1) As a valid rule of our legal system, killing another human being is prohibited and punished with 20 years imprisonment.
2) Jane has been found by this court to have killed another human being.
Therefore
C)Jane has done something prohibited and is to be punished with 20 years imprisonment.
1b) Valid rule: For all x, if x kills a human being, then the punishment is 20 years.3
1c) For all x, if x kills a human being AND does not act in self-defence AND does not act under duress AND is not insane, then the punishment is 20 years.
In particular, these approaches distinguish between a strict implication, formally represented as «A → B» and interpreted like the implication of classical logic, and a weak or defeasible implication, represented as «A ~ B» for «A weakly or defeasibly implies B». Informally, we can think of the former as saying: «If A, then definitely B». The latter could be rendered in a number of ways, each with subtly different shades of meaning (and possibly as a result subtly different logical behaviour),6 e.g. «If A, then normally B», «If A, typically B», «If A, then unless stated otherwise, B» or «If A, then assume B until proven otherwise».
Arguments that try to establish if a specific norm is best understood as introducing a strict implication or a weak implication, and if the latter, what exactly the meaning and scope of the «~» is, are difficult to represent adequately in argumentation logics. In logical terms, this is because they evoke the meta-theory of the logic under consideration, or maybe even the informal rules of translating a text into its logical form. The problem here is that we only know that «~» is supposed to represent something like the natural language «it generally follows» when we look at the semantic meta-language of the formalism, or the informal explanation and motivation. Distinguishing object- and meta-language issues is of course a common and necessary technical device in formal logic. Unfortunately, in legal discourse object and meta-language issues often appear side by side, and any more or less artificial separation between the two means that some aspects of a legal decision can’t be expressed any longer, or not expressed adequately.
The effect can be mitigated, of course. We can for instance introduce new argumentation schemata that take the rules of statutory interpretation as their antecedent. In the case of the «exclusio alterius» rule, we could then use this new premise to «undercut» an argument that reasons for making an exception. However, what gets lost in this analysis is that the issue is strictly about meaning and appropriateness of the «~» in the legal rule under discussion. If it were possible to enrich our language so that it can talk directly about the meaning and extension of defeasibility (while avoiding the inconsistencies that often arise when object and meta language are conflated) then an entire class of legal arguments could be represented more faithfully.
2.
«That’s it, folks» ^
So far we have shown how we can express some of the underlying issue through a logic with defeasibility.
1) As a valid rule of our legal system, killing another human being is normally prohibited and punished with 20 years imprisonment.
2) Jane has been found by this court to have killed another human being.
Therefore
C) Jane has done something prohibited and is to be punished with 20 years imprisonment.
Formally
1* ∀x KilledHuman(x) ~ Punished(x)
2* KilledHuman (Jane)
C* Punished (Jane)
In the defeasible systems discussed above, 3 is derived from the premises because there was no argumentative move by the opposition that defeated the premise. To understand this though, we have to look at the meta-theory of the proposed logic that explicates the meaning of «~»
In a series of influential papers, Richard Holton has proposed a different reconstruction. According to him, ethical or legal arguments are subject to a that’s it premise that states that locally, all relevant exceptions have been taken into consideration. A formal rendition of our initial argument in Holton’s approach looks in first approximation like this:
i. ∀x(KilledHuman(x) ∧... ∧ Fn(x) ∧ That’s it → Punished(x))
ii. KilledHuman(Jane) ∧...Fn(Jane)
iii. That’s it
(C) Va
We now have added a condition (iii) that states that the exceptions to the general rule that are explicitly stated (the conjunction of conditions … ∧Fn(x)) are also the only exceptions that need to be considered – a direct formal representation of the «exclusio alterius» rule of statutory interpretation. Intuitively, the universal premise is now true. In every case of a potentially falsifying instance of (1), there is a defeating reason. That is, if a killing is not wrong, then there is a reason for why it is not – that it was an act of self-defence, for example. So in that case That’s it was falsely asserted, and it follows logically that the relevant instance of (i) is true.
What is the logical structure of That’s it arguments? We will call «simple arguments» arguments the type of problematic argument with false universal premise with which this paper started. Every simple argument is uniquely specified by the name of the agent or action (in our case, Jane), the predicates F1, . . . , Fn occurring in the singular premise (In our example, killing someone) and a predicate V occurring in the conclusion (in our case, the 20 years punishment) . Let s(a, F1, . . . , Fn, V ) refer to the simple argument from (1) and (2) to (C) displayed above. Likewise, every that’s it argument is uniquely specified by such a name and such predicates. We shall use t(a, F1, . . . , Fn, V ) to refer to the That’s it argument from (i), (ii) and (iii) to (C) displayed above.
Definition 1 : t(a,F1,...,Fn,V) supersedes t(b,G1,...,Gm,V′) if
(i) F1(a) ∧...Fn(a) entails G1(b) ∧...Gm(b)
(ii) G1(b) ∧...Gm(b) does not entail F1(a) ∧...Fn(a)
(iii) V(a) and V (b) are incompatible
(Typically, the second condition is only satisfied if a = b.)
Supersession is thus both a matter of the logical relations between the predicates involved in the arguments, and of the properties had by the individuals involved. We obtain a plausible example by taking a to be an individual who killed in self- defence, b = a, and F1, F2, V and V ′ to be «killed», «acted in self-defence», «is guilty», and «is not guilty», respectively.
Constraint 1: An occurrence of That’s it in the argument t(a, F1, . . . , Fn, V) is true if there is no sound argument superseding t(a, F1, . . . , Fn, V).
Since definitions need to be non-circular, on the orthodox view, we are calling this biconditional a «constraint». It certainly does constrain the truth-values that occurrences of That’s it receive in arguments, given an interpretation of the rest of the language. What we do not know, before further investigation, is whether there is always a unique assignment of truth-values to That’s it that satisfies the constraint. If (and only if) it did, the constraint would arguably deserve to be called a «definition».
At this point, we depart from Holton’s formal account while trying to remain true to the spirit of his argument. To ensure that premise 3), the that’s it proposition, is not rendered false just because there have been some cases where self-defence was accepted as an exception, but expresses instead the idea that in the case under consideration, self defence has not been raised, we have to be able to quantify inside the That’s it proposition. To enable us to do so, we shall introduce a family of That’s it predicates. Given a stock of basic predicates in the language, these can be taken to be complex predicates. Specifically, whenever F1, . . . , Fn as well as V are basic predicates,
then TvF1 ,...,Fn is a complex predicate. To a first approximation, TvF1 ,...,Fn is true of an agent x if and only F1,...,Fn cover between them all the facts legally or procedurally relevant to whether V applies to x. In terms of argument supersession, it is natural to modify Constraint 1 as follows:
Constraint 2. TvF1 ,...,Fn is true of x if there is no sound argument superseding t(x,F1,...,Fn,V).
This account allows us to quantify into the That’s it clause. On this revised account, a That’s it argument has the following syntactic form:
(1) F1(a)∧...Fn(a)
(2) ∀x(F1(x)∧...∧Fn(x)∧Tv F1,...,Fn(x) →V(x))
(3) Tv F1,...,Fn(a)
(C) V(a)
From now on, we will use «t(a, F1, . . . , Fn, V)» to refer to such an argument, rather than one in which That’s it contains no quantifiable variable.
Such arguments are easily seen to be valid:
F1(a)∧...∧Fn(a)∧TF1,...,Fn(a)→V(a))
follows from (2) by universal instantiation, and together with (1) and (3) entails (C).
The modification avoids the problem of false universal premises. The instance of the scheme concerning Jane’s killing looks as follows:
(1) Jane killed someone.
(2) For all x (x killed someone ∧T guiltykilled (x) → x is guilty)
(3) Tguilty killed(Jane)
(C) Jane is guilty.
The problem with the earlier version of this argument was that if the that’s it premise (3) is true, then premise (2) is false. Or in other words, as soon as an exception to the general rule has been argued successfully anywhere, it also has to be considered in the specific case at hand. This allows us to understand the evolution of the law and its tendency to add new exceptions to established rules over time. But crucially it does not account for the observation of stasis that we described above, and that legal decisions remain valid even if an exception is argued in a later case, or if an available exception was not argued. With other words, it did not really resolve the problem that often, legal decisions seem to be based on a prima facie false universal rule. In this version, in contrast, premise (2) may well be true even if there are other cases, such as say the case of a John Doe who killed in self-defence:
3.
Discussion ^
The resulting logic is capable of handling interesting and important examples of legal reasoning. In some respects however, it is in the version introduced above not capable of correctly analysing a class of legal arguments that can be represented convincingly in argumentation frameworks. Our approach allows us to explain why it can be the case that John is acquitted or murder while Jane is convicted, even though the legal rule that justifies both decisions simply states that killing is prohibited: John successfully pleaded an exception, self-defence, that Jane did not. However, what about the situation where John did act in self-defence (so that the exception is in principle available to him) and he also raises it in a procedurally correct way, and his defence is nonetheless unsuccessful because his actions were deemed excessive? Intuitively, we have then an exception from an exception which, if argued by the prosecution, reinstates the original guilty verdict.
- 1 Work on this paper was supported by the Arts and Humanities Research Council (grant number AH/M009610/1).
- 2 For a comprehensive informal discussion of defeasibility in law from a jurisprudential perspective see d’Almeida, Allowing for Exceptions: A Theory of Defences and Defeasibility in Law. Oxford University Press, Oxford 2015.
- 3 For the purpose of this paper, we do not distinguish practical syllogisms, that is syllogisms where the universal remise is a normative statement and the conclusion a mandate for action, from a syllogism simpliciter. For our purpose this distinction does not matter and could confuse the discussion.
- 4 See e.g. Prakken, Incomplete arguments in legal discourse: a case study. Proceedings of the JURIX 2002 conference. IOS, Amsterdam 2002, pp. 93–102.
- 5 See e.g. Prakken, Logical Tools for Modelling Legal Argument. A Study of Defeasible Reasoning in Law. Kluwer Dordrecht, 1997; Prakken/Reed/Walton, Argumentation schemes and generalisations in reasoning about evidence. Proceedings of the Ninth International Conference on Artificial Intelligence and Law, Edinburgh 2003. ACM, New York 2003, pp. 32–34; Bench-Capon, Argument in artificial intelligence and law. Artificial Intelligence and Law 1997, Volume 5, pp. 249–261; Governatori, On the relationship between Carneades and defeasible logic. Proceedings of the 13th International Conference on Artificial Intelligence and Law, Pitsburgh 2011, ACM, New York 2011, pp. 31–40.
- 6 See also Prakken/Sartor, The three faces of defeasibility in the law. Ratio Juris 2004, Volume 17, pp. 118–139.
- 7 See e.g. Prakken, Analysing reasoning about evidence with formal models of argumentation. Law, Probability & Risk 2004, Volume 4, pp. 33–50; Bex/Prakken/Reed/Walton, Towards a formal account of reasoning about evidence: argumentation schemes and generalisations. Artificial Intelligence and Law 2003, Volume 11(2-3), pp. 125–165.
- 8 For an example see Andrus v. Glover Const. Co., 446 U.S. 608, 616–617, 27 May 1980 (citing Continental Casualty Co. v. United States, 314 U.S. 527, 533, 5 January 1942)
- 9 In the US, this is made explicit in art 416.1489 of the Code of Federal Regulations.
- 10 Horty, Argument construction and reinstatement in logics for defeasible reasoning. Artificial intelligence and Law 2001, Volume 9, pp. 1–28; Prakken, Intuitions and the modelling of defeasible reasoning: some case studies. arXiv preprint cs/0207031 2002, https://arxiv.org/abs/cs/0207031 (accessed 12. January 2016).