NER for Austrian court decisions

Code to repeat the results of Annotating Entities with Fine-Grained Types in Austrian Court Decisions.

Variables and config

All the variables are stored in config.toml and are read during the execution of the code. Please, check this file and modify the entries. The datasets with original general purpose NER annotations (nif_annotations_folder), with manual annotations (manual_annotations_folder), and with manual verified samples (manual_verified_folder) are available at https://doi.org/10.5281/zenodo.4625767. Please, download them and then paste the correct pathes to those dataset into the config.toml.

How to run

1. Create a virtual environment with Python3.7

pip3.7 install virtualenv
python3.7 -m virtualenv MyEnv
source MyEnv/bin/activate

2. Install dependencies

pip3.7 install -r requirements.txt

3. Pre-train the model on wic-tsv-de

python3.7 HyperBertCLS.py

4. Repeat results of experiments

python3.7 evalute_on_manually_verified.py 

5. Produce final results

python3.7 cybly_type_verification.py