Effectively Using Monotonicity Analysis for Paraphrase Identification

We analyse in this paper the role of monotonicity for learning to identify sentence-level paraphrasing. Our approach is based in a system architecture which consists of two components. The first component is the features set definition module which takes care of the order of the elements for the analysis of monotonicity as well as the use of semantic heuristics to recognize false paraphrasing. The learning phase is carried out by the second module which makes uses of supervised learning algorithms such as logistic regression and support vector machines for the definition of the classifier model. The results of the experimentation conducted show how the set of features that we propose in this paper leads to decent accuracy. In fact, the results of the experimentation using monotonic and non-monotonic features show how our approach is a plausible alternative to cope with the syntactic and semantic diversity of a data set.