Toward Automatic Tutoring of Math Word Problems in Intelligent Tutoring Systems

Math Word Problem (MWP) solving, which involves solving math problems in natural language, is a prevalent approach employed by Intelligent Tutoring Systems (ITS) for teaching mathematics. However, one major drawback of ITS is the complexity of encoding all potential solutions for each problem supported, which is both time-consuming and labour-intensive. In this study, we propose a novel method for automatically converting the statement of a previously unseen MWP into the internal representation of an ITS, thereby simplifying the task of adding new MWPs by only requiring the problem statement. To accomplish this, we propose the use of large pre-trained language models to translate the problem into Python code, which can then be easily imported into an ITS. Experimental results indicate that this approach is effective and suitable for the task, and as language models continue to improve, the accuracy rates are expected to increase further