Jusletter IT

During his distinguished career, Erich Schweighofer managed to balance academic research interests in legal AI and the formal semantics of legal reasoning with practical, high impact work for the Agriculture and Rural Development Directorate of the European Commission. It is therefore only appropriate to try to bring these two spheres of his intellectual life together, in an attempt to provide a formal analysis of the semantics of a European Union regulation, and the particularly heated debate between member states, eventually resolved by the ECJ, that it created. This paper attempts to provide such an analysis, while at the same time arguing for the benefits that can be achieved through such a formal, philosophical analysis for the project of European legal harmonisation.

The example that forms the centre of this paper are the attempts by the European Union to create legal protection for geographical indications (GI), and in particular the debate concerning the designation «Feta», contested between Greece, Germany and Denmark [2,8,13]. We develop a simple formalism that displays possible disambiguation of the semantic of the relevant norms and the court decision that settled the debate. This will allow us to show that the court in its reasoning took up themes and ideas that are also discussed in analytical philosophy, in particular the discussion on the semantics of natural kind terms following the work of Saul Kripke [9].

Before we begin with the analysis though, some preliminary discussions about the background of the discussion within the logic community are necessary. Classical logic, the logic that is still in one way or the other builds the foundation of most AI systems, is an extensional language. That means that two terms that refer to the same object (have the same extension) can be replaced in all contexts without changing the truth value of the sentence. If Peter is taller than Paul, and Paul = The serving president, the Peter is also taller than the serving President. While in mathematics and the natural sciences, most contexts that are of interest for AI researchers are extensional and can be adequately modelled using classical logic, there are numerous examples in natural language that violate that principle, many of them of particular interest for law. If Peter e.g. attempts to kill Paul, we may not be entitled to replace «Paul» by «the serving president», since Peter may not be aware of the fact that the two are identical. The legal implications are obvious: depending on the legitimacy of the substitution, we may be able to charge Peter for both attempted murder and treason, or only for attempted murder. Modal terms such as «believes that», «intends to» or «promises to» introduce referential opacity into language, and mere co-extensionality is not enough to guarantee that the truth value of a statement remains the same after substitution of co-referential terms. Since the 1950s and the work by Saul Kripke, intensional contexts like this are normally analysed within the framework of a «possible worlds semantics». To permit substitution, it is not sufficient that two terms contingently refer to the same object inthis world, they also have to have the same referent in all relevant alternative worlds. In the case at hand, if Peter knows that Paul is the serving President, then Paul and «the serving president» refer to the same individual in all possible worlds that are consistent with what Paul believes. In this case, substitution is permissible – if he attempts to kill Paul, he also attempts to kill the serving President.

This elegant and intuitive solution raised however immediately several new problems. One of them is the issue of transworld identity: how do we «pick out» Paul and «the serving President» in other possible worlds to see if they refer indeed in all of them to the same individual? Saul Kripke, inNaming and Necessity , invites us to consider the following thought experiment [9]: Imagine a French citizen, Pierre. Pierre, proudly French and monolingual, believes on the basis of some films he saw the following: «Londres est joli.» («London is beautiful.») Much later, Pierre moves to London without realizing that London = Londres. In England, he immerses himself in British culture and learns English the way a native speaker would learn it, without translating from French. Pierre lives at this time in a very unattractive part of London, which he now associates with the word «London», so he comes to believe that London is not beautiful. According to Kripke, Pierre now believes both a) that London is «joli» and b) that London is not beautiful. Are there now two Londons in our possible worlds? Or just one – but then which one? Is Pierre holding a contradictory belief? Part of Kripke’s answer was to argue that proper names behave in an interesting way differently from other linguistic entities. They are «rigid designators» that pick out in every possible world the same object [3]. While the referent of «the British capital» might change (it could have been York, or Lancaster) and hence pick different objects in different worlds, «London» and «Londres» pick in all possible worlds the same object, that is London.

As an obiter, Kripke also considers that terms other than proper names may be candidates for rigid designation, in particular natural kind terms such as «gold» or «water», and the list has been expanded by other analytical philosophers [4,6]. Molino defines natural kinds as «a family of entities possessing properties bound by natural law; we know of natural kinds in the form of categories of minerals, plants, or animals, and we know that different human cultures classify natural realities that surround them in a completely analogous fashion» [10]. It is in particular the last part of Molino’s analysis that makes natural kinds an issue for the European Union and the common market, as we will see later.

Hilary Putnam developed Kripke’s idea further, again in the form of a thought experiment, the «Twin earth» experiment [11; 4]. Assume that there is a planet exactly like earth in virtually all respects, our ‹Twin Earth›. On that Twin Earth, we can find a Twin equivalent of every person and thing here on Earth. However, there is one small difference. In place of water, on Twin Earth there is a liquid that while superficially identical to H2O, is chemically different, which we express by assigning it the chemical formula ‹XYZ›. The Twin Earthlings, in a language they call ‹English› call XYZ ‹water›. Let us now assume in addition that neither earthlings nor twin-earthlings have yet discovered the chemical make-up of their respective liquids, that is they can’t of know that the liquids they called ‹water› were H2O and XYZ respectively. The way earthlings experience H2O, and twin-earthlings experience XYZ is hence identical. For Putnam, this raises the following question: when an earthling and his twin on Twin Earth say ‹water› do they mean the same thing? Again, part of the argument is the intuition that water is a rigid designator, and therefore picks the same object in each possible world, and that it is hence necessarily H2O [11,12,7]

Umberto Eco finally gave this though experiment a twist, that makes its relevance for legal discourse, and European law in particular, abundantly clear [5]. He send astronauts on a mission to boldly go where no man has gone before. During their journey, they reach Twin Earth, and, firmly believing in the doctrine of rigid designation, drink a liquid which the natives of the planet call «water». Unfortunately, it turns out to be hydrochloric acid, and the explorers find an untimely end.

Does this thought experiment show that Putnam’s analysis is so counterintuitive that it should be abandoned? Whatever we think of Eco’s planet, it is for sure not a «deontologically ideal» alternative to our world. Unfriendly natives fraudulently bringing the lives of heroic explorers at risk? We surely can’t have that! And so, we would expect in «real» worlds the law to prevent this kind of disaster. Concerns like this, if a bit less drastic, are indeed the objective of the European Commission regulation 1829/2002and a whole family of related legislation, administrative decrees and court judgements of a similar kind [1,13]. If we now think of possible worlds in a deontic context as jurisdictions, then we can make a direct connection between Kripke’s and Putnam’s thought experiments, and the structure of legal rules that prescribe a uniform word usage in a European context. What these norms have in common is that they «rigidify» the use of words in the labelling of foodstuff, that they prescribe specific expressions as names for specific products.

Commission regulation 1829/2002 contains a specific kind of normative statements: norms which prescribe or prohibit the use of specific expressions for certain products. It inter alia prohibits the use of the name «Feta» for cheese not made from goat’s and sheep’s milk. What distinguishes this norm from more conventional legal regulations is that they do not prescribe so much an action (under a description?) but rather a description (for an action?). That this kind of norms has not attracted more interest is surprising, given that this phenomenon is much older and much more widespread than our examples might imply. Trade mark protection is another major field in which descriptions and names are «rigidified» – don’t sell «Coca-Cola» if it is not Coca-Cola, or Irn Bru if it is not Irn Bru, or you face some of the best paid lawyers in the world. A considerable number of what Herbert Hart calls «secondary rules» too mandate specific language use: they confer titles to judges and policeman, and names to courts, government departments or regions within a state. Sometimes they even decree the use of specific words for an oath or the national anthem. As secondary norms they give structure to a legal system, and form the background against which primary norms are interpreted, they (should) form an important part in any systematic formal analysis of legal argumentation.

In what follows, I will use very loosely (quasi)formal representations of the norms in questions, not primarily in order to develop a new formalism, but to help the reader in understanding which parts of the norm are problematic, where ambiguities arise and generally which part of a norm is discussed in a given paragraph. The use of these structural expressions therefore helps the flow of the argument, but the emphasis is nonetheless not in syntactical structure, but on the semantic theory of certain expressions.

Before we give a formal account, we need a bit more background information about the problem that the norm tried to address [see 2;13]. In the 1990s, Greek Feta producers realised that their export of Feta cheese to other European countries was plummeting. When they investigated the reason, they found that this was not because of a drop in demand for Feta cheese. Rather, much more products were sold under the label «Greek Feta cheese» than produced in Greece at that time. Even more outrageous, more cheese was sold than the production of sheep’s milk in Europe could possibly allow. In fact, most of the Greek Feta cheese in circulation was produced in Germany and Denmark, using cow’s milk instead of the original recipe.

If Putnam’s fictitious explorer had instead of looking for the stars decided to explore the nearby neighbourhood travelling through Europe, what would their findings have been? One answer could have been: «Feta is not necessarily sheep’s cheese», as an empirical finding. This was indeed the opinion of the German Government, to which we will return below. They argued against the Greek claim that far from using an inappropriate expression to refer to cow’s cheese, the Greek had become victim of an ambiguity. As good analytical philosophers, the German representatives proposed an analytical distinction: «Feta» as a word of the German language would refer to both sheep and cow’s cheese, as long as it had the same flavour. «φετα» on the other hand might well be an expression of the Greek language, and designate sheep’s cheese only.

The commission was not impressed. Arguably, the fact that on the packaging of «German Feta», people in Greek costumes would dance in front of interested looking sheep did not help their case. Consequently, the Commission maintained that a) Feta is (de-re) sheep’s cheese and that b) this was the case in all member states. German manufacturers as a result simply lied when selling their product as Feta.¹ .The Greek Government applied as a result of their findings for a «protection of origin order», a means available to them under European consumer protection law. To gain the required protection, they had to show that consumers buying «Greek Feta» would normally assume that they would get indeed Greek Feta, and with that a product that a) is physically different from other products on the market and b) owes this difference to a form of production/raw materials specific to a geographical region. In that case, they were (initially) able to argue that both conditions were met, and the registration of Greek Feta as a sheep milk product from Greece was approved.

This created a new norm: You ought not to sell something as «Feta», unless it is Feta. What logical form does this norm have? Let us first contrast it with a more familiar form of norm, a norm requiring a simple action, for instance:

We might formalise the first in classical possible world systems as

But to analyse the second sentence, we need to find a way to represent the syntactical structure of the «as»-qualification, and give it a semantic interpretation. One might be tempted to use again modal logic:

But this is in every respect a poor solution. It departs considerably from the syntactic structure of the natural language sentence, and introduces a strange «Feta-object», which furthermore has no logically connections to the property of «being Feta». But even if we disregard this for the moment, it could not work. Assuming that «Feta» is an adequate translation of the German «Schafskäse» (and hence Feta = Schafskäse a necessary analytical truth), it would follow from 2b) that if you sell something as «Schafskäse», it has to be Feta. But this isnot what the regulation requires from you. To make this clearer, let’s go back to Putnam’s explorers. If you insist on producing mineral water containing sources of a dangerous substance, e.g. hydrochloric acid, your customers might want to know. Likewise, to insert a label in Gaelic if you intend to sell them in Italy won’t do. Indeed, consumer protection regulation of the European Union would rule out something like this as well.² This consumer protection norm has the same surface structure as the regulation under discussion so far. But if the Gaelic word and the Italian are (ex hypothesis) translations of each other, then they designate the same individuals inall possible worlds. This however means that they are substitutable in classical modal logics.

As mentioned above, we are dealing here with the problem that legal norms are typically language-relative. To preserve the meaning of the norm, and to preserve the syntactical structure of the natural language expression, we have to enrich our logic. A very natural way of doing this is to introduce a formal quotation operator. We add to our vocabulary a new function constant, the quotation function «», and require syntactically that

As long as we do not allow quantifying into the brackets, this should not pose any problems. (This becomes relevant if we think of the obligation for member statesto develop a classification system e.g. for dangerous substances. This might well create the obligation , say for Germany, to introducesome single expression for HCl (e.g. «HCl», «Salzsäure», «Echt giftiges Mistzeug» etc.) so that every producer is under the obligation to use this and only this word in labelling his product)

For the time being, we might then re-formulate 2b)

This quotation function seems to be a necessary addition to our formal instruments in any case, and it has applications above and beyond the very specific problem discussed here. Are we now safe? This depends on some crucial semantic choices. The problem now shifts to the interpretation of the sentence «Fx». Let’s assume we have opted for a semantic with «nested domains». This would mean that «Feta» is not a rigid designator, that there is no non-trivial necessary truth connected with it. If this were the case, it would mean that «Feta» can have different referents in different possible worlds, and from the outset, this seems to be a rather innocent assumption. «The Greeks could have used cows instead of sheep» seems to be a rather innocent statement about possible events in the past. However, this would mean that we can think of a possible world where is fulfilled (and hence, this world is a deontologically preferable alternative to one where it is violated), and nevertheless, what they call «Feta» in this world is cow’s cheese. This however is exactly what the regulation prohibits. It seems that we must assume that «Feta is sheep’s cheese» is a necessary truth, and hence, that Feta is a rigid designator.

With the quotation function available to us, we can avoid this problem,if we are prepared to treat legal norms as linguistic entities. The deontic operators would then become predicates that take specific objects, quoted expressions, as arguments. To see norms as linguistic entities comes rather natural to the lawyer, or so I would argue. We typically deal with books, court reports, statutes, and learn to present cases in court, and to draft contracts and submissions. Not only that, but legal normas are typically «linguistically relative». That is to say, two norms of different legal systems can be perfect translations of each other, and are nonetheless not identical. «A contract made under duress is void» is a norm of Scots contract law, but not of German law, even though its perfect translation: «ein unter Drohung gemachter Vertrag ist nichtig» is. But as there is a norm that states that all norms in German law have to be in German, the two must be distinguished.

If we can therefore accept that legal norms are linguistic entities, we can introduce formulations of the kind:

«Ought» is a one place predicate, which takes a linguistic object, the quotes sentence «S(‹F›,x) ( Fx» as argument. The norm is de-dicto in the sense that it does not presuppose the existence of Feta, nor does it allow substitution of co-referential or even synonymous expressions. Normally, we would assume from our discussion that this is the structure of legal norms proper, expressing the idea that they are language-relative in the sense discussed. The relative German and Greek norms about Feta and φετα are no less contradictory then Pierre’s de-dicto believes about Londres and London

However, European law is different from these de-dicto norms of the individual legal systems, as it operates in a multi-language environment. EU law is not only higher in hierarchy of norms – the legal aspect of transnational law – but also speaks about the individual legal systems which it comprises in a uniform language – the meta-linguistic or philosophical aspect of transnational law. In taking an outsider position towards individual legal systems, it is able to «see through» the referential opacity created by the quotation marks, in the same way as we as observers can see that Pierre means «in reality» the same city. The Commission decreed that feta-de-re is cheese made from sheep’s milk, and that therefore labels like Feta and φετα must be used only to name sheep’s cheese. So while national implementations of the norm can be represented as de-dicto norms, the meta-level EU norm is best understood as a de-re norm, or in fact a bundle of de re norms, one for each language in the EU. Every norm in this bundle is a de-re norm, about Feta as an object, not the word or description «Feta». We can achieve this result by moving the definition of Feta outside the «» bracket. Technically, this makes it necessary to quantify into the brackets, also allowing us to express sentences such as: Peter thinks of Paul(de-re) under the description «my wife’s lover» that his shot will kill him, but he does not believe of Paul under the description «serving President» that his shot will kill him. Similarly, we can express the idea that it is illegal to sell something that is not Feta (de-re) under the expression «Feta».

To be able to express this formally, we need two different types of variables, de-re and de-dicto variables and . The first type behaves just like the variables of classical logic and takes only objects-de-re as arguments, the other take linguistic entities as object. Consequently, a sentence of the form «xo Poxo» is true if we can find a name which when substituted for the bound variable yields a true sentence. Similarly, each predicate carries a type = « »or «o», indicating whether its argument position is interpreted referentially ( ) or substitutionally (o).

With 6 and 5, we now have a de-re and a de-dicto version of norms that prescribe the use of specific expressions. 6) in particular comes closest in our reconstruction to the logical form that represents the reasoning of the Commission when protecting as «Feta» cheese that , «in reality» or de-re, is Feta.

For the purpose of this paper, I will give only a very short version of the formal semantics, insofar it deviates from classical logic.

Semantics

A universe for L is an arbitrary set D. For each d D we assume a fictitious object name d D which is either an object parameter of L or a new symbol which is not element of L.D be the set of these names. We get the formulas and terms of LD from the terms and formulas of L by substituting object names for one or more free object variables. (the additional names are not part of the language of L but of the interpreted language QDL They are meta-theoretical tools to simplify the semantic rules for the quantifiers. Therefore, they do not appear in de-dicto positions, where the linguistic forms of the expressionsas used by a speaker play a role) Expressions without object names are called pure expressions.

Unlike standard interpretations, we do not interpret the predicate parameters, but the elemental de-re predicates.

Let Pe1,...er be a pure formula in basic notation, P predicate parameter of type t1,...tr, so that for each i = 1,...r:

Let x1,....xk be the alphabetically first free de-re variables in Pe1...er (0 k r). Then we call this formula anelemental predicate in x1,...xk and introduce the metatheoretical abbreviation Ek . From Ek we derive En1,...nk by substituting terms ni for xi

Aninterpretation for L overD is a functionf that fulfils the 4 conditions:

The truth-values of elemental sentences are determined by the following rules

From the rules for complex sentences, only the rules for quantification differ from the usual definitions, and are therefore given:

Assigning formal structures to European Union norms provides a plethora of interesting challenges. Many of the preconditions for formal analysis that are taken for granted when analyzing the laws of individual nation states are absent, or in need of revision. This paper focused on but one of these aspects, the ability of EU norms to «talk about» the norms of member states in their respective languages. To model this feature, we had to enrich the formal language with a quotation function that allowed us to preserve the multilingual character of EU law, not in order to «solve» questions like the one the ECJ faced in the Feta decision, but to be able to represent more faithfully the respective arguments, and to see more clearly what linguistic intuitions were driving them. We saw that all parties in the debate drew on intuitions which in this form have also been discussed in the literature on possible world semantics and cross world identity in analytic philosophy. Subtly different intuitions gave rise to different formal representations of the same underlying norm. This is only the first step for a more detailed and more systematic «parallel reading» of analytical philosophy and legal argumentation on the issue of natural kinds. The hopes for such a study are twofold: In cases like the one discussed here, the law has self-reflexively to account for its own understanding of the «meaning of meaning». What, in legal terms, does it mean to use a term? This in turn can lead to a richer, more rigorous understanding of some of the perennial questions of legal theory. Can we for instance generalize our discussion to legal terms? Are they too rigid designators, and do they too display the systematic ambiguities that we encountered in the Feta example? Secondly, by providing a formalism with very straightforward computational properties, and, unlike many previous examples of disquotational logics for intensional contexts, a well motivated formal semantic, the paper also hopes to make a small contribution to the continuing task to develop appropriate AI tools for reasoning in multi-lingual domains.

BEIER, F., The need for protection of indications of source and appellations of origin in the common market. in: Industrial Property (35) 1977
DAVIS, A., Case report Denmark v. Commission. in: Consumer Law Journal (7) 1999 p. 243-244
DONELLAN, K., Reference and definite description. in: Philosophical Review 75 (1966)
DONELLAN, K., Kripke and Putnam on Natural Kind Terms. In: Ginet, C. (ed.) Knowledge and Mind. Oxford University Press, Oxford 1983
ECO, U., Travels In Hyperreality, Harvest Books, New York 1990
ELDER, C., Realism, Naturalism and culturally generated kinds. in: Philosophical Quarterly 39 (1989)
FUSTEGUERAS, A., On the semantics of natural kind words. in: Theoria (11) 1996 pp. 143-159
KNAAK, R., Case Law of the European Court of Justice on the protection of Geographical Indication and Designations of Origin Pursuant to the EC Regulation No 2081/92›, in: ICC 32 (2001)
KRIPKE, S., Naming and Necessity. Basil Blackwell, Oxford 1980
MOLINO, J., Toward an Evolutionary Theory of Music and Natural Language, The Origins of Music. MIT Press, Cambridge, Mass:, 2000
PUTNAM, H., The meaning of «meaning». in ibid. Philosophical papers. Cambridge University Press, Cambridge 1975
PUTNAM, H.: Natural Kinds and Human Artefacts. In: Mind (91) 1982
SCHWAB, B., ‹The Protection of Geographical Indications in the European Economic Community›, in: E.I.P.R. 17(1995)

---

Burkhard Schafer, Professor of Computational Legal Theory, University of Edinburgh, School of Law, SCRIPT, South Bridge, Old College, Edinburgh EH8 9Yl UK,b.schafer@ed.ac.uk ;www.law.ed.ac.uk/staff/burkhardschafer_69.aspx

---