Using Natural Language Processing Technology to Explore the Semantic Evolution of "Textile" Related Words in Bamboo and Silk Manuscripts

The semantic evolution of textile-related vocabulary preserved in bamboo and silk manuscripts remains difficult to investigate because of archaic language forms, incomplete textual records, and the absence of temporally comparable semantic representations. This study develops a computational framework that integrates an ancient Chinese BERT model with dynamic word vector alignment to quantify long-term semantic variation in historical textile corpora. After constructing period-specific corpus subsets and performing context-sensitive representation learning, semantic spaces from different historical stages are aligned to enable cross-temporal comparison, while cosine-distance-based semantic drift metrics are employed to characterize lexical evolution. Experimental results show that the proposed approach achieves a global alignment error of 0.242 and limits the semantic trajectory lengths of twenty representative textile terms to below 3.12, demonstrating improved stability and comparability over conventional alignment methods. Beyond revealing the historical development of textile terminology, the framework provides a reproducible methodology for large-scale semantic analysis of domain-specific archives and historical technical documents. Such semantic representation and alignment strategies may also facilitate knowledge organization and intelligent information retrieval for multidisciplinary engineering literature, including electromagnetic and antenna-related digital resources, thereby supporting cross-domain knowledge discovery and long-term technical heritage preservation.