Overcoming the Lack of Parallel Data in Sentence Compression

   Abstract