A Corpus-Based Comparative Study of Translator Styles in English Translations of Guwen Guanzhi

Abstract

This study adopts a corpus-based approach to systematically investigate and compare translator styles across three English versions of Guwen Guanzhi, a canonical anthology of classical Chinese prose: two human translations and an AI-generated translation produced by Deepseek. Drawing upon fifteen parallel texts selected from the anthology, a self-compiled tripartite bilingual corpus was constructed to facilitate quantitative analysis. A set of quantitative indicators at both lexical and syntactic levels was computed and compared, including average sentence length, sentence-length variability, average word length, lexical diversity, and type-token ratio. The findings reveal distinct stylistic profiles among the three translations. The first human translation exhibits the greatest sentence-length variability, creating a rhythmically dynamic reading experience. The second human translation displays the longest average sentences and the highest average word length, reflecting a formal, information-dense scholarly style. The Deepseek translation achieves the highest lexical diversity among the three versions whilst notably lacking prosodic dynamism, suggesting that algorithmic generation prioritizes lexical variation over rhythmic coherence. This study contributes to the empirical research on classical Chinese prose translation and offers insights into the stylistic implications of AI-assisted literary translation, highlighting both the potential and limitations of machine-generated translations in preserving the aesthetic and rhetorical qualities of classical texts.