Learning task-dependent distributed representations by backpropagation through structure
While neural networks are very successfully applied to the processing of fixed-length vectors and variable-length sequences, the current state of the art does not allow the effi...