|
--- |
|
license: mit |
|
tags: |
|
- flair |
|
- token-classification |
|
- sequence-tagger-model |
|
language: de |
|
widget: |
|
- text: "ab dryn matten, gelegen ze Niderlentz, hinden in langen matten eychen, waren ze etlichen zitten Jennis Huͤbers von Niderlentz, und hat sie gekoͧft von Walther Renold" |
|
--- |
|
|
|
# Königsfelden NER |
|
Tagger for Medieval Scripts. |
|
|
|
Part of the developments at the [Digital Humanities@University of Bern](https://www.dh.unibe.ch/). |
|
Developed by Ismail Prada Ziegler based on different ground truth (see below). |
|
|
|
A model for historical German developed by Ismail Prada Ziegler as part of a research project at the University of Bern, Digital Humanities. |
|
|
|
|
|
## Performance |
|
|
|
| | PER | ORG | LOC | Micro-Avg | |
|
| :---: | :---: | :---: | :---: | :---: | |
|
| Precision | 88.76% | 69.01% | 81.96% | 84.19% | |
|
| Recall | 88.99% | 62.97% | 85.10% | 84.02% | |
|
| F1-Score | 88.88% | 65.85% | 83.50% | 84.10% | |
|
|
|
|
|
## Data Set |
|
|
|
[Königsfelden Charters](https://zenodo.org/record/5179361), 14th-17th Century, Early New High German, 623k tokens training, dev, test data. |
|
|
|
|
|
## Notice |
|
|
|
This model was a prototype with no extensive experimenting, better results can be likely be achieved on this dataset. |
|
|
|
The documents in the dataset are published [here](https://koenigsfelden.sources-online.org). |
|
|
|
|
