The project

Formations

Early modern East Asian intellectual life did not conform to the modern disciplinary divisions by which it is now studied. It was not unusual for a scholar who revered Zhu Xi 朱熹 (1130–1200) also to compose essays and poetry in the manner of Su Shi 蘇軾 (1037–1101), debate the authority of past precedents, appraise poems and paintings, and pursue medical, religious, ethical, or aesthetic disciplines of bodily, mental, and spiritual cultivation.

Philosophy, literature, religion, history, aesthetics, and embodied practice belonged to intersecting fields of learning whose boundaries shifted with scholarly affiliation, patronage, officeholding, textual circulation, and changing standards of judgment. Inherited labels are therefore treated as historical claims requiring explanation, not as stable entities defined a priori.

The central question is how historical actors created recognizable traditions and communities by reading, copying, and reprinting books, constructing lineages, translating texts, imitating prestigious styles and forms, and selectively appropriating ideas from rival traditions. Neo-Confucian intellectuals might publicly condemn Buddhist and Daoist teachings while continuing to borrow their vocabulary, contemplative techniques, cosmology, or sense of taste.

Circulation and appropriation

These exchanges took place amid doctrinal controversy, reputational risk, ostracization, and institutional competition, and nonetheless shaped what later appeared as coherent Confucian, Buddhist, or Daoist traditions. Recovering and making sense of these processes also demands attention to the linguistic and stylistic range of Literary Sinitic (wenyan 文言 or hanmun 漢文), the retrospective construction of intellectual genealogies, the movement between close reading and defensible generalization, and the circulation of knowledge, taste, and sensibility that contemporaries did not necessarily name.

Ideas were transmitted selectively to followers and readers, while prestigious forms were copied, altered, and recast with the creative interventions of their admirers. Informal circles sometimes acquired the solidity of recognized schools, while more inchoate communities dispersed before receiving lasting names.

The complete bamboo

The title draws on Su Shi’s commemorative account of Wen Tong, whom he praised as possessing the complete bamboo in the breast. The phrase describes more than a preconceived image of the finished painting. Wen Tong had internalized the bamboo’s full life—from germination and flourishing to decay and renewal—so completely that his brush could express its dynamic existence and myriad manifestations without lapsing into mechanical imitation.

Su Shi’s point was that the bamboo could not be painted successfully by assembling it joint by joint and leaf by leaf; its form had to be grasped as an organic unity, then realized with sufficient speed and discipline that mind and hand remained in accord. Artistic failure arose when inward understanding and outward execution parted company.

This is the challenge Playing Heaven proposes to address. The past is reconstructed from dispersed and uneven traces—a doctrinal claim in one text, a stylistic imitation in another, a political association recorded later, a translation choice preserved in a bilingual edition, an aesthetic judgment embedded in a poem, a loose affiliation retrospectively elevated into lineage. Human researchers can interpret each piece closely, but it is exceedingly difficult to hold all of them in view at once.

The historical question before the method

The project develops a domain-specific compound AI system in response to these questions, combining transformer-based and non-transformer machine learning with rule-based methods according to the task at hand. Its individual components perform distinct historical operations, including transcription, translation, entity recognition and resolution, search and discovery, historiographical comparison, and higher-order interpretive tasks. These operations draw on proprietary and open-weight models, vector embeddings, and knowledge distillation techniques. The research environment prioritizes source criticism, domain expertise, and the requirement that every result can be returned to a specific location in an authoritative primary or secondary source.

The method →

Research operations

A compound system combining transformer-based and non-transformer machine learning with rule-based methods according to the task at hand. Its components perform distinct historical operations. The historical question guides the selection of sources, the structure of the data, the choice of computational method, and the range of possible interpretations.

1. i

   Transcription

   Optical character recognition for woodblock and manuscript Manchu and Literary Sinitic; every transcribed unit keeps its image coordinates.

2. ii

   Translation

   Historically informed, context-sensitive translation between Manchu, Literary Sinitic, Korean and English; the source stays beside the rendering.

3. iii

   Entity recognition and resolution

   Deciding whether differently recorded persons, places and titles refer to the same referent, with the evidence for each decision retained.

4. iv

   Search and discovery

   Vector embeddings and rule-based measures to find latent proximities and documentary absences across genres, languages and generations.

5. v

   Historiographical comparison

   Secondary scholarship distilled by the same process, so that arguments can be linked to the primary sources they rest on, raw and distilled.

6. vi

   Higher-order interpretation

   Section → essay → provisional profile. Each level cites the level beneath it; statement and inference stay distinguishable.