Event co-occurrences for prompt-based generative event argument extraction.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Prompt learning enhances Pre-trained Language Model (PLM) training for Event Argument Extraction (EAE).
• Existing prompt-based methods overlook complex event interdependencies within sentences.
• A need exists for models that can effectively handle multiple co-occurring events and their relationships.

Purpose of the Study:

• To propose a novel approach, Event Co-occurrences Prefix Event Argument Extraction (ECPEAE), for EAE.
• To address the limitations of previous methods in capturing complex event structures and relationships.
• To improve the accuracy and robustness of EAE models, especially in complex linguistic contexts.

Main Methods:

• ECPEAE incorporates event co-occurrence information using a prefix module.
• A present event bias module emphasizes relevant template information during attention calculation.
• An adjustable copy mechanism mitigates noise from additional information in attention layers.

Main Results:

• ECPEAE achieves state-of-the-art performance on the ACE2005-EN and ERE-EN datasets.
• The model demonstrates effectiveness in handling complex event relationships.
• ECPEAE shows adaptability to low-resource environments with varying training data sizes.

Conclusions:

• ECPEAE effectively models complex event co-occurrences and relationships for improved EAE.
• The proposed modules enhance template information integration and output constraint.
• ECPEAE offers a robust and adaptable solution for Event Argument Extraction, even in data-scarce scenarios.