Natural language processing (NLP) has captivated the attention of educational researchers over the past three decades. In this study, a total of 2,480 studies were retrieved through a comprehensive literature search. We used neural topic modeling and pre-trained language modeling to explore the research topics pertaining to the application of NLP in education. Furthermore, we used linear regression to analyze the changes in the hotness of each topic. The results of topic modeling in different periods were visualized as topic word co-occurrence networks. This study revealed that the application of NLP in education can be broadly categorized into: (1) Technology innovations in language learning, (2) Mining and analysis (Practices, trends, factors, and challenges), (3) Student learning, (4) Interact and evaluation, and (5) Models and algorithms. Moreover, the field followed evolutionary trajectories in language learning, interaction and assessment, educational research and analysis, and learning technologies. Additional examination of the topic word co-occurrence, we found that research topics in this field have transitioned from being primarily technology-oriented to adopting a problem-oriented approach, reflecting a growing trend of diversification.