ArgumentPrompt: Activating Multi-category of Information for Event Argument Extraction with Automatically Generated Prompts

Shenpo Dong, Wei Yu, Hongkui Tu, Xiaodong Wang, Yunyan Zhou, Haili Li, Jie Zhou, Tao Chang

Abstract: Event Argument Extraction (EAE) task is a critical and challenging subtask for event extraction. The current mainstream approaches are based on manually designed relevant questions for extracting specific argument roles from the text. However, manually crafting templates take time and fail to cover all possible argument roles. To design better and faster questions related to the EAE task, we embrace automatically generated prompts and propose a method called ArgumentPrompt (AMP): activating multi-category of information for event argument extraction with automatically generated prompts. AMP applies automatically generated prompts to eliminate the demand of the time-consuming and labor-intensive question design for the target extraction. To improve the quality of prompts, we mainly apply multi-category information for prompts and a dual-module to augment the guidance of prompts to language models. Experiments demonstrate that AMP can achieve a performance gain of 1.5% and 3.1% on ACE2005 and RAMS, respectively.