Enhancing Relation Extraction via Adversarial Multi-task Learning

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6190–6199 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6190 Enhancing Relation Extraction via Adversarial Multi-task Learning Han Qin♠∗, Yuanhe Tian♥∗, Yan Song♠† ♠The Chinese University of Hong Kong (Shenzhen) ♥University of Washington ♠hanqin@link.cuhk.edu.cn ♥yhtian@uw.edu ♠songyan@cuhk.edu.cn Abstract Relation extraction (RE) is a sub-field of information extraction, which aims to extract the relation between two given named entities (NEs) in a sentence and thus requires a good understanding of contextual information, especially the entities and their surrounding texts. However, limited attention is paid by most existing studies to re-modeling the given NEs and thus lead to inferior RE results when NEs are sometimes ambiguous. In this paper, we propose a RE model with two training stages, where adversarial multi-task learning is applied to the first training stage to explicitly recover the given NEs so as to enhance the main relation extractor, which is trained alone in the second stage. In doing so, the RE model is optimized by named entity recogni- tion (NER) and thus obtains a detailed understanding of entity-aware context. We further propose the adversarial mechanism to enhance the process, which controls the effect of NER on the main relation extractor and allows the extractor to benefit from NER while keep focusing on RE rather than the entire multi-task learning. Experimental results on two English benchmark datasets for RE demonstrate the effectiveness of our approach, where state-of-the-art performance is observed on both datasets.‡ Keywords: relation extraction, multi-task learning, adversarial learning 1. Introduction Relation extraction (RE) is an important task in natu- ral language processing (NLP), which has been widely used in many downstream NLP applications such as summarization (Wang and Cardie, 2012), question an- swering systems (Xu et al., 2016a) and text mining (Distiawan et al., 2019). The object of RE is to detect the relation between two given named entities (NEs) in the input sentence, where a good understanding of the contextual information is important to achieve a satis- fying performance. In doing so, most previous studies (Xu et al., 2015; Miwa and Bansal, 2016; Xu et al., 2016b; Zhang et al., 2018; Guo et al., 2019; Sun et al., 2020; Yu et al., 2020; Mandya et al., 2020; Chen et al., 2021; Tian et al., 2022) leveraged the depen- dency tree of the input sentence and model the con- textual information along the shortest dependency path between the two entities and showed promising per- formance on RE. However, although NEs are usually given for RE, a good modeling of them is also highly important especially when they are ambiguous in some scenarios. For example, the example sentence in Figure 1 has three different NEs (i.e., “president”, “cabinet”, and “room”), where the NE “room” is paired with the other two NEs with separate relations: “room” is the lo- cation of “president” and owned by “cabinet”. In this case, with the same sentential context, “room” has am- biguities when it is …

Content truncated to 500 words