Introduction
In the previous, first post of this series of posts on the Structured Consultation Tool (SCT), we outlined the motivation for the SCT. In this second post, we discuss argumentation and a particular form of argument that is central to policy-making, Practical Reasoning (reasoning about what to do), along with some elements of the formal representation. The objective is to introduce participants in the User Consultation Board to elements of argumentation that are particularly relevant to the project. The topic could be discussed at much greater length, but in the post we present the high points. In the final post of this series, we will provide some indicative screen shots of a SCT prototype to give a sense of how users will work with the proposed tool.
In policy-making, arguments are central since, given the deliberative context of the consultation, contributors respond to some point of the proposed legislation either by arguing for or against that point, or providing alternatives (which may or may not be construed as incompatible). The arguments may take a range of forms such as giving reasons against a point, giving a definition, adding a premise, identifying anomalies, giving a counter-example, or stating conditions under which the rule is inapplicable, among others. While contributors are aware that they are deliberating, they do not usually systematically address issues raised by other contributors, much less formalise the arguments as might a logician so as to enable further reasoning over the responses. By the same token, without some formalisation, further automated processing for reasoning is infeasible. The latter is rather important given the sheer amount and complexity of information users can submit. As most contributors are not trained logicians or computer scientists, they cannot be expected to provide systematic, formal, machine-readable arguments. Given this, we must attempt to bridge the gap between the deliberative inputs that the respondents provide and the systematic, formal representations that can be used for further automated processing such as for reasoning. To this end, the IMPACT Project and the SCT use a formal theory of argumentation using Argumentation Schemes.
In the following, we give an overview about arguments and argumentation schemes using familiar examples to give a flavour of the main ideas. We initially outline familiar notions of deductive and defeasible arguments. We then introduce Argumentation Schemes (AS), which are accessible, prototypical, defeasible reasoning patterns. As we point out, ASs are useful for they provide fine-grained information about what users of the SCT agree with or disagree with about the policy under discussion; this is in marked contrast to ePetitions, which are all-or-nothing, or to other policy tools that do not structure the arguments, but leave this to manual analysis after the data has been gathered.
Arguments
To clarify some of the main issues that are being addressed in the SCT and the IMPACT Project, let us first briefly review some foundational issues in logic. Our general point here is that where we can translate reasoning in natural language into a formal representation, we can then take the further step of reasoning with that formal representation quickly, systematically, and transparently over large volumes of information, which might otherwise be beyond any one individual's reasoning ability. In effect, we have proposed that what has been done for arguments in standard logic can (and should) also be done for important arguments in policy formulation. However, there are several points to make along the way.
First, let's review what an argument is in logic. Arguments generally are understood as inference patterns - premises and rules from which we infer a claim. In Classical Propositional Logic, sentences as wholes are considered as the basic component: the sentence Jill is happy can be represented as the proposition P, where we do not consider the structure of the sentence or its parts. Similarly, suppose Bill is happy is Q, and the rule If Jill is happy, then Bill is happy is P -> Q. We have reasoning patterns with premises, a rule, and a conclusion, for example, the inference pattern that logicians call Modus Ponens:
(1)
Premise: Jill is happy
Rule: If Jill is happy, then Bill is happy
Claim: Bill is happy.
We formalise this, maintaining the reasoning pattern. Formalisations may be taken, for our purposes, as templates that need particular values to be assigned to the variables (slots in the template structure).
(2)
Premise: P
Rule: P -> Q
Claim: Q
The reason for the symbolic form is that it allows us to see reasoning at a level of forms, patterns of reasoning, rather than strictly in terms of the particular content of sentences. Any pair of sentences can be substituted in for P and Q in (2). Moreover, we have programming languages in which we can express such symbolic forms and reason with them. This is an example of a deductive argument, in that where the premises are true and given the rule, the claim must follow; no additional statements can change this inference.
Note that in the IMPACT Project, no one other than the system developers need be aware of the formal level of representation; our point is only to show and justify this in the scope of the project. In particular, end users of the survey (as we see in the next post) only see natural language expressions. However, as they are associated internally with a formal representation, we can further process them.
In Classical Predicate Logic, matters are more complex since we do represent some aspects of the particular structure of sentences, yet we find similar reasoning, deductive patterns:
(3)
Premise: Socrates is a man.
Rule: All men are mortal.
Claim: Socrates is mortal.
We can translate the elements of the sentence into formal expressions of the logical language: verbs (is mortal, is a man) - predicates (mortal', man'), nouns (Socrates) - arguments (socrates'), and quantifiers (Some, All) - quantifiers (Forall).
(4)
Premise: man'(socrates')
Rule: Forall x [man'(x) -> mortal'(x)]
Claim: mortal'(socrates')
This reasoning pattern can be abstracted to a general form. Where P and Q are any predicates in the logical language, and we instantiate x for any individual, we have the argument:
(5)
Premise: P(x)
Rule: Forall x [P(x) -> Q(x)]
Claim: Q(x)
We know that we can translate from these reasoning patterns expressed in natural language into symbolic reasoning patterns that a machine can reason with. However, there is a problem that is highly relevant to policy formulation. In particular, policy formulation contains arguments that can be defeated by further information, meaning what we had inferred we can no longer infer.
For example, suppose that in (1), we find out that Jill is poor; while logically it would still follow that Bill is happy, we as reasoners might not accept this inference, say where Bill is a bit of a snob. Moreover, debates may have conflicting or contradictory information: though one person might assert all the statements in (3), someone else might counter that Socrates is not a man, but a woman, in which case it does not stictly follow from the rule that Socrates is mortal. Furthermore, some reasoning patterns resemble (5), but do not license the same inference:
(6)
Premise: Jill is a woman.
Rule: Most women are happy.
Claim: Jill is happy.
While we might presume that on balance the claim follows from the premise and rule, it is easy to imagine a situation where Jill isn't happy; she happens to be an exception to the generalisation. Logicians call the pattern in (6) Defeasible Modus Ponens. In this pattern, the presumptive claim can defeated by a counter-example. Finally, in a deductive logic, if we have even one contradiction, anything at all can be inferred, so reasoning is rendered useless. We need other ways of representing reasoning to address these sorts of problems.
Argumentation Schemes
So far, we have only considered deductive arguments, those arguments where from given premises and rule, the claim must follow. We have seen that there are arguments that are not deductive, which are referred to as presumptive or defeasible arguments; that is, we have arguments from which we presume we can infer the claims unless and until we receive information that suggests otherwise. Given that in the real world there are very few rules that have no exceptions, most reasoning in most contexts is presumptive reasoning; in policy formulation, this is very likely to be so. In addition to this general class of arguments, we are also interested in more particular forms of arguments, especially those that can be used to represent arguments about policy. Such particular forms are called Argumentation Schemes, which are patterns of reasoning that permit claims to be drawn defeasibly. Such schemes are associated with conditions characteristic of the scheme, which are normally true, but if false will lead us to withdraw the presumptive claim.
As we said at the onset, Argumentation Schemes in policy formulation are very useful for they provide clear, fixed, and fine-grained discussion points that users of the SCT may specifically agree or disagree upon. In this way, the SCT returns very specific information to the policy analysts about exactly what respondents object to. Unlike the all-or-nothing approach of ePetitions, users of the SCT can agree with some portions and disagree with others.
Here, we consider two schemes, Practical Reasoning and Expert Opinion. Practical Reasoning relates to determining what people should do in a given situation, which is often central to policy-making consultations; Expert Opinion is what is often used to back up or support particular premises of an argument. We outline each of these schemes in terms of two levels levels of representation, as they appear in natural language and as they appear as a template; the natural language version standing as an instantion of the schema.
The following is an example derived from an ePetition on fox hunting in the United Kingdom. We only represent the premises, leaving the rule implicit. Where the premises are true, the claim presumptively follows (as indicated by using should).
(7)
Premise 1a: The ban on fox hunting negatively affects the livelihoods of those who make a living from fox hunting;
Premise 2a: Repealing the ban on fox hunting creates more jobs in the countryside;
Premise 3a: Creating more jobs in the countryside promotes prosperity.
Claim: We should repeal the ban on fox hunting.
We may make a schema for Practical Reasoning, where we have variables that need to be assigned values. For instance, suppose that R = The ban on fox hunting...., A = Repealing the ban on fox hunting, G = creates more jobs in the countryside, and V = prosperity.
(8)
Premise 1a: The current circumstances are R;
Premise 2a: Doing action A realises goal G;
Premise 3a: The goal G promotes value V;
Claim: We should do action A.
As in the move between (1) and (2), we have created a formal representation that can then be implemented in a program and automatically reasoned with (more accurately, further formalisation is required than given, but this is not relevant to our purposes).
The Expert Opinion argumentation scheme may be used to argue for or against a particular statement of another scheme, which we give as a template and then as an instantiated example.
(9)
Premise 1b: E is an expert in subject domain S
Premise 2b: S contains proposition A;
Premise 3b: E asserts that it is true that A;
Claim: A
We connect our argumentation schemes -- the claim of this Expert Opinion argument is Premise 1a of the previous Practical Reasoning argument. Other premises of the Practical Reasoning argument might also find support from an expert. For illustration, we use made up individuals and domain knowledge.
(10)
Premise 1b: Professor James is an expert on UK rural economic research.
Premise 2b: UK rural economics research contains the proposition that the ban on fox hunting negatively affects the livelihoods of those who make a living from fox hunting.
Premise 3b: Professor James asserts that it is true that the ban on fox hunting negatively affects the livelihoods of those who make a living from fox hunting.
Claim: The ban on fox hunting negatively affects the livelihoods of those who make a living from fox hunting.
Given these arguments, one might be persuaded to repeal the ban on fox hunting.
Alternatively, one might object to particular statements within the arguments, thereby denying that the presumptive claim -- that the ban on fox hunting should be repealed -- follows. Such objections relate to the conditions under which the scheme can properly be used; they are often presented as questions which, if answered negatively, represent objections to a statement in an argument. Objections stand as attacks on arguments such that if the objection stands, then the argument is defeated; that is, the claim that was presumptively implied no longer holds.
For instance, one might object to Premise 1b of (10), claiming that Professor James is not an expert on UK rural economic research; one might then support this claim by showing that he has not been a member of any professional research organisation for 10 years and has no qualification. Or, one might object to Premise 3a, citing research that jobs which are created in the countryside are so low paying that they are only marginally better than government support, and thereby do not promote prosperity. Note in particular, that the argumentation schemes provide clear, fixed, and fine-grained discussion points, such as those concerning current circumstances, actions, goals, values, expertise, domains, and so on; objections are directed at these points specifically. It is this aspect of argumentation that structures and makes coherent the debate about policy. In this way, it returns very specific information to the policy analyst about exactly what respondents object to.
There is an additional layer to the overall processing of the arguments, where we evaluate a large network of arguments and the statements for or against particular statements. However, as the SCT in the IMPACT Project does not carry out such evaluations, we leave this aside for the time being.
In this post, we outlined the SCT's conceptual technology - connected and analysed Argumentation Schemes. In the next post, we present an overview of a prototype of the SCT.
