A Study on Semantic Similarity of English Translations of Costume Terms in Jīn Píng Méi Based on Natural Language Processing

Abstract

This study aims to provide guidance for translating costume terms in Chinese classics, with a specific focus on exploring the semantic similarity of two English translations of costume terms in 《金瓶梅》(Jīn Píng Méi). To achieve this objective, the study compiled a corpus of 35 groups of costume terms (covering clothes, headwear, and footwear) and their corresponding translations from the two versions. It then used the multilingual BERT (mBERT) model, a key natural language processing tool, with relevant technical processes (including text preprocessing, semantic vector mapping, and cosine similarity calculation) to analyze semantic differences between the translations. The main results are as follows: the overall average cosine similarity of the translated terms is 0.69, indicating moderately high consistency. Footwear terms have the highest similarity (0.82) due to their strong practical attributes, while ceremonial and clothes terms show lower similarity (affected by cultural load and complex components). Additionally, about 33% of daily casual terms have high similarity, and around 11% of terms have low similarity, mainly caused by misinterpretation of cultural symbols.