Dataset for the bilingual Rule of St. Benedict (manuscripts: OCCC197, CCCC178, BLCTaiv)

This repository contains ground truth for three witnesses of the bilingual (Latin / Old English) Rule of St. Benedict: - Oxford, Corpus Christi College, MS 197- London, British Library, Cotton Titus A.iv- Cambridge, Corpus Christi College, MS 178 used in the experiments for the oral paper "Using Computer Vision to Measure Character Proportions, Inter-Letter Spacing and Word Separation in Bilingual Latin / Old English Manuscripts" (Malamatenia Vlachou Efstathiou ; Mark Faulkner; Elisabetta Magnanti), presented at the [Biblissima+ Cluster 3 conference "Intelligence artificielle et la reconnaissance de formes et d'écritures manuscrites"](https://biblissim-ia-26.sciencesconf.org/) (Biblissima-IA 2026, Campus Condorcet, Paris). Repository structure: 📁 `bilingual_RSB_GT.zip`This folder contains the ground truth dataset used for Handwritten Text Recognition (HTR), created from selected folia of the three manuscripts above. Folder structure:bilingual_RSB_GT├── images└── annotations - `images/`: High-resolution selected images downloaded from each library's open-access viewer. Image names follow the pattern `<manuscript_siglum>_f<number>`, corresponding to the page/folio number in the source viewer. ➤ Credit:- Oxford, CCC 197 — images courtesy of Corpus Christi College, Oxford, via [Digital Bodleian](https://digital.bodleian.ox.ac.uk/objects/5d0e482b-2222-455a-b75e-d8ca73e93c6b/)- London, BL, Cotton Titus A.iv — © British Library Board, images via [British Library Digitised Manuscripts](http://www.bl.uk/manuscripts/FullDisplay.aspx?ref=Cotton_MS_Titus_A_IV)- Cambridge, CCC 178 — images courtesy of the Parker Library, Corpus Christi College, Cambridge, via [Parker Library on the Web](https://parker.stanford.edu/parker/catalog/pm669yz1553) `annotations/`: XML-ALTO annotation files created with [eScriptorium](https://escriptorium.inria.fr/).Layout: Annotations follow the [Segmonto](https://segmonto.github.io/) ontology. Additionally, we annotate lines with custom tags according to the language: #EnglishLine and #LatinLine, respectively. Transcription: Transcription is graphemic, so allographic variants are not distinguished: for example, insular, round, and long "s" are all transcribed uniformly as "s". Annotated by Malamatenia Vlachou Efstathiou (Latin partition) and by Mark Faulkner and Elisabetta Magnanti (Old English partition). 📁 `dataset.zip`This folder contains the dataset used in the experiments, in the format of text lines and associated annotations and metadata.Folder structure:dataset├── images└── annotation.jsonChapter selection: Of the 70 chapters of the Rule of St. Benedict, we selected and transcribed 18 chapters across all three manuscripts. The unit of organization is the chapter, not the manuscript page: each chapter is transcribed separately for its Latin and English partitions, since the two languages occupy separate (though often adjoining) folio spans. - `images/`: Each subfolder corresponds to one partition (Latin or English) of one selected chapter, named `<manuscript_siglum>_<folio_start>-<folio_end>` after the folio span it covers, and contains the polygonal line extractions (with alpha transparency) for that span. For example, `OCCC197_7v-8v` contains the Latin partition of a chapter spanning folios 7v–8v, and `OCCC197_8v-9r` contains the English partition of the following chapter, spanning folios 8v–9r. `annotation.json`: Contains the annotation and metadata for each line. `annotation.json` structure example:```json"<image_id>": { // corresponds to the image names in the images folders "split": "train", "label": "A beautiful calico cat.", // transcription text of the line "script": "A", // identifier for the scribal hand "source": "OCCC197_8v.tif", "line": "LatinLine" // "Latin" or "English" chapter": "1"} ```Finally, we share a .csv file with structured metadata: `bilingualRSB_metadata.csv`, including hand and partition (English vs. Latin) and number of lines for each chapter.