Beyond Single-Event Extraction: Towards Efficient Document-Level Multi-Event Argument Extraction

Recent mainstream event argument extraction methods process each event inisolation, resulting in inefficient inference and ignoring the correlationsamong multiple events. To address these limitations, here we propose amultiple-event argument extraction model DEEIA (Dependency-guided Encoding andEvent-specific Information Aggregation), capable of extracting arguments fromall events within a document simultaneouslyThe proposed DEEIA model employs amulti-event prompt mechanism, comprising DE and EIA modules. The DE module isdesigned to improve the correlation between prompts and their correspondingevent contexts, whereas the EIA module provides event-specific information toimprove contextual understanding. Extensive experiments show that our methodachieves new state-of-the-art performance on four public datasets (RAMS,WikiEvents, MLEE, and ACE05), while significantly saving the inference timecompared to the baselines. Further analyses demonstrate the effectiveness ofthe proposed modules.