Improving Event Detection via Open-domain Trigger Knowledge

• DOI: 10.18653/v1/2020.acl-main.522
• Publication Date: 2020-07-29T14:14:43Z
• Authors: Meihan Tong; Bin Xu; Shuai Wang; Yixin Cao; Lei Hou; Juanzi Li; Jun Xie
• Keywords: Overfitting; Leverage (statistics); Open domain; Benchmark (surveying); Structuring; Heuristics

Event Detection (ED) is a fundamental task in automatically structuring texts. Due to the small scale of training data, previous methods perform poorly on unseen/sparsely labeled trigger words and are prone overfitting densely words. To address issue, we propose novel Enrichment Knowledge Distillation (EKD) model leverage external open-domain knowledge reduce in-built biases frequent annotations. Experiments benchmark ACE2005 show that our outperforms nine strong baselines, especially effective for The source code released https://github.com/shuaiwa16/ekd.git.