A Joint Probabilistic Classification Model of Relevant and Irrelevant Sentences in Mathematical Word Problems