The Search for Agreement on Logical Fallacy Annotation of an Infodemic

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4430–4438 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4430 The Search for Agreement on Logical Fallacy Annotation of an Infodemic Claire Bonial,1 Austin Blodgett,2 Taylor Hudson,3 Stephanie M. Lukin,1 Jeffrey Micher,1 Douglas Summers-Stay,1 Peter Sutor4 & Clare R. Voss1 1U.S. Army Research Lab, Adelphi, MD 20783 2Institute for Human & Machine Cognition, Pensacola, FL 32502 3Oak Ridge Associated Universities, Oak Ridge, TN 37831 4University of Maryland, College Park, MD 20742 claire.n.bonial.civ@army.mil Abstract We evaluate an annotation schema for labeling logical fallacy types, originally developed for a crowd-sourcing annotation paradigm, now using an annotation paradigm of two trained linguist annotators. We apply the schema to a variety of different genres of text relating to the COVID-19 pandemic. Our linguist (as opposed to crowd-sourced) annotation of logical fallacies allows us to evaluate whether the annotation schema category labels are sufficiently clear and non-overlapping for both manual and, later, system assignment. We report inter-annotator agreement results over two annotation phases as well as a preliminary assessment of the corpus for training and testing a machine learning algorithm (Pattern-Exploiting Training) for fallacy detection and recognition. The agreement results and system performance underscore the challenging nature of this annotation task and suggest that the annotation schema and paradigm must be iteratively evaluated and refined in order to arrive at a set of annotation labels that can be reproduced by human annotators and, in turn, provide reliable training data for automatic detection and recognition systems. Keywords: logical fallacies, misinformation, inter-annotator agreement metrics 1. Introduction With the outbreak of the COVID-19 pandemic, a par- allel “infodemic” has emerged, defined by the World Health Organization as “too much information includ- ing false or misleading information in digital and phys- ical environments during a disease outbreak.”1 We seek to identify such false or misleading information, or misinformation, in our efforts to develop a semantic search framework for COVID research, in which users can pose unconstrained natural language queries and receive a range of answers with explanations of their relevance (Bonial et al., 2020). Misinformation can take many forms and has become a quickly growing area of NLP research; here, we focus on an approach for annotating and automatically identifying logical fallacies. This area of research is particularly challeng- ing because logical fallacies can be subtly encoded in the structure of a document across multiple sentences, making it difficult for human annotators or systems to recognize fallacies, yet also further motivating the need for assistance in finding such hidden, powerful sources of misinformation. There are a growing number of relevant annotation schemas that could be leveraged for this task, many focused on the annotation of misinformation markers generally, including some logical fallacies (see §7). However, there are relatively few that are focused on logical fallacies as the primary phenomena of inter- est. After surveying existing schemas, we adopt the 1https://www.who.int/health-topics/ infodemic Argotario fallacy annotation schema of Habernal et al. (2017) and …

Content truncated to 500 words