Title: A paraphrase-based exploration of cohesiveness criteria
Abstract:This paper proposes an empirical approach to the development of a computational model for assessing texts according to cohesiveness. We argue that the NLG technologies for the generation of structural...This paper proposes an empirical approach to the development of a computational model for assessing texts according to cohesiveness. We argue that the NLG technologies for the generation of structural paraphrases can be used to efficiently create what we call a cohesion-variant parallel corpus, which would serve as a good resource for empirical acquisition of cohesiveness criteria. We also present our pilot case study, in which we took a particular type of paraphrasing that separates a relative clause from a sentence. We have so far created a cohesion-variant parallel corpus containing 499 cohesive instances and 841 incohesive instances. Based on this corpus, we conducted a preliminary experiment on cohesion evaluation, obtaining encouraging results.Read More