By David Meyer
August 4, 2017

When Facebook shows you an automated translation of someone’s post, the tech behind that translation is now based entirely on neural networks – essentially, brain-like systems that are among the building blocks of today’s artificial intelligence efforts.

Facebook announced the move Thursday. Previously, it had been using a combination of technologies, also including good old-fashioned phrase-based machine translation models.

As the company noted, phrase-based machine translation doesn’t work so well when translating between languages that order words in very different ways, because the technique relies on breaking sentences down into phrases.

The neural network model, on the other hand, makes it possible to “take into account the entire context of the source sentence and everything generated so far, to create more accurate and fluent translations,” wrote Facebook’s Juan Miguel Pino, Alexander Sidorov and Necip Fazil Ayan.

Get Data Sheet, Fortune’s technology newsletter.

That essentially makes for “more accurate and fluent translations,” they wrote, noting an average relative increase of 11 percent in a commonly-used metric used for scoring the accuracy of machine translations.

Neural networks are computing systems that simulate the highly interconnected, flexible nature of biological brains in order to “think” in a similar way to how we think. They’re very useful for image and speech recognition, and other applications where you might get better results from training a system to learn for itself than from giving it set rules.

Google also uses neural networks to power some of the machine translations in its Google Translate service. For both companies, as well as rivals such as Microsoft, Apple and Amazon, the ability for their nascent AI technologies to understand context is crucial to their hopes for building human-like virtual assistants, bots and other futuristic interfaces for their users.


You May Like