The *Analects*, compiled during China's Warring States Period, presents Confucius' teachings, significantly influencing modern society. Over 110 English translations exist, reflecting the text's global importance. Understanding each translation's nuances is crucial for guiding future translators and aiding reader selection. Previous research, often subjective or focusing on theoretical exploration, has explored variations among translations. This study addresses these limitations by employing NLP to quantitatively analyze semantic similarities, offering a more objective and data-driven approach to understanding the nuances of translation choices. The study's integration of NLP methods within the digital humanities provides a new paradigm for translation research.

Scholars have explored variations in *Analects* translations, attributing discrepancies to translators' experiences, backgrounds, Sinology expertise, bilingual proficiency, purposes, and translation strategies. However, these studies are often subjective. Corpus-based studies have provided some objectivity but lacked pragmatic considerations crucial for enriching reader understanding. This research builds upon these studies, leveraging NLP’s capabilities in text generation, data mining, phonetics, sentiment analysis, and semantic similarity computation to offer a more empirical and unbiased analysis of the textual features of various *Analects* translations.

Five high-acceptability English translations of *The Analects* (by Lau, Legge, Jennings, Slingerland, and Watson) were selected based on previous research assessing reader engagement metrics. A parallel corpus was created by aligning these translations sentence-by-sentence, handling missing sentences with "None" placeholders. Word2Vec, GloVe, and BERT algorithms were used to compute semantic similarity between corresponding sentences across the translations. Python 3.6 was used for implementation, with the code and corpus made publicly available. The algorithms calculate similarity scores, reflecting the degree of semantic agreement between sentence pairs from different translations. The average similarity score of the three algorithms for each sentence pair was calculated for a more robust and reliable result.

The overall semantic similarity trend across the three NLP algorithms (Word2Vec, GloVe, and BERT) was consistent, despite differences in absolute values. The majority (approximately 78%) of sentence pairs showed high semantic similarity (above 80%), indicating that the core meaning of *The Analects* is largely preserved across translations. However, a significant minority (approximately 22%) showed low semantic similarity (below 80%), highlighting substantial variations in interpretation and translation choices. 33 sentence pairs were classified as "Abnormal," primarily due to missing sentences in some translations. Analysis of high-similarity sentence pairs revealed that Slingerland and Watson’s translations exhibited the highest degree of similarity, possibly due to their reliance on previous translations and inclusion of extensive supplementary materials. Jennings’s translation showed a lower level of similarity to others, likely due to his unique stylistic choices and restructuring of sentences to enhance readability. Further analysis of low-similarity sentence pairs focused on high-frequency words, revealing that core conceptual words (e.g., 君子, 小人, 仁, 道, 礼) and personal names significantly contributed to semantic variations. Analysis of specific word choices for key concepts, such as “君子” (often translated as "gentleman" or "superior man"), showed considerable variation among translators, reflecting different interpretations and stylistic preferences.

The findings demonstrate the effectiveness of NLP in quantifying semantic similarities and differences across *Analects* translations. The high proportion of high-similarity sentence pairs suggests a general agreement on the core meaning, but the low-similarity pairs reveal significant variations reflecting different translational choices and interpretative nuances. The impact of core conceptual words and personal names on semantic representation is notable. The variations observed highlight the challenges of translating culturally specific concepts and the importance of understanding the complexities of the original text and the historical context of translation. The study provides objective data to support previous subjective observations regarding translational variations.

This study provides a novel approach to analyzing *Analects* translations using NLP. The results highlight both the consistent core message across translations and significant variations in interpretation. Future research could explore additional factors contributing to translation differences beyond core concepts and personal names. The methodology used can be applied to other translated texts to evaluate semantic similarity and inform future translation practices. The findings offer valuable insights for both readers seeking to enhance their understanding of *The Analects* and translators seeking to refine their strategies.

This study focused on five translations and three NLP algorithms, and the results may not be fully generalizable to all translations of *The Analects* or other classical texts. The analysis primarily considered sentence-level similarities, potentially overlooking subtleties at other linguistic levels. Future studies could explore additional factors that influence translation differences, such as translator background and intended audience, using more comprehensive methods and a larger dataset.