Automatically Solving Number Word Problems by Semantic Parsing and Reasoning
- Shuming Shi ,
- Yuehui Wang ,
- Chin-Yew Lin ,
- Xiaojiang Liu ,
- Yong Rui
The 2015 Conference on Empirical Methods in Natural Language Processing (EMNLP 2015) |
This paper presents a semantic parsing and reasoning approach to automatically solving math word problems. A new meaning representation language is designed to bridge natural language text and math expressions. A CFG parser is implemented based on 9,600 semi-automatically created grammar rules. We conduct experiments on a test set of over 1,500 number word problems (i.e., verbally expressed number problems) and yield 95.4% precision and 60.2% recall.