Treasures Outside Contexts: Improving Event Detection via Global Statistics

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2625–2635 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 2625 Treasures Outside Contexts: Improving Event Detection via Global Statistics Rui Li, Wenlin Zhao, Cheng Yang∗, Sen Su Beijing University of Posts and Telecommunications, Beijing, China {lirui,zhaowenlin,yangcheng,susen}@bupt.edu.cn Abstract Event detection (ED) aims at identifying event instances of specified types in given texts, which has been formalized as a sequence la- beling task. As far as we know, existing neural-based ED models make decisions re- lying on the contextual semantic features of each word in the input text, which we find is easy to get confused by varied contexts in the test stage. To this end, we come up with the idea of introducing a set of statistical features from word-event co-occurrence frequencies in the entire training set to cooperate with the contextual features. Specifically, we propose a Semantic and Statistic-Joint Discriminative Network (S2-JDN) consisting of a semantic feature extractor, a statistical feature extractor, and a joint event discriminator. In experiments, S2-JDN effectively exceeds ten recent state-of- the-art (SOTA) baseline methods on ACE2005 and KBP2015 benchmark datasets. Further, we perform extensive experiments to investi- gate the effectiveness of S2-JDN. 1 Introduction Event detection (ED) is an important information extraction task in the NLP field, which aims to identify event instances of specified types in given texts. Associated with each event mention is a phrase, i.e., the event trigger1 evoking that event. More precisely, the task involves identifying event triggers and classifying them into the specific types. For instance, according to ACE2005 annotation guideline, in the sentence “A police officer was killed in New Jersey today”, an ED model should be able to recognize the word “killed” as a trigger for the event type “Die”. With the development of deep learning, ED has been formalized as a sequence labeling task and ∗Corresponding author. 1The event trigger is usually a single verb or nominaliza- tion, and some papers also refer to the multi-word trigger as event nugget. In this paper, we uniformly use “trigger”. Two texts in the training set: Trigger Event T1: There have been too many civilians killed … Killed Die T2: … a sister were wounded late Wednesday … Wounded Injure Four variants of T1 and T2 in testing: DMBERT Ours V1: There have been too many civilians killed. Die √ Die √ V2: … a sister were killed late Wednesday… Injure × Die √ V3: There have been too many civilians wounded . . . Die × Injure√ V4: EU foreign policy supremo Javier Solana likewise slammed the attack, saying, “There have been too many civilians wounded . . .” Other × Attack× Figure 1: A case study of SOTA model DMBERT (Wang et al., 2019) and our proposed model. We train DMBERT and our proposed model on ACE2005, and select two texts (T1 and T2) with “killed” and “wounded” as triggers in the training set. For testing, we change the contexts of the two triggers in …

Content truncated to 500 words