Graph Based Decoding for Event Sequencing and Coreference Resolution

Proceedings of the 27th International Conference on Computational Linguistics, pages 3645–3657 Santa Fe, New Mexico, USA, August 20-26, 2018. 3645 Graph-Based Decoding for Event Coreference and Sequencing Resolution Zhengzhong Liu and Teruko Mitamura and Eduard Hovy Language Technologies Institute Carnegie Mellon University Pittsburgh, PA 15213, USA {liu, teruko, hovy}@cs.cmu.edu Abstract Events in text documents are interrelated in complex ways. In this paper, we study two types of relation: Event Coreference and Event Sequencing. We show that the popular tree-like decoding structure for automated Event Coreference is not suitable for Event Sequencing. To this end, we propose a graph-based decoding algorithm that is applicable to both tasks. The new decoding algorithm supports flexible feature sets for both tasks. Empirically, our event coreference system has achieved state-of-the-art performance on the TAC-KBP 2015 event coreference task and our event sequencing system beats a strong temporal-based, oracle-informed baseline. We discuss the challenges of studying these event relations. Title and Abstract in Chinese 基于图的事件共指消解以及依序关系解码算法 文本中提及的事件之间常存在着复杂的关系。在这篇文章中，我们主要研究两种关系： 事件间的共指关系以及依序关系。我们指出，常被应用在事件指代消歧上的传统的树结 构解码方式并不适用于解决事件依序问题。为了解决这个问题，我们提出了一种适用于 两种情况的新的图结构解码算法。这个算法让我们可以为这两个任务设计灵活的特征。 在实验上，我们的事件共指消解系统在TAC-KBP 2015的共指消解任务上达到了目前最佳 的水平。同时，我们的依序关系系统的结果超过了一个基于时间分析并有部分正确答案 的基线系统。最后我们分析了结果，并讨论了事件关系判别任务的一些挑战。 1 Introduction Events are important building blocks of documents. They play a key role in document understanding tasks, such as information extraction (Chambers and Jurafsky, 2011), news summarization (Vossen and Caselli, 2015), story understanding (Mostafazadeh et al., 2016). Conceptually, events correspond to state changes and normally include a location, a time interval, and several entities/participants. In a text, events are realized as text spans, normally as verbs and nouns that indicate state changes (Vendler, 1957). The text spans are often referred to as event mentions or event nuggets (We use the term event mention in this paper.). The textual mentions of events have rich relations among them, and collectively convey the meaning of one or more related documents. In this paper, we study two different types of relation: Event Hopper Coreference (EH) and Event Sequencing (ES). Event Coreference: There is a rich literature on the Event Coreference problem (Liu et al., 2014; Cybulska and Vossen, 2014; Lu et al., 2016; Peng et al., 2016; Lu and Ng, 2017; Araki, 2018). By analogy to entity coreference, the “same” conceptual event may be realized by multiple text spans (event mentions). The coreference problem aims at identifying these relations to recover events from the text spans. The Event Hopper Coreference task in the TAC-KBP evaluation campaign defines coreference links as follows (Mitamura et al., 2018): Two event mentions are considered coreferent if they refer to the conceptually same underlying event, even if their arguments are not strictly identical. For example, This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/ 3646 Figure 1: Example of Event Coreference and Sequence relations. Red lines are coreference links; solid blue arrows represent Subevent relations; dotted green arrows represent After relations. mentions that share similar temporal and location scope, though not necessarily the same expression, are considered to be coreferent (Attack in Baghdad on Thursday vs. Bombing in the Green Zone last week). This means that the event arguments of coreferential events mentions can …

Content truncated to 500 words