The evaluation of translation teaching involves multi-dimensional abilities such as semantic understanding, terminology control, syntactic transformation, discourse cohesion and cultural adaptation. Traditional manual scoring, automatic indicator evaluation and large language model annotation methods are often difficult to simultaneously give attention to diagnostic finness, feedback continuity and teaching decision support. In order to solve this problem, this paper designs an intelligent evaluation system for translation teaching empowered by reinforcement learning (RL-IETS). Rl-iets converts student translations, revision records, error labels, teacher annotations, and platform behavior logs into learning state representations, and dynamically selects scoring, error diagnosis, feedback generation, learning path recommendation, and teacher intervention strategies through the state-action-reward mechanism. The experimental results show that the average evaluation accuracy of the system reaches 91.8%, the comprehensive value of error diagnosis reaches 89.4%, the profit of personalized path optimization reaches 83.6%, and the comprehensive score of teachers ‘digital literacy is improved from 67.8 to 86.5. The research shows that the system can improve the adaptive level of translation teaching evaluation, and provide a traceable practice path for teachers ‘digital literacy development.
