Automatic Scribe Attribution for Medieval Manuscripts
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
We propose an automatic method for attributing manuscript pages to scribes. The system uses digital images as published by libraries. The attribution process involves extracting from each query page approximately letter-size components. This is done by means of binarization (ink-background separation), connected component labelling, and further segmentation, guided by the estimated typical stroke width. Components are extracted in the same way from the pages of known scribal origin. This allows us to assign a scribe to each query component by means of nearest-neighbour classification. Distance (dissimilarity) between components is modelled by simple features capturing the distribution of ink in the bounding box defined by the component, together with Euclidean distance. The set of component-level scribe attributions, which typically includes hundreds of components for a page, is then used to predict the page scribe by means of a voting procedure. The scribe who receives the largest number of votes from the 120 strongest component attributions is proposed as its scribe. The scribe attribution process allows the argument behind an attribution to be visualized for a human reader. The writing components of the query page are exhibited along with the matching components of the known pages. This report is thus open to inspection and analysis using the methods and intuitions of traditional palaeography. The present system was evaluated on a data set covering 46 medieval scribes, writing in Carolingian minuscule, Bastarda, and a few other scripts. The system achieved a mean top-1 accuracy of 98.3% as regards the first scribe proposed for each page, when the labelled data comprised one randomly selected page from each scribe and nine unseen pages for each scribe were to be attributed in the validation procedure. The experiment was repeated 50 times to even out random variation effects.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it