Learning N-gram Language Models from Uncertain Data

   Abstract