Detailed Information on Publication Record

CZECH supermodel_SGS. Practice set: 947 pages; Validation set: 93 pages. Results: Practice set: 2.98% Validation set: 5.83% Researchers in the Student Grant Competitions (SGS) project mastered the work on the Transkribus platform within a few months. They mastered the methods of preparation, recording, segmentation and performed a number of partial experiments in creating their own transcription models. They gained knowledge and skills and valuable know-how in manuscript transcription. Based on five manuscript documents, Lukáš Němec created Finale 2.0 (ID: 213733) with an error rate of only CER 6.56%. Another great model, Agreg-8 (ID: 207993), was created by Vojtěch Říha with an error rate of only 2.86%. We also included in the study the description and results of the preparation of the model ID 58379 with an error rate of only 4.10% for the transcription of typewritten documents of Klára Pohlová (POHLOVÁ, 2024) from the SGS project in 2023. Her partial model ID 58379 was attached to the supermodel ID78289 (SUPERMODELP SUPERMODELP 2024). Selected collections are available within the Manuscriptorium digital library (Vojtěch Říha): 1. Czech prayer book (CZECH, 1733-1766) 2. Cesta Svatocellenská (ROAD, 1733-1766) 3. Joyful journey (JOYFUL, 1829-1884) 4. Prayers, songs and litanies (PRAYERS, 1826) 5. Prayer book (PRAYER, 1700-1750) Manuscript documents for the creation of the model were selected from other possible ones (Lukáš Nemec): 1. Gallaš, Josef Heřman Agapit [Manuscript]: Mythical stories about the gods and goddesses of the Moravian Slavs. (GALLAŠ, 1820) 2. Gallaš, Josef Heřman Agapit: [Manuscript]. [Physical] monuments of the city of Hranice and its surroundings. (GALLAŠ, 1808-1811) 3. Gallaš, Josef Heřman Agapit [Manuscript] Walaši v kraji Přerovském (GALLAŠ, 1801-1804) 4. Polášek, František [Manuscript]: Pravé poznanie Boha aneb troje hodiny o perfectičkych bohích [Manuscript] (POLÁŠEK, 1800-1900) 5. Jaroš, Otakar [Manuscript]: Nauka o terenu [School notebook, squared/lined paper]. (JAROŠ) In the SGS 2024 project, we finally created an aggregated CZECH supermodel_SGS (ID 220865) based on the above-mentioned partial models, prepared by students Lukáš Němec and Vojtěch Říha, with errors of only 5.86%. With our model, it is possible to transcribe similar manuscripts with an accuracy of 94.17%. The basis of the CZECH supermodel_SGS supermodel is: 1. partial model, the author of which is student Lukáš Němec. This is a Finale 2.0 model (ID: 213733). It was built on the basis of smaller own five working models: -Physical Absolut (IČO 213213); Wallachian absolute (IČO 211773); Mystic Absolute (ID 210053); Francis II. (ID 204714); Ota 14 (IČO 182965); 2nd partial model, the author of which is student Vojtěch Říha. This is the Agreg-8 model (ID: 207993); 3. 15 manuscript pages of GT from the document Protocoly Matice slezské (ID:1663382).