Cross-lingual biaffine graph-based dependency parser

This repository provides the implementation of cross-lingual biaffine graph-based dependency parser which utilizes multilingual cased BERT.

How to run the program:

1. Set up a new conda environment with Python 3.
2. Download torchtext 0.9.0, torch, nltk, transformers, and matplotlib using pip command.
3. Download ud-treebanks-v2.9 from Universal Dependencies 2.9 and place it inside external_resources folder.
4. Create figures, logs, and trained_models folders to store the results of the experiments.
5. Set the settings and parameters of the candidate model in .json file and store it inside the configs folder. You can find existing examples inside the folder.
6. Run the training using python main.py <name of the config file>, for example: python main.py czech_serbian_few if you store the config inside configs/czech_serbian_few.json.