New frontiers in supervised word sense disambiguation: building multilingual resources and neural models on a large scale