Non-entity event argument extraction on structural representation

摘要

Currently almost all work of event argument extraction focused on extracting entity-based arguments. However, non-entity arguments also play an important semantic role in most events. Inspired by the task of speculation and negation scope detection, this paper proposes a novel structural representation-based method to extract those non-entity arguments, which regards the selected syntactic subtrees of a sentence as the candidates of non-entity arguments and directly explores the syntactic relations between the event triggers and the syntactic subtrees to recognize nonentity arguments. The experimental results show that our method improves PCS by 12.6%, compared to the baseline.