Towards Resource-Constrained Event Extraction: A Knowledge-Augmented Framework for Overcoming Challenges in Vietnamese NLP

Event Extraction (EE) is a crucial task in Natural Language Processing (NLP), instrumental in capturing meaningful activities and contributing to the tracking of narratives and developments within textual documents. Extensive research has been dedicated to improving the accuracy of event trigger identification and argument role classification, spanning from traditional machine learning to modern deep learning architectures. Recently, driven by the rapid advancements of Large Language Models (LLMs), these models have been applied to EE, primarily through data augmentation or fine-tuning approaches. However, the computational and resource overhead associated with LLMs remains a significant challenge. Furthermore, existing state-of-the-art methods predominantly focus on high-resource languages such as English and Chinese, leaving low-resource languages, like Vietnamese, largely under-explored due to their unique linguistic ambiguities. Consequently, our research direction focuses on leveraging a Small Language Model-based (SLM-based) approach, augmented with external knowledge, to address the EE task in Vietnamese. The ultimate objective is to develop a compact model capable of effectively addressing core EE challenges, such as rare events, semantic ambiguity, and long-range dependencies—thereby establishing an efficient and robust framework specifically tailored for the Vietnamese low-resource language domain.