西醫新法([1872];Bodleian Sinica 1826) / Xi yi xin fa [New methods of Western medicine]: Transcription dataset
Authors/Creators
Description
《西醫新法》(Xi yi xin fa)是討論外科與包紮的晚清漢譯醫學文獻,署嘉約翰(John Glasgow Kerr)口譯、林湘東筆述;Bodleian Sinica 1826 的館目供定刊地 [Canton]、年代 [1872]。本資料集依25幅影像提供44個指定物理面的 TXT、JSON、TEI XML、疑讀與來源對應,可供晚清醫學翻譯、包紮史及圖文關係研究。文字為 AI 暫定轉錄並經局部原圖修訂,未經人類全文雙校或專家審定。
Xi yi xin fa [New methods of Western medicine] is a late-Qing Chinese work on surgery and bandaging, credited to John Glasgow Kerr for oral translation and Lin Xiangdong for written rendering. The catalogue for Bodleian Sinica 1826 supplies [Canton] and [1872]. Based on 25 images, the dataset provides TXT, JSON and TEI XML for 44 designated physical faces, with uncertainties and source mappings for research on medical translation, bandaging history and image-text relationships. The provisional AI transcription includes selected source-image corrections; full human double collation and expert review have not been performed.
Notes
本版書目與權利說明 / Current bibliographic and rights notes
本紀錄對應 v0.2.1;前版 v0.2.0 及其 DOI 10.5281/zenodo.23971701 保留。來源原件、館號、書目題署、語言與分層權利沿用前版;歷史原文、具著作權的自有說明/編排及館方影像條款仍分別適用。原有分層條件未因本次只發布文字而移除。文言正文為lzh,英文說明為eng。固定新版GitHub發行入口為 https://github.com/poplolioi11/bodleian-sinica1826-xiyi-xinfa-1872/releases/tag/v0.2.1;前版DOI以IsNewVersionOf連接,來源關係保留其原有語義。
This record describes v0.2.1; v0.2.0 and its DOI 10.5281/zenodo.23971701 remain the previous version. Source witnesses, shelfmarks, bibliographic attributions, languages and layered rights continue from that record. Historical text, copyrightable original documentation/arrangement and library-image conditions remain distinct; text-only publication does not remove the recorded source conditions. Literary Chinese is lzh and English documentation is eng. The fixed new GitHub release is https://github.com/poplolioi11/bodleian-sinica1826-xiyi-xinfa-1872/releases/tag/v0.2.1; the previous DOI is linked as IsNewVersionOf while source relations retain their existing semantics.
前版書目歷史說明 / Previous-version bibliographic notes (0.2.0)
以下各段僅記錄前版方法或書目狀態,不代表本輪修訂的現況。 / The following passages describe the previous version only, not this revision.
Source: https://digital.bodleian.ox.ac.uk/objects/d07550d1-b969-4d5a-a5f3-5bd10634b21c/ IIIF: https://iiif.bodleian.ox.ac.uk/iiif/manifest/d07550d1-b969-4d5a-a5f3-5bd10634b21c.json Official fixed catalogue: https://github.com/bodleian/east-asian-mss/blob/c177d5a6db8585168b744d38a6bff02186d1968f/collections/chinese/Sinica%201826.xml Formats: TXT, JSON, TEI XML; scope: 25 captures, 44 numbered physical faces (six figure faces, 37 main-text faces, one original-frame/blank-body face); four cover surfaces excluded. Historical public-domain text and faithful mechanical transcription acquire no new exclusive restriction. Copyrightable original documentation and arrangement: CC BY 4.0. Bodleian images are excluded and separately carry CC BY-NC 4.0. The English bracketed title is an editorial translation. Repository: https://github.com/poplolioi11/bodleian-sinica1826-xiyi-xinfa-1872
Methods
本輪 v0.2.1:局部非盲原圖覆核 / Current targeted review
本輪以同一固定底本回看9組指定讀者問題,候選在判讀前已知,屬局部非盲原圖覆核。本版採用12處逐項文字操作,改動12個Unicode碼點;保留未採納及未解候選、原有字形編碼、單位、讀序與版本歷史,不做全域換字或正規化。 刻點編碼問題僅核對7個診斷圖區;程式列出的1093個位置只是定位清冊,不作全數已辨圖的宣稱,原有碼列保留。 本輪未新做全文獨立二校,也未完成人類全文雙校、專家審定或準確率量測。本輪原生AI實際模型識別、推理設定及完整用量未獲回報,保持未知;不以0或算法名稱填補,不宣稱三維追溯已完整。格式與字串回讀通過只表示工程檢查結果。影像、裁片、私有程式、提示與完整內部紀錄的既有排除界線維持,不增加由影像完整重跑的承諾。
This revision checks 9 specified reader-question groups against the same fixed source witness with the candidates already known: a targeted, nonblind source-image review. It adopts 12 individually specified literal operations affecting 12 Unicode codepoints, while retaining rejected and unresolved proposals, existing encoding, unit identities, reading order and version history without global replacement or normalization. The punctuation-encoding question used seven diagnostic image regions; the1,093 program-listed positions are a location inventory, not a claim of individual visual adjudication, and existing codepoints are retained. This round is not a new independent full-book collation. Full human double collation, expert review and accuracy measurement have not been performed. Exact native-AI model identifiers, reasoning settings and complete counters for this selected review were not reported and remain unknown; neither zero nor an algorithm label is substituted, and full provenance completion is not claimed. Format and fixed-string readback establish engineering properties only. Existing exclusions for source images, crops, private programs, prompts and complete internal records remain; no new promise of complete image-to-text reproduction is added.
前版方法歷史說明 / Previous-version methods (0.2.0)
以下各段僅記錄前版方法或書目狀態,不代表本輪修訂的現況。 / The following passages describe the previous version only, not this revision.
# Methods / 處理方法
The fixed initial transcription was preserved as the v0.1.0 baseline. Separate source-only second image observations were collected, then compared over the 44 physical faces using literal page streams while retaining both original unit segmentations. No Unicode, variant-character or punctuation normalization was applied.
Nine Gemini jobs were submitted. Four yielded usable body transcription covering 10 source captures and 15 physical faces; five did not yield usable body transcription. The reported actual Gemini model was `gemini-3.8-flash-high`. Native source-only supplements supplied the remaining 29 faces. Exact native model identities, effort and complete counters are unknown and are not recorded as zero. This mixed observation coverage must not be described as a successful complete Gemini transcription or a guarantee from different models.
Subsequent AI source adjudication was nonblind: both readings and their differences were exposed. The resulting edition applies 215 literal operations in 154 original units. The initial 472 identifiers, ordering and all source hashes are retained. All 1,130 comparison decisions are preserved privately; 215 candidates remain unresolved. Source-visible marks, their approximate Unicode representation, and their printed or later-added origin are separate questions.
The 69 initial uncertainty declarations retain their original strings and positions in private revision history. Current ranges derive from the actual operations. Two reordered cases retain a broader, explicitly provisional replacement context rather than a guessed one-to-one glyph location. Two further uncertainty records were added, and one existing record gained candidates, giving 71 current records. The 15 original edge observations retain unknown extent and uncertain ownership where applicable.
JSON, TXT and TEI XML are deterministic exports of the fixed revised strings. Source identity, schema, round-trip, privacy and package checks evaluate engineering properties, not reading correctness. No human full-text double collation, expert approval, peer review or accuracy measurement has been performed.
Source scans, crops, dedicated programs, prompts and complete private working records are not distributed. A complete method for rerunning image-to-text transcription from source images is not provided. The public revision summary is deliberately limited to text changes, source references and evidence scope.
本版為 AI 輔助歷史文獻修訂。兩稿已曝光的回圖判斷,不屬盲校或人類審定;未解點位、字形與頁邊歸屬仍保留。本工作沒有進行生物研究或現代臨床推導。
The per-face `historical_initial_observation` preserves initial-reader metadata verbatim, including old glyph descriptions, batch limits and pending statements. It is historical evidence rather than the status of this revision. The adjacent `revision_status` records the present provisional edition and counts; unresolved source-edge locator labels do not establish page ownership.
各面 historical_initial_observation 原樣保存初錄頁註;其中舊字形、批次範圍與 pending 只描述初錄當時,不代表本版現況。本版狀態另列 revision_status,未因此清除疑讀、比較未解處或邊區歸屬。