Chinese Textual Collation (Jiaokan): How Scholars Read Variant Texts

A text survives through witnesses, not as an abstraction

Work, edition and copy are different levels

Textual collation—jiaokan 校勘—begins by comparing witnesses: manuscripts, impressions, quotations or other material carriers of wording. A named work is not the same thing as an edition, and an edition is not the same as one physical copy. Missing leaves, repairs, handwritten changes and rebinding may distinguish two copies of the same printing. Start with the material clues in our guide to Chinese thread-bound books, then describe exactly what is being compared.

A variant is evidence before it is an error

A different graph, phrase, order or omission is a variant reading. It may arise through copying or cutting, but it may also preserve an alternate formulation, editorial clarification or distinct textual state. Edward Shaughnessy's study of variants in the Tsinghua Ming xun manuscript classifies several mechanisms while showing why comparison need not reduce every difference to corruption. Record the difference before deciding what it means.

Variants arise through many paths

Graphic similarity, sound and layout can redirect a reading

Characters can be confused because their written forms resemble one another, because one graph is used for a similar sound, or because a line, column or repeated phrase invites omission or duplication. Early manuscripts also preserve orthographic practices that do not map neatly onto later standardized characters. The histories of dictionaries and rhyme books help explain why graph, pronunciation and meaning must be considered together rather than corrected by visual resemblance alone.

Damage, normalization and interpretation also change texts

A damaged leaf may invite completion; taboo avoidance, house style or a later standard can prompt substitution; an editor may smooth difficult grammar or harmonize a parallel passage. Matthias Richter's Harvard seminar on early Chinese texts cautions that features often treated as accidents may sometimes be intentional. The cause of a variant is therefore a hypothesis to argue, not a label to apply automatically.

Collation begins by identifying the witnesses

Describe provenance before comparing words

Give each witness a clear siglum and record date or date range, material form, edition statement, provenance, completeness, page or folio system, paratexts and later interventions. A catalogue title alone is not enough: the guide to Chinese bibliographies distinguishes title, juan, edition and copy. Such description makes the comparison repeatable and prevents an attractive digital image from floating free of its source object.

A base text is an operational choice

The base text supplies the line to which other readings are attached; it is not automatically the oldest, most famous or ‘best’ witness. It may be chosen for completeness, accessibility, historical importance or because a project studies that textual state. State the reason and preserve the base reading even when the edited text adopts another one. Harvard's guide to reading a variant apparatus shows how the reading text and recorded alternatives work together.

Compare readings in context

One character belongs to a sentence and a textual system

Compare the whole phrase, grammar, argument, rhyme, parallel construction, commentary and recurrent vocabulary. A locally smooth reading may be secondary smoothing; a difficult reading may be early, or simply mistaken. Quotations and linguistic reference works can help, but they must be dated and evaluated as witnesses in their own right. Collation is strongest when several independent kinds of evidence converge.

Older is valuable, not automatically decisive

An early manuscript may preserve an early state while a later copy preserves a sound reading lost elsewhere. A recent print may descend from an old source, and an old witness may contain recent corrections or local errors. Xiaofei Tian's study of Tao Yuanming and manuscript culture emphasizes the fluidity of title, content and authorship across medieval transmission. Date matters, but relationship and transformation matter too.

An apparatus separates observation from judgment

Record who reads what, and where

A compact collation note should identify the location, the adopted or base reading, each relevant alternative and the witness that carries it. Consistent sigla, punctuation and an explained convention matter more than decorative complexity. Harvard's introduction to variant notation demonstrates how even changes often called non-substantive can illuminate transmission, scribal habits and punctuation.

Correction and conjectural emendation are not the same

Selecting a reading attested by a witness is different from proposing an unattested correction. Conjectural emendation can be necessary when every surviving reading is defective, but it should be marked openly and justified by language, context, parallel passages and a plausible path of corruption. David B. Honey's review of Huainanzi textual criticism also summarizes limits of simple family trees when witnesses are contaminated, authorial variants remain possible or no single definitive text is assumed.

Paratexts and quotations are evidence with limits

Prefaces, colophons and tables of contents map editorial acts

A preface may name a base copy or editorial aim; a colophon may record comparison, correction or ownership; a table of contents can preserve an order lost in the body. Yet each statement has an author and occasion. Read these materials with the guide to Chinese prefaces and afterwords rather than treating them as neutral laboratory logs.

Quotations and excerpts can preserve lost readings

A quotation in commentary, anthology or encyclopedic leishu may preserve wording absent from surviving standalone books. It may also shorten, adapt, normalize or misattribute its source. Our guide to classified books explains why an excerpt proves that wording circulated somewhere in its source chain, not that it reconstructs a whole lost work or its original order.

Print and digital tools change the scale, not the duty of checking

A recut edition may look the same while carrying new readings

Later printers could reuse a layout, recut blocks, replace damaged leaves or combine sheets from different impressions. Page resemblance is therefore a clue, not proof of identity. The article on illustrated Chinese books shows how blocks, recutting and impressions interact, while the history of private libraries helps trace ownership and handwritten intervention.

Searchable text is an aid, not the witness

The Chinese Text Project links transcriptions to scanned sources where available, and its OCR instructions explicitly describe automatically created text from transmitted copies. Search and OCR can locate parallels or candidate differences; they cannot replace checking the scan, layout and catalogued source. Save stable identifiers, images and access dates alongside every transcription used.

Use eight questions before adopting a reading

Make the chain of reasoning visible

Ask: 1) what physical or quoted witness is this; 2) what edition and copy does it belong to; 3) how is it related to the others; 4) what exactly varies; 5) could layout, sound, graph form or editing explain it; 6) what contextual evidence bears on the choice; 7) is the chosen reading attested or conjectural; and 8) how will rejected readings remain visible?

A good edition preserves disagreement instead of hiding it

The goal is not always to recover one authorial original. It may be to document several textual states, explain an editorial history or offer a readable text with a transparent apparatus. Harvard's Library of Chinese Humanities provides examples of scholarly editions and translations that note textual variants. This method connects the histories of individual collections to the wider field of Chinese Poetry and Literature without silently turning editorial judgment into fact.