Towards Computational Chinese Paleography

Chinese paleography, the study of ancient Chinese writing, is undergoing a computational turn powered by artificial intelligence. This position paper charts the trajectory of this emerging field, arguing that it is evolving from automating isolated visual tasks to creating integrated digital ecosystems for scholarly research. We first map the landscape of digital resources, analyzing critical datasets for oracle bone, bronze, and bamboo slip scripts. The core of our analysis follows the field’s methodological pipeline: from foundational visual processing (image restoration, character recognition), through contextual analysis (artifact rejoining, dating), to the advanced reasoning required for automated decipherment and human-AI collaboration. We examine the technological shift from classical computer vision to modern deep learning paradigms, including transformers and large multimodal models. Finally, we synthesize the field’s core challenges – notably data scarcity and a disconnect between current AI capabilities and the holistic nature of humanistic inquiry – and advocate for a future research agenda focused on creating multimodal, few-shot, and human-centric systems to augment scholarly expertise.

💡 Research Summary

The paper presents a comprehensive position on the emerging field of computational Chinese paleography, charting its evolution from isolated visual‑task automation to the construction of integrated digital ecosystems that support scholarly research. It begins by outlining the unprecedented growth in excavated materials—oracle bones, bronze inscriptions, bamboo slips, and silk manuscripts—and the consequent pressure on traditional manual methods. The authors map the current landscape of digital resources, comparing publicly available datasets for each script type in terms of size, annotation quality, and standardization.

A central contribution is the formulation of a three‑stage methodological pipeline: (1) foundational visual processing, (2) contextual analysis, and (3) advanced reasoning. In the visual stage, the paper reviews the shift from classical computer‑vision techniques (edge detection, handcrafted features) to deep‑learning approaches such as CNNs for image restoration, super‑resolution, and character detection. It highlights the use of diffusion models for noise reduction and synthetic data generation, which is crucial given the scarcity of labeled images.

The contextual stage addresses artifact rejoining, deduplication, period classification, and language modeling. The authors discuss graph‑based matching algorithms that align fragmented pieces based on shape, material, and glyph patterns, as well as transformer‑based temporal models that learn stylistic evolution across dynastic periods. For bamboo slip texts, they propose Vision‑Language Transformers that jointly segment characters and model their sequential context, enabling more accurate reconstruction of broken texts.

In the advanced reasoning stage, the paper explores knowledge‑graph construction that integrates visual features, phonological hypotheses, semantic roles, and archaeological metadata. Automated decipherment is treated as a multimodal inference problem: radical‑level representations, phonetic components, and semantic clues are learned simultaneously using multi‑task transformers, while large multimodal models (e.g., CLIP, Flamingo) provide cross‑modal verification of candidate readings. Human‑AI collaboration is emphasized through interactive interfaces where AI proposes plausible interpretations and scholars validate or revise them, creating a feedback loop that continuously refines the models.

The authors identify two overarching challenges. First, data scarcity: high‑quality, annotated images and deciphered transcriptions are limited, making few‑shot, meta‑learning, and synthetic data generation essential. Second, a conceptual gap between AI’s focus on visual form and paleography’s holistic integration of phonology, semantics, and historical context. To bridge this gap, the paper advocates for human‑centric, multimodal, few‑shot systems that act as intelligent assistants rather than replacements.

Finally, a forward‑looking research agenda is outlined. It calls for standardizing data formats and metadata, building open‑source repositories with API access, fine‑tuning large multimodal models on domain‑specific corpora, and developing collaborative platforms that bring together archaeologists, historians, linguists, and AI researchers. The ultimate goal is not full automation but the creation of a digital research ecosystem that amplifies scholarly insight, enabling deeper historical, linguistic, and script‑evolution studies across China’s millennia‑long written heritage.