King's College London

Research portal

Classical logic, argument and dialectic

Research output: Contribution to journalArticle

M. D'Agostino, S. Modgil

Original languageEnglish
Pages (from-to)15-51
Number of pages37
JournalARTIFICIAL INTELLIGENCE
Volume262
Early online date20 Jun 2018
DOIs
Accepted/In press24 May 2018
E-pub ahead of print20 Jun 2018
PublishedSep 2018

Documents

King's Authors

Abstract

A well studied instantiation of Dung's abstract theory of argumentation yields argumentation-based characterisations of non-monotonic inference over possibly inconsistent sets of classical formulae. This provides for single-agent reasoning in terms of argument and counter-argument, and distributed non-monotonic reasoning in the form of dialogues between computational and/or human agents. However, features of existing formalisations of classical logic argumentation (Cl-Arg) that ensure satisfaction of rationality postulates, preclude applications of Cl-Arg that account for real-world dialectical uses of arguments by resource-bounded agents. This paper formalises dialectical classical logic argumentation that both satisfies these practical desiderata and is provably rational. In contrast to standard approaches to Cl-Arg we: 1) draw an epistemic distinction between an argument's premises accepted as true, and those assumed true for the sake of argument, so formalising the dialectical move whereby arguments' premises are shown to be inconsistent, and avoiding the foreign commitment problem that arises in dialogical applications; 2) provide an account of Cl-Arg suitable for real-world use by eschewing the need to check that an argument's premises are subset minimal and consistent, and identifying a minimal set of assumptions as to the arguments that must be constructed from a set of formulae in order to ensure that the outcome of evaluation is rational. We then illustrate our approach with a natural deduction proof theory for propositional classical logic that allows measurement of the ‘depth’ of an argument, such that the construction of depth-bounded arguments is a tractable problem, and each increase in depth naturally equates with an increase in the inferential capabilities of real-world agents. We also provide a resource-bounded argumentative characterisation of non-monotonic inference as defined by Brewka's Preferred Subtheories.

Download statistics

No data available

View graph of relations

© 2020 King's College London | Strand | London WC2R 2LS | England | United Kingdom | Tel +44 (0)20 7836 5454