Paper: Paraphrase Generation as Monolingual Translation: Data and Evaluation

ACL ID W10-4223
Title Paraphrase Generation as Monolingual Translation: Data and Evaluation
Venue International Conference on Natural Language Generation
Session Main Conference
Year 2010
Authors

In this paper we investigate the auto- matic generation and evaluation of senten- tial paraphrases. We describe a method for generating sentential paraphrases by using a large aligned monolingual cor- pus of news headlines acquired automat- ically from Google News and a stan- dard Phrase-Based Machine Translation (PBMT) framework. The output of this system is compared to a word substitu- tion baseline. Human judges prefer the PBMT paraphrasing system over the word substitution system. We demonstrate that BLEU correlates well with human judge- ments provided that the generated para- phrased sentence is sufficiently different from the source sentence.