Facebook will Instantly Translate Your Posts into 44 Languages

Today, Facebook has a user base of 1.5 billion people from all across the world. However, only about half can speak English, while the rest speak in dozens of different languages. This means that they are effectively cut off from interacting with English speakers as well as from each other. It is a curious case of what is meant to be social media being asocial. This is slowly changing – today, if you come across a post written in a foreign language, you could translate it somewhat semi-effectively using the translate option on the platform.

To address this limitation when it comes to language, from this month, millions of Facebook users will be given the option of translating their own posts from their native tongue instantly to any of the 44 other languages that Facebook’s translation engine supports. These posts will then appear on the News Feeds of friends and fans in their native language.

For the first time, the social media giant is testing what it calls the ‘multilingual composer’ across Facebook’s general population. Although the testing at this initial stage is limited, the ultimate point of the exercise is to arrive at a point where all Facebook users can talk to each other, regardless of their native tongue or language preference. Necip Fazil Ayan, who is the head of the company’s translation wing, says that the development of such revolutionary technology is his main agenda and that it was, in fact, his main motivation for joining Facebook.

 

Built for Businesses and Individuals with a Multilingual Audience

The development of the multilingual composer is in response to Facebook’s recognition of the fact that many celebrities and businesses with a global audience require this service. Every day, around 5,000 celebrity types and businesses publish almost 10,000 posts in several different languages. The posts are then viewed by fans and followers about 70 million times daily, often in a foreign language.

Ayan is an avid follower of several international footballers, such as Brazilian star Ronaldinho, an early adopter of the new composer. Ronaldinho posts in Portuguese, Spanish and English, with Ayan saying he only sees the posts in English. Soon, millions of others will be able to post in a similar way.

Ayan’s team developed the composer to specifically cater to people who have a multilingual audience. The translation capability allows posters to edit the automated translations or offer their own. The ultimate aim is to have the entire process automated for all Facebook users.

 

Facebook translation technology

 

A Big Step Towards Completely Automated Translation of Online Content

Although machine translation is still far from perfect, it is rapidly improving. At present, Facebook automatically translates posts across 45 commonly spoken languages, handling the task with algorithmic models which are heavily reliant on language statistics (basically an analysis of common words and phrases used in natural language). However, English-German translations now lean toward the use of deep neural networks, which are networks consisting of hardware and software which mimic the web of neurons present in the human brain, providing more accurate translation capabilities.

Due to rapid improvements in the design of deep neural nets, they have become incredibly adept at performing certain tasks, such as identifying spoken language or recognizing faces within photographs, thanks to the vast amounts of data that they can analyze. Now, these networks are also helping to improve the understanding of natural language and machine translation; the software truly understands what the words and sentences that it translates mean. Facebook plans to push the technology across its entire automated translation engine.

A similar transformation is taking place across the internet. For example, Microsoft’s Skype has a translation service that relies on neural networks. According to Joseph Sirosh, who leads Microsoft’s team that works on cloud computing services which are involved in machine learning, the technology will be added to the company’s other automated translation services as well. Although neural networks are still some way from completely mastering machine translation, research shows that they provide a clear path towards that goal.

Ayan says that to achieve this goal, it is necessary to collect more data, because it is what neural networks thrive on. Facebook’s multilingual composer has a big role to play in this. Because actual humans can edit the automated translations as well as add their own, additional data is generated. This is particularly useful in helping users post and read in the languages which are not among the 45 which the company currently supports.

 

Conclusion

The data that Facebook collects from the new multilingual composer is useful in training the network’s natural language processing systems. Facebook claims that this will not only improve communication across the diverse groups of people from around the world who speak different languages, but will help to improve the machine translation model with the addition of words from less common languages. The development of the multilingual composer marks a significant step in the removal of language barriers from popular social networks and will also help to streamline communication and interaction in the near future.