Optimized Machine Translation of Technical Acronyms Using Large Language Models: A Workflow-based Approach

Machine translation systems often struggle with the accurate interpretation of technical acronyms, particularly in domain-specific contexts where the same acronym may have multiple meanings. A novel workflow was developed to enhance acronym detection and translation accuracy by modifying the LLaMA model through targeted fine-tuning and hierarchical attention mechanisms. This workflow, tested on a diverse dataset of domain-specific texts, demonstrated substantial improvements in both translation accuracy and computational efficiency, outperforming baseline models across several metrics. The integration of caching and post-processing techniques further optimized the system for scalability, making it suitable for handling large volumes of technical documents. Results indicate that the workflow significantly reduces errors in acronym disambiguation, ensuring precise translations in critical fields such as medicine, engineering, and computer science. The approach is poised to enhance the quality of machine translations where technical terminology must be accurately conveyed, reducing ambiguity and improving communication across specialized industries.