The following packages should be installed:

• PyTorch: >= 1.12.1
• Transformers: >= 4.2

Please modify data/model-conf/data.ini according to paths of your own dataset.

You can download the processed PTB-train, where incompatible instances are removed. The meaning of compatibility can be viewd in our paper.

Try the following commands to train second-order Joint2o model:

python src/parse.py -c data/model-conf/bert-ptb/joint2o-bi.ini -d 0

To make predictions:

python src/pred.py -d 0 -c data/model-conf/bert-ptb/joint2o-bi.ini \
    --path exp/bert-e25-ptb/joint2o-bi-s1/model \
    --data_con data/ptb/test.pid \
    --data_dep data/ptb/test.conllx \
    --pred exp/joint2o-bi-s1/test.pred \
    --decode satta

If you have any questions, feel free to contact me via emails.

If you are interested in our work, please cite

@inproceedings{gu-etal-2024-high-order,
  title     = {High-order Joint Constituency and Dependency Parsing},
  author    = {Yanggan, Gu  and
               Yang, Hou  and
               Zhefeng, Wang  and
               Xinyu, Duan  and
               Zhenghua, Li},
  booktitle = {Proceedings of LREC-COLING},
  year      = {2024},
  address   = {Torino, Italia},
  publisher = {International Committee on Computational Linguistics}
}