More than Classification: A Unified Framework for Event Temporal Relation Extraction

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 9631–9646 July 9-14, 2023 ©2023 Association for Computational Linguistics More than Classification: A Unified Framework for Event Temporal Relation Extraction Quzhe Huang1,2, Yutong Hu1,2, Shengqi Zhu3, Yansong Feng1∗, Chang Liu1,4, Dongyan Zhao1,5 1Wangxuan Institute of Computer Technology, Peking University, China 2School of Intelligence Science and Technology, Peking University 3University of Washington 4Center for Data Science, Peking University 5National Key Laboratory of General Artificial Intelligence {huangquzhe,huyutong,fengyansong,liuchang97,zhaody} @pku.edu.cn sqzhu@uw.edu Abstract Event temporal relation extraction (ETRE) is usually formulated as a multi-label classifica- tion task, where each type of relation is simply treated as a one-hot label. This formulation ignores the meaning of relations and wipes out their intrinsic dependency. After examining the relation definitions in various ETRE tasks, we observe that all relations can be interpreted us- ing the start and end time points of events. For example, relation Includes could be interpreted as event 1 starting no later than event 2 and ending no earlier than event 2. In this paper, we propose a unified event temporal relation extraction framework, which transforms tem- poral relations into logical expressions of time points and completes the ETRE by predicting the relations between certain time point pairs. Experiments on TB-Dense and MATRES show significant improvements over a strong baseline and outperform the state-of-the-art model by 0.3% on both datasets. By representing all rela- tions in a unified framework, we can leverage the relations with sufficient data to assist the learning of other relations, thus achieving sta- ble improvement in low-data scenarios. When the relation definitions are changed, our method can quickly adapt to the new ones by simply modifying the logic expressions that map time points to new event relations. The code is re- leased at https://github.com/AndrewZhe/ A-Unified-Framework-for-ETRE. 1 Introduction In order to fully understand natural language utter- ances, it is important to understand the temporal information conveyed in the text, especially the relations between the events (Pustejovsky et al., 2003a, 2010). Such temporal relations play an es- sential role in downstream applications, such as ∗Corresponding author. Annotation Scheme Labels Interval One-hot Unified Representation TB-Dense Before [1,0,0,0,0,0] Includes [0,0,1,0,0,0] Matres Before [1,0,0,0] Vague [0,0,0,1] time 𝑡! " < 𝑡! # 𝑡! " < 𝑡! # ∧ 𝑡! " > 𝑡! # ∧ 𝑡$ " ≤ 𝑡! # 𝑡! " < 𝑡! # (𝑡! "≤ 𝑡! # ∧ 𝑡$ " > 𝑡$ #) ∨ (𝑡! "< 𝑡! # ∧ 𝑡$ " ≥ 𝑡$ #) Figure 1: Examples of labels from TB-Dense and MA- TRES and their Interval, One-hot and Unified represen- tations. and represent the intervals of event 1 and event 2 in the timeline. t∗ s and t∗ e represent the start and end time points of an event. question answering, event timeline generation, and information retrieval (Choubey and Huang, 2017; Han et al., 2019). The Event Temporal Relation Extraction (ETRE) task is proposed to address the extraction of temporal relations between event pairs from text. Researchers …

Content truncated to 500 words