Code for Document-Level Event Argument Linking as Machine Reading Comprehension

Dependencies

  • pytorch=1.6
  • transformers=3.1.0
  • spacy=3.0 # conflicts with transformers

Datasets

Training Instructions

  • Run python preprocess.py for preprocessing, which will generate a data.pk file (We have released the back translation results in the file back_translation_result.txt to ease the preprocessing steps).
  • Run train.py for model training, which will save the model at the models directory.
  • Run evaluate.py for model evaluation.