Document-Level Event Role Filler Extraction using Multi-Granularity Contextualized Encoding

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8010–8020 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 8010 Document-Level Event Role Filler Extraction using Multi-Granularity Contextualized Encoding Xinya Du and Claire Cardie Department of Computer Science Cornell University Ithaca, NY, USA {xdu, cardie}@cs.cornell.edu Abstract Few works in the literature of event extrac- tion have gone beyond individual sentences to make extraction decisions. This is prob- lematic when the information needed to recog- nize an event argument is spread across multi- ple sentences. We argue that document-level event extraction is a difficult task since it re- quires a view of a larger context to deter- mine which spans of text correspond to event role fillers. We first investigate how end-to- end neural sequence models (with pre-trained language model representations) perform on document-level role filler extraction, as well as how the length of context captured affects the models’ performance. To dynamically aggre- gate information captured by neural represen- tations learned at different levels of granularity (e.g., the sentence- and paragraph-level), we propose a novel multi-granularity reader. We evaluate our models on the MUC-4 event ex- traction dataset, and show that our best system performs substantially better than prior work. We also report findings on the relationship be- tween context length and neural model perfor- mance on the task. 1 Introduction The goal of document-level event extraction1 is to identify in an article events of a pre-specified type along with their event-specific role fillers, i.e., arguments. The complete document-level extrac- tion problem generally requires role filler extrac- tion, noun phrase coreference resolution and event tracking (i.e., determine which extracted role fillers belong to which event). In this work, we focus only on document-level role filler extraction. Fig- ure 1 provides a representative example of this task. Given an article consisting of multiple para- graphs/sentences, and a fixed set of event types 1The task is also referred to as template filling (MUC-4, 1992). Related Work Machine reader reads through the document Perpetrator Individual four terrorists Perpetrator Organization - Target Newspaper El Espectador Victim Teofilo Forero Castro, Luis Carlos Galan Sarmiento Weapon car bomb, dynamite [S1] ... by special urban troops, four terrorists have been arrested in soacha. [S2] They are responsible for the car bomb attack on the Newspaper El Espectador, to a series of bogota dynamite attacks, to the freeing of a group of paid assassins. [S3] The terrorists are also connected to the murder of Teofilo Forero Castro, … [S4] General Ramon is the commander of the 13th infantry brigade. [S5] He said that at least two of those arrested have fully confessed to having taken part in the accident of Luis Carlos Galan Sarmiento in soacha, Cundinamarca. [S6] .. triumph over organized crime, its accomplices and its protectors. ... Figure 1: The document-level event role fillers extrac- tion task. (e.g., terrorist events) and associated roles (e.g., PERPETRATOR INDIVIDUAL, VICTIM, WEAPON), we aim to identify those spans of text that denote the role fillers for …

Content truncated to 500 words