Can AI Models Capture Natural Language Argumentation?

Can AI Models Capture Natural Language Argumentation?

Leila Amgoud (Institut de Recherche en Informatique de Toulouse, France) and Henri Prade (Institut de Recherche en Informatique de Toulouse, France)
DOI: 10.4018/jcini.2012070102

Abstract

Formal AI models of argumentation define arguments as reasons that support claims (which may be beliefs, decisions, actions, etc.). Such arguments may be attacked by other arguments. The main issue is then to identify the accepted ones. Several semantics were thus proposed for evaluating the arguments. Works in linguistics focus mainly on understanding the notion of argument, identifying its types, and describing different forms of counter-argumentation. This paper advocates that such typologies are instrumental for capturing real argumentations. It shows that some of the forms cannot be handled properly by AI models. Finally, it shows that the use of square of oppositions (a very old logical device) illuminates the interrelations between the different forms of argumentation.
Article Preview

Introduction

Argumentation is a social activity of reason in which a proponent agent tries to convince an opponent one that a certain statement is true (or false) by putting forward arguments. While reasoning looks for the truth of a statement, argumentation looks only for persuading agents. Indeed, the proponent may succeed to persuade the opponent even if himself is not convinced by the statement.

Argumentation is an interdisciplinary topic. It has been studied by philosophers like Hamblin (1970), Rescher (1977), Perelman and Olbrechts-Tyteca (1969) and Toulmin (1958). Patterns of argumentation are studied in a pedagogical perspective for identifying fallacies in reasoning and avoiding them (Blackburn, 1989). Argumentation has also become an Artificial Intelligence keyword since early nineties. It is particularly used for nonmonotonic reasoning (e.g., Dung, 1995, Simari & Loui, 1992) and for modeling dialogues between agents (e.g., Amgoud, Dimopoulos, & Moraitis, 2007, Prakken, 2005). See also Bench-Capon, and Dunne (2007), Besnard and Hunter (2008), and Rahwan et al. (2009) for descriptions of research on argumentation in AI. Whatever the application is, the same kind of argumentation model is considered. It consists of a set of arguments supporting statements and attacks among those arguments. Acceptability semantics are then used in order to evaluate the arguments and to decide on which statements to rely on. In all existing models, an argument has mainly three parts: a conclusion, a set of premises (called support) and a link between the support and the conclusion.

Besides, argumentation has been extensively studied by linguists like Salavastru (2007) and Apothéloz (1989, 1993) (Quiroz, Apothéloz, & Brandt, 1992). The main focus here is on the notion of argument and its different types in real dialogues. In Apothéloz (1989, 1993), four argumentative types are defined. Two of them are arguments and two others are rejections of arguments. In addition, Apothéloz defined four modes of counter-argumentation. Each of them may be divided into at least two distinct cases.

Our aim in this paper is to analyze the typologies of arguments and the four modes of counter-argumentation proposed in Apothéloz (1989, 1993) and Quiroz et al. (1992), and to investigate whether they can be captured by the argumentation models developed in AI. Comparing research originating in the two communities (computer science and linguistics) is important since it allows a better understanding of work in both communities and may lead to the development of richer models of argumentation.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 13: 4 Issues (2019): Forthcoming, Available for Pre-Order
Volume 12: 4 Issues (2018): 3 Released, 1 Forthcoming
Volume 11: 4 Issues (2017)
Volume 10: 4 Issues (2016)
Volume 9: 4 Issues (2015)
Volume 8: 4 Issues (2014)
Volume 7: 4 Issues (2013)
Volume 6: 4 Issues (2012)
Volume 5: 4 Issues (2011)
Volume 4: 4 Issues (2010)
Volume 3: 4 Issues (2009)
Volume 2: 4 Issues (2008)
Volume 1: 4 Issues (2007)
View Complete Journal Contents Listing