Event Detection with Multi-Order Graph Convolution and Aggregated Attention

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 5766–5770, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 5766 Event Detection with Multi-Order Graph Convolution and Aggregated Attention Haoran Yan†‡, Xiaolong Jin†‡, Xiangbin Meng§, Jiafeng Guo†‡ and Xueqi Cheng†‡ †CAS Key Laboratory of Network Data Science and Technology, Institute of Computing Technology, Chinese Academy of Sciences (CAS); ‡School of Computer and Control Engineering, The University of CAS {yanhaoran17s, jinxiaolong, guojiafeng, cxq}@ict.ac.cn §School of Computer Science, Beijing University of Posts and Telecommunications mengxb@bupt.edu.cn Abstract Syntactic relations are broadly used in many NLP tasks. For event detection, syntactic rela- tion representations based on dependency tree can better capture the interrelations between candidate trigger words and related entities than sentence representations. But, existing studies only use first-order syntactic relations (i.e., the arcs) in dependency trees to identify trigger words. For this reason, this paper pro- poses a new method for event detection, which uses a dependency tree based graph convolu- tion network with aggregative attention to ex- plicitly model and aggregate multi-order syn- tactic representations in sentences. Experi- mental comparison with state-of-the-art base- lines shows the superiority of the proposed method. 1 Introduction As an important information extraction task, event detection aims to find event mentions with spe- cific event types from given texts. Each event mention is identified by a word or phrase called event trigger, which serve as the main word(s) to the corresponding event. For example, in the sen- tence presented in Figure 1, an event detection sys- tem should be able to recognize the trigger word “fired” corresponding to the event type “Attack”. Among existing methods for event detection, sequence based ones (e.g., Chen et al. (2015); Nguyen et al. (2016)) only use the given sen- tences, which suffer from the low efficiency prob- lem in capturing long-range dependency; On the contrary, dependency tree based methods utilize the syntactic relations (i.e., arcs) in the depen- dency tree of a given sentence to more effectively capture the interrelation between each candidate trigger word and its related entities or other trig- gers. For example, Nguyen and Grishman (2018) and Liu et al. (2018) treat each dependency tree as a graph and adopt Graph Convolution Network (GCN) (Kipf and Welling, 2016) to extract event triggers. The syntactic relations between trigger words and related entities may be of first order, embod- ied as the direct arcs in the dependency tree. They may also be of high-order (i.e., the paths more than one hop in the dependency tree). Particularly, ac- cording to our statistics on the benchmark ACE- 2005 dataset, about 51% (4977/9793) of event re- lated entities need more than one hop to get to the corresponding trigger words and related entities in their dependency trees. For example, in Figure 1, it needs at least 4 hops (i.e., “fired”–“evidence”– “blood”–“soldiers”) to figure out that the word “fired” means “shot” instead of “dismissed”. However, the above dependency …

Content truncated to 500 words