... the candidate sentence with that of thefirst ranked candidate. This is because we try to avoid a very large or small compression ra-tio, and the first candidate is generally a goodcandidate with ... sentences. In the second step, weuse a discriminative training approach to cap-ture sentence level global information fromthe candidates and rerank them. For evalua-tion, we compare our system output ... to Sentence Compression of Spoken UtterancesDong Wang, Xian Qian, Yang LiuThe University of Texas at Dallasdongwang,qx,yangl@hlt.utdallas.eduAbstractThis paper presents a two -step approach...