The Four Books: Multi-Model English Translation and Multi-Dimensional Style Assessment Dataset]

It integrates English translations from four representative large language models (DeepSeek, ERNIE Bot (Wenxin Yiyan), ChatGPT-5, and Google Translate), with authoritative human translations by James Legge and Wu Guozhen serving as references. The dataset is designed to support research in translation studies, particularly focusing on digital translator style.The core value of this dataset lies in its provision of a suite of multi-dimensional, quantifiable style assessment metrics, encompassing:Semantic Level: BERTScore-based semantic similarity data between translations and reference translations.Pragmatic Level: Evaluation of translation consistency and conceptual accuracy for key Confucian terms such as "benevolence (Ren)", "righteousness (Yi)", and "propriety (Li)".Discourse and Syntactic Level: Discourse features including average sentence length, parallelism/antithesis preservation, coherence scores, and detailed syntactic complexity data (e.g., mean length of T-unit, clause ratio) calculated based on Spacy.Utilizing this dataset, researchers can systematically analyze and compare the stable stylistic features exhibited by different digital translators when processing classical texts.

It provides an empirical foundation for the theoretical construction of "digital translator style" and offers references for model selection and optimization in classical text translation.