Solving Arithmetic Mathematical Word Problems: A Review and Recent Advancements

Abstract

This paper studies the research problem of solving mathematical word problems (MWPs) and reviews the related research and methodologies. Word problems are any numerical problems written in natural languages like English, based on any subject domain (mathematics, physics, chemistry, biology, etc.), and MWPs relate to word problems in the mathematics domain. Solving MWPs has been a long-lasting open research problem in the field of natural language processing (NLP), machine learning (ML), and artificial intelligent (AI); however, unlike other research problems in NLP, ML, and AI, it has not made much progress. MWPs which can be easily solved by second-grade students can often pose serious challenges to MWP solvers due to its diverse problem types and varying degree of complexities. Understanding such problems written in natural language requires proper reasoning toward equation formation and answer generation. We restrict the review in this survey only to research on solving arithmetic word problems from elementary school level mathematics. We analyzed all the important methodologies proposed by researchers along with the datasets they used for training and evaluation. We studied the technical aspects of the system components and the algorithms relevant to their research along with the scopes, constraints, and limitations. This review paper also discusses the performance of different MWP solvers and provides observations on related datasets.