Recent approaches to data-to-text generation have shown great promise thanks
to the use of large-scale datasets and the application of neural network
architectures which are trained end-to-end. These models rely on representation
learning to select content appropriately, structure it c