MATHion: Solving Math Word Problems with Logically Consistent Problems

Anonymous

Abstract: Solving the math word problems (MWPs) is a challenging task. Some existing MWP solvers retrieve textually similar problems and draw on their solutions to solve the given problem. However, textually similar questions are not guaranteed to have similar solutions. And questions could share the same solution but with different descriptions. Therefore in this work, we investigate the logical consistency among different problems and propose a novel framework named MATHion which solves math word problems with logically consistent problems. Experimental results show that our method outperforms many strong baselines, including some pre-trained language model-based methods. Further analysis shows that our retrieval method can learn the logical similarity between questions and plays a key role in our model's performance.