How Facebook’s Convolutional Neural Network is Changing Language Translation for the Better

Language translation remains a key focus of Facebook’s mission to connect users across the world in meaningful, seamless ways. With almost 1.5 billion active daily users, the platform reaches far beyond US borders with the ability to translate over 100 languages.

As the world continues to globalize, marketing translation services become more important for brands. For Facebook, connecting people as well as supporting marketing targeting utilities for businesses offers an enormous opportunity to achieve goals.

In pursuit of lowering language barriers among users, the Facebook Artificial Intelligence Research (FAIR) team recently published results on the speed and accuracy of language translation by a convolutional neural network (CNN) versus the more conventional recurrent neural network (RNN).

NYU professor Yann LeCun, well-known for his CNN research and considered one of the founders of the field, heads the FAIR team. While others focused on the RNN approach to artificial intelligence visual data processing, LeCun championed CNN as a more useful method.
The Difference Between CNN and RNN

Facebook’s CNN process produces localized translations nine times faster than RNN, with greater speed and accuracy. The key difference between CNN and RNN comes down to the way the information is processed.

Utilizing internal memory, RNN works sequentially, translating one word at a time, from left to right or vice versa. It uses the current word to anticipate and predict the next until a complete sentence has been transformed from one language to the other. With less research done on CNNs, RNNs have come to be the dominant mode of language translation processing.

CNN, on the other hand, works on tasks in parallel. By focusing on all the words in a sentence at once, CNNs can analyze their relationships in context, similarly to how humans use spoken or typed speech. The actual work exploits the GPU’s ability to perform parallel tasks rather than how an RNN uses the CPU for its speed.

For the FAIR team, CNNs represent a more scalable approach over RNNs due to the dynamics of language. There are well over 3,000 languages in the world, and each has its own often complex rules for vocabulary, syntax, and grammar. The contextual nature of CNN language processing promises a more efficient approach for expansion over RNNs.
Facebook’s Mission Drives Its Research

As the premier social media platform, Facebook drives research in multiple fields to improve user experiences. With communication and information-sharing at its core, Facebook’s desire for quality language translation serves to underpin its artificial intelligence research.

The implications for Facebook alone are impactful. With its attendant applications like Instagram and Oculus VR, the social media giant demonstrates the dual opportunity of connecting users while offering financial benefits through improved business opportunities in marketing goods and services.

In addition, Facebook offers its source code and trained systems in the open-source environment through Github. This enables more researchers to bring their own customizations and ideas into the mainstream. The benefits of Facebook’s decision to offer its research in the open source community will mean higher performance, better translations, and faster results as CNN research catches up, and surpasses, RNN.

About the contributor :Rae Steinbach,

“Rae is a graduate of Tufts University with a combined International Relations and Chinese degree. After spending time living and working abroad in China, she returned to NYC to pursue her career and continue curating quality content. Rae is passionate about travel, food, and writing, of course.”