Code README

Code and datasets for the Coling 2025 paper "Fusion meets Function: The Adaptive Selection-Generation Approach in Event Argument Extraction". This codebase is based on PAIE (Prompting Argument Interaction for Event Argument Extraction), with modifications to allow for the combination of selective and generative methods.

Usage

By default, running the engine.py file will use the BART-base model with the WikiEvent dataset. However, you can easily switch between different configurations as follows:

Switching Models

To switch between BART-base and BART-large models, open the config_parser.py file and locate the model_name_or_path parameter.

parser.add_argument("--model_name_or_path", default="./bart", type=str,
                        help="pre-trained language model")

Switching Datasets

To switch between the RAMS and WikiEvent datasets, open the config_parser.py file and locate the dataset_type parameter. Set it to "rams" or "wikievent" as desired.

parser.add_argument("--dataset_type", default="wikievent", choices=["rams", "wikievent"], type=str,
                        help="dataset type document-level(rams/wikievent)")
parser.add_argument("--role_path", default='./data/dset_meta/description_wikievent.csv', type=str, 
                    help="a file containing all role names. Read it to access all argument roles of this dataset")
parser.add_argument("--prompt_path", default='./data/prompts/prompts_wikievent_full.csv', type=str, 
                    help="a file containing all prompts we use for this dataset")

Adjusting Selection and Generation Ratio

The model's behavior can be adjusted by modifying the self.loss_ratio parameter in the models/paie.py file. This parameter controls the ratio between selective and generative losses. Adjust it as needed to achieve the desired balance between selection and generation.

class PAIEModel(nn.Module):
    def __init__(self):
        # Other model initialization code

        # Adjust the loss ratio to control selection vs. generation balance
        self.loss_ratio = 0.5  # Modify this value as needed

Running the Code

Once you have configured the desired model, dataset, and loss ratio, you can run the code by executing engine.py. The code will use the specified settings to perform event arguments extraction.