Artificial Intelligence (AI) has permeated more aspects of our lives in recent years than many people are even aware of. AI is no longer the stuff of fiction; Machine Learning technologies and Deep Learning algorithms are replacing older forms of technology.
In terms of translation, this means a move away from Statistical Machine Translation (SMT) and Rule-Based Machine Translation (RBMT). Today, the media translation industry is moving toward AI driven technologies of two different varieties: Neural Machine Translations (NMT) and Convolutional Network Translations (CNN).
Google’s Translation Breakthrough
In September 2016, Google announced that they were switching to a single multilingual system based on artificial neural networks. By November of the same year, Google Translate was being powered by AI tech aptly named Google Neural Machine Translation (GNMT). GNMT uses artificial neural networks designed to process information from left to right through multiple connected receptors, one word or phrase at a time.
The improvement to their translation technology was profound. The large neural network is capable of deep learning. This allows it to continuously adapt and improve the translations it produces via millions of linguistic examples and a wider context, in order to generate the most relevant translation.
As expected, the network improves over time as each connection and translation is strengthened through continued use. What was not expected, however, was the ability of this system to provide a coherent translation between two language pairs it had never tried.
The success of this ‘zero-shot’ translation caused programmers to wonder if the system had created its own language – some sort of ‘interlingua’, to successfully translate Korean to Japanese. It has since been found that this is not the case. The system first translated from Korean to English, and then from English to Japanese (all of which it had been taught to do).
Although not quite as spectacular as creating a whole new language, it did show that the system was able to transfer its “translation knowledge” from one language pair to another with surprising efficiency, and without explicit instruction to do so.
Facebook’s Translation Method
Early last year, Facebook released news of their own translation technology, purported to be nine times faster than GNMT. Rather than relying on neural networks that mimic the human brain by processing information from left to right, they created a system to simulate an animal’s visual cortex. Known as Convolutional Neural Networks (CNN), it allows their network to look at entire datasets at once.
One feature of the architecture is multi-hop attention, a method similar to the way an individual approaches a sentence when translating it. Instead of just viewing the sentence once and writing down the full translation without further reference, the network takes recurrent “glimpses” of the content to select which words to translate next. This is very similar to how humans do it, looking back at specific keywords when creating a translation.
The Impact on General AI
These two different approaches to machine translation have increased our general knowledge on how artificial neural networks work. Translation is just one of the verticals at the forefront of this technology; other applications for general AI applications are near limitless.
CNN technology is being used to detect melanomas and enhance security camera and drone capabilities. Other artificial Neural Networks are being used to diagnose disease in patients, spot patterns in consumer behavior, and reduce power outages in large and complex power systems.
Both Facebook and Google’s open source sharing of their own neural network research and development will only quicken the pace on our journey toward general AI.
About the Author: 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.