The study proposes a personalized teaching content generation and adaptation method based on generative adversarial network in language education reform in response to the problems in language education, and designs a language learner portrait model. By introducing the self-attention mechanism of Gaussian deviation as a generator, student portrait features are collected to ensure personalized teaching content generation. In addition, the feature vectors of learner portraits and learning resources are extracted using normalization to improve the adaptability of teaching content. The designed reform method is applied to language teaching in a city experimental school to evaluate the application effect of the method in this paper.The FID and IS values of the PBESGAN model converge to 9.35 and 7.81, respectively, before iterating for 100 rounds, and the authenticity and diversity of the language teaching content is more generated is improved. In addition, the F1 values of the model in the teaching resource recommendation fitness experiment are improved by 18.33% to 42.08% compared with the comparison algorithm, which can provide reliable teaching resources for students. At the end of teaching, the language ability of students in class 2 is significantly improved compared with class 1, with a difference of 6.27 points between the two language test scores. Students’ satisfaction with the teaching content recommended by the model of this paper is as high as 4.76 points, which demonstrates the accuracy and variety of the language teaching content generated by the model of this paper.
