Summary

This is the code repository for the ACL 2020 paper "Extracting Headless MWEs from Dependency Parse Trees: Parsing, Tagging, and Joint Modeling Approaches".

Dependencies

• fire
• Cython
• numpy
• pytorch
• transformers

Data format

The data is formatted similarly as UD, except that the MISC column is now overided with B/I/O tags corresonding to the MWE spans. (A future version should make this format less confusing. )

How to train a new model

./scripts/exec.sh $SEED $LAN $MODE $BERT $PWEIGHT where $SEED denotes a random parameter initialization, $LAN is the treebank code, $MODE is the training mode choosing among parsing tagging jointparsing jointtagging jointdecoding, $BERT is a boolean value deciding whether to use pretrained bert models or not, $PWEIGHT is the coefficient of parsing module weight in joint training.

Refer to the same script for other related hyperparameters used in our experiments.

Referenced code

• https://github.com/jiesutd/NCRFpp
• https://github.com/nikitakit/self-attentive-parser
• https://github.com/allenai/allennlp

Reference

@inproceedings{shi-lee-2020-extracting,
    title = "Extracting Headless {MWE}s from Dependency Parse Trees: Parsing, Tagging, and Joint Modeling Approaches",
    author = "Shi, Tianze  and
      Lee, Lillian",
    booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.acl-main.775",
    pages = "8780--8794",
}