Reading the Registers of Honorius III without Manual Line Transcription: Forced-Aligned HTR, End to End, Modular or Hybrid

Tamas Kovacs; Daniel Luger

cs.CVOct 26, 2026

Reading the Registers of Honorius III without Manual Line Transcription: Forced-Aligned HTR, End to End, Modular or Hybrid

Authors: Tamas Kovacs, Daniel Luger

Organizations: University of Graz Department of Digital Humanities · Institut für Österreichische Geschichtsforschung University of Vienna

Abstract

Unlocking medieval archives at scale is bottlenecked by the cost of line-level diplomatic transcriptions and manual abbreviation expansion. Focusing on the registers of Pope Honorius III (1216-1227) over 5,000 letters surviving in full only as manuscript images - we show that existing, line-break-free expert transcriptions (edition drafts) suffice to train end-to-end hand-written text recognition (HTR) that jointly transcribes and expands heavily abbreviated Latin. Using forced alignment on 860 unaligned edition entries across 394 pages of Reg. Vat. 9, we automatically bootstrap 13,122 training lines without manual alignment of segmented lines. While the generic CATMuS baseline struggles on this domain (18.9% CER) and a modular space-free HTR + word-segmentation pipeline (cf. Camps et al. 2021) plateaus at 15.6% WER, our from-scratch hybrid Conformer - combining CTC with bidirectional subword attention decoders, iterative self-realignment and n-best LM rescoring - achieves 4.7% CER and 8.7% WER (8.4% WER with ROVER system combination), nearly halving errors on clause-final words (22.7% to 12.2% WER). These figures are measured on the test lines that align with the transcription (86% of all test lines for the best hybrid). Furthermore, we address a tension between paleographic fidelity and editorial normalization: training on targets that preserve scribal name initials reduces the WER against the manuscript text on the 31 affected test lines from 13.3% to 9.2%, at a small cost against the expanded transcription (WER +0.3 points, CER +0.4). Finally, extensive ablations show that none of an auxiliary diplomatic CTC head, a copy mechanism, a permutation decoder or masked-image pretraining improves on direct end-to-end supervision, while strict padding-mask propagation and per-line beam termination make inference batch-invariant.

Figures & tables

Study	Source	Training data	Test	Target	Result
ICR crowd pipeline ( Nieddu et al., 2021 )	Reg. Vat. 12	20 crowd-annotated pages	39 pages	letters + abbr. symbols	CER 19.18%
Transkribus model ( Nieddu et al., 2021 )	Reg. Vat. 12	9,000 expert lines, other material	same 39 pages	–	CER 15.90%
ICR expert lines ( Ammirati et al., 2023 ; Ammirati et al., 2025 )	Reg. Vat. 12	$\approx$ 20 folios	5 folios	expanded	$\leq$ 8 errors / 80-char. line
CATMuS, Kraken ( Clérice et al., 2024 )	8th–16th c.	146,885 / 138,900 lines	seen / held-out mss.	graphemic	CER 4.7 / 13.1%
Camps et al. (2021) , full systems	BnF lat. 14525	1,524 lines, one scribe	folio 45v	expanded	CER 6.89–8.60%, WER 31.81–54.68%
Torres Aguilar and Jolivet (2023)	charters, registers	$\approx$ 105,000 lines	held-out mss.	semi-diplomatic	accuracy $\geq$ 0.85 / $\geq$ 0.90

Table 1: Published results on the papal registers and related material, and the recognizers of this study (held-out pages of the same register). Test sets, targets and metrics differ; the comparison is indicative.

Baseline and decoding
Recognizer	WER	CER	final	OOV	$\alpha$	$\beta$
	%	%	%	%	%	%
C, greedy CTC	14.7	7.1	22.7	26.9	10.5	13.6
C retrained with mask propagation	14.3	6.9	19.8	26.6	8.4	11.0
C + word trigram LM	13.0	6.8	19.9	26.0	9.9	11.6
Attention decoders (on masked C)

Table 2: All variants on the 39 test pages (words aligned per line). Final, OOV: WER on clause-final and out-of-vocabulary words;

\alpha

\beta

: cadence errors at clause ends (Section 3): non-canonical endings read as canonical (

\alpha

), canonical endings read as non-canonical (

\beta

). Guarded ROVER: best single hybrid with two other systems (Section 3). Hybrids use the batch-invariant beam; last group trained on silver v2. a Second seed: WER 10.6%, CER 5.4%, final 14.6%.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

	entries	words	lines	accepted	silver
				%	lines
Year 1 (Luger), fol. 1–130	535	132,943	12,166	83.4	8,556
Year 2, part (Singh Bottanova), fol. 137–203	325	70,876	6,966	80.8	4,566
train pages (315)			15,282	82.5	10,525
dev pages (40)			1,846	84.8	1,313
test pages (39)			2,004	79.3	1,284

Appendix

Table 3: Transcriptions, recognized lines, accepted lines (distance

\leq 0.35

) and silver lines (

\geq 15

characters, distance

\leq 0.25

), by transcription and by split.

Recognizer	line CER	compared CER	WER	page CER	aligned
	%	%	%	%	%
A CATMuS Medieval 1.5	19.5	18.9	54.3	26.4	79
A ′ = A + rule expansion	18.0	17.5	48.1	–	80
B CATMuS fine-tuned	8.5	8.3	18.1	12.2	78
C Conformer-CTC	7.8	7.1	14.5	15.9	89
C+G (C with Gaussian logit head)	7.9	7.3	15.4	14.9	88

Appendix

Table 4: Recognizers on three error bases (%); WER with words aligned per page. A ′ : A with rule-based expansion of the CATMuS abbreviation signs. Aligned: share of the 2,004 test lines accepted by the alignment (distance

\leq 0.35

); line, compared CER and WER use the accepted lines with at least 15 reference characters (compared CER: pages averaged). C+G: C with a heteroscedastic Gaussian logit head and a latent Gaussian bottleneck ( Kendall and Gal, 2017 ) , trained with Monte Carlo CTC (8 samples, KL weight

10^{-3}

) to obtain per-frame uncertainty; it is no better than C and is not used elsewhere. The hybrids are scored against the expanded transcription (the re-aligned ones as in Table 1 ); their page CER is not computed (–).

data	train lines	epochs	dev CER	CER	WER
			%	%	%
10%	1,052	60	10.7	11.5	28.9
25%	2,631	50	8.1	9.2	20.6
50%	5,262	60	6.9	8.0	17.0
100%	10,525	57	5.9	7.1	14.7

Appendix

Table 5: C trained on random samples of the silver training lines: epochs to early stopping, best development CER, and test CER and WER (words aligned per line).

Figure 1: CER and WER of C against the number of silver training lines: converged runs and early snapshots after epochs 3, 6 and 12.

	C				hybrid	+ NER
Position (reference clause ends)	words	WER	abbreviated	in full	WER	WER
		%	%	%	%	%
clause-final word	619	22.3	25.4	19.6	16.4	17.3
penultimate word	600	19.5	21.1	18.1	16.8	16.2
3rd–4th word from the end	1,160	20.3	16.0	25.0	17.1	17.2
all other words	17,974	13.9	13.6	14.1	11.9	12.3

Appendix

Table 6: Word error by position relative to clause ends: C, for words abbreviated on the page and words written in full, and the hybrid without and with the NER head (sigla as written), against the expanded transcription (words aligned per line together with the diplomatic reading, which gives 22.3% for clause-final words against 22.7% in Table 2 ).

Recognizer	WER	CER	WER	CER	WER	CER	WER	$\alpha$	$\beta$
	expanded ref.		sigla ref., all		sigla ref., 31 lines		names
hybrid, names expanded (E10)	12.5	6.4	8.9	4.5	13.3	10.1	32.1	8.1	10.1
hybrid, sigla as written	12.8	6.8	8.7	4.3	9.2	5.0	29.6	7.8	10.5
hybrid, sigla + NER head	12.8	6.8	8.7	4.3	8.2	4.9	29.4	7.2	10.6
hybrid, silver v2 + R2L + NER head	10.5	5.2	8.1	4.0	9.2	5.9	31.0	11.7	5.0
+ LM rescoring, sigla guard	9.8	5.1	7.2	3.8	7.7	5.2	31.7	11.1	4.9

Appendix

Table 7: Hybrid recognizers and name initials on the 39 test pages. Expanded ref.: the common basis of Table 2 (names expanded by the editor); sigla ref.: the transcription with name initials as written, on all test lines and on the lines the rule changes. Names: WER on words the name rule selects in the sigla reference.

\alpha

\beta

on the common basis. Below the line: the NER fine-tuning applied to the R2L hybrid of silver v2, alone, with LM rescoring under the sigla guard, and after MWER training with the same rescoring. Without the guard, LM rescoring gives 8.2% / 5.3% WER/CER on the 31 lines (MWER: 7.7% / 5.7%).

Model	batch	lines changed	CER	WER	$\alpha$	$\beta$
		vs. batch 1	%	%	%	%
C as trained	1	0	7.08	14.48	10.5	13.6
	16	465	7.01	14.39	11.2	13.2
	32	598	7.14	14.48	12.1	13.4
	32, random	810	7.08	14.44	9.5	13.2
C weights, masked inference	1	0	7.09	14.49	10.5	13.6

Appendix

Table 8: Batch dependence of the output: lines changed against decoding one line at a time, and the metrics of each batch layout (words aligned per page). Batches are grouped by line width unless marked random. Hybrids: joint 5-best decoding under the original stopping rule of the beam search (a batch stops when all its lines are finished) and the fixed rule (each line stops on its own).

Historical Document Processing

Historical document processing uses AI to transcribe, structure, restore, and analyze manuscripts, books, newspapers, and archival records.

See all 76 papers

Oct 27, 2026

Mining the Nineteenth Century for the Thirteenth: Register-Scale Forced Alignment and HTR Bootstrapping via Legacy Regesta and Antiquarian Editions

Handwritten text recognition (HTR) for the registers of Pope Honorius III depends on modern expert transcriptions, which exist only for regnal years 1-2 of Reg. Vat. 9. For the remaining registers, nineteenth-century scholarship offers Pressutti's regesta and partial editions. On fol. 1-203, where both layers exist, we measure what this legacy layer can contribute, on transcriptions that we first audit against it, and find a division of labor across three tiers. At the orthographic tier, raw editions are at best neutral for adapting a recognizer to the second regnal year and harmful without replay, whereas projection onto curial spelling with replay of modern source lines turns them into zero-label adaptation targets: pooled over all 134 year-2 sides, the WER falls by 1.50 points (95% CI -1.84 to -1.15), closing 33% of the gap to an oracle; the spelling changes at the year split, through the scribe or the transcriber, whereas the hand changes visibly only at fol. 191r. At the lexical tier, the corpus-wide legacy lexicon contains 82.2% of the out-of-vocabulary words of unedited entries and flags recognition errors with 66.7% precision, and labels from pre-trained medieval NER models, distilled into the recognizer's NER head, reach F1 82.1% on an expert validated benchmark, level with the best tagger applied after recognition. At the topological tier, Pressutti's regests, the closing datum clauses, and the regests' dates converted to the chancery's Roman form find 73.9% of entry starts in the recognized line stream, and edition agreement estimates a CER of 7.80% beyond fol. 203. Antiquarian scholarship thus cannot replace modern transcription, but it can steer, check, and segment HTR across the untranscribed registers.

Tamas Kovacs, Daniel Luger

University of Graz Department of Digital Humanities · Institut für Österreichische Geschichtsforschung University of Vienna

Historical Document Processing Forced Alignment

Ask PDF

Jun 23, 2026

TrOCR for Medieval HTR: A Systematic Ablation Study with Cross-Dataset Validation

Fine-tuning transformer-based handwritten text recognition (HTR) models on medieval manuscripts is challenging because these models are pre-trained on modern text and must adapt to a very different visual domain. This paper studies how three controllable fine-tuning choices (contrast normalization, data augmentation, and layer freezing) affect recognition accuracy when adapting TrOCR to small historical datasets. We run controlled experiments on a 13th-century Italian manuscript (I-CT 91 "Cortonese") and replicate the same experimental grid on the public READ-16 benchmark as robustness evidence. On Cortonese, our best configuration achieves 8.03% character error rate (CER). Statistical comparisons across 13 configurations show that freezing up to three encoder layers or six decoder layers does not significantly harm accuracy, while deeper freezing becomes progressively detrimental. Removing contrast normalization (CLAHE) yields 7.84% CER, comparable to a domain-specialized baseline, suggesting strong optimization can reduce reliance on image preprocessing. Cross-dataset validation on READ-16 shows that decoder freezing thresholds transfer more robustly than encoder thresholds, and combined freezing strategies require dataset-specific re-validation. Finally, we use Grad-CAM gradient attributions and decoder cross-attention maps to diagnose error patterns and failure modes revealed by the ablations. Source code is available at https://github.com/LaudareProject/TrOCR-analysis

Sachin Sharma, Michele Flammini, Federico Simonetta

GSSI – Gran Sasso Science Institute L’Aquila, Italy

Fine-Tuning VLM Adaptation Online Handwriting Recognition Optical Character Recognition

Ask PDF

May 1, 2026

Democratizing the medieval English legal tradition

The record of the beginning of the most widespread legal system in the world is contained in millions of pages of handwritten text. Most of the records of the first centuries of the Anglo-American legal system are hand-written in a highly abbreviated form of medieval Latin which only a few dozen scholars in the world are trained to read. In this interdisciplinary project, we construct a dataset of 4029 lines of text across 193 medieval criminal and civil cases. We then use the dataset to train an open-source end-to-end pipeline for transcribing these manuscripts. We first train standard neural network architectures for line segmentation and handwriting recognition (R-Blla and CNN+LSTM with CTC decoding, respectively) and show that they can already achieve 79% word accuracy, despite the relatively small training set and the challenge of expanding abbreviations. We then demonstrate that simple post-processing significantly boosts accuracy: adding an n-gram language model to the CTC decoder improves word accuracy to 82%, while asking Gemini Pro 3 to correct mistakes boosts accuracy to 88%. Finally, we compare the CNN+LSTM architecture with TrOCR, a transformer-based OCR architecture, demonstrating that TrOCR shows comparable word accuracy but worse character accuracy due to its over-willingness to guess, making it harder for humans to infer the correct reading. We incorporated our pipeline into a web portal (glyphmachina.com), opening up the English legal tradition to legal scholars, medievalists, and students.

Michael Zhang, Elise Wang, Charlotte Whatley +2

University of Chicago, Chicago IL 60637 · California State University, Fullerton CA 92831 · University of Wisconsin-Madison, Madison WI 53706 +2

Online Handwriting Recognition Optical Character Recognition Historical Document Processing

Ask PDF

Abstract

Historical Document Processing

Explore similar work

Mining the Nineteenth Century for the Thirteenth: Register-Scale Forced Alignment and HTR Bootstrapping via Legacy Regesta and Antiquarian Editions

TrOCR for Medieval HTR: A Systematic Ablation Study with Cross-Dataset Validation

Democratizing the medieval English legal tradition