Prompt-based data augmentation for semantically-precise event relation classification

The process of recognizing and classifying the relationships between events mentioned in the text is a crucial task in natural language processing (NLP) known as event relation extraction. If temporal relations and causality are largely studied in the literature, other types of relations have found less interest. Our study specifically concentrates on four types of event relations: causality, enabling, prevention, and intention. Our main contribution consists of the use of a state-of-the-art language model (GPT-3) to extend an existing small dataset with synthetic examples to address the challenge of insufficient training data. We evaluate the quality of these generated samples by training an event relations extraction system, showing improved performances in classifying event relations.