Neural Machine Translation systems are more and more effective. However, they are still far from reaching the human level. One of the reason is that the machine is using text only, lacking of general context. I will present our recent research work on integrating visual information as context into an NMT system. I will then discuss about the quantitative and qualitative aspects of the obtained results.
I participated in the Conference on Deep Learning : from theory to applications at Technicolor in Cesson-Sévigné (near Rennes, France), September 6th.