Paraphrase Based Similar Expression Generation

In this paper we propose a novel paraphrase based method to improve the performance of the similar expressions generation that has been widely used in writing assistant system. The users’ queries are paraphrased to multiple expressions which cover abundant example sentences in the system. Three paraphrase methods are presented to satisfy the users’ different intentions. First of all, we employ a statistical collocation model to generate collocations for two-word queries. Then we use a phrase based paraphrase model and a dependency based paraphrase model to substitute contiguous phrases and long-distance collocations in multi-word queries (> 2 words). The experimental results indicate that the proposed method can effectively help the users to correct their inappropriate expressions.