Pruning Sparse Non-negative Matrix N-gram Language Models

   Abstract