Generating Temporally-ordered Event Sequences via Event Optimal Transport

Proceedings of the 29th International Conference on Computational Linguistics, pages 1875–1884 October 12–17, 2022. 1875 Generating Temporally-ordered Event Sequences via Event Optimal Transport Bo Zhou1,2, Yubo Chen1,2, Kang Liu1,2,4, Jun Zhao1,2, Jiexin Xu3, Xiaojian Jiang3, Qiuxia Li3 1School of Artificial Intelligence, University of Chinese Academy of Sciences 2National Laboratory of Pattern Recognition, CASIA 3China Merchants Bank, 4Beijing Academy of Artificial Intelligence {bo.zhou,yubo.chen,kliu,jzhao}@nlpr.ia.ac.cn {jiexinx,jiangxiaojian,annielqx}@cmbchina.com Abstract Generating temporally-ordered event se- quences in texts is important to natural language processing. Two emerging tasks in this direction are temporal event ordering (rearranging the set of events to correct order) and event infilling (generating an event at a specified position). To tackle the two related tasks, the existing method adopts a vanilla sequence-to-sequence model via maximum likelihood estimation (MLE). However, applying this approach to these tasks will cause two issues. One issue is that the MLE loss emphasizes strict local alignment and ignores the global semantics of the event. The other issue is that the model adopts a word-level objective to model events in texts, failing to evaluate the predicted results of the model from the perspective of event sequence. To alleviate these issues, we present a novel model to tackle the generation of temporally-ordered event sequences via Event Optimal Transport (EOT). First, we treat the events in the sequence as modeling units and explicitly extract the semantics of the events. Second, to provide event sequence-level evaluation of the predicted results of the model, we directly match events in sequences. Extensive experimental results show that our approach outperforms previous models on all evaluation datasets. In particular, the accuracy is improved by 7.7%, and the Macro F1 is improved by 7.2% on one of the datasets. 1 Introduction Generating temporally-ordered event sequences in texts is crucial to many artificial intelligence ap- plications, such as discourse understanding (Nie et al., 2019), dialog generation (Wu et al., 2018) and stock prediction (Ding et al., 2016). Temporal event ordering and event infilling are two challeng- ing tasks in this line of work. The former refers to the rearrangement of an unordered sequence of events to an ordered sequence of events, and the e3 (a) Ordering Task: (b) Infilling Task: (c) Unified Model: e2 e4 e1 e2 e3 e1 e2 e1 e2 e3 e1 e2 e3 e2 e1 e1:The British Prime Minister proposed a referendum. e2:Britain held a referendum on Brexit. e3:The referendum triggered panic in the stock market. e4:Investors sold stocks. e4 e4 e4 e1 e4 e4 Generation Model Figure 1: The temporal event ordering task, event infill- ing task and a single generation model handling these two tasks. latter refers to inferring missing events in an in- complete sequence of events. Figure 1 shows an example of the temporal event ordering task and event infilling task in (a) and (b), respectively. For the temporal event ordering, given an unordered sequence (e3, e2, e4, e1) as input, the output is or- dered sequence (e1, e2, e3, e4). For the event infill- ing, given an incomplete sequence (e1, e2, e4) …

Content truncated to 500 words