Transkribus models

Descriptions, stats, and sample training data (where available) from public Transkribus models, to help scholars find a model to work with.

If you have a model you'd like to have listed here (along with contact info so people can request access), please email Quinn at qad@stanford.edu.

18c Danish Administrative Writing

Language: Danish Description: This model was made in a collaboration between Associate Professor Nina Javette Koefoed from Aarhus University and City Archivist Søren Bitsch Christensen and Archvist Kristian Pindstrup of Aarhus City Archive.

It is based on the model 18th Century Danish Prison Records v.5.1, which is also used as base model. The primary training material consists of 18th Century protocols from the penitentiary on Møn. To increase the universality of the model, further training material has been added from the models "Various Danish Hands c. 1780 v. 1.13" by Associate Professor Johan Heinsen from Aalborg University and "17th Century Norwegian Court Records - Model Version 5" by former Senior Advisor Steve Jackson from the National Archival Services of Norway.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	623268	103155	7.17%	5.93%

No training image samples available.

18c Danish Administrative Writing (PyLaia)

Language: Danish Description: This model was made in a collaboration between Associate Professor Nina Javette Koefoed from Aarhus University and City Archivist Søren Bitsch Christensen and Archvist Kristian Pindstrup of Aarhus City Archive.

It is based on the model 18th Century Danish Prison Records v.5.1, which is also used as base model. The primary training material consists of 18th Century protocols from the penitentiary on Møn. To increase the universality of the model, further training material has been added from the models "Various Danish Hands c. 1780 v. 1.13" by Associate Professor Johan Heinsen from Aalborg University and "17th Century Norwegian Court Records - Model Version 5" by former Senior Advisor Steve Jackson from the National Archival Services of Norway.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	PyLaia	623268	103155	3.90%	6.00%

No training image samples available.

18th Century Danish 1.03

Language: Danish Description: This model was trained by associate professor Johan Heinsen, Aalborg University. It was trained on handwriting from various military court records as well as court cases relating to inmates in various convict labour institutions in the eighteenth century. An example of the material can be found here: https://ao.sa.dk/ao/data.ashx?bid=80654798

Because the material deals mostly with soldiers and male convicts, the model has difficulty with pronouns. It also struggles dealing with hands that are highly slanted or overly compact.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	PyLaia	288719	40646	5.61%	6.48%

No training image samples available.

ASMi

Language: Italian Description: Model trained on documents c.1560-1620 from the Archvio di Stato di Milano

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	PyLaia	11325	1082	0.91%	20.13%

Ground Truth Examples

9	14	52

Acta_17 (extended)

Language: German, Latin

Description: The HTR-model was trained on the basis of more than 1,500,000 words from about 1300 different writers during the period 1580-1750. It can handle the languages German, Low German and Latin and is able to decipher simple german and latin abbreviations. Besides the usual chancery writings, the training material also contains a selection of concept writings and printed material of the period. The entire training material is based on legal texts or court writings from the Responsa of the Greifswald Law Faculty and the Wismar High Court.

Validation sets are based on a chronological selection of the years: 1580 - 1750 . GT & validation set was produced by Dirk Alvermann, Elisabeth Heigl, Anna Brandt, and Antje Laasch.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	1632534	276418	10.52%	7.19%

No training image samples available.

Acta_17 HTR+

Language: German, Latin

Description: The HTR-model was trained on the basis of more than 500,000 words from about 1000 different writers during the period 1580-1705. It can handle the languages German, Low German and Latin and is able to decipher simple german and latin abbreviations. Besides the usual chancery writings, the training material also contained a selection of concept writings and printed material of the period. The entire training material is based on legal texts or court writings from the Responsa of the Greifswald Law Faculty.

Validation sets are based on a chronological selection of the years: 1580 - 1705 . GT & validation set was produced by Dirk Alvermann, Elisabeth Heigl, Anna Brandt.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	594628	102545	9.43%	6.30%

No training image samples available.

Acta_17 PyLaia

Languages: German, Latin

Description: The PyLaia model was trained on the basis of more than 500,000 words from about 1000 different writers during the period 1580-1705. It can handle the languages German, Low German and Latin and is able to decipher simple German and Latin abbreviations. Besides the usual chancery writings, the training material also contained a selection of concept writings and printed material of the period. The entire training material is based on legal texts or court writings from the Responsa of the Greifswald Law Faculty. Validation sets are based on a chronological selection of the years: 1580 - 1705 . GT & validation set was produced by Dirk Alvermann, Elisabeth Heigl, Anna Brandt.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	Pylaia	594628	102545	7.10%	5.16%

No training image samples available.

Admiraliteit Zeeland 1605-1609 compleet

Languages: Dutch

Description: Civil affairs, 1605-1609, mainly customs

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	142715	20008	1.20%	1.12%

No training image samples available.

Annuaires_Propietaires_Adr_Paris_1898_1923

Languages: French

Description: Annuaires_Propietaires_Adr_Paris_1898_1923

Directory of owners and properties in Paris and the Seine department. Link in the BNF catalog: https://catalogue.bnf.fr/ark:/12148/cb32697229h Paris Time Machine Consortium - TGIR Humanum EHESS / CNRS / LATTICE / INRIA

169 pages in total of volumes 1898 and 1923 of the directory, the choice is made by alphabetical criterion because it is the mode of organization of the information in this document. 118 pages for training and 51 pages for validation.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	72674	13285	0.29%	0.29%

Ground Truth Examples

1	83	101

B2022 English Model M4

Languages: English

Description: English with some Latin characters. By the Beyond 2022 project, Trinity College Dublin. Early 17th to late 19th century, 40 hands using Transkribus base model and 750,000 word ground truth. The model is particularly effective on secretary and copperplate texts and can be used as a base model for more cursive scripts.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	759625	101297	7.84%	5.15%

No training image samples available.

BBM Bulliot French C19th handwritten 2021

Languages: French

Description: Trained on he notebooks of Jacques-Gabriel Bulliot (1817-1902), who initiated the first excavations at the Bibracte archaeological site between 1867 and 1895 (https://bbm.hypotheses.org).

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	146466	23507	10.57%	7.73%

No training image samples available.

Ground Truth Examples

474	782

BM_Kortrijk_M1+

Languages: Dutch

Description: Amsab-ISG, A World of Diamond

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	11915	1665	0.22%	2.53%

Ground Truth Examples

1	83	101

Behrens_Peter_final

Languages: German

Description: Peter Behrens' handwriting based on letters to Richard and Ida Dehmel between the years 1899-1919.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	10504	1851	0.58%	3.04%

No training image samples available.

BnF_Newseye_M2+

Languages: French

Description: The model works well with French script from late 18th century to mid of 20th century. For normal running text in French newspapers from that time error rates much below 1% were measured.

The model was created in the NewsEye project and is based on training data coming from the digital library Gallica of the French National Library (BnF).

Note: the model is trained on French language documents (French dailies, 1850-1945) and will therefore be less performant on other languages.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	434100	67073	4.72%	2.73%

No training image samples available.

Charos V/ Charles V

Languages: Early modern Spanish - 15th century (1543)

Description: This HTR-model is based on the personal handwriting of Charles V of Spain. It is based upon political-institutional texts/ a letter of instructions to his son. It is based upon the book: Rachael Ball and Geoffrey Parker (eds.), Cómo ser rey. Instrucciones del emperador Carlos V a su hijo Felipe. Mayo de 1543. Spanish-English bilingual edition; jointly published with the CSA and The Hispanic Society of America; 2014; ISBN 978-84-15245-45-2

Examples of the handwriting can be found at: https://doi.org/10.5281/zenodo.5381739

Please cite this HTR+-model as: Rachael Ball, Geoffrey Parker, the Hispanic Society of America, C.A Romein, Transkribus Public HTR+ Model: Carlos V/ Charles V Cómo ser rey_1543

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	9279	1603	0.39%	6.94%

Ground Truth Examples [

Charter Scripts XIII-XV_M1

Languages: German, Latin, French

Description: Combination of GT of different charter scripts. From different projects and institutions. Aiming at building a generic model.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	938325	55401	7.82%	6.32%

No training image samples available.

Combined_Full_VKS_2

Languages: Church Slavonic

Description: Training data consist of parts of the Russian Church Slavonic Great Reading Menology (16th century), Old Church Slavonic Codex Suprasliensis (11th century), and the 11th century manuscript of the Catecheses of Cyril of Jerusalem. This is a generic model suitable for transcribing a variety of Old Cyrillic script styles including uncial and semi-uncial.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	393079	75422	4.44%	3.94%

Ground Truth Examples

1	311	720

Danish 1870-1950

Language: Danish

Description: General model for Danish Handwriting from late 19th and 20th Century. Based on the model RoyalDanishLibrary_20thCentury+ and parish council minutes from Aarhus City Archives.

Created by Jan Mattias Jonsson Agger at Aarhus City Archives based on work by volunteers at the City Archives and the work from Jakob K. Meile & the staff at the Royal Danish Library

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	348716	81959	5.23%	4.28%

No training image samples available.

Danish 1870-1950 v3.5

Language: Danish

Description: General Model for Danish handwriting from 1870-1950.

Newer incrementation of Danish 1870-1950 with added material and further experimentation with base models.

Using material from The Royal Danish Library, Aarhus City Archive, Faxe Archive, Næstved Archive and Gentofte Archive.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	1603622	372704	7.68%	5.91%

No training image samples available.

Danish 1870-1950 v3.5 PyLaia

Language: Danish

Description: General Model for Danish handwriting by various hands from c. 1870-1950.

The same data as Danish 1870-1950 v3.5 but using the PyLaia engine instead of HTR+.

Created by Kristian Pindstrup & Jan Mattias Jonsson Agger at Aarhus City Archives. Using available material from The Royal Danish Library as well as material from the work of the volunteers at Aarhus City Archive, Faxe Archive, Næstved Archive and Gentofte Archive who has transcribed various city and parish council minutes.

More info in Danish on the project at http://retrodigitalisering.dk

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	Pylaia	1603622	372704	5.20%	4.70%

No training image samples available.

Danish Fraktur SB 19th century v.2.35

Language: Danish

Description: Danish Fraktur SB 19th century v.2.35 Based on 500+ pages from Royal Danish Court & State Calendar several years, pages from Danish High Court Proceedings, a few pages from Danish news-papers and other from 19th century, all together based on the NZZ Gold Standard model. Trained on a total of 745.249 Fraktur words in Danish and German.

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	30983	5349	0.28%	0.97%

Ground Truth Examples

4	37	55	56

DAT 18. Jh M3b_Pylaia

Language: German

Description: 19th century newspaper. Fraktur. 130 pages OCR correction and proofing. Print style for umlaut varies throughout (superscript "e" vs öäü). Since no applicable Pylaia model was available for training, the training set includes the HTR model data for DAT 18. Jh M1 and M2.

Text	Tech	Words	Lines	CER train	CER validation
Print	Pylaia	102875	17029	0.50%	0.30%

Ground Truth Examples

110	166	181	201

Devanagari mixed M1

Language: Hindi

Description: Devanagari script from several books late 19th early 20th century.

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR	25938	4258	4.12%	5.59%

Ground Truth Examples

3	50	105	200

Devanagari_nagara_M1

Language: Hindi

Description: Devanagari script, early 20th century, NKP collection, p. 8-40, 60-80, 100-115

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR	19243	2092	0.46%	1.77%

Ground Truth Examples

1	38	68

DiJeSt for Hebrew Script Languages

Language: Hebrew, Yiddish, Ladino, Judeo-Arabic

Description: Trained on a mix of historic Hebrew scripts and languages, in the framework of the project DiJeSt. For details: http://dijest.net/gtmodel

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	774192	79483	1.80%	1.56%

Ground Truth Examples

1	41	65	1322

Dutch_Gothic_Print

Language: Dutch (16th, 17th, 18th century)

Description: This model is based on printed texts in the Gothic font that was used in the Low Countries, during the 16th, 17th and 18th century. The type of sources used for this model, are books of ordinances, which contained the norms ('laws') at the time. This model has been the result of one of the KB National Library of the Netherlands Researcher-in-Residence position 2019. The project was called 'Entangled Histories'. For more information regarding the background of the model and how to cite it, please visit: https://lab.kb.nl/dataset/entangled-histories-ordinances-low-countries

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	51143	7143	0.22%	1.71%

No training image samples available.

Dutch_Gothic_Print_Pylaia

Language: Dutch Gothic

Description: This model is based on printed texts in the Gothic font that was used in the Low Countries, during the 16th, 17th and 18th century. The type of sources used for this model, are books of ordinances, which contained the norms ('laws') at the time.

Text	Tech	Words	Lines	CER train	CER validation
Print	Pylaia	51143	7143	0.20%	2.00%

No training image samples available.

Dutch manuscript poetry 1603-1636

Language: Dutch

Description: Model trained on an extensive manuscript of early modern poetry, in separate hands (of which one is the most important) using different types of writing and special lay-outs (e.g. chronograms). Author of the manuscript is a rhetorician (vernacular poet) from Mechelen, present-day Belgium, active in the first decades of the seventeenth century. Training was based on a word count of over 50,000 words (more than 200 folios of text).

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	51788	7414	1.43%	4.78%

No training image samples available.

Dutch Margaretha Turnor 17th Century

Language: Dutch

Description: This is the first model created by the Utrecht Archives. It is based on a thousand letters Margaretha Turnor wrote to her husband during the late 17th century. She managed the castle of Amerongen, while her husband worked abroad as a diplomat for the Dutch Republic. Her letters provide an insight into family life in the Dutch Republic as well as the political situation in the country.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	36289	4888	0.71%	1.83%

Ground Truth Examples

36	48	65	167

Dutch Mountains (18th Century)

Language: Dutch

Description: Combination of the 18th Century models from the Amsterdam City Archives (3500+ scans of 15 notarial handwritings) and the National Archives of the Netherlands (3500+ scans of VOC handwritings).

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	1384893	235985	9.04%	5.67%

Ground Truth Examples

1	6521	6792	6962

Dutch Mountains (18th Century)

Language: Dutch

Description: Combination of the 18th Century models from the Amsterdam City Archives (3500+ scans of 15 notarial handwritings) and the National Archives of the Netherlands (3500+ scans of VOC handwritings). PyLaia

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	PyLaia	1384893	235985	9.04%	5.67%

Ground Truth Examples

1	6521	6792	6962

Dutch Notarial Model 18th Century

Language: Dutch

Description: This is the first 18th Century general model created by the City Archives of Amsterdam. It is based on thousands of scans from in total 15 different notaries who worked in Amsterdam during the 18th Century.

All notaries (except Van Hoorn and Van Esterwege) have 10 scans validation included (2671 scans training, 130 validation)).

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	622904	116698	7.11%	5.32%

No training image samples available.

Dutch_Romantype_Print

Language: Dutch Roman Print

Description: This model is based on printed texts in the Roman font that was used in the Low Countries, during the 16th, 17th and 18th century. The type of sources used for this model, are books of ordinances, which contained the norms ('laws') at the time.

This model has been the result of one of the KB National Library of the Netherlands Researcher-in-Residence position 2019. The project was called 'Entangled Histories'.

For more information regarding the background of the model and how to cite it, please visit: https://lab.kb.nl/dataset/entangled-histories-ordinances-low-countries

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	88105	13013	0.74%	1.20%

No training image samples available.

Dutch_Romantype_Pylaia

Language: Dutch Roman Print

Description: This model is based on printed texts in the Roman font that was used in the Low Countries, during the 16th, 17th and 18th century. The type of sources used for this model, are books of ordinances, which contained the norms ('laws') at the time.

This model has been the result of one of the KB National Library of the Netherlands Researcher-in-Residence position 2019. The project was called 'Entangled Histories'.

For more information regarding the background of the model and how to cite it, please visit: https://lab.kb.nl/dataset/entangled-histories-ordinances-low-countries

Text	Tech	Words	Lines	CER train	CER validation
Print	Pylaia	88105	13013	0.30%	1.40%

No training image samples available.

Dutch XVII Century

Language: Dutch

Description: Chronicling Novelty and the Amsterdam City Archives present a first collaboration for a general model for handwritings from the Low Countries in the 17th Century.

The Amsterdam City Archives contributed with 14 17th Century Amsterdam notarial handwritings, consisting of each maximum 290 scans training and each 10 scan validation: Bruijningh, Mathijsz, Westfrisius, Schaef, De Winter, Van der Groe, Anthony van de Ven, Van Sevenhoven, Des Pommare, Borsselaer, Pondt, De Vos, Joost van de Ven and Venkel.

Chronicling Novelty (https://chroniclingnovelty.com/) contributed with 20 17th Century handwritings from chronicles from nowadays Belgium and the Netherlands, consisting of minimum 10 and maximum 290 scans training material per handwriting, if possible with a validation set of 10 scans. The following chronicles are included: Antwerpen (1643), Den Bosch (1604), Brugge (1675, 1684), Brussel (1681), Gent (1666, 1668, 1668), Hauwert (1636), Ieper (1695), Leeuwarden (1671, 1697), Mechelen (1657, 1665), De Rijp (1652), Rotterdam (1648, 1658, 1663, 1690) and Zwolle (1681).

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	1891025	331753	6.93%	6.28%

Ground Truth Examples

4051	4244	4480	4834

Dutch XVII Century

Language: Dutch

Description: Chronicling Novelty and the Amsterdam City Archives present a first collaboration for a general model for handwritings from the Low Countries in the 17th Century.

The Amsterdam City Archives contributed with 14 17th Century Amsterdam notarial handwritings, consisting of each maximum 290 scans training and each 10 scan validation: Bruijningh, Mathijsz, Westfrisius, Schaef, De Winter, Van der Groe, Anthony van de Ven, Van Sevenhoven, Des Pommare, Borsselaer, Pondt, De Vos, Joost van de Ven and Venkel.

Chronicling Novelty (https://chroniclingnovelty.com/) contributed with 20 17th Century handwritings from chronicles from nowadays Belgium and the Netherlands, consisting of minimum 10 and maximum 290 scans training material per handwriting, if possible with a validation set of 10 scans. The following chronicles are included: Antwerpen (1643), Den Bosch (1604), Brugge (1675, 1684), Brussel (1681), Gent (1666, 1668, 1668), Hauwert (1636), Ieper (1695), Leeuwarden (1671, 1697), Mechelen (1657, 1665), De Rijp (1652), Rotterdam (1648, 1658, 1663, 1690) and Zwolle (1681).

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	PyLaia	1890993	341783	11.70%	9.80%

Ground Truth Examples

4051	4244	4480	4834

Edelfelt M13+

Language: Swedish

Description: 1867, 1873, 1881-1884, 1888-1901

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	438817	67412	3.77%	3.80%

Ground Truth Examples

1113	2389	2575	2710

English Writing M1

Language: English

Description: Based on Jeremy Bentham and secretaries from early 19th century. 50.000+ words.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR	N/A	N/A	4.54%	3.66%

No training image samples available.

Estonian Court Records 19thC

Language: Estonian

Description: Model is based on Uue-Põltsamaa court records (vallakohus, gemeindegericht) 1852-1864 from Estonian National Archives.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	56190	10204	0.68%	3.55%

Ground Truth Examples

36	42	68	173

EvenkiRussian-RychkovArchive-v1.0

Language: Evenki, Russian

Description: Bilingual manuscripts by Konstantin Rychkov, 1910s. Parallel Evenki (Tungusic) and Russian texts. Cyrillic script with extensions.

This model is trained on Rychkov archive Folder 5 pp. 001-233. No Base Model. Validation set 18 pages. 60 epochs. Acute accent disregarded.

Model trained in the INEL project ("Grammatical Descriptions, Corpora, and Language Technology for Indigenous Northern Eurasian Languages") https://inel.corpora.uni-hamburg.de/

Manuscripts from the archive of Konstantin Rychkov provided by the Institute of Oriental Manuscripts RAS http://www.orientalstudies.ru/

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	40561	9968	2.76%	2.91%

No training image samples available.

Français ANOM

Language: French

Description: Documents from series F3, C11A and C11E. 17th and 18th Centuries

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	128175	16392	6.43%	8.05%

No training image samples available.

French_18thC_Print

Language: French

Description: This model is based on printed texts in French (Romantype Font) that was used in Flanders (Low Countries), during the 18th century. The type of sources used for this model, are books of ordinances, which contained the norms ('laws') at the time.

This model has been the result of one of the KB National Library of the Netherlands Researcher-in-Residence position 2019. The project was called 'Entangled Histories'. The books used for this specific model, have been provided by the Bodleian Library Oxford (RECUEIL DES ÉDITS, DÉCLARATIONS, LETTRES-PATENTES, &c. ENREGISTRÉS AU PARLEMENT DE FLANDRES).

For more information regarding the background of the model and how to cite it, please visit: www.https://lab.kb.nl/dataset/entangled-histories-ordinances-low-countries

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	38487	3883	0.09%	0.74%

No training image samples available.

French_18thC_Pylaia

Language: French

Description: This model is based on printed texts in French (Romantype Font) that was used in Flanders (Low Countries), during the 18th century. The type of sources used for this model, are books of ordinances, which contained the norms ('laws') at the time.

This model has been the result of one of the KB National Library of the Netherlands Researcher-in-Residence position 2019. The project was called 'Entangled Histories'. The books used for this specific model, have been provided by the Bodleian Library Oxford (RECUEIL DES ÉDITS, DÉCLARATIONS, LETTRES-PATENTES, &c. ENREGISTRÉS AU PARLEMENT DE FLANDRES).

For more information regarding the background of the model and how to cite it, please visit: www.https://lab.kb.nl/dataset/entangled-histories-ordinances-low-countries

Text	Tech	Words	Lines	CER train	CER validation
Print	Pylaia	38487	3883	0.10%	0.91%

No training image samples available.

German Fraktur 18th Century - WrDiarium_\M9

Language: 18th ct. Austrian German

Description: Creator: Austrian Centre for Digital Humanities and Cultural Heritage at the Austrian Academy of Sciences. The model is based on c. 3.000 pages from the „Wien[n]erisches Diarium“ / „Wiener Zeitung“ (1703-1799). The model was created within the project „Das Wien[n]erische Diarium: Digitaler Datenschatz für die geisteswissenschafttlichen Diziplinen“ (The Wien[n]erisches Diarium: A digital data treasury for the humanities) and is provided by Claudia Resch and Dario Kampkaspar. See: https://digitarium.acdh.oeaw.ac.at/

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	829447	125537	1.16%	0.79%

Ground Truth Examples

1	1203	1317	1451

German_Kurrent_17th-18th

Languages: German, Latin, French

Description: The model was trained with manuscripts of the 17th, 18th and few dates of the 19th century. Occasionally, printed material from the same period were also included. In total, the Ground Truth is based on about 1500 individual handwritings, which come from the council minutes of the University of Greifswald, the assessor votes of the Wismar High Court, the minutes of the Pomeranian government of Stralsund, the responsa of the Greifswald Law Faculty and other archival holdings, such as private letter collections.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	PyLaia	1839841	289857	6.00%	5.50%

No training image samples available.

German_Kurrent_XIX-XX_M6-1

Language: German

Description: Large train and test set for german kurrent (19. century). GT from different projects and partners.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	6343706	943019	6.75%	7.48%

No training image samples available.

German_Kurrent_XIX-XX_M6-2

Language: German

Description: Large train and test set for german kurrent (19. century). GT from different projects and partners.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	PyLaia	6343689	1027960	4.00%	4.60%

No training image samples available.

German_Kurrent_XIX_M2

Language: German

Description: Large train and test set for german kurrent (19. century). GT from different projects and partners. Biased towards Swiss Documents. Capable to recognize printed pages.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	5100439	764457	4.40%	6.53%

No training image samples available.

German_Kurrent_XIX_pylaia

Language: German

Description: Large train and test set for german kurrent (19. century). GT from different projects and partners. Biased towards Swiss Documents.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	Pylaia	5100439	764457	4.60%	6.90%

No training image samples available.

German_Kurrent_XVI-XVII_M1

Language: German

Description: Generic model for the recognition of Kurrent scripts. Different Documents and Styles.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	1579208	248255	11.82%	8.42%

No training image samples available.

Gjentofte 1881-1913 Denmark 1000 epochs

Language: Danish

Description: Gentofte 1908- 1913 s. 132-231 (215-115) ((only 1-6), Gentofte 1881-1892 s 2-15 (50050), Gentofte 1881-92 fil 94-140 (500045), Gentofte 1893-1903 s. 190-289(50058), Gentofte 1903-1908 forside-95(50094), Gentofte 1903-1908 s. 96-195 (50093), Gentofte 1903-1908 s.196-283 (50092), Gentofte 1908-1913 forside-104 50074), Gentofte 1867-1880 s 258-bagside (50034)

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	154388	28230	2.46%	4.43%

Ground Truth Examples

1	3	136	203

Glagolitic printings

Language: South Slavic Glagolitic

Description: The model is the result of a collaboration between the University Library Tübingen and the Slavic Department Freiburg (Achim Rabus). Selected pages from different printings of the Tübingen-Urach tradition have been taken as GT.

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	28212	5093	3.00%	4.24%

No training image samples available.

Gothenburg_police_reports_1868-1902

Language: Swedish

Description: This model is trained from reports from the Gothenburg Police Detective department 1868-1902, held at the Swedish National Archives in Gothenburg. The groundtruth for the model training consists of transcribed spreads from 1873, 1880, 1888, and 1896.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	165060	27931	2.85%	2.70%

Ground Truth Examples

318	364	385	422

Gothic_Book_Scripts_XIII-XV_M4

Language: Latin, German

Description: Retraining of different gothic book hands (handwriting). XIII-XVct.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	875728	141633	4.61%	4.92%

No training image samples available.

HIMANIS Chancery M1+

Language: Latin, French

Description: As part of the HIMANIS project (lead by D. Stutzmann, C. Kermorvant & E. Vidal), the text edition provided by P. Guérin and encoded in TEI by the Ecole nationale des Chartes (http://corpus.enc.sorbonne.fr/actesroyauxdupoitou/) and the one by J. Viard were aligned at line level and used to train this comprehensive model for French and Latin Chancery documents. More information on the project can be found at: http://himanis.huma-num.fr/himanis/

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	665988	39410	6.87%	5.33%

Ground Truth Examples

1	1010	1225	1358

Italian Administrative Hands, 1550-1700

Language: Italian

Description: The Italian Administrative Hands model features a variety of Italian-language documents from state archives in Milan, Venice, Florence, Pisa, and Genoa. The training set represents a spectrum of humanistic, italic and cursive hands characteristic of administrative records, employed by secretaries and newswriters. The model has been trained to perform well with a mix of quantitative and qualitative information as well as many common proper nouns for the period, such as locations in Europe and contemporary rulers. Administrative documents often employ common superscript abbreviations, which the accompanying documentation treats in greater detail. The model can also be used with Latin, Spanish and French documents to some extent. The model represents a collaboration between Jake Dyble (Exeter/Pisa), Antonio Iodice (Exeter/Genoa), Sara Mansutti (Cork), and Rachel Midura (Virginia Tech). Documentation at https://emdigit.org/tool/2020/07/21/italian-administrative-hands.html.

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	67361	7193	9.15%	10.80%

No training image samples available.

Manuscripts of Ethiopia and Eritrea

Language: Classical Ethiopic (Geʽez script)

Description: Model for the transcription of Manuscripts of Ethiopia and Eritrea in Classical Ethiopic (Gǝʿǝz). Trained as part of the Beta maṣāḥǝft project and in order to feed a workflow to import transcriptions into the project's database. Transcriptions for the training have been kindly provided by

• Alessandro Bausi for ESum039, ff. 16vb-29va;
• Antonella Brita for DAS002, 101va-110ra;
• Dorothea Reule for ESqdq004, ff. 97ra-101vb, 104ra-109rb.
• Nafisa Valieva for BLorient718, ff. 1ra-7vb, images British Library.
• Several parts of manuscripts transcribed by Jeremy Brown and pertaining to the Miracle of the Cannibal of Qemer.

Importing of images and transcriptions in Transkribus has been done by Pietro Liuzzo

The project Beta maṣāḥǝft: Manuscripts of Ethiopia and Eritrea (Schriftkultur des christlichen Äthiopiens und Eritreas: eine multimediale Forschungsumgebung) is a long-term project funded within the framework of the Academies' Programme (coordinated by the Union of the German Academies of Sciences and Humanities) under survey of the Akademie der Wissenschaften in Hamburg. The funding will be provided for 25 years, from 2016–2040. The project is hosted by the Hiob Ludolf Centre for Ethiopian Studies at the Universität Hamburg. It aims at creating a virtual research environment that shall manage complex data related to the predominantly Christian manuscript tradition of the Ethiopian and Eritrean Highlands.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	HTR+	53830	16028	5.57%	5.16%

No training image samples available.

Medieval Protocolbook 's-Hertogenbosch by Townclerck Petrus de Os sr., 1497-1542

Language: Latin, Dutch

Description: Protocolbook of the city of 's-Hertogenbosch, written by magister Petrus de Os sr., townclerck from 1497 until 1542. This protocolbook contains the minutes of the charters on voluntary jurisdiction

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	46638	3876	1.00%	4.11%

Ground Truth Examples

4	32	53	87

NeoLatin_Ravenstein_1643-1772

Language: Latin

Description: This model is based on the transcription of the "Litterae Annuae Parochiae Ravensteijn SJ ab Anno 1643 ad Annum 1772". The annual letters were kept at the Archivum Neerlandicum Societatis Iesu (Berchmannianum, Nijmegen) . These are now at the Catholic Document Center (Katholiek Documentatie Centrum (KADOC)) in Leuven, Belgium (inventarisnummer 15.606).

Tom Gribnau photographed the manuscript; the transcriptions were made by Pim Boer, Leo Nellissen. This belongs to the publication: Tom Gribnau, Pim Boer, Leo Nellissen, Paul Begheyn SJ & Charles Caspers, Martiaal en theatraal. De jezuïeten in Ravenstein (1643-1772). Inleiding en vertaling van de jaarbrieven Nijmgen: Uitgeverij Valkhof Pers 2019; ISBN 978 90 5625 514 5) More information can be found at: https://jaarbrieven.blogspot.com/ and https://www.stilus.nl/litterae/

T2I from the transcripts to Transkribus model has been done by Dr. C.A. Romein

Text	Tech	Words	Lines	CER train	CER validation
Handwritten	HTR+	64435	6864	4.51%	3.58%

Ground Truth Examples

22	145	66	114

Noscemus GM 2.2

Language: Latin, Greek, German et al.

Description: The "NOSCEMUS General Model" is able to read printed Latin text, especially from the 15th, 16th, 17th and 18th century. The model was released by Stefan Zathammer and it is based on training data coming from the Digital Sourcebook of the NOSCEMUS project (https://www.uibk.ac.at/projects/noscemus/)

For the 2nd revised and updated version a substantial amount of new pages was added, including prints from the 15th century, texts set in Fraktur and texts with a considerable amount of Greek passages.

Although the model is tailored towards transcribing (Neo-)Latin texts set in Antiqua-based typefaces, it is also, to a certain degree, able to handle Greek words and words set in (German) Fraktur.

The NOSCEMUS project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 741374).

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	345878	52459	0.39%	0.79%

No training image samples available.

Noscemus GM 3.0

Languages: Latin, Greek (German, Italian, English)

Description: The "NOSCEMUS General Model" is able to read printed Latin text, especially from the 15th, 16th, 17th and 18th century. The model was released by Stefan Zathammer and it is based on training data coming from the Digital Sourcebook of the NOSCEMUS project (https://www.uibk.ac.at/projects/noscemus/)

For the 3rd revised and updated version a substantial amount of new pages was added, including prints from the 15th century and especially Greek texts.

Although the model is tailored towards transcribing (Neo-)Latin texts set in Antiqua-based typefaces, it is also, to a certain degree, able to handle Greek words and words set in (German) Fraktur.

The NOSCEMUS project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 741374).

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	448464	66575	0.40%	0.98%

No training image samples available.

Pylaia_NeoLatin_Ravenstein

Language: NeoLatin

Description: This model is based on the transcription of the "Litterae Annuae Parochiae Ravensteijn SJ ab Anno 1643 ad Annum 1772". The annual letters were kept at the Archivum Neerlandicum Societatis Iesu (Berchmannianum, Nijmegen) . These are now at the Catholic Document Center (Katholiek Documentatie Centrum (KADOC)) in Leuven, Belgium (inventarisnummer 15.606).

Tom Gribnau photographed the manuscript; the transcriptions were made by Pim Boer, Leo Nellissen. This belongs to the publication: Tom Gribnau, Pim Boer, Leo Nellissen, Paul Begheyn SJ & Charles Caspers, Martiaal en theatraal. De jezuïeten in Ravenstein (1643-1772). Inleiding en vertaling van de jaarbrieven Nijmgen: Uitgeverij Valkhof Pers 2019; ISBN 978 90 5625 514 5) More information can be found at: https://jaarbrieven.blogspot.com/ and https://www.stilus.nl/litterae/

T2I from the transcripts to Transkribus model has been done by Dr. C.A. Romein.

Text	Tech	Words	Lines	CER train	CER validation
Handwriting	Pylaia	64435	6864	1.30%	4.00%

Ground Truth Examples

22	145	66	114

Transkribus print 0.1

Language: Danish, Dutch, German, Finnish, French, Latin, Swedish

Description: A first attempt to create a large model for printed historical documents. The model should be able to read historical Danish, Dutch, German, Finnish, French, Latin, and Swedish with good quality. Note: If you have printed material in other languages use this model as basemodel for training to speed up the creation of your special models.

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	3661935	592933	2.18%	1.50%

No training image samples available.

Transkribus Typewriter 0.1

Language: English, German, Dutch, Finnish

Description: General model for typewritten documents

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	654961	73144	1.80%	1.28%

No training image samples available.

Typewritten/print_early_1900

Language: Dutch

Description: The National Archives Netherlands have made this model for typewritten and print writings which were used in the Netherlands during the late 19th and early 20th century.

For more information regarding the background of the model read this article (Dutch): https://www.nationaalarchief.nl/beleven/nieuws/klussen-aan-nieuw-tekstherkenningsmodel-transkribus-platform

Text	Tech	Words	Lines	CER train	CER validation
Print	HTR+	72170	13700	1.65%	3.24%

Ground Truth Examples

59	94	108	191