This repository contains the code for our ACL 2023 paper Event Extraction as Question Generation and Answering .

We adapted the preprocessing scripts from the Dygiepp repo. The main difference is that we retrieve the character offsets of the annotations as well as sentences. Please refer to ./data_process/README.md for details.

The code is based on Python 3.8+, and the scores reported are based on experiments on a single AWS p3.2xlarge instance.

To install the required dependencies:

pip install -r requirements.txt

bash ./train_event_trigger_model.sh

The trained model will be saved in ./model_checkpoint/trigger_model by default.

bash ./train_qg_bart.sh for the BART backbone.

bash ./train_qg_t5.sh for the T5 backbone.

The trained model will be saved in ./model_checkpoint/qg_model_bart or ./model_checkpoint/qg_model_t5 for BART and T5 backbone respectively by default.

bash ./train_argument_extraction_bart.sh for the BART backbone.

bash ./train_argument_extraction_t5.sh for the T5 backbone.

The trained model will be saved in ./model_checkpoint/eae_model_bart or ./model_checkpoint/eae_model_t5 for BART and T5 backbone respectively by default.

bash evaluate_e2e_predicted_triggers_bart.sh for the BART backbone.

bash evaluate_e2e_predicted_triggers_t5.sh for the T5 backbone.

If you find the code in this repo helpful, please cite our paper:

@inproceedings{lu-etal-2023-event,
    title = "Event Extraction as Question Generation and Answering",
    author = "Lu, Di  and
    Ran, Shihao  and
    Tetreault, Joel  and
    Jaimes, Alejandro",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.acl-short.143",
}