Math Word Problem Generation with Memory Retrieval

Abstract

AbstractThe task of math word problem generation (MWPG), which generates a math word problem (MWP) given an equation and several topic words, has increasingly attracted researchers’ attention. In this work, we propose a memory retrieval model to better take advantage of the training data. We first record training MWPs into a memory. Later we use the given equation and topic words to retrieve relevant items from the memory. The retrieved results are then used to complement the process of the MWP generation and improve the generation quality. In addition, we also propose a low-resource setting for MWPG, where only a small number of paired MWPs and a large amount of unpaired MWPs are available. Extensive experiments verify the superior performance and effectiveness of our method.KeywordsMath word problem generationMemory retrievalLow-resource