Extracting Event Temporal Relations via Hyperbolic Geometry

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8065–8077 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 8065 Extracting Event Temporal Relations via Hyperbolic Geometry Xingwei Tan1, Gabriele Pergola1, Yulan He1 2 1Department of Computer Science, University of Warwick, UK 2Alan Turing Institute, UK {Xingwei.Tan, Gabriele.Pergola, Yulan.He}@warwick.ac.uk Abstract Detecting events and their evolution through time is a crucial task in natural language un- derstanding. Recent neural approaches to event temporal relation extraction typically map events to embeddings in the Euclidean space and train a classifier to detect temporal relations between event pairs. However, em- beddings in the Euclidean space cannot cap- ture richer asymmetric relations such as event temporal relations. We thus propose to em- bed events into hyperbolic spaces, which are intrinsically oriented at modeling hierarchical structures. We introduce two approaches to en- code events and their temporal relations in hy- perbolic spaces. One approach leverages hy- perbolic embeddings to directly infer event re- lations through simple geometrical operations. In the second one, we devise an end-to-end architecture composed of hyperbolic neural units tailored for the temporal relation extrac- tion task. Thorough experimental assessments on widely used datasets have shown the ben- efits of revisiting the tasks on a different geo- metrical space, resulting in state-of-the-art per- formance on several standard metrics. Finally, the ablation study and several qualitative anal- yses highlighted the rich event semantics im- plicitly encoded into hyperbolic spaces.1 1 Introduction Successful understanding of natural language de- pends, among other factors, on the capability to accurately detect events and their evolution through time. This has recently led to increasing interest in research for temporal relation extraction (Cham- bers et al., 2014; Wang et al., 2020) with the aim of understanding events and their temporal orders. Temporal reasoning has been proven beneficial, for example, in understanding narratives (Cheng et al., 2013), answering questions (Ning et al., 2020), or summarizing events (Wang et al., 2018). 1Source code is available at https://github.com/ Xingwei-Warwick/hyper-event-TempRel. e1: tendered e0: said e5: buy e2: expected e3: cleared e4: permits West German and French authorities have (e3: cleared) Dresdner Bank AG's takeover of a majority stake in Banque Internationale de Placement. The approval, which had been (e2: expected), (e4:permits) West Germany's second-largest bank to acquire shares of the French investment bank. Dresdner Bank (e0:said) it will (e5: buy) all shares (e1: tendered) by shareholders on the Paris Stock Exchange at the same price from today through Nov. 17. Figure 1: Events annotated with temporal relations from a document excerpt. Arrow lines represent the Before relations, while red dashed lines the Vague ones. However, events that occurred in text are not just simple and standalone predicates, they rather form complex and hierarchical structures with different granularity levels (Fig. 1), a characteristic that still challenges existing models and restricts their per- formance on real-world datasets for temporal rela- tion extraction (Ning et al., 2018a,c). Addressing such challenges requires models to not only recog- nize accurately the events and their hierarchical and …

Content truncated to 500 words