Introduction to Computational Models of Natural Argument Chris Reed, Floriana Grasso & Giuseppe Carenini Universities of Dundee, Liverpool and British Columbia
[email protected],
[email protected],
[email protected]
Argumentation is becoming entrenched in a number of areas of AI as a powerful means of approaching and framing problems, and of developing novel solutions. A prime example is in multi-agent systems (MAS), where argumentation has been proposed as a means of structuring inter-agent communication, linking the definition of language protocols to the design of structures in belief databases. Thus, it has been suggested, agents may put forward arguments that try to develop appropriate plans, or to affect the beliefs of other agents. The approach has been so successful that it has made its way into undergraduate textbooks in the area (Wooldridge, 2002). Recent work in argumentation theory has had a direct impact on research in MAS, with theoretical descriptions of types of dialogues proposed in (Walton and Krabbe, 1995) - negotiation, persuasion, deliberation, and others - being used operationally in systems such as (Reed, 1998) and (McBurney and Parsons, 2002). Some of the earliest work built on Dung's (1995) exploration of the idea of 'acceptability' of arguments, as a crude means of assessing arguments received from other agents. A preliminary discussion of a small example was presented in (Parsons and Jennings, 1996), which was then widely expanded and refined. One question that was not asked of that work concerns the accuracy of the characterisation of acceptability classes; it is this question that is posed by White in this volume. By providing a formal characterisation of the Parson & Jennings system, White constructs an analysis that critiques and refines the notion of acceptability. A related task is undertaken by Cayrol and Lagasquie-Schiex in this volume, tackling Dung's framework directly, and introducing a more fine-grained classification of acceptability classes.
have been submitted to it. One alternative that is available for more subtle assessment draws on the field of rhetoric. Traditionally, rhetoric and argumentation have been somewhat at odds, often misrepresented in the two camps. Often, the more formal tendencies of argumentation theorists and informal logicians have been at odds with the more hearer-centric approach of rhetoricians. More recently, however, work by authors such as Tindale (1999) has sought to close the gap. As a result, rhetorical theory has found its way into MAS research. Ramchurn et al., in this volume, seek to integrate various types of rhetorical appeals into inter-agent argumentation, and show further that such argumentation has computational benefits. This echoes work in argumentation theory by Woods and Walton (1989) and much subsequent writing by Walton alone, showing how the traditional fallacies of the ad baculum, ad populum and so on, have legitimate roles to play in particular forms and contexts of argumentative discourse. Where Woods and Walton have made these claims in theoretical work, Ramchurn et al. approach them experimentally.
By itself, however, acceptability - even when refined - is a rather broad-brush technique available to an agent that needs to assess or evaluate arguments that
Guerini et al. in this volume tackle the problem of persuasion head-on, integrating not just rhetoric, but also social and cognitive psychological models. The
Although perhaps the least computational of the philosophical and social components of argumentation, the field of rhetoric has had a number of applications in AI, including Quintilian and Cicero in (Reed and Long, 1997), Perelman in (Grasso, 1998) and many others in (Crosswhite et al., 2003). Perhaps the most obvious domain of application of computational theories and models of rhetoric is at the human computer interface, where the aim in many cases is to produce language tailored to the user at hand.
domain is similar to Oberlander et al.'s (2000) ILEX system, focusing on dynamic information presentation in a museum, but the language generation involves a much richer set of rhetorical heuristics for structuring the text. Carenini, in this volume, takes a similar rhetorical approach to the problem of producing carefully tailored persuasive text. Where Guerini et al. focus on general rhetorical techniques for producing effective argument - in essence, formalising a subset of the Aristotelian topoi - Carenini instead is driven by precise modelling of a specific audience, much more in the relativistic style of Perelman (Perelman & Ohlbrechts-Tyteca, 1969). Both Guerini and Carenini examined the role of rhetorical argument in effecting change (of belief or action) in general for a user. One special application of such change is in education, where argument can be used as a device for explanation as much as for persuasion. Horacek, in this volume, explores the problem of providing approriate hints to students in the context of deductive argumentation. But applications of argumentation to computer aided learning do not stop at such a 'monological' view of argument (albeit one that may include the implicit contraposition of Johnson's dialectical tier (Johnson, 2002)). As Yuan et al, in this volume, have demonstrated, theories of dialogue also have a role to play in structuring educational discourse. They show how Mackenzie's (1979) dialectical system DC can be used not just as normative specification, as was intended, but as an operational definition of a computer-mediated communication system oriented towards pedagogy. They go on to identify the strengths and weaknesses of the system from a purely experimental point of view. Perhaps one of the oldest uses of argumentation and argument-like structures in AI is in knowledge representation. In an elegant paper, Lin and Shoham (1987) demonstrate that many popular methods of handling nonmontonic structures, including default logic and circumscription, can be recast as structures of argument. One popular method of characterising the relationships expressed in argumentation is by using probabilistic structures (Zukerman, et al., 2000). Green, in this volume, adopts the approach for analysing medical text, and demonstrates that it is a useful tool for argument reconstruction. This marks
a significant refinement of the more typical, qualitative approach to reconstruction characterised by van Eemeren et al. (1993), for example. As an alternative to probabilistic models, nonclassical logics, have become extremely popular, particularly in specify agent behaviour. Boella et al. describe an approach that is founded upon the popular BDI approach, but that is extended to handle normative statements. They show that argumentation has a role to play in both representation and decision making in the context not only of beliefs, but also of desires, intentions and obligations. The role of argumentation in AI systems is extremely broad, exploiting results in monological and dialogical theory, in conflict-based dialectic, in persuasion and rhetoric, and in the structures of informal logic. It crops up in knowledge representation, belief revision, defeasible reasoning, dialogue modelling, inter-agent communication, computer-based learning, natural language generation and more. But perhaps this is to be expected; the study of how humans reason alone, how they express their reasoning to others, and how they reason collaboratively, is perhaps a cornerstone for work in both modest practical applications of AI technology, and in grander, theoretical, GOFAI research. Maybe the surprise is that argumentation research wasn't recgonised and exploited earlier. But it is quite clear that argumentation theory is a major influence as a collaborator and stimulator of AI research, and that that influence is set to increase. References [Dung, 1995] Dung, P. M., "On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games", Artificial Intelligence 77, 321--357, 1995. [van Eemeren et al., qqq] van Eemeren, F.H., Grootendorst, R., Jackson, S. & Jacobs, S. "Reconstructing Argumentative Discourse", University of Alabama Press, 1993. [Crosswhite et al., 2003] Crosswhite, J., Fox, J., Reed, C.A., Sclatsas, T. & Stumpf, S. "Computational Models of Rhetoric", in Reed, C.A. & Norman, T.J. (eds) Argument and Computation, Kluwer (to appear). [Grasso, 1998] Grasso, F. "Exciting Avocados and Dull Pears: Combining Behavioural and Argumentative Theory for Producing Effective
Advice", Proceedings of the 20th Annual Meeting of the Cognitive Science Society (COG-SCI'98), Madison, WI, 1-4 August 1998, pp. 436-441.
[Perelman & Ohlbrechts-Tyteca, 1969] Perelman, Ch. & Ohlbrechts-Tyteca, L. The New Rhetoric, Notre Dame Press, 1969.
[Johnson, 2002] Johnson, R., Manifest Rationality, LEA, 2002.
[Tindale, 1999] Tindale, C. Acts of Arguing: A Rhetorical Model of Argument, SUNY Press, 1999.
[Lin and Shoham, 1987] Lin, F, and Shoham, Y.. "Argument systems: a uniform basis for nonmonotonic reasoning". Proceedings of the 1st International Conference on Principles of Knowledge Representation and Reasoning, 245-255, Toronto, Canada, May 1989. Morgan Kaufmann.
[Reed, 1998] Reed, C.A. "Dialogue Frames in Agent Communication" Proceedings of the 3rd International Conference on Multi Agent Systems (ICMAS98), IEEE Press, Paris pp246-253. 1998.
[Mackenzie, 1979] Mackenzie, J. "Begging the Question in Dialogue" Journal of Philosophical Logic 8, 117--133, 1979. [McBurney and Parsons, 2002] McBurney, P. & Parsons, S., "Games that agents play", Journal of Logic, Language and Information 11 (3), 315-334, 2002. [Oberlander et al., 1998] J. Oberlander, M. O'Donnell, A. Knott and C. Mellish. "Conversation in the museum: experiments in dynamic hypermedia with the intelligent labelling explorer." New Review of Hypermedia and Multimedia, 4, 11--32, 1998. [Parsons and Jennings, 1996] Parsons, S.D. & Jennings, N.R. "Negotiation through Argumentation", Proceedings of ICMAS'96, 1996.
[Reed and Long, 1997] Reed C.A. and Long, D.P. (1997) "Content ordering in the generation of persuasive discourse". Proceedings of the 15th International Joint Conference on Artificial Intelligence (IJCAI97), Morgan Kaufmann, Nagoya, Japan pp1022-1027, 1997. [Walton and Krabbe, 1995] Walton, D.N. & Krabbe, E.C.W., Commitment in Dialogue, SUNY Press, 1995. [Woods and Walton, 1989] Woods, J. & Walton, N.D., Fallacies:Select papers 1972-1982, Dordrecht, 1989. [Wooldridge, 2002] Wooldridge, M. An Introduction to Multi-Agent Systems, John Wiley, 2002. [Zukerman et al., 2000] Zukerman, I., McConachy, R. & Korb, K. "Using argumentation strategies in automated argument generation" in Proceedings of INLG2000, 55--62, 2000.
