Please register for the CV & Online Identity Workshop below.

Reasoning is an inherently relational activity. Ideas are not simply connected by association - the acceptance of one carries implications for the acceptability of others. I characterize belief as both a mental state and representation of what may or may not be the case about things in the world, whether objects, events, or representations themselves. Reasoning is then the conscious consideration of the linkages between beliefs which determine relations of consistency, support, undermining, and the like. Language enables the communication of mental states between minds, and argumentation is the expression of reasoning in language by which we can affect the beliefs of others. This linguistic expression of reasoning has a basic level of invariant structural ingredients: candidate beliefs are expressed via some form of statements, locutions, or propositions, and are connected by directional relations of support or attack.

Argument mining, a machine learning task, aims to identify these argumentative units and relations in text, necessarily resulting in data with a network structure. Taking a complex systems approach to these data opens up potential applications ranging from investigations of belief dynamics to collective intelligence, cognitive dissonance, polarization, group formation, and so forth.

Before charging ahead with hopeful computational analyses to draw conclusions about concepts of interest in the study of social systems, we must have some strong basis for drawing the link between phenomenon and representation. The present work builds just this foundation, moving from an elementary consideration of the relationship between thought and language to an account of argumentative units and relations as representing what I define as reasoned belief. I then lay out some categorizations of forms of argumentative activity which might aid in identifying when certain data have relevance to only a restricted scope of these activities. I provide an overview of differing ontologies which specify what exactly argumentative units and relations are, resulting in different network representations. I highlight some major considerations in selecting a data model for moving from studies of individuals’ thought processes to collective and interactive systems. Finally, I discuss some examples of existing datasets in context of the preceding sections.