A methodology for generating natural language paraphrases

In this paper we address the topic of generating automatically accurate, meaning preserving and syntactically correct paraphrases of natural language sentences. The design of methods and tools for paraphrasing natural language text is a core task of natural language processing and is quite useful in many applications and procedures. We present a methodology and a tool developed that performs deep analysis of natural language sentences and generate paraphrases of them. The tool performs deep analysis of the natural language sentence and utilizes sets of paraphrasing techniques that can be used to transform structural parts of the dependency tree of a sentence to an equivalent form and also change sentence words with their synonyms and antonyms. In the evaluation study the performance of the method is examined and the accuracy of the techniques is assessed in terms of syntactic correctness and meaning preserving. The results collected are very promising and show the method to be accurate and able to generate quality paraphrases.