Facebook’s New Artificial Intelligence Technique for Language Translation

Normally, when Facebook is in the news we change our passwords and update our privacy settings so our coworkers don’t see what we’re really doing all day! Today, though, we’re talking about Facebook’s latest artificial intelligence breakthrough. Just a few days ago, the tech mogul’s Artificial Intelligence research team (FAIR) published a paper explaining that Facebook has developed new technology that will significantly enhance their abilities to translate quickly between two languages, especially when one language is somewhat more obscure.

In the past, we have written articles about Neural Machine Translation (NMT) which allows machines to translate between two languages when they have “knowledge” of both languages. For example, “We are going out,” in English to “Salimos,” in Spanish. However, to do such translations, the language for both sentences already had to be stored on the machine’s server so it could recall the information. Then, the machine could recall the translation more quickly and build its own vocabulary over time with the help of a human. Until now, Google translate has primarily dominated the online translation scene with their system being the most successful, reliable, and learning quickly.

Now, Facebook has developed something called Unsupervised Neural Machine Translations (UNMT) which gives machines learning algorithms that allow for the translation of sentences without pairing the sentences themselves. It may seem like magic but Facebook’s algorithms can actually understand language and translate in real time without recalling stored words in their systems. Systems now at leading research institutions are actually able to learn languages, and it’s going to revolutionize the translation industry, we believe.

How will this change the industry? Normally, humans are paid to do manual translations, which can be very time consuming and expensive. Of course, we still believe that a human translation, or at least a once over for proofreading is absolutely fundamental before publishing anything online or in print, but that doesn’t mean we don’t think technology can be of great assistance. Facebook’s publication however explains that new training systems would allow artificial intelligence to identify a house in a photo without seeing or hearing the word “house” itself. The technology scans images to learn the language, instead of relying on words, so humans will no longer have to input manually the actual translation.

This means that maybe, machines could actually help us to translate somewhat forgotten languages into more mainstream languages. Almost like reading hieroglyphics, machines could teach us what humans have forgotten through the use of neural signals in the brain when certain words are heard by descendants or poor speakers of lost languages.

Facebook has done developers a favor, as well, by opening access to the coding for this new technology for free. Anyone can work to improve the system, making it a collaborative global effort that will hopefully benefit humanity as a whole. The company is collaborating with academic institutions to continue improving the technology, and will continue to share its findings online. We’re not sure if we believe Mr. Zuckerberg when he says he wants to bring people together, but this certainly seems like a genuine effort to improve communication across borders and use machines to connect us in real life.

Facebook has made the nitty-gritty details available online, and we’ll be sharing them soon on our social media outlets, but for now, keep an eye out for news about more UNMT developments. We will certainly be standing by to see how we can integrate this new technology into our current systems to ensure that our translators are as efficient and cost-effective as possible!