Research on Machine Understanding Math Word Problems: From the Perspective of Discourse Comprehension Models

Abstract

With the rapid development of artificial intelligence (Al) technology, machine understanding math word problems (MWPs) has received growing attention. However, existing methods of automatic understanding MWPs are hardly integrated into cognitive intelligent systems used for individual learning. To address the integration problem, this paper firstly clarified the relationship between understanding MWPs and discourse comprehension. According to the trait of discourse comprehension models, the existing methods were divided into knowledge schema-based methods and mental processing-based methods. Then we shortly presented the construction-integration model and proposed a conceptual framework for machine understanding MWPs. The proposed conceptual framework was established from long and short-term memory, cognitive computing services, formal representation models, and human-computer interaction. Finally, we draw a conclusion that integrating cognitive models of human understanding discourse into the process of machine understanding MWPs is conducive to developing a humanized cognitive intelligence system for personalized learning.