Podrobný výpis o publikaci

SLO: On May 17, 2024, we launched the creation of a new aggregated Slovak supermodel in the Transkribus web application. The basis for the creation of a supermodel for certain Slovak printed historical documents and typewritten documents was the partial models of task solvers in the Skriptor project (Matej Bela University in Banská Bystrica and the State Science Library in Banská Bystrica), as well as transcriptions prepared by students of the University of Silesia in Opava in within the Student Grant Competition and educational activities. As part of the Skriptor project, Michaela Mikušková and Lucia Nižníková completely processed the demanding segmentation and manual transcription of 92 s. GT of historical printed book J.A. Comenius' Orbis Pictus (1798 edition). Dušan Katuščák, as part of the Skriptor project, educational activities and the SGS student grant competition at the University of Silesia, and the management of the diploma thesis in Opava, processed a whole range of historical newspapers, magazines and books from the 19th and early 20th centuries to GT quality. Klára Kováčová-Pohlová (Diplomová práce, 2024, FPF SU Opava) and Matej Šmida (UMB Banská Bystrica) processed typewritten documents, using samples of various fonts in Slovak, Czech, German languages (ca. 150 pp.) Nikola Halfarová, Terezie Gajdošová, Lenka Málková, Nikol Taufrová, Nela Kociánová (4th year, FPF SLU) in the subject of prof. Dušan Katuščák Digitization II. they prepared ca. 80 s. of GT transcriptions from various historical prints from the 18th and 19th centuries written in Czech (Svabian). The model is marked ID78289 SLOVAK Supermodel print&typewriter (SSPT1) we used 542 pages in Ground Truth quality (GT 37897 lines and 200697 words). 59 pages for validation of the new model (Validation set). In the model ID78289 SLOVAK Supermodel print&typewriter (SSPT1) the values Train set: 1.00% and Validation set: 1.00% were achieved. So it means the "accuracy" of the transcription is 99%. The creation of the SSM1 model on the Transkribus server took 21 hours and 52 minutes. The creation process was set to 100 cycles and ended after 100 cycles (epochs). The SSPT1 model is the first attempt in Slovakia and the Czech Republic to create an aggregated tool through which it would be possible to automatically make available certain types of printed and typewritten documents that are similar to the fonts used in its creation. ID78289 SLOVAK We do not consider the Supermodel print&typewriter (SSPT1) to be a definitive universal model for the transcription of historical prints and typewriters of Slovak provenance of all types and periods. The variety of fonts and styles is diverse, and the creation of an optimal aggregate model is a task for other researchers and enthusiasts in the years to come. Copyright: CC BY-NC-SA. Of course, such automatic transcription will not immediately bring satisfactory results. However, it can be "rough" to gradually automatically transcribe additional pages, manually correct them to GT status, and then use larger GT datasets to refine a new model based on our SSPT1. After the creation of hundreds and thousands of pages of GT, it will be possible to proceed with the creation of further generations of new models based on SSPT1. Development could continue for printing and typewriting with models of new generations SSPT2, SSPT3 etc. The challenge for researchers is also the development and creation of a new aged supermodel, which would include both manuscripts and prints and typescripts. This Slovak supermodel could be shared within the Transkribus community of experts and included in some big Transkribus Community supermodel, etc.