pointer-net-for-nested

The official implementation of ACL2022: Bottom-Up Constituency Parsing and Nested Named Entity Recognition with Pointer Networks

Setup

Environment

conda create -n parsing python=3.7
conda activate parsing
while read requirement; do pip install $requirement; done < requirement.txt 

Download preprocessed PTB, CTB7, GENIA from: link

For ACE04 and ACE05, send me e-mails.

Run

python train.py +exp=ft_10 datamodule=a model=pointer 
a={ptb, ctb7}

python train.py +exp=ft_10 datamodule=genia model=pointer model.use_prev_label=True 

python train.py +exp=ft_50 datamodule=b model=pointer model.use_prev_label=True
b={ace04, ace05}

multirun example:

python train.py +exp=base model=pointer datamodule=ptb,ctb7 seed=0,1,2 --mutlirun

evaluation:

python evaluate.py +load_from_checkpoint=your/checkpoint/dir

Contact

Please let me know if there are any bugs. Also, feel free to contact [email protected] if you have any questions.

Citation

@misc{yang2021bottomup,
      title={Bottom-Up Constituency Parsing and Nested Named Entity Recognition with Pointer Networks}, 
      author={Songlin Yang and Kewei Tu},
      year={2021},
      eprint={2110.05419},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Credits

The code is based on lightning+hydra template. I use FastNLP for loading data. I use lots of built-in modules (LSTMs, Biaffines, Triaffines, Dropout Layers, etc) from Supar.