An Applied Study of Using Natural Language Processing to Optimize Reading Comprehension Teaching in Elementary School Languages

JingZhang
Xichang Ethnic Preschool Teachers' College, China.

Abstract:

Under the background of the integration and development of the “Double Reduction” policy and education informatization, there is an urgent need to break through the traditional mode of reading comprehension teaching in elementary school language to realize the intelligent identification and personalized support of the comprehension process. Reading comprehension, as a complex linguistic cognitive activity, covers vocabulary identification, grammatical analysis, semantic construction and chapter integration, etc. Existing teaching is generally characterized by homogeneous strategies, sloppy assessment, and lagging intervention, which makes it difficult to effectively respond to students' diverse comprehension barriers. Based on natural language processing technology, this study integrates BERT, Transformer and graph neural network to construct a reading comprehension optimization system for elementary school language, forms a semantic modeling framework based on lexical-syntactic-chapter three-dimensional parsing, and constructs lexical annotation, syntactic dependency, and discourse aggregation modules to accurately identify semantic deviations of students in the comprehension process. semantic deviations and structural barriers in the comprehension process. The experimental results show that the system dynamically models the students' comprehension paths through the semantic residual formula and structure matching algorithm, and generates personalized learning recommendation paths by combining the GAT structure. The experiment is based on a sample of 120 third-grade students from an elementary school in Beijing, and comparing the traditional teaching group with the intelligent intervention group, the latter's performance is significantly improved in key indicators such as semantic residuals, feedback timeliness, and comprehension stability, with the average semantic bias reduced by 48.6%, and the response time of the strategy shortened by 64.4%. The results show that the system has strong cognitive adaptation ability and teaching intervention accuracy. This study provides feasibility verification on the integration path of natural language processing and language subjects, and provides theoretical support and empirical evidence for the deep landing of intelligent language teaching.