Why are NLP Models Fumbling at Elementary Math? A Survey of Automatic Word Problem Solvers

Anonymous

Abstract: From the latter half of the last decade, there has been growing interest in developing algorithms for automatically solving mathematical word problems (MWP). It is an exciting language problem which demands not only surface level text pattern recognition but requires coupling with mathematical reasoning as well. In spite of the dedicated effort, we are still miles away from building robust representations of elementary math word problems. In this paper, we critically examine the various models that have been developed for solving word problems, their pros and cons and the challenges ahead. In the last two years, a lot of deep learning models have come out with competing results on benchmark datasets. We take a step back and analyse why, in spite of this, the predominantly used experiment and dataset designs are a stumbling block and provide a road-map for the future.

Paper Type: long