Language Change in Ancient Chinese: A Computational Approach

This talk explores computational methods for analyzing language change in Ancient Chinese, spanning the Tang through Qing dynasties, with an emphasis on language periodization. Traditional approaches to Chinese periodization rely heavily on close reading of historical texts which demands extensive expertise. Recent computational studies have tried to predict text dynasties using automated systems, yet these efforts prioritize classification model design and benchmark creation over linguistic analysis. To address this gap, we introduce a new diachronic corpus of Biji (笔记, “written notes”), which is rich in conversational language and covers diverse topics, along with an additional dataset of Chinese Buddhist conversations to test model robustness. We systematically assess different computational feature representation methods including character and word n-grams, static embeddings, and contextualized embeddings on the Biji dataset. The results indicate that the task is very challenging, with contextualized embeddings (especially when used with SVM classifiers) delivering superior performance. Furthermore, we explore model performance on out-of-domain texts and investigate whether contrastive learning techniques can enhance the representational quality of contextualized embeddings. Finally, through a neighboring dynasty classification task, we address the issue of language periodization and offer an analysis of the top features utilized by the models.

This is a 2024-2025 Colloquium event. Registration for Colloquium events is not required. There are two ways to receive details and virtual meeting information: Sign up for the Colloquium listserv: you will receive an email about every event, including virtual meeting information when the event is remote or hybrid Request to receive information for the next upcoming event only.

Presenter

Zuoyu Tian, Assistant Professor, Macalester College