LLM-Driven Professional English Training: A space-Oriented Adaptive Learning Framework

Abstract

With the increasing complexity of space missions and the need for international collaboration, professional English has become a key competency in the cultivation of aerospace engineering talents. However, existing English teaching methods are difficult to dynamically adapt to professional contexts, individual ability differences, and task complexity, resulting in inaccurate learning paths and low efficiency of knowledge transfer. To this end, this paper introduces large language models (LLMs) and reinforcement learning strategies to construct a personalized English learning path planning framework for the space field to solve the problem of inaccurate matching between "task-ability-context". The specific methods include: (1) modeling learner portraits based on multidimensional ability assessment; (2) designing a task type-language ability matching mapping matrix and dynamically generating learning task sequences through Bayesian optimization; (3) using context embedding and professional terminology ontology to construct a semantic reinforcement mechanism to enhance professional perception ability. The experimental results show that the success rate of personalized path planning driven by a large language model reached 86.50%, which is a significant improvement compared to the static path group (74.10%) and the manual path group (79.30%). This shows that the learning path is more closely matched to the task ability, and the learners are more accurate and confident in completing the task.