Critical Interpretation of Arguments using Graphs and Representations

Abstract

The complexity of argumentative reasoning requires the systematic evaluation of conflicting perspectives, where different parts of an argument refer to claims subject to different levels of rigour and supporting evidence. Argumentation is at the centre of many high-impact decision-making processes, both at institutional and personal levels, for example, within legal and democratic systems and for policy-making. However, despite its ubiquity and importance, the argumentation process remains mostly informal, and the tools that are available to support the critical assessment of arguments are beyond the reach of the general public. Argumentation classification schemes [5] are categorical frameworks used to structure and qualify the constituting parts of arguments, in order to support their critical interpretation. However, while categorical frameworks for the critical interpretation of arguments are available, the barriers for their application remain significant, as the per- son on the receiving end of the argument needs to mentally decompose, structure and classify the underlying argumentation relations, in order to deliver a critical assessment on the quality of the argument. This is an intrinsic limitation of the medium/substrate of most arguments, namely text and speech, which do not allow for an efficient interpretation of these elements. More recently, the evolution of natural language processing (NLP) and, in particular, argumentation mining is lowering the barriers to automatically classifying and structuring arguments [6, 7], bringing the opportunity to introduce representation models that can facilitate critical argumentative reasoning for end users. As NLP evolves, argumentation mining models are expected to become more robust, ultimately resulting in the commodification of argument interpretation. This increasing capability brings the opportunity to revisit and reconceptualise our relationship, with mechanisms that can enhance our ability to interpret argumentative discourse. This involves incorporating an interpretative layer capable of representing and organising the structures and categories currently obscured, flattened, and hidden in the text. In direct relation to this trend, this work proposes a representation and visualisation model that organises the arguments according to their scheme types [5, 8], allowing end-users to visualise structural relations and their categorical mapping. The intent of this representation and visualisation model is to elicit the relationships and categories during the argument evaluation process, efficiently improving the alignment between end users and frameworks of critical argumentative reasoning. There are two proposed models of structured argument representations and visualisation: one emphasises the hierarchical argumentation structure between sentences, known as the Argument Graph Model, and the other prioritises the representations of the argumentation scheme types, named the Aggregate Representation of Arguments. Each model targets a set of complementary properties that can be used to facilitate critical reasoning over arguments. Both representations are instantiated into an argument visualisation software and empirically evaluated through an end-user study, comparing them with contrasting baselines. These models are found to have a significant positive impact by supporting end-users in their critical argumentation assessment

Date of Award	31 Dec 2024
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Andre Freitas (Supervisor) & Markel Vigo (Supervisor)