Quantifying Stylistic Evolution in Classical Chinese Literature Using Word Embeddings and Trajectory Analysis

The stylistic evolution of classical Chinese literature reflects profound transformations in society, culture, and linguistic practices over centuries. However, the vast scale and complexity of the corpus have posed significant challenges for traditional qualitative analysis. To address these limitations, this study develops a computational framework that leverages advanced natural language processing (NLP) techniques to analyze the temporal stylistic evolution of classical Chinese literature. Using word embeddings (word2vec and BERT) combined with dimensionality reduction (UMAP), the framework uncovers stylistic patterns across over 10,000 texts from the Tang, Song, Ming, and Qing dynasties. Preprocessing steps, including tokenization, normalization, and the removal of infrequent characters and stopwords, address the unique challenges posed by the compact and context-dependent nature of classical Chinese. UMAP visualizations reveal distinct stylistic clusters for each dynasty, with Tang poetry forming cohesive clusters and Qing literature displaying significant stylistic variability. Temporal trajectories derived from UMAP embeddings capture stylistic changes over time, with a novel trajectory curvature metric introduced to quantify abrupt shifts. Results show that Tang poetry exhibits smooth stylistic transitions, while Qing literature demonstrates fragmented trajectories, reflecting rapid stylistic changes. Quantitative evaluation using trajectory length, curvature, silhouette scores, and Davies-Bouldin Index validates the reliability of these findings. A focused case study on metaphorical expressions, such as the shifting meanings of “moon” in Tang poetry, highlights the adaptability of embedding-based methods for diachronic analysis. This framework provides a robust and reproducible approach to exploring long-term literary trends, emphasizing the dynamic interplay between historical context and stylistic change.