OneEE: A One-Stage Framework for Fast Overlapping and Nested Event Extraction

Proceedings of the 29th International Conference on Computational Linguistics, pages 1953–1964 October 12–17, 2022. 1953 OneEE: A One-Stage Framework for Fast Overlapping and Nested Event Extraction Hu Cao1∗, Jingye Li1∗, Fangfang Su1, Fei Li1, Hao Fei2, Shengqiong Wu2, Bobo Li1, Liang Zhao3, Donghong Ji1† 1 Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, China 2 School of Computing, National University of Singapore, Singapore 3 Department of Computing and Mathematics, University of São Paulo, Brazil {whucaohu, theodorelee, lifei_csnlp, dhji}@whu.edu.cn Abstract Event extraction (EE) is an essential task of information extraction, which aims to extract structured event information from unstructured text. Most prior work focuses on extracting flat events while neglecting overlapped or nested ones. A few models for overlapped and nested EE includes several successive stages to ex- tract event triggers and arguments, which suffer from error propagation. Therefore, we design a simple yet effective tagging scheme and model to formulate EE as word-word relation recog- nition, called OneEE. The relations between trigger or argument words are simultaneously recognized in one stage with parallel grid tag- ging, thus yielding a very fast event extraction speed. The model is equipped with an adaptive event fusion module to generate event-aware representations and a distance-aware predictor to integrate relative distance information for word-word relation recognition, which are em- pirically demonstrated to be effective mecha- nisms. Experiments on 3 overlapped and nested EE benchmarks, namely FewFC, Genia11, and Genia13, show that OneEE achieves the state- of-the-art (SoTA) results. Moreover, the infer- ence speed of OneEE is faster than those of baselines in the same condition, and can be fur- ther substantially improved since it supports parallel inference.1 1 Introduction Event Extraction (EE) is a fundamental yet chal- lenging task in information extraction research (Miwa and Bansal, 2016; Katiyar and Cardie, 2016; Fei et al., 2020b; Li et al., 2021b; Fei et al., 2022a). EE facilitates the development of practical applica- tions such as knowledge graph construction (Wei et al., 2019b; Bosselut et al., 2021), biological pro- cess analysis (Miwa et al., 2013), and financial mar- ket surveillance (Nuij et al., 2013). The goal of EE *Equal contribution †Corresponding author 1The codes at https://github.com/Cao-Hu/OneEE Wang Yawei reduced shares in Sanju environmental protection. Event: Share Reduction trigger subject object Citic Securities acquired 100% equity of Guangzhou Securities. Event: Investment trigger subject object Event: Share Transfer trigger subject target proportion Degraded CGN stimulated ICAM-1 expression in monocytes. Event: Positive Regulation trigger Theme Event: Gene Expression trigger Theme (a) Flat Event (b) Overlapped Event (c) Nested Event Figure 1: Examples of three kinds of events, including a flat event (a), overlapped events (b), and nested events (c). Different event mentions are denoted in distinct colors. Triggers are marked with red boxes while argu- ments are underlined. is to recognize event triggers as well as the associ- ated arguments from texts. As an example, Figure 1(a) illustrates a Share Reduction event in- cluding a trigger “reduced” and a …

Content truncated to 500 words