(no subject)

[gusl]

Erik Krabbe about normative persuasive argumentation and the role of computer systems
from Erik Krabbe - Symposium on Argument and Computation
Group: Argument and Computational Societies Position Paper

In Walton and Krabbe (1995: 175), four features are proposed as desiderata for what are there called "dialogue systems" and here "models for the persuasion dialogue":
(1) a model should be realistic (i.e., descriptively accurate);
(2) it should have normative bite (i.e., one should be able to use the model to effectively criticize behavior in arguments: not anything goes);
(3) it should be rigorously formulated (the ideal being the mathematical preciseness of a "formal dialogue game");
(4) it should be easy to apply to ordinary contexts of arguments (the applications include tasks of interpretation, evaluation, and criticism).
It is then shown that various models conform to various desiderata to a different extent. There is a "large gap here, with on the one hand theories of dialogue that are mathematically precise but not very realistic [Hamblin, 1970, 1971; Lorenzen&Lorenz, 1978; Mackenzie, 1979, 1981, 1985, 1990; Barth&Krabbe, 1982; Hegselmann, 1985; Walton&Krabbe, 1995] and on the other hand theories that have no pretence at mathematical precision but fare better when assessed for descriptive accuracy and applicability [Grice, 1975, Van Eemeren and Grootendorst, 1992]. We do not think that a choice should be made. A good theoretical foundation of informal logic [or argumentation theory] needs both mathematically precise models and readily applicable models that are formulated in ordinary language. The precise or rigorous models are to elucidate the concepts used by the theory that formulates the second type of models, which can be looked upon as a kind of interface between the mathematical models of dialogue and actual dialogues." (Walton&Krabbe, 1995: 176.)


4. Complexities, Profiles and Computational Societies

The construction of formal dialectic systems (or, formal models for persuasion dialogue) that are to cover some significant section of real life argument, and thus to elucidate some useful concepts of argumentation theory, is a highly complicated and arduous task. As one struggles along, one gets the idea that a simpler approach could more readily yield results: it should be possible to give rather precise specifications of some theoretical notions, such as concepts of different types of commitment, or of various fallacies, without going all the way of defining a complete system. Here the method of dialogue profiles, first prominently used by Walton in his discussions of the spouse-beating question (1989a: 68-69, 1989b: 37-38), and after used by both of us on a number of occasions, provides an attractive alternative. "Profiles of dialogue are tree-shaped descriptions of sequences of dialectic moves that display the various ways a reasonable dialogue could proceed. With reference to such profiles, one may discuss various dialectic fallacies and critical moves without having to go through all the technical preliminaries necessary for the complete definition of a dialogue system. Moreover one may gain insight in the constraints that should be imposed upon the rules of such a system." (Krabbe, 1992: 277.)

It may be apparent from the last sentence in this self-quote, that it was not my intention to abandon the construction of full-fledged formal dialectic systems. On the contrary, the use of profiles was seen as a means to bridge the gap between our intuitions and hunches on one side and formal specifications on the other. Thus the study of profiles may prepare the way towards full-fledged dialectic systems.

It seems to me that those who work in computing science on the problem of specifying protocols for multi-agent systems that include argumentation face the same problem of being confronted with an overwhelming complexity. Their articles, too, often provide specifications of concepts on an intermediate level of formality. But, since machines do not understand informal concepts, they must have retained an interest in ultimately achieving full-fledged formal specifications. In these respects their position parallels that of the formal dialectician. So, maybe we can all work together on profiles of dialogue!

To end this paper, I want to stress where formal dialectic (the construction of models for persuasion dialogue) may hope to get some help from computing science. This concerns the crucial matter of testing formal dialectic systems. For, how does the formal dialectician know that the system he or she proposes has the properties such a system is supposed to have, and no properties it is not supposed to have? One has to think things through and to diligently debug the system, and then hope that no inconsistencies or anomalies remain, and that if unhappily there remain any such flaws, that someone sometime will point them out. The only other test is whether the transfer to informal models of argumentation shows the system to be a fruitful one, or whether the system is useful in explaining or illustrating a number of concepts that need to be used informally. But a system may work fine, and nevertheless contain anomalies! Here computer science can be of great service, both by strengthening the debugging process as implementations of a dialectic system are actually tested in multi-agent environments, and by providing correctness proofs of various kinds. Argumentation theorists may look forward to these developments.