A lightweight AI model with ethics priority: Differential English teaching practice in junior middle schools under the dimension of cultural competence

Abstract

Abstract: In view of the long-standing dilemma of "pseudo-stratification" in junior middle school English stratification teaching and the ethical controversy caused by artificial intelligence technology, this study constructs a lightweight teaching framework with ethics priority. By integrating federated learning and non-inductive data acquisition technology, a collaborative model of "three-layer dynamic grouping and four-dimensional strategy matrix" was innovatively designed, and a 6-month controlled experiment was carried out among 326 students covering eastern cities, central counties and towns, and western villages. The empirical results show that the hierarchical accuracy rate is improved to 89%, the grammatical error rate is significantly reduced by 31%, and the cultural understanding is increased by 44%. At the same time, the hierarchical informed agreement reduces the parent data withdrawal request rate to 0.9%. The framework uses mobile phone sensors to achieve low-cost deployment. Through dynamic diagnosis and ethical compliance design, it provides a scalable differentiated teaching path for resource-weak areas, and promotes the transformation of educational equity from concept to empirical evidence.