Design of Big Data-Based Intelligent Machine Translation Model for Low-Resource Languages

In the context of globalization, cross-cultural communication between low-resource languages and English is becoming increasingly frequent. The demand for bilingual communication and translation of relevant academic documents among international students from low-resource language countries in university English teaching has significantly increased. However, due to the scarcity of corpora and unique grammatical rules, the intelligent translation performance of low-resource languages with English is generally low. Existing intelligent translation models mostly rely on large-scale parallel corpora, which suffer from low translation accuracy, semantic loss, and frequent grammatical errors in low-resource language scenarios, making it difficult to meet the actual needs of university English teaching and cross-cultural communication. This paper first reviews relevant research on machine translation of low-resource languages and its application in big data; second, it designs a big data-enhanced low-resource language-English translation model, including three core parts: a low-resource corpus acquisition and expansion module, a translation model architecture incorporating transfer learning, and a model optimization strategy, to achieve accurate capture of low-resource language features; finally, comparative experiments are conducted to verify the model’s performance. Experimental results show that when the parallel corpus size is reduced to 2,000 sentence pairs (extremely low-resource scenario), the METEOR value of the proposed model remains at 29.8, while the METEOR values of Transformer, mBART-50, and XLM-R decrease to 22.3, 25.7, and 24.5, respectively. This further expands the advantages of the proposed model. In real-world tests such as university English textbook excerpts and intercultural dialogues, its semantic fidelity and grammatical correctness are superior to existing mainstream translation models.