefits.