Research on the Application of Cross-Language Transfer Learning Model in English Translation for Low-Resource Scenarios

In the context of digital education and intelligent language systems, developing effective natural language processing (NLP) methods for low-resource languages remains a critical but challenging goal. Traditional neural machine translation (NMT) models, while effective in high-resource settings, often struggle with limited parallel data and the linguistic variability of underrepresented languages. To address these limitations, we propose a framework that combines formal linguistic modeling with adaptive learning strategies tailored for low-resource English translation. The approach introduces a structured representation of language learning, a model architecture that integrates syntactic and cognitive features, and an instructional policy informed by pedagogical principles. Central to the framework is the LexiStruct model, which enhances encoder-decoder translation with syntactic priors and task-aware attention mechanisms. This is complemented by the CurricStruct strategy, which applies an adaptive curriculum to improve learner modeling and language generalization. Experiments on benchmark datasets demonstrate consistent improvements over multilingual baselines under constrained data conditions. These results suggest that incorporating structured linguistic knowledge and curriculum adaptation can support more effective translation in targeted low-resource scenarios.