Abstract
We present a method for improving machine translation (MT) evaluation by targeted paraphrasing of reference sentences. For this purpose, we employ MT systems themselves and adapt them for translating within a single language.
We describe this attempt on two types of MT systems -- phrase-based and rule-based. Initially, we experiment with the freely available SMT system Moses.
We create translation models from two available sources of Czech paraphrases -- Czech WordNet and the Meteor Paraphrase tables. We extended Moses by a new feature that makes the translation targeted.
However, the results of this method are inconclusive. In the view of errors appearing in the new paraphrased sentences, we propose another solution -- targeted paraphrasing using parts of a rule-based translation system included in the NLP framework Treex.