Profiles of dialogue: a method of argument fault diagnosis and repair.

• Author: Walton, Douglas
• Position: Report

Studies of arguments can be heuristically grouped into those that describe how actual arguments are advanced and those that normatively prescribe how arguments should function. The descriptive approaches often seek to identify fallacies or some other problem of logic or communication. Such analyses may examine news reports, Internet sources such as Debatepedia, legal trials, parliamentary debates, or congressional transcripts. Alternatively, abstract normative models of argumentation have been built to formalize structures with dialogue rules (protocols) that stipulate what kinds of moves (speech acts) are permissible and obligatory in sequences of argumentative exchanges. Normative argument analysts are increasingly turning to the field of computer science to encode and graphically represent arguments. There are now developments in artificial intelligence and multiagent systems to model complex exchanges where arguments are both advanced and contested. This paper shows how to apply the profiles of dialogue method to the texts of real arguments with the goal of bringing the descriptive and normative approaches closer together.

In the informal logic textbooks, fallacies were traditionally taken to be arguments that are inherently fallacious by virtue of their structure or relations between claims and premises. There has been a gradual revolution in the subject starting with (Hamblin, 1970) that such arguments can sometimes be reasonable. Proving beyond doubt whether or not an argument can properly be diagnosed as defective is the key issue facing argument analysts. Formal dialogue models provide part of the answer by using the context in which the argument occurs in modeling the conditions under which an example of argumentation is not fallacious. These models provide rules to determine whether the dialogue proceeds in the correct order, whether arguers ask necessary preliminary questions, and whether arguers' commitments are properly established. Using the method of profiles of dialogue developed in this paper, an interpretative reconstruction of a naturally occurring argument can be derived using evidence from the text and compared to the requirements of the abstract, formal model. I will show how formal argumentation systems found in artificial intelligence can be used to evaluate arguments and to support the profiles of dialogue method. In the next to last section, a twelve-step procedure is presented explaining how to employ the method as a way to help an argument analyst diagnose, and even repair, faulty arguments such as informal fallacies. The last section shows how this method has both normative and descriptive parts.

NORMATIVE AND DESCRIPTIVE APPROACHES

Reed, Wells, Rowe, and Devereux (2008, p. 2) identify the distinction between normative and descriptive aspects of argumentation by differentiating prescriptive normative structures defined by abstract procedural rules for formal dialogue systems from representations of actual arguments in actual dialogue histories. It has also been observed that there is quite a gap between formulations of abstract normative argumentation structures in formal models, and actual practices of argumentation in public debates and political controversies (Bruschke, 2004). These latter kinds of studies tend to be more descriptive and empirical in their methods.

The technique of argument diagramming is a normative method of informal logic. An argument diagram is composed of two elements (Freeman, 1991): (1) a set of nodes (vertices, points) representing the premises and conclusions in the sequence of argumentation in a given case, and (2) a set of arrows (lines) representing inferences from premises to conclusions. Each arrow basically represents an inference. However, the arrows also represent arguments from premises to conclusions. In applying this technique, the analyst designates the nodes in the diagram as premises or conclusions, based on his or her interpretations of the natural language text of discourse. In constructing the diagram, the analyst needs to designate one point that is taken to represent the ultimate proposition to be proved or contested.

Van Eemeren and Grootendorst (1992) advocated the use of the type of dialogue called the critical discussion, which has the goal of resolving a conflict of opinions. The disagreement is set at the opening stage when opponents mutually recognize they disagree. Van Eemeren and Grootendorst also distinguish a confrontation stage in which the participants define the goals of the discussion and agree on the dialogue rules. The rules are meant to represent a way of codifying the Gricean conversational principles (Grice, 1975). In the argumentation stage both parties bring forward arguments to support their viewpoints and have an opportunity to criticize the arguments put forward by the other side. The argumentation stage of the critical discussion is an adversarial type of dialogue in which there is a good deal of strategic maneuvering on both sides, enabling each side to find the weak points in each other's argumentation (Van Eemeren, 2010). The closing stage comprises both parties examining whether one side has been victorious over the other. The dialogue rules offer a normative model of argumentation.

The formal dialectical approach (Walton and Krabbe, 1995), describes a dialogue as an ordered sequence of moves in which participants take turns in making a move containing a speech act. This defines a dialogue as essentially a sequence of pairs of moves by each side that are defined by a set of rules. At each move, depending on the rules of dialogue, participants draw on their commitment stores to insert propositions into, or remove them from, the ongoing dialogue. The rules of a dialogue type define the permissible moves, the types of locutions involved in a move, the regulation of commitment insertion and deletion, and sequences of moves that fulfill the goals of the dialogue.

FORMAL DIALOGUE SYSTEMS

As an advance on the traditional, or standard, treatment of fallacies, Hamblin (1970) took a new dialectical approach that linked fallacies to formal dialogue systems rather than invariant structural forms of message types. Traditionally in the informal logic textbooks, informal fallacies were treated as kinds of arguments that are fallacious. This traditional approach was called the standard treatment of fallacies by Hamblin (1970), who took a new dialectical approach that linked fallacies to formal dialogue systems. This approach had the implication that the forms of argument associated with the traditional fallacies are not necessarily fallacious, leaving us with task of judging in particular cases whether a given argument is fallacious or not. But Hamblin also proposed a framework for how to tell whether or not a given argument in a natural language text is fallacious. He proposed that the context of the argument can be modeled dialectically using formal dialogue systems, and he constructed several simplified formal systems of this sort. Using formal dialogue systems was not altogether new, however, his program of applying formalized mathematical dialogue systems to the study of informal fallacies was a revolutionary step forward. Although he built mathematical models of formal dialogues based on the traditional Greek models, and such systems as the Obligation Game of the Middle Ages, he made no serious attempt to take the next step of classifying different types of dialogues and modeling their properties. That step was subsequently taken in studies by Walton (1989), as well as Walton and Krabbe (1995). This formal dialectical approach advocated that argumentation is based on communication between agents (in most cases, people) in a dialogue structure. In such a structure one agent makes a move, such as putting forward an argument, and the other agent takes a turn responding, such as criticizing the argument. A dialogue is comprised of an opening stage, an argumentation stage and a closing stage. In this approach, six types of dialogue were initially recognized (Walton and Krabbe, 1995): information-seeking, inquiry, persuasion, negotiation, deliberation and eristic dialogue.

Dialogues are formal structures defined by a set of initial conditions, a set of goals for each agent, and collective dialogue goals binding both (or all all) the agents. An agent can be a human or a machine entity, such as a computer program designed to take part in multiagent communications on the Internet. It is assumed that each agent has a commitment store, which is defined as a set of propositions it has gone on record in the dialogue as accepting. Each stage has rules (protocols) governing the permissible or obligatory activation of a speech act put forward.

For example, in a persuasion dialogue the proponent has a thesis (claim) that it needs to prove by a sequence of logical reasoning in order to win. To win, the respondent needs to cast sufficient doubt on the proponent's proof by undercutting it, or, alternatively, by producing a stronger argument for the negation of the proponent's claim. The collective goal of a persuasion dialogue is to resolve the conflict of opinions expressed at the opening stage by testing the one argument against the other. The global burden of proof, called the burden of persuasion in a persuasion dialogue, requires that the proponent has a thesis to be proved. The default burden of persuasion applicable in a typical case is set by the preponderance of evidence standard, essentially meaning that the stronger argument wins. Meeting this burden of persuasion is determined by coming up with a strong enough argument using a chain of argumentation in which individual arguments in the chain follow the normative rules of the dialogue type (Prakken, 2006). Determining which argument is stronger requires an evaluation of the entire sequence of argumentation on each side. A subsequent evaluation of the...