Event Detection without Triggers

Proceedings of NAACL-HLT 2019, pages 735–744 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 735 Event Detection without Triggers Shulin Liu, Yang Li, Xinpeng Zhou, Tao Yang, Feng Zhang Tencent AI Lab {forestliu, youngyli, xinpengzhou}@tencent.com {rigorosyang, jayzhang}@tencent.com Abstract The goal of event detection (ED) is to detect the occurrences of events and categorize them. Previous work solved this task by recogniz- ing and classifying event triggers, which is de- fined as the word or phrase that most clear- ly expresses an event occurrence. As a con- sequence, existing approaches required both annotated triggers and event types in train- ing data. However, triggers are nonessential to event detection, and it is time-consuming for annotators to pick out the “most clear- ly” word from a given sentence, especially from a long sentence. The expensive anno- tation of training corpus limits the applica- tion of existing approaches. To reduce man- ual effort, we explore detecting events with- out triggers. In this work, we propose a novel framework dubbed as Type-aware Bias Neural Network with Attention Mechanisms (TBN- NAM), which encodes the representation of a sentence based on target event types. Ex- perimental results demonstrate the effective- ness. Remarkably, the proposed approach even achieves competitive performances com- pared with state-of-the-arts that used annotat- ed triggers. 1 Introduction This work tackles the task of event detection (ED), whose goal is to detect the occurrences of prede- fined events and categorize them. For example, consider the following sentence “In Baghdad, a cameraman died when an American tank fired on the Palestine Hotel.”, an ideal event detection sys- tem should recognize two events, Death and At- tack(suppose that both Death and Attack are in the predefined event set) . Previous work typically solved this task by rec- ognizing and classifying event triggers. Accord- ing to ACE (Automatic Context Extraction) even- t evaluation program, event trigger is defined as the word or phrase that most clearly expresses an event occurrence. Take the following sentence as an example: S: In Baghdad, a cameraman died when an American tank fired on the Palestine Hotel. “died” is the trigger word of Death event, and “fired” is the trigger word of Attack event. The majority of existing approaches modeled this task as word classification (Ji and Grishman, 2008; Liao and Grishman, 2010; Hong et al., 2011; Li et al., 2013; Nguyen and Grishman, 2015; Liu et al., 2016b,a; Chen et al., 2017), which predicted whether each word in a given sentence is an event trigger and what type of event it triggered. As a consequence, these approaches required both an- notated triggers and event types for training. However, event triggers are nonessential to this task. Remind that the goal of event detection is to recognize and categorize events, thus trigger- s could be viewed as intermediate results of this task. Furthermore, it is time-consuming for anno- tators to pick out the “most clearly” word from a given sentence, especially from a long sentence, which limits the application …

Content truncated to 500 words