Explaining Math Word Problem Solvers
This reveals a critical flaw in automated solvers for educational applications, showing they may overfit to specific words rather than solving problems logically.
The study investigated the reasoning of neural network-based math word problem solvers by removing parts of the input and found that models often rely on superficial patterns, achieving correct answers even with nonsense questions, indicating a lack of semantic understanding.
Automated math word problem solvers based on neural networks have successfully managed to obtain 70-80\% accuracy in solving arithmetic word problems. However, it has been shown that these solvers may rely on superficial patterns to obtain their equations. In order to determine what information math word problem solvers use to generate solutions, we remove parts of the input and measure the model's performance on the perturbed dataset. Our results show that the model is not sensitive to the removal of many words from the input and can still manage to find a correct answer when given a nonsense question. This indicates that automatic solvers do not follow the semantic logic of math word problems, and may be overfitting to the presence of specific words.