Facebook published 9 times more accurate language translation tool research results

Facebook recently published the results of research using CNN (Convolutional Neural Networks), an AI with parallel processing capability to perform complex tasks. We already know the ability to translate 45 languages of Facebook using the See Translation option . But the accuracy level is still not perfect.

The research team said the new system can translate faster and more accurately than traditional language translation software up to 9 times . In addition, the source code is also provided in an open source format so that anyone interested can verify or use it to research and build more accurate translation tools. Scientists developing this project belong to the FAIR group, short for Facebook AI Research.

Artificial intelligence can break the language barrier

" Language translation is very important in the mission to help people connect with each other in the open world, allowing anyone to see posts or videos in their preferred language, with the fastest translation speed and accuracy. maybe ", Facebook said on a blog post.

Artificial neural networks have been used for years in the areas of image recognition to voice or online search. Last fall, Google also revealed a new translation system based entirely on artificial neural networks. But Facebook has a completely different approach than the other big guys.

CNN is not completely new technology but has never been used to translate languages ​​before. Language translation often uses RNN (Recurrent Neural Networks), processing language one-sided (for example, from left to right or right to left). CNN is paying attention to words in contextual sentences the same way people think. Previous RNN method is still "enough" when it comes to accuracy and speed, but there are certain limits that can now be solved by parallel processing model of CNN.

How to handle information of CNN

CNN was invented by Yann Lecun, a deep-learning researcher, currently working for the Facebook AI lab, which proved to be very effective when identifying images. Many people have realized that such networks can be used as basic technology for machine translation, including researchers at DeepMind, the Google AI lab in London, but according to Manning, Facebook's translation system. is the most impressive proof.

According to new research results, Facebook can handle many elements in the sentence at the same time, instructing the system to use less and less computing capacity, thereby faster and faster translation. Facebook hopes to bring this method to use on a large scale, supporting translation of more than 6,500 languages around the world.

Christopher Manning, a Stanford University professor specializing in machine translation, looked at the research report and called it " impressive achievement ," especially because it could teach the model to translate faster than the current system. Facebook also pointed out that its engineering team is preparing to incorporate the technology into the company's social networks, serving more than 1.8 billion people worldwide.

The tool promises to be integrated into Facebook applications

Facebook has not revealed anything about the plan to integrate this technology into their products, but when social networks choose CNN's processing structure as a foundation, it will be interesting to look forward to what is coming. When the language barrier is removed, the world can become smaller. We will soon see the vast impact of this technology, especially on Facebook applications like Messenger, Instagram or WhatsApp.

Interested readers can learn more about CNN and Facebook's report at this address.