Zero-Shot Dependency Parsing with Worst-Case Aware Automated Curriculum Learning

This codebase contains code to reproduce experiments from our ACL 2022 paper. (camera-ready version coming soon). It integrates the method developed by Zhang et al 2020 (which was implemented within the MT-DNN framework) into the MaChAmp library. Detailed instructions for how to run models can be found in the machamp repository. The configurations of our different models are in config/. The language configurations can be created with scripts/create_params.py, given a UD treebanks location. The main training script is in scripts/train_acl22.sh

@inproceedings{delhoneux2022worst,
  title = "Zero-Shot Dependency Parsing with Worst-Case Aware Automated Curriculum Learning",
  author = "de Lhoneux, Miryam and
  Zhang, Sheng and
  S{\o}gaard, Anders",
  year = "2022",
  url = "https://openreview.net/pdf?id=h0lckggpp4X",
  booktitle = "Proceedings of ACL",
}