Create environment:
conda create -n multitask_parser python=3.7
conda activate multitask_parser
Install software as of 11/05/2021:
pip install -r requirements.txt
Or install AllenNLP:
pip install allennlp
pip install allennlp-models
./scripts/download_iwpt_data.sh
For running a single-task model:
./scripts/train_eud_parser.sh fr_sequoia enhanced transformer bert-base-multilingual-cased enhanced_kg_parser transformer no_delex
For using the multitask components:
train_eud_multitask_parser.sh fr_sequoia enhanced transformer bert-base-multilingual-cased enhanced_kg_parser enhanced_dependencies
Launch a batch job:
sbatch --gres=gpu:rtx6000:1 ./scripts/train_eud_parser.sh en_ewt:bg_btb:pl_pdb enhanced transformer bert-base-multilingual-cased enhanced_kg_parser transformer
To pre-process the blind test data, we use Trankit.
conda create -n trankit python=3.7
conda activate trankit
pip install trankit==1.0.1
To train a trankit model, see scripts/trankit.py
If you use this code for your research, please cite our paper as:
@misc{barry2021dcuepfl,
title={The DCU-EPFL Enhanced Dependency Parser at the IWPT 2021 Shared Task},
author={James Barry and Alireza Mohammadshahi and Joachim Wagner and Jennifer Foster and James Henderson},
year={2021},
eprint={2107.01982},
archivePrefix={arXiv},
primaryClass={cs.CL}
}