Dynamically Updating Event Representations for Temporal Relation Classification with Multi-category Learning

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1352–1357 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1352 Dynamically Updating Event Representations for Temporal Relation Classification with Multi-category Learning Fei Cheng1, Masayuki Asahara2, Ichiro Kobayashi3, and Sadao Kurohashi1 1Graduate School of Informatics, Kyoto University 2National Institute for Japanese Language and Linguistics 3Ochanomizu University {feicheng,kuro}@i.kyoto-u.ac.jp, masayu-a@ninjal.ac.jp, koba@is.ocha.ac.jp Abstract Temporal relation classification is a pair-wise task for identifying the relation of a tempo- ral link (TLINK) between two mentions, i.e. event, time and document creation time (DCT). It leads to two crucial limits: 1) Two TLINKs involving a common mention do not share in- formation. 2) Existing models with indepen- dent classifiers for each TLINK category (E2E, E2T and E2D) 1 hinder from using the whole data. This paper presents an event centric model that allows to manage dynamic event representations across multiple TLINKs. Our model deals with three TLINK categories with multi-task learning to leverage the full size of data. The experimental results show that our proposal outperforms state-of-the-art models and two transfer learning baselines on both the English and Japanese data. 1 Introduction Reasoning over temporal relations relevant to an event mentioned in the document can help us un- derstand when the event begins, how long it lasts, how frequent it is, and etc. Starting with the Time- Bank (Pustejovsky et al., 2003) corpus, a series of temporal competitions (TempEval-1,2,3) (Verha- gen et al., 2009, 2010; UzZaman et al., 2012) are attracting growing research efforts. Temporal relation classification (TRC) is the task to predict a temporal relation (after, before, in- cludes, etc.) of a TLINK from a source mention to a target mention. Less effort has been paid to explore the sharing information across ‘local’ pairs and TLINK categories. In recent years, a variety of dense annotation schemas are proposed to over- come the ‘sparse’ annotation in the original Time- bank. A typical one is the Timebank-Dense (TD) corpus (Chambers et al., 2014), which performs 1Time-to-Time (T2T) is not included in this paper, as we focus on event centric representations. a compulsory dense annotation with the complete graph of TLINKs for the mentions located in two neighbouring sentences. Such dense annotation in- creases the chance of pairs sharing common events and demands of managing ‘global’ event represen- tations across pairs among TLINK categories. However, globally managing event representa- tions of a whole document takes an extremely heavy load for the dense corpora. Timebank-Dense contains around 10,000 TLINKs in only 36 docu- ments and is 7 times denser than the original Time- bank. Thus, we propose a simplified scenario called Source Event Centric TLINK (SECT) chain. For each event ei in a document, we group all TLINKs containing the common source event ei into the ei centric TLINK chain and align them with the chronological order of the target mentions appear- ing in the document. We assume that our system is capable of learning dynamic representations of the centric event ei along the SECT chain via a …

Content truncated to 500 words