Here is the English translation of the provided paragraph while maintaining the original format: ---

This paper discusses the impact of artificial intelligence (AI) technology on film and television translation under the backdrop of globalization, emphasizing the importance of optimizing translation paradigms for cross-cultural communication. It analyzes the current challenges in film and television translation, such as linguistic diversity, cultural differences, and synchronization issues. The study explores the application of AI technology in translation, employing quantitative statistics and empirical analysis to highlight the accuracy of large language models (LLMs) in textual expression while acknowledging issues like mistranslation, structural deviations, and insufficient emotional conveyance. The findings suggest that LLMs enhance the speed of translation and facilitate cross-cultural communication but also introduce new challenges due to technical limitations and cultural disparities.

1. AI Technology and Film/Television Translation

The advancement of AI technology has revolutionized the film and television industry, particularly in creative production, editing efficiency, and personalized content creation. AI's evolving role across various stages of production holds significant implications for the cross-cultural dissemination of audiovisual works. The field of translation has progressed from manual efforts to automated tools and now AI-assisted solutions. The integration of AI in translation has improved efficiency and quality, driving the development of the industry. Comparative studies across multiple dimensions examine the differences between human and machine translation, alongside the role and trends of computer-assisted translation (CAT) in subtitling. This research aims to enrich the study of AI-assisted audiovisual translation by conducting quantitative and qualitative analyses of translated texts using LLMs.

2. Research Methodology

The study outlines the methodology for AI-assisted film/television translation. First, it discusses the selection of LLMs and datasets, including four mainstream Chinese LLMs—Baichuan-7B, ChatGLM-6B, InterLM-7B, and Qwen-7B—trained on extensive textual data and excelling in cross-cultural semantic understanding and generation. The dataset comprises clips from films and TV shows spanning historical, contemporary, and mythological genres to ensure textual diversity. Second, evaluation metrics are introduced, combining rule-based and statistical quantitative analyses (BLEU and ROUGE scores) with qualitative textual assessments from linguistic and cultural perspectives. This dual approach provides a comprehensive framework for evaluating subtitle translation by LLMs.

3. Research Process and Results Analysis

The study details the process and findings of AI-assisted translation. Representative clips from *Farewell My Concubine*, *Dying to Survive*, and *Black Myth: Wukong* were selected for their cultural depth and emotional richness. Data collection involved downloading official bilingual subtitles, followed by cleaning and annotation for model evaluation. Environment setup for the selected LLMs was critical. Quantitative results and textual analyses were then compared across models, with visualizations to highlight differences.

Quantitative analysis revealed significant disparities in *Dying to Survive*'s results compared to the other two works, likely due to its simpler, repetitive modern-dialogue dataset. *Farewell My Concubine* and *Black Myth: Wukong* showed similar performance, suggesting stylistic consistency. Variations in BLEU and ROUGE scores across LLMs were attributed to metric focus and subtitle segmentation differences.

Qualitative analysis of translated texts demonstrated that LLMs generally convey accurate semantics, occasionally outperforming human translations. However, issues like verbosity, mistranslation, emotional misalignment, and misplaced emphasis were noted. The study highlights LLMs' strengths in handling cultural elements while underscoring how sample data and evaluation dimensions influence outcomes.

4. Opportunities and Challenges in AI-Driven Cross-Cultural Communication

AI presents new opportunities for cross-cultural exchange by enhancing translation speed, depth, and reach, enabling global audiences to engage with content in real time. It breaks language barriers and optimizes international collaboration workflows. However, challenges persist, including consistency, cultural acceptability, and algorithmic biases. To address these, human-AI collaboration must be refined, algorithms improved, corpora expanded, and cultural adaptability prioritized for diverse audiences.

5. Conclusion

AI translation has transformed audiovisual localization, shifting from manual processes to a feedback-driven, technology-centric model. While it accelerates language conversion and broadens dissemination, depth and reliability still hinge on cultural understanding. Future efforts should focus on synergistic human-AI workflows, algorithmic optimization, and rigorous validation to enhance cross-cultural communication and global dialogue.

--- The translation preserves the original structure, headings, and paragraph breaks while adapting the content into natural, academic English. Technical terms (e.g., BLEU/ROUGE scores, LLMs) are retained, and culturally specific titles (e.g., *Farewell My Concubine*) are presented in standard English forms.