Facebook Develops a Faster and More Accurate Translation System

Facebook said on Tuesday its machine-learning experts have developed a neural network that translates languages at speeds greater than nine times and more accurately than current translation systems that use standard text translation.

The scientists who developed the new system work in the company’s FAIR group, an acronym for Facebook Artificial Intelligence Research.

“Neural networks are designed to emulate the human brain,” says Michael Auli, a researcher at FAIR who is behind the new syustem. “One of the problems the neural network can solve is to translate a sentence from one language to another, and it can be used to do things like summarizing the text.

There are many types of neural networks. The standard approach to date has been based on the use of repetitive neural networks to translate text, which look at each word individually and then predict what the resulting word will be in the new language. They learn the sentence while reading it. But researchers on Facebook followed another technique called “adaptive neural networks” that look at words in groups instead of each word separately.

“For example, a translator using the neural network can look at the first five words of the sentence, in order to figure out the meaning of the word,” he says. In other words, the system works in parallel with itself.

Facebook has not yet said how it plans to integrate the new translation technology with its consumer products, but until then it has introduced the technology to the public as an open source so that other programmers can benefit from it.

The Facebook network now has more than two billion active monthly users who speak different languages of the world, and supports the translation network for more than 45 languages.

Reference