The company has recently announced that they have been testing out a new method of translating using deep learning AI technology which apparently is capable of achieving “state of the art accuracy” at nine times the speed offered up by more traditional methods of language translation technology. This is thanks to the use of recurrent neural networks (RNN) versus the traditional methods of convolutional neural network (CNN) which has been in use for many years.
This works by analyzing the data sequentially by working from left to right through a sentence to translate it word by word. CNNs on the other hand looks at different aspects of data simultaneously which is apparently more suited towards GPU hardware. Speaking to The Verge, Facebook AI engineers Michael Auli and David Grangier note that their work is more research based at the moment, but there are plans to eventually implement it into the social network.
According to Grangier, “We’re currently talking with a product team to make this work in a Facebook environment. There are differences when moving from academic data to real environments in terms of language. The academic data is news-type data; while conversation on Facebook is much more colloquial.”