Design and Development of Chinese Cultural Translation System Based on Fine-Tuned LLM Model

The growing demand for accurate and context-aware translation of Chinese cultural expressions necessitates advanced systems capable of preserving both linguistic meaning and cultural integrity. Traditional machine translation models often struggle with idioms, metaphors, and culturally rich phrases due to limited contextual understanding and semantic adaptability. To address these challenges, this study proposes an Quantized Low-Rank Adaption based M2M-100 translation framework model to enhance the accuracy of Chinese-to-English cultural translations. The Parallel English Translations of Chinese dataset OPUS, which contains diverse idiomatic expressions and culturally significant phrases, is used to fine-tune the model for better contextual adaptation. The suggested model effectively learns semantic representations, improving translation fluency while preserving cultural nuances. Experimental results demonstrate that the proposed approach significantly outperforms existing baselines, achieving a BLEU score of 45.8, ROUGE-L of 61.3, and METEOR score of 38.6, indicating substantial improvements in translation quality and cultural preservation. The results indicate that integrating with a domain-specific cultural dataset significantly enhances translation quality, making it well-suited for cross-cultural communication. This study highlights the potential of advanced neural network-based approaches in bridging linguistic and cultural gaps, thereby contributing to more accurate and context-aware translations.