Elamur Research⌕
☰
home / papers / 2026 / Capturing Ancient Chinese Sense Induction with Automatic Pipelines

Capturing Ancient Chinese Sense Induction with Automatic Pipelines

Paper · 2026 · Guan-Yu Tseng, Chunki Lim, Chih-Han Lin et al.

Summary

Computational resources that trace how word meanings evolve are scarce for Ancient Chinese, a gap this paper addresses. The authors built an unsupervised sense-induction pipeline on the Chinese Text Project (ctext) corpus. Every sentence containing a target character is encoded with GujiBERT-fan, a BERT-style model pre-trained on the Siku Quanshu, and the token vector of the character is kept. The pipeline then removes OCR noise, caps samples per document, over-clusters with k-means, merges near-identical centroids, and recursively splits oversized clusters. Humans label the resulting clusters with a hierarchical coding scheme.

Clustering runs at two levels. A global pass covers all periods, and independent local passes run per dynasty. This design avoids two failure modes: in incremental methods, errors from sparse pre-Qin data propagate forward, and in joint methods, Ming–Qing volume swamps earlier periods. A pilot on 手 ("hand") across eight periods shows a drift from bodily to abstract and grammatical senses. Appendix runs on 水 and 道 suggest the method generalizes to other characters. The approach offers a scalable link between philology and diachronic NLP.

Key points

  • Data and pilot scope: The 手 pilot covers more than 185,000 occurrences split into eight periods, from pre-Qin to the Republican era. Each document contributes at most 200 instances, so long texts cannot dominate.
  • Hyperparameters: Initial K is 20, and clusters merge when centroid cosine similarity exceeds 0.95. The split ratio is 0.1 for global clustering and 0.3 for local clustering. Sub-K is fixed at 6 globally and set dynamically from period size locally.
  • Choice of K: A one-factor-at-a-time sensitivity analysis compared settings. K=25 gave a slightly higher global Silhouette score than K=20 (0.054 vs. 0.049), but it ended with almost the same cluster count after merging (16 vs. 17). The authors therefore kept K=20.
  • Generalization to 水: The global run (n = 313,184) produced 12 clusters. Concrete senses dominate throughout, while literary uses grow from the Song-Yuan period onward.
  • Generalization to 道: The global run (n = 331,910) produced 22 clusters in nine macro-categories.
    • Philosophy accounts for 41% of uses.
    • A separate Speech Act group (15%) reflects grammaticalization into a quotative marker. It is absent in the pre-Qin period and grows from the Song-Yuan period on.
    • The Republican period has the highest Silhouette score (0.1718).

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksLiterature and Literary Theory · 541 works

🧭 Topics 1Research subjects and languages the work is about.Word sense disambiguation · 87 works

Cite

Cite as Guan-Yu Tseng, Chunki Lim, Chih-Han Lin, Tung-Le Pan, Yu-Chieh Wang, Lang-Ching Yeh and Shu-Kai Hsieh. 2026. Capturing Ancient Chinese Sense Induction with Automatic Pipelines. Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/4ku4whfwarht

@inproceedings{tseng2026capturing,
  title = {{Capturing Ancient Chinese Sense Induction with Automatic Pipelines}},
  author = {Guan-Yu Tseng, Chunki Lim, Chih-Han Lin, Tung-Le Pan, Yu-Chieh Wang, Lang-Ching Yeh and Shu-Kai Hsieh},
  booktitle = {Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026},
  year = {2026},
  doi = {10.63317/4ku4whfwarht},
  url = {https://doi.org/10.63317/4ku4whfwarht}
}

References

15 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Margareta Ackerman and Sanjoy Dasgupta. 2014. Incremental clustering: The case for extra clusters. arXiv preprint arXiv:1406.6398. — [arXiv:1406.6398](https://arxiv.org/abs/1406.6398) find
  2. Pierpaolo Basile, Annalina Caputo, Tommaso Caselli, Pierluigi Cassotti, and Rossella Varvara. 2020. DIACR-Ita @ EVALITA2020: Overview of the EVALITA2020 Diachronic Lexical Semantics (DIACR-Ita) Task, pages 411–419. find
  3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. BERT: pre-training of deep bidirectional transformers for language understanding. CoRR, abs/1810.04805.
  4. William L. Hamilton, Jure Leskovec, and Dan Jurafsky. 2016. Diachronic word embeddings reveal statistical laws of semantic change. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1489–1501, Berlin, Germany. Association for Com…
  5. Bin Li, Yiguo Yuan, Jingya Lu, Minxuan Feng, Chao Xu, Weiguang Qu, and Dongbo Wang. 2022. The first international Ancient Chinese word segmentation and POS tagging bakeoff: Overview of the EvaHan 2022 evaluation campaign. In Proceedings of the Second Workshop on Language Technologies for Historical… find
  6. Francesco Periti and Stefano Montanelli. 2024. Lexical semantic change through large language models: a survey. ACM Computing Surveys, 56(11):1–36.
  7. Valerio Perrone, Simon Hengchen, Marco Palma, Alessandro Vatri, Jim Q. Smith, and Bar-Lexical semantic bara McGillivray. 2021. change for ancient greek and latin. CoRR, abs/2101.09069. find
  8. Peter J. Rousseeuw. 1987. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics, 20:53–65. find
  9. Dominik Schlechtweg, Anna Hätty, Marco Del Tredici, and Sabine Schulte im Walde. 2019. A wind of change: Detecting and evaluating lexical semantic change across times and domains. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 732–746, Florence, Ita… find
  10. Dominik Schlechtweg, Barbara McGillivray, Simon Hengchen, Haim Dubossarsky, and Nina Tahmasebi. 2020. SemEval-2020 task 1: Unsupervised lexical semantic change detection. In Proceedings of the Fourteenth Workshop on Semantic Evaluation, pages 1–23, Barcelona (online). International Committee for Com…
  11. Language Resource References Donald Sturgeon. 2011. Chinese text project: a dynamic digital library of premodern chinese. https://ctext.org/. ↗
  12. Dongbo Wang, Chang Liu, Zhixiao Zhao, Si Shen, Liu Liu, Bin Li, Haotian Hu, Mengcheng Wu, Litao Lin, Xue Zhao, and Xiyu Wang. 2023. Gujibert and gujigpt: Construction of intelligent information processing foundation language models for ancient texts. find
  13. Francesco Periti, Sergio Picascia, Stefano Montanelli, Alfio Ferrara, and Nina Tahmasebi. 2025. Studying word meaning evolution through incremental semantic shift detection. Language Resources and Evaluation, 59:1363–1399. find
  14. Francesco Periti and Nina Tahmasebi. 2024. A systematic comparison of contextualized word embeddings for lexical semantic change. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers),… find
  15. A. Appendix: Generalizability Analysis on 水 (shuǐ) and 道 (dào) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tseng-2026-capturing-ancient-chinese-sense-induction/