Narrative Optimization for Chinese Enterprise English Websites via Multilingual Pretraining and Business Corpus Finetuning

This paper proposes an English text optimization model that integrates multilingual pre-training and business corpus fine-tuning to address the problems of low narrative quality and inconsistent language style in the English official websites of Chinese companies in international communication. The model is based on a multilingual Transformer architecture, combines corporate English corpus for domain adaptation training, and introduces a structured prompt mechanism to achieve coordinated control of narrative logic, language style and pragmatic intention. By constructing an English official website dataset covering multiple industries, five groups of model control experiments are designed to conduct a comprehensive evaluation from the dimensions of language quality, narrative coherence and pragmatic consistency. The experimental results show that the proposed method is significantly better than the un-fine-tuned model in terms of indicators such as BLEU, ROUGE-L, and BERTScore, and shows stronger structural control ability and language naturalness in the generation of brand culture and market communication texts. This study not only verifies the adaptability of multilingual models in business narrative scenarios, but also provides a feasible optimization path for automatic English generation in cross-cultural contexts, which has good theoretical promotion and engineering application value.