This repo provides the scripts to reproduce models and scores for Event Corefernce Resolution for Contentious Politics Events paper.
In order to run the scripts, first create an environment and activate;
conda create --name ecr4 python=3.8 && conda activate ecr4
Before moving forward to train or test models, first you need the download the data and code repos.
The dataset is shared in coreference_data repo of our project, emerging-welfare.
git clone https://github.com/emerging-welfare/coreference_data.git
git clone https://github.com/emerging-welfare/ECR4-Contentious-Politics.git
Then go to ECR4 repo and install the required packages;
pip install -r requirements.txt
To train a model from scratch, go to scripts folder and run train.py. For example you can run the following command in order to train a model for english, spanish and portuguese mix dataset;
python train.py --train_dataset_paths en_es_pr --dev_dataset_paths en_es_pr --test_dataset_paths en_es_pr
A model checkpoint will be saved to models. Then you can test the model in any language with following line of code;
python test.py --model_checkpoint ../models/<checkpoint_name>.ckpt --test_dataset_paths en_pr
You will be able to see results default and fine-tuned threshold for validation and test set of each language, respectively.