Continual Relation Learning via Episodic Memory Activation and Reconsolidation

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6429–6440 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 6429 Continual Relation Learning via Episodic Memory Activation and Reconsolidation Xu Han1∗, Yi Dai1∗, Tianyu Gao1, Yankai Lin2, Zhiyuan Liu1† , Peng Li2, Maosong Sun1, Jie Zhou2 1 State Key Lab on Intelligent Technology and Systems, Institute for Artificial Intelligence, Department of Computer Science and Technology, Tsinghua University, Beijing, China 2Pattern Recognition Center, WeChat AI, Tencent Inc., China {hanxu17,daiy17,gty16}@mails.tsinghua.edu.cn {liuzy,sms}@mail.tsinghua.edu.cn {yankailin,patrickpli,withtomzhou}@tencent.com Abstract Continual relation learning aims to continually train a model on new data to learn incessantly emerging novel relations while avoiding catas- trophically forgetting old relations. Some pio- neering work has proved that storing a hand- ful of historical relation examples in episodic memory and replaying them in subsequent training is an effective solution for such a chal- lenging problem. However, these memory- based methods usually suffer from overfitting the few memorized examples of old relations, which may gradually cause inevitable confu- sion among existing relations. Inspired by the mechanism in human long-term memory for- mation, we introduce episodic memory acti- vation and reconsolidation (EMAR) to contin- ual relation learning. Every time neural mod- els are activated to learn both new and memo- rized data, EMAR utilizes relation prototypes for memory reconsolidation exercise to keep a stable understanding of old relations. The ex- perimental results show that EMAR could get rid of catastrophically forgetting old relations and outperform the state-of-the-art continual learning models. The code and datasets are re- leased on https://github.com/thunlp/ ContinualRE. 1 Introduction Relation extraction aims at detecting relations be- tween entities from text, e.g., extracting the rela- tion “the president of” from the given sentence “Newton served as the president of the Royal So- ciety”, which could serve as external resource for various downstream applications (Dong et al., 2015; Xiong et al., 2017; Schlichtkrull et al., ∗ indicates equal contribution † Corresponding author 2018). The conventional RE methods (Riedel et al., 2013; Zeng et al., 2014; Lin et al., 2016) mostly focus on recognizing relations for a fixed pre-defined relation set, and cannot handle rapidly emerging novel relations in the real world. Some researchers therefore explore to detect and learn incessantly emerging relations in an open scenario. As shown in Figure 1, their ef- forts can be formulated into a two-step pipeline: (1) Open Relation Learning extracts phrases and arguments to construct patterns of specific rela- tions, and then discovers unseen relation types by clustering patterns, and finally expands sufficient examples of new relation types from large-scale textual corpora; (2) Continual Relation Learn- ing continually uses those expanded examples of new relations to train an effective classifier. The classifier is trained on a sequence of tasks for han- dling both existing and novel relations, where each task has its own relation set. Although continual relation learning is vital for learning emerging re- lations, there are rare explorations for this field. A straightforward solution is to store all histor- ical data and re-train …

Content truncated to 500 words