Arguments as Narratives

Aspects of narrative coherence are proposed as a means to investigate and identify arguments from text. Computational analysis of ar-gumentation largely focuses on representations of arguments that are either abstract or are constructed from a logical (e.g. propositional or first order) knowledge base. Argumentation schemes have been advanced for stereotypical patterns of defeasible reasoning. While we have well-formedness conditions for arguments in a first order language, namely the patterns for inference, the conditions for argumen-tation schemes is an open question, and the identification of arguments ‘in the wild ’ is problematic. We do not understand the ‘source’ of rules from which inference follows; formally, well-formed ‘arguments ’ can be expressed even with random sentences; moreover, argument indicators are sparse, so cannot be relied upon to identify arguments. As automated extraction of arguments from text increas-ingly finds important applications, it is pressing to isolate and integrate indicators of argument. To specify argument well-formedness conditions and identify arguments from unstructured text, we suggest using aspects of narrative coherence