Generating,Reasoning and Ranking:Multitask Learning Framework for Math Word Problem Generation
A math word problem(MWP)is a narrative which reflects the underlying logic of math equations.Successful MWP generation has wide prospect in language generation and educational field.This paper proposes a multitask learning based MWP generation framework.We devise three novel tasks,including number relation extraction,number ranking and sentence substitution prediction.These tasks are jointly trained with generation objective and supervise the learning of MWP decoder,so as to enhance the model's comprehension of arithmetic logic and condition.Experiments demonstrate the effectiveness of our proposed method in equation consistency of generated MWPs.
NETL
NSTL
万方数据
