MLW Zettelmaterial
Creators:
Bayerische Akademie der Wissenschaften (BAdW)
Publication Date:
2023
Data Category:
Dataset Description:
General information:
The data set comprises a total of 114,653 images (18,9 GB), corresponding to 3,507 distinct lemmas.
All images are in RGB, but not uniform in size, i.e. height, and width differ from image to image.
Additionally, the information on the corresponding lemma is available for each image in a separate json file.
Structure:
Most record cards follow the same structure being composed of three main parts.
- The first one (1), and the one deemed most challenging, is the lemma, which is always located in the upper left corner of the record card.
- The second part (2) is the index of the text where the lemma is found.
- The third part (3) contains a text extract in which the word (corresponding to the lemma) occurs in context.
- A total of 2,420 lemmas (69%) were found to appear on ten record cards or less
- 854 lemmas (24.4%) are present on between 10 and 100 record cards
- 233 lemmas (6.6%)can be found on more than 100 record cards
- 1,123 lemmas (approximately 36.7%) had only one record card
- Lemma lengths range from one character up to a maximum of 19 characters.
- The average length of the lemmas lies between five and six characters.
- The data set is available on huggingface: https://huggingface.co/datasets/misoda/MLW_data
- images.zip contains the images, while the labels are stored in labels.json
- Koch, P., Nuñez, G. V., Arias, E. G., Heumann, C., Schöffel, M., Häberlin, A., & Aßenmacher, M. (2023). A tailored Handwritten-Text-Recognition System for Medieval Latin. arXiv preprint arXiv:2308.09368.
Variables:
Details: