Describir: Learning Distributed Selective Attention Strategies with the Sigma-if Neural Network