Incorporating Zoning Information into Argument Mining from Biomedical Literature

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6162–6169 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6162 Incorporating Zoning Information into Argument Mining from Biomedical Literature Boyang Liu, Viktor Schlegel, Riza Batista-Navarro, Sophia Ananiadou Department of Computer Science, The University of Manchester Manchester, United Kingdom {boyang.liu-2@postgrad., viktor.schlegel@, riza.batista@, sophia.ananiadou@}manchester.ac.uk Abstract The goal of text zoning is to segment a text into zones (i.e., Background, Conclusion) that serve distinct functions. Argumentative zoning, a specific text zoning scheme for the scientific domain, is even considered as the antecedent for argument mining by many researchers. Surprisingly, however, little work is concerned with exploiting zoning information to improve the performance of argument mining models, despite the relatedness of the two tasks. In this paper, we propose two transformer-based models to incorporate zoning information into argumentative component identification and classification tasks. One model is for the sentence-level argument mining task and the other is for the token-level task. In particular, we add the zoning labels predicted by an off-the-shelf model to the beginning of each sentence, inspired by the convention commonly used biomedical abstracts. Moreover, we employ multi-head attention to transfer the sentence-level zoning information to each token in a sentence. Based on experiment results, we find a significant improvement in F1-scores for both sentence- and token-level tasks. It is worth mentioning that these zoning labels can be obtained with high accuracy by utilising readily available automated methods. Thus, existing argument mining models can be improved by incorporating zoning information without any additional annotation cost. Keywords: Argument Mining, Text Zoning, Biomedical Literature 1. Introduction Argument mining (also known as argumentation min- ing) is a task that aims at analyzing the argumen- tative structure of discourse. This task can be di- vided into four sub-tasks (Eger et al., 2017): argumen- tative component recognition (separating argumenta- tive units from non-argumentative units), argumenta- tive component classification (i.e., the distinction be- tween claims and premises), argumentative relation recognition (finding relations between argumentative components) and argumentative relation classification (i.e., whether argumentative components support or at- tack each other). Based on granularity, argument min- ing can be divided into sentence-level and token-level tasks. The former means that the boundary of an argu- mentative component is the same as the sentence, while the latter means that the length of argumentative com- ponents can range from less than a clause to several sentences. In scientific literature, arguments (see examples in Fig- ure 1) are fundamental building blocks whose main purpose is to persuade others to accept the academic opinion (i.e., claims) proposed by the authors. This means that the key to understanding a scientific paper is to find its argumentative structure. However, with the growth of literature (Wang and Lo, 2021), it is ex- tremely time-consuming to manually analyse the latest research and understand the argumentative structure of related literature on such a large scale. Automatic iden- tification of the argumentative structure in scientific lit- erature is important for researchers and practitioners …

Content truncated to 500 words