Acquiring lexical paraphrases from a single corpus