A Neural Transition-based Model for Argumentation Mining

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 6354–6364 August 1–6, 2021. ©2021 Association for Computational Linguistics 6354 A Neural Transition-based Model for Argumentation Mining Jianzhu Bao1,2∗, Chuang Fan1,2∗, Jipeng Wu1,2, Yixue Dang3 Jiachen Du1,2, Ruifeng Xu1,4† 1Harbin Institute of Technology (Shenzhen), China 2Joint Lab of China Merchants Securities and HITSZ 3China Merchants Securities Co., Ltd. 4Peng Cheng Laboratory, Shenzhen, China {jianzhubao, fanchuanghit}@gmail.com wujipeng@stu.hit.edu.cn, dangyixue@cmschina.com.cn jacobvan199165@gmail.com, xuruifeng@hit.edu.cn Abstract The goal of argumentation mining is to au- tomatically extract argumentation structures from argumentative texts. Most existing meth- ods determine argumentative relations by ex- haustively enumerating all possible pairs of ar- gument components, which suffer from low ef- ficiency and class imbalance. Moreover, due to the complex nature of argumentation, there is, so far, no universal method that can address both tree and non-tree structured argumenta- tion. Towards these issues, we propose a neu- ral transition-based model for argumentation mining, which incrementally builds an argu- mentation graph by generating a sequence of actions, avoiding inefficient enumeration op- erations. Furthermore, our model can han- dle both tree and non-tree structured argumen- tation without introducing any structural con- straints. Experimental results show that our model achieves the best performance on two public datasets of different structures. 1 Introduction Argumentation mining (AM) aims to identify the ar- gumentation structures in text, which has received widespread attention in recent years (Lawrence and Reed, 2019). It has been shown beneficial in a broad range of fields, such as information retrieval (Carstens and Toni, 2015; Stab et al., 2018), auto- mated essay scoring (Wachsmuth et al., 2016; Ke et al., 2018), and legal decision support (Palau and Moens, 2009; Walker et al., 2018). Given a piece of paragraph-level argumentative text, an AM system first detects argument components (ACs), which are segments of text with argumentative meaning, and then extracts the argumentative relations (ARs) between ACs to obtain an argumentation graph, where the nodes and edges represent ACs and ARs, ∗Equal Contribution †Corresponding Author [ Either State or Federal law should provide some penalty for passing an NSF or “closed account” check that includes a presumption of guilt. ]AC1 [ The check either bounced or it did not. ]AC2 [ There’s not a lot of grey area here. ]AC3 [ That said, either State or Federal law should also allow the drafter of the check a “safe harbor” wherein they ··· ]AC4 [ Michigan has such a procedure and it seems relatively equitable. ]AC5 AM AC2 Fact AC3 Value AC1 Policy Reason Argumentative Text Argumentation Graph Reason Reason AC4 Policy AC5 Value Reason Figure 1: An example of argumentation mining from the CDCP dataset (Park and Cardie, 2018). Policy, Fact, and Value represent the types of ACs and Reason refers to the types of ARs. Note that, the CDCP dataset we use is preprocessed by Niculae et al. (2017). respectively. An example of AM is shown in Fig- ure 1, where the text is …

Content truncated to 500 words