The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration.