EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Unified Event Argument Extraction Framework with Layered Attention and Retrieval-Aware Prompting

Bruce Dorian, Jiwoo Park

Journal of the Artificial Intelligence Engineers Consortium · 2026

DOI: 10.66238/jaiec54 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of the Artificial Intelligence Engineers Consortium, 2026. Verfasst von Bruce Dorian und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Event argument extraction is a critical subtask of information extraction that aims to identify and classify entities that participate in specific events mentioned in unstructured text. Despite recent advancements in natural language processing, extracting arguments accurately remains challenging due to the complex syntactic structures of long sentences, the implicit relationships between event triggers and arguments, and the scarcity of high-quality annotated data. To address these persistent challenges, this paper proposes a novel, unified event argument extraction framework that synergistically integrates a layered attention mechanism with a retrieval-aware prompting strategy. The layered attention module is designed to capture hierarchical structural semantics from word-level to sentence-level representations, enabling the model to dynamically focus on contextually relevant spans while ignoring noisy syntactic artifacts. Concurrently, the retrieval-aware prompting module leverages a dense retrieval system to fetch structurally similar historical event templates from an external datastore. These retrieved instances are then seamlessly integrated into the prompt formulation process, providing explicit grounding and guiding the generative language model to produce accurate argument roles. Extensive evaluations on standard benchmark datasets demonstrate the superiority of the proposed framework over existing state-of-the-art baselines. Through comprehensive ablation studies and error analyses, we reveal that the combination of hierarchical context modeling and retrieval-augmented guidance significantly mitigates issues related to long-range dependencies and out-of-domain generalization.

Zitieren

APA

Dorian, B., & Park, J. (2026). A Unified Event Argument Extraction Framework with Layered Attention and Retrieval-Aware Prompting. Journal of the Artificial Intelligence Engineers Consortium. https://doi.org/10.66238/jaiec54

Vancouver

Dorian B, Park J. A Unified Event Argument Extraction Framework with Layered Attention and Retrieval-Aware Prompting. Journal of the Artificial Intelligence Engineers Consortium. 2026. doi:10.66238/jaiec54

BibTeX
@article{Dorian2026,
  author = {Dorian, Bruce and Park, Jiwoo},
  title = {A Unified Event Argument Extraction Framework with Layered Attention and Retrieval-Aware Prompting},
  journal = {Journal of the Artificial Intelligence Engineers Consortium},
  year = {2026},
  doi = {10.66238/jaiec54}
}
Kostenlos starten →