The dominant sequence transduction models are based on complex recurrent or convolutional neural networks which include an encoder and decoder component. The best performing models also connect the encoder and decoder via an attention mechanism.