A Novel Joint Framework for Multiple Chinese Events Extraction

A Novel Joint Framework for Multiple Chinese Events Extraction Nuo Xu 1,2, Haihua Xie 2, Dongyan Zhao 1 1 Wangxuan Institute of Computer Technology, Peking University, Beijing 100080, China 2 State Key Laboratory of Digital Publishing Technology, Peking University Founder Group Co., Ltd., Beijing 100871, China xunuo2019@pku.edu.cn, xiehh@founder.com, zhaodongyan@pku.edu.cn Abstract Event extraction is an essential yet challenging task in information extraction. Previous approaches have paid little attention to the problem of roles overlap which is a common phenomenon in practice. To solve this problem, this paper defines event relation triple to explicitly represent relations among triggers, arguments and roles which are incorporated into the model to learn their inter-dependencies. A novel joint framework for multiple Chinese events extraction is proposed which jointly performs predictions for event trig- gers and arguments based on shared feature representations from pre-trained language model. Experimental comparison with state-of-the-art baselines on ACE 2005 dataset shows the superiority of the proposed method in both trigger classification and argument classification. 1 Introduction Event extraction (EE) is of utility and challenge task in natural language processing (NLP). It aims to identify event triggers of specified types and their arguments in text. As defined in Automatic Content Extraction (ACE) program, the event extraction task is divided into two subtasks, i.e., trigger extraction (identifying and classifying event triggers) and argument extraction (identifying arguments and labeling their roles). Chinese event extraction is a more difficult task because of language specific issue in Chinese (Chen and Ji, 2009). Since Chinese does not have delimiters between words, segmentation is usually a necessary step for further processing, leading to word-trigger mismatch problem (Lin et al., 2018). The approaches based on word-wise classification paradigm commonly suffer from this. For instance, two characters in one word “打死” (hit and die) trigger two different events: an “Attack”event triggered by “打” (hit) and a “Die” event triggered by “死” (die). It is hard to extract accurately when a trigger is part of a word or cross multiple words. To avoid this issue, we formulate Chinese event extraction as a character-based classification task. In addition, another interesting issue in event extraction which is rarely followed requires more efforts. It is the roles overlap problem that we concern in this paper, including the problems of either roles sharing the same argument or arguments overlapping on some words. There are ©2020 China National Conference on Computational Linguistics Published under Creative Commons Attribution 4.0 International License Proceedi ngs of t he 19t h Chi na Nat i onal Conf erence on Comput at i onal Li ngui st i cs, pages 950- 961, Hai nan, Chi na, Oct ober 30 - Novermber 1, 2020. ( c) Techni cal Commi t t ee on Comput at i onal Li ngui st i cs, Chi nese I nf ormat i on Processi ng Soci et y of Chi na CCL2020 Computational Linguistics Figure 1: Examples of roles overlap problem multiple events existing in the one sentence, which commonly causes the roles overlap problem and is …

Content truncated to 500 words