A Multi-task Event and Argument Trigger Detection in Hindi using POS Tagging as an Auxiliary Task

Anonymous

Abstract: The event, as well as argument trigger detection, are essential sub-tasks of the event extraction system. Lots of effort has been devoted to improving the performance of trigger detection systems. But, the effect of low-level tasks like Parts-of-Speech (POS) tagging as an auxiliary task in multi-task learning of event and argument trigger detection has not been understood well in literature. In our current work, we propose a BERT-based multi-task architecture that learns a shared representation from two sequence labeling tasks, trigger detection (both event and argument), and POS tagging in a multi-task setup using POS tagging as an auxiliary task. We show that our proposed approach achieves a significant performance boost as compared to single-task models. We perform our experiment in the Hindi language, unlike previously proposed works.