Zero-Shot Cross-Lingual Event Argument Extraction with Language-Oriented Prefix-Tuning

Event argument extraction (EAE) aims to identify the arguments of a given event, and classify roles that those play. Due high data demands training EAE models, zero-shot cross-lingual has attracted increasing attention, as it greatly reduces human annotation effort. Some prior works indicate generation-based methods have achieved promising performance for monolingual EAE. However, when applying existing EAE, we find two critical challenges, including Language Discrepancy Template Construction. In this paper, propose novel method termed Language-oriented Prefix-tuning Network (LAPIN) address above challenges. Specifically, devise Prefix Generator module handle discrepancies between source target languages. Moreover, leverage Language-agnostic Constructor design templates can be adapted any language. Extensive experiments demonstrate our proposed achieves best performance, outperforming previous state-of-the-art model by 4.8% 2.3% average F1-score on multilingual datasets.