Using Cross-Linguistic Data Formats to Enhance the Annotation of Ancient Chinese Documents Written on Bamboo Slips

Michele Pulini, Johann-Mattis ListNAACL 2025

TL;DR

This paper introduces a novel workflow for creating interactive digital editions of ancient Chinese bamboo slip manuscripts. By integrating established cross-linguistic data formats and computational linguistics tools, the authors enable both manual scholarly analysis and computational exploration of these historically rich but challenging texts.

What Problem Does This Paper Address?

Ancient Chinese documents, particularly those written on bamboo slips from over 2000 years ago, are invaluable for understanding the evolution of the Chinese language, paleography, and history. However, these documents present significant challenges:

  • Character Variation: Bamboo slip scripts exhibit immense variation in character forms and combinations compared to the standardized "regular script" used in later texts. This makes it difficult to identify modern counterparts and their Unicode values.
  • Digital Curation Gap: Most existing editions are merely scans with annotations, lacking structured data that would allow for computational analysis or interactive exploration. Scholars often lack the technical expertise to create such digital resources.

This paper aims to bridge this gap by developing a workflow that transforms these scanned manuscripts into rich, digitally accessible resources.

Methods

The authors propose a four-stage workflow:

  1. Digital Annotation: This initial stage involves marking individual characters on the scanned manuscript with bounding boxes. The tool Recogito is used for this, facilitating semantic annotation and allowing characters to be identified later.
  2. Linguistic Analysis: This stage has two parts:
  3. Character Analysis: Each annotated character is analyzed for its external and internal structure, modern counterparts, and pronunciation (using Middle Chinese and reconstructed Old Chinese readings). Ideographic description sequences are employed to systematically analyze character components, especially for unusual forms.
  4. Text Analysis: The two manuscript versions are unified into a single coherent text, with the AD version taking precedence and the SB version filling gaps. Words and phrases are identified and semantically glossed using interlinear morphemic glossing conventions.
  5. CLDF Integration: The analyzed data is modeled using the Cross-Linguistic Data Formats (CLDF) specification. CLDF is chosen for its suitability for cross-linguistic data, including wordlists, dictionaries, and feature collections, and its extensibility to corpus data. This format allows for computational querying and easy import into applications.
  6. Deployment with CLLD: The CLDF data is deployed using the CLLD toolkit, a Python library that provides an interactive web framework. This creates a digital edition where original characters can be explored through graphemic, phonetic, and semantic analysis.

Datasets

The primary dataset used is the Cáo Mò zhī zhèn (曹沫之陳), a recently rediscovered ancient Chinese philosophical dialogue on ethics and warfare. Two manuscript copies on bamboo slips were used:

  • Shanghai Museum manuscript (SB): First published in 2004.
  • Anhui University copy (AD): First published in 2022.

Both manuscripts were radiocarbon dated to the mid-third century BCE (SB) and late fourth century BCE (AD). The high-quality scans from the Anhui University collection served as the starting point for the digital edition.

Key Insights and Ideas

  • Bridging the Manual-Computational Divide: The workflow successfully creates a resource that serves both traditional scholarly analysis (manual transcription, glossing) and modern computational linguistics (data querying, analysis).
  • Leveraging Existing Standards: The adoption of CLDF, originally designed for cross-linguistic data, proves effective for ancient Chinese texts, demonstrating the power of standardized formats for diverse linguistic data.
  • Handling Character Variation: The use of ideographic description sequences provides a systematic way to analyze and represent characters that deviate significantly from modern forms, preserving valuable information about scribal practices.
  • Interactive Exploration: The CLLD-based web application offers a unique way to explore the data, integrating images with detailed linguistic annotations and allowing users to compare variants directly.

Key Results

The paper presents a functional digital edition of the Cáo Mò zhī zhèn. While specific quantitative results like accuracy metrics are not the focus, the paper demonstrates:

  • The successful creation of an interactive web application that integrates original manuscript images with detailed linguistic annotations (character analysis, phonetic readings, semantic glosses).
  • An example of frequent word analysis (Table 1), showing the identification of grammatical markers, dialogue indicators (e.g., 曰 "say"), and thematic words (e.g., 戰 "war").
  • An example of comparing individual word occurrences across editions (Figure 3), highlighting the utility for studying textual variation.

Limitations and Future Work

The authors acknowledge several limitations and suggest future directions:

  • Preliminary Workflow: The proposed workflow is still in its early stages.
  • Longer Texts: Extending the method to significantly longer texts may present additional challenges.
  • Critical Apparatus: Further development is needed to integrate critical apparatus into the digital edition.
  • Detailed Checks: Future work will involve more thorough checks of glosses and translations, and direct referencing of sources used for individual character analysis.
  • Broader Application: The authors hope this work serves as a proof of concept for developing digital critical editions of other early Chinese manuscripts and encourages discussion on consistent annotation workflows for ancient languages.
  • Computational Linguistics Applications: The annotated data could be used to develop models for handwritten text recognition, create corpus-based dictionaries, and improve understanding of ancient Chinese pronunciation.

Summary updated

BibTeX
@inproceedings{pulini2025using,
  author = {Michele Pulini and Johann-Mattis List},
  title = {{Using Cross-Linguistic Data Formats to Enhance the Annotation of Ancient Chinese Documents Written on Bamboo Slips}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2025},
}
Figure