Improving the English Translation Quality of Chinese Silk Cultural Terms Using Prompt Tuning

Abstract

Accurate translation of domain-specific terminology is essential for preserving technical semantics and enabling reliable international knowledge exchange in engineering disciplines, including electromagnetic waves, antennas, and propagation, where standardized descriptions and interdisciplinary communication are increasingly important. To address semantic collapse and cultural information loss in the English translation of Chinese silk cultural terms, this study proposes a prompt tuning framework incorporating a cultural parameter decoupling mechanism. A gated attention strategy is introduced to separate cultural features from underlying semantic representations, while a three-dimensional dynamic prompt template and a hierarchical knowledge correction mechanism are developed to preserve historical symbolism, craft characteristics, and customary functions during translation. Experiments conducted on 12,800 Chinese–English parallel corpora demonstrate that the proposed method reduces the cultural mistranslation rate to 6.1% while achieving high robustness against spelling errors, dialect interference, abbreviation ambiguity, and symbol omission. Compared with conventional neural machine translation models, the framework exhibits superior terminology accuracy and cultural semantic fidelity through coordinated optimization of parameter decoupling and knowledge-guided correction. Beyond silk cultural translation, the proposed methodology provides a transferable strategy for preserving specialized semantic information in technical documentation and multilingual knowledge systems, offering potential reference for cross-domain terminology management and intelligent engineering information processing.