University of New South Wales Law Journal
DR SIMON EVANS[*]
Categorisation is fundamental to legal decision making. In constitutional law, jurists must regularly decide whether or not laws and facts fall within the categories defined by the Constitution: is this law a law ‘with respect to … taxation’? Is that corporation a ‘trading corporation’? Is performance of this function an exercise of ‘judicial power’? Is that nation a ‘foreign power’? For the most part, however, judges and legal advisers approach categorisation unreflectively. They observe that their practices work; they assume that their practices are coherent. In this article, I draw on the extensive literature in psychology and cognitive science to encourage jurists to rethink some of their assumptions about how they go about the business of categorisation.
In particular, I argue that what judges do is often rather different from what they say that they do. The standard semantic model that Australian judges use, in constitutional cases to resolve categorisation questions, is the connotation-denotation model. That model rests on several propositions. One is of particular interest here. That is the proposition that the meaning of a constitutional term (its connotation) is determined by a set of necessary and sufficient features. A person, purpose or activity falls within the meaning of the term if, and only if, it possesses all of the ‘essential features’ or ‘essential characteristics’ that define that term. I shall argue that the semantic model that judges actually use frequently differs from this simple proposition. In particular, they often employ definitions that do not consist of ‘essential features’.
One response to the divergence between rhetoric and practice might be to characterise the divergence from the standard model as a deviation from orthodoxy and argue that judges should adhere more faithfully to the standard model. However, I shall argue that departures from the standard model are often inevitable and appropriate. For all its appeal – apparent simplicity, stability and roots in classical philosophy – the ‘essential features’ element of the standard model is unworkable. It does not represent how we think or speak. It cannot provide answers to many of the questions of categorisation that arise in constitutional law. Judges should, therefore, abandon the search for ‘essential features’. Instead, they should recognise that departures from the standard model are entirely defensible. Categorisation does not in fact require ‘essential features’. In ordinary cognition and language use, categorisation regularly depends on exemplars, prototypes and explanations or theories that identify and explain the meaning of concepts without reference to ‘essential features’. So it does also in constitutional categorisation. There is no need to regret the inability to identify ‘essential characteristics’ if a constitutional concept can be adequately stabilised using these other techniques of everyday categorisation.
The standard semantic model articulated by the High Court in Australian constitutional law is the connotation-denotation model. The following description of the model by Dawson J, in Street v Queensland Bar Association, is typical:
The essential meaning of the Constitution must remain the same, although with the passage of time its words must be applied to situations which were not envisaged at federation. Expressed in the technical language of the logician, the words have a fixed connotation but their denotation may differ from time to time. That is to say, the attributes which the words signify will not vary, but as time passes new and different things may be seen to possess those attributes sufficiently to justify the application of the words to them.
This passage identifies five key propositions involved in the connotation-denotation model.
1. The meaning of a term (its connotation) is distinct from its application (its denotation).
2. The connotation of a term was fixed at the time the Constitution was enacted.
3. The connotation of a term is determined by the essential features (‘attributes’ for Dawson J) of the term.
4. The denotation of a term can change.
5. The denotation of a term at any time after the Constitution was enacted is determined by the range of things that at that later time possess the essential features or attributes specified in the connotation of the term.
The distinction between connotation and denotation (or meaning and application) is firmly entrenched; and, indeed, according to McHugh J, sufficiently entrenched to render irrelevant the views of philosophers who regard the distinction as ‘outdated’. I do not enter this debate in this article. Nor do I enter the debates about originalism and non-originalism, or about the possibility of constitutional evolution – whether the connotation of a term is fixed at 1900 or can change in light of contemporary circumstances.
Rather, my exclusive focus in this article is the third proposition in the list above, that the connotation or meaning of a term is determined by the essential features or attributes of the term. This proposition is the core of the standard semantic model in Australian constitutional interpretation. It appears in constitutional and non-constitutional contexts throughout the decisions of the High Court. The use of this principle in Attorney-General (NSW) ex rel Tooth & Co Ltd v Brewery Employees Union of New South Wales, was summarised by Kirby J in Grain Pool of Western Australia v Commonwealth. Justice Kirby described Justice Isaacs’s view on the process of interpretation of ‘trade marks’ in section 51(xvii) of the Constitution as ‘as a search for the “really essential characteristics”, the “universal element”, the “fundamental conception” or the “essential particulars”’, and not the ‘procedural or substantive incidents of “trade marks” as they had developed in England up to 1900’. Justice Kirby noted that Higgins J described interpretation as a search for the ‘essential differentia’.
More recently, the search for essential features has been explicitly invoked in interpretation of ‘trial … by jury’ (s 80); ‘duties … of excise’ (s 90); ‘aliens’ (s 51(xix)); ‘trade marks’ (s 51(xviii)); ‘writ of … prohibition’ (s 75(v)); representative and responsible government; ‘judicial power’ (s 71 ); ‘trade and commerce … among the States’ (s 51(i)); ‘industrial disputes’ (s 5 1(xxxv)); and other constitutional terms. It appeared prominently in the reasoning in Brownlee v The Queen (which I discuss in Part IV) and Grain Pool of WA.
Thus, the meaning of words and concepts in constitutional interpretation has long been, and remains, identified with a set of essential features. It should be noted, however, that although this approach to identifying part of the connotation is the standard semantic model in constitutional interpretation, it is not universally applied. Sometimes, it is asserted that the particular concept has a nature that is not amenable to a classical definition. For example, in Resch v Federal Commissioner of Taxation, Dixon J regarded it as inapplicable to political concepts, in that case ‘subjects of taxation’:
[Section 55 of the Constitution] is concerned with political relations, and must be taken as contemplating broad distinctions between possible subjects of taxation based on common understanding and general conceptions, rather than on any analytical or logical classification.
Such concepts did not have an essence that could be determined by the Court in the usual manner. However, Justice Dixon’s preference for analytical and logical classification is clear. As will be seen in Part IV, there are other instances where the Court does not faithfully employ the standard approach to identifying the meaning of words and concepts. But even in those instances there is sufficient homage paid to the normativity of the ‘essential features’ approach to continue to identify it as the standard approach.
The ‘essential features’ approach is even more clearly the standard or classical approach to categorisation outside the legal context. Before I explore the origins and some of the implications of this approach, however, it is necessary to introduce some terminology that I will use in discussing categorisation. I distinguish between terms (or words), properties (or features), concepts and categories.
• A category is a set of things, real or abstract.
• A concept is a mental representation of a category. The category is the set of things which the concept represents.
• A term (or a word) is a symbol used to identify a concept. The concept is the term’s meaning.
• A feature (or property) is an attribute, quality or characteristic of a thing.
An example may assist. The category HORSE includes all horses, living and dead, historical and mythical. The concept horse is an individual’s mental representation of the category. The term ‘horse’ then serves to identify the concept horse when a speaker attempts to communicate the concept to a hearer. A particular horse may have the feature, property or attribute ‘brown’ or ‘bay’.
This terminology does not say anything about the relationship between properties and concepts, or, in other words, between properties and the meaning of terms. In particular, it does not assume that meaning is defined in terms of properties or features.
The classical approach to categorisation can be traced to Aristotle’s Metaphysics. Aristotle distinguished between the essential features of a thing and its merely accidental features. The essential features are inherent in the thing. They are what makes it a thing of that kind – a member of a particular category. The accidental features are incidental to making it a thing of that kind. For example, the essential features of the category ‘horse’ might include the attributes ‘animal’ and ‘quadruped’. On the other hand, it might be an accidental feature of a particular horse that it is brown – although a particular horse (and some others) has the feature ‘brown’, it is not necessary to it being a member of the category ‘horse’. On this approach, the definition of a term is a list of necessary and sufficient features. The definition of the concept horse (equivalently, of the term ‘horse’) will include the features ‘animal’ and ‘quadruped’, as both features must be present in order for a thing to be a horse; but the definition will not include the feature ‘brown’.
It is important to note that, as Taylor puts it, ‘[f]eatures are a matter of all or nothing’. A thing either has a feature or it does not. A thing has the feature ‘animal’ or it does not (it cannot be ‘half animal’ or ‘three-quarter animal’). A feature is involved in the definition of a category or it is not. The definition of horse, for example, involves the feature ‘animal’ or it does not.
Two things follow on this approach. First, categories have clear boundaries. A thing is a horse or it is not and there are no ‘almost’-horses. Second, the members of a category have equal status. All horses are equally horsey – no horse is a more typical horse than any other. As a result, the classical approach to categorisation allows for clear judgments about whether or not a thing is a member of a category, and about whether or not a term that represents a category has been correctly used to refer to a thing.
There is a clear link between the classical ‘essential features’ approach to categorisation and the constitutional model that identifies the connotation or meaning of a term with the essential features of the concept to which the term corresponds. Why do these logical and legal semantic models converge in this way? A partial explanation lies in the functional similarities between concepts and rules. The reasons for the use of concepts in cognition are similar to the reasons for rule-based decision making. Rosch argues that humans often categorise common information in order to extract ‘maximum information with the least cognitive effort’. In law, Sunstein argues
[r]ules … are often simple summaries of good decisions in individual cases. In carrying out this task, they reduce costs, ease choice, limit the errors encountered in particular decisions, produce coordination, and make it unnecessary to debate issues of value and fact every time someone does something having social consequences.
Rules and concepts each simplify a universe of particulars by grouping instances that are relevantly similar.
Moreover, rule-based decision making is a particular application of human cognitive capacities. It differs from some (composing free-form verse, driving a car) in that categorisation is a central and essential aspect of the activity, rather than simply an incidental aspect of a cognitive process that is directed at some other end. Nonetheless, these similarities between rules and concepts, noted above, suggest that a convergence of semantic models is not unsurprising, though of course they do not suggest convergence on the classical model unless that is the model used in ordinary cognition.
Why then might there be convergence on the classical model rather than on any other model? There are at least four reasons.
Psychological essentialism. Both semantic models reflect what cognitive scientists have described as ‘psychological essentialism’. This is the ‘traditional and pervasive view about concepts … that things have essences that are deeper and more basic to a kind’. People commonly believe that the features of things(in particular natural kinds) are accounted for ‘by some “essential nature” which the thing shares with other members of the natural kind’ whether or not they can identify that essence. If psychological essentialism is a pervasive phenomenon, it is unsurprising that semantic models in legal and non-legal domains converge in the way demonstrated by the classical approach to categorisation and the legal model that identifies connotation or meaning with essential features.
Metaphysical essentialism. The classical model also makes the stronger claim that concepts do in fact have an essence, not merely that many people act as if they do. This essence is independent of language use and human intentions. This coheres with classical legal thought – a conception of the judicial role in which judges do not themselves create the boundaries of legal categories but rather expound boundaries that already exist prior to adjudication. Category membership is determined by criteria that are independent of the idiosyncratic opinions of the judge. A semantic model that assumes that the concepts invoked in legal decision making have an essence, and are real and external to the judge, is more readily justified (and, therefore, more likely to be adopted) than one that denies this.
Bright lines. Not only are the criteria of category membership under the classical model external to the decision maker, they define clear boundaries. A thing is or is not a member of a category. This law either is or is not a law with respect to taxation; that corporation either is or is not a trading corporation; and so on. These judgments need not invoke a vaguely defined criterion of similarity to determine category membership. Hence, the classical semantic model could be seen to advance the legal system’s aspiration (particularly in classical legal thought) for definition and certainty.
Context independence. Finally, the classical model tends to define concepts abstractly, independently of the situations in which instances of the concept are found. This is capable of reinforcing the notion that legal categories are the creatures of the (impersonal) law and not the idiosyncratic creations of individual judges responding to particular cases before them.
Some of these reasons for the convergence of the two approaches also suggest why the classical ‘essential features’ approach is a desirable methodology for judicial decision making. But, desirable or not, whether the classical model can provide a solution to the legal problem of categorisation depends on whether it can deliver definitions in terms of essential features or necessary and sufficient conditions.
In fact, the classical approach has proven remarkably unfruitful in delivering definitions. It has some successes, particularly in producing definitions of natural kinds (for example, odd number, prime number) but many more natural kinds lack this kind of definition (for example, bird, fish). The difficulties are even more pronounced for artefacts (for example, hammer, bridge) and social categories (for example, jury, trading corporation). Even old chestnuts like bachelor, which is said to have four defining features, ‘human’, ‘male’, ‘adult’ and ‘never married’, prove problematic on closer inspection. For example, is the Pope a bachelor? Standard usage allows one to say that he is hardly a typical member of the category. But classical categories do not have more and less typical members. On the classical model a person is a bachelor if he possesses all of the defining features and he is not a bachelor if he does not – there is no question of typicality.
Often when a classical definition can be produced, it is not a useful definition. It is possible to define human as an animal with two additional features – ‘featherless’ and ‘bipedal’. But that tells language users very little about what it is to be human. Sometimes the problem lies deep in the elements of the definition: some features that look like abstract primitives, and, hence, are capable of being elements of a classical definition, turn out to be complex social constructs. (‘Never married’ as a feature in the putative definition of bachelor makes sense only in a society where there is a social institution of marriage.) In other cases, the problem is that the primitive features that determine meaning are accessible only to experts and not to ordinary language users.
Moreover, the stability and clarity that the classical model promises can prove fleeting when new artefacts and social categories emerge over time. To take an example from the constitutional realm, is it ‘a characteristic feature and an essential feature’ of ‘telephonic, telegraphic and other like services’ that ‘any member of the public [has] the right to avail himself of them for both sending and receiving’? That they are ‘secret or private’? The answer may plausibly differ before and after the invention of radio broadcasting. The teleological reconciliation achieved by the High Court in R v Brislan; Ex parte Williams is well known:
In dealing with such a question it must not be forgotten that it is a constitutional power intended to provide for the future and bearing upon its face an attempt to cover unknown and unforseen developments. A wide operation should be given to such a power.
But this reasoning stretches the commitment to essential characteristics that the classical model requires.
In the second half of the 20th century, the classical model of categorisation came under challenge, first from philosophers of language and later from experimental linguists and cognitive scientists. Their work added to the already identified difficulties with the classical model, suggesting that it is not, in fact, the default model used in human cognition.
In his Philosophical Investigations, Wittgenstein demonstrated that some categories are not defined by a set of necessary and sufficient common features on the basis of which their members can be clearly distinguished from non-members. Wittgenstein’s leading example is the concept game identified by the term ‘game’. It is worthwhile reproducing this influential passage in full.
Consider for example the proceedings that we call ‘games’. I mean board-games, card-games, ball-games, Olympic games, and so on. What is common to them all? – Don’t say: ‘There must be something common, or they would not be called “games”’ – but look and see whether there is anything common to all. – For if you look at them you will not see something that is common to all, but similarities, relationships, and a whole series of them at that. To repeat: don’t think, but look! – Look for example at board-games, with their multifarious relationships. Now pass to card-games; here you find many correspondences with the first group, but many common features drop out, and others appear. When we pass next to ball-games, much that is common is retained, but much is lost. Are they all ‘amusing’? Compare chess with noughts and crosses. Or is there always winning and losing, or competition between players? Think of patience. In ball games there is winning and losing; but when a child throws his ball at the wall and catches it again, this feature has disappeared. Look at the parts played by skill and luck; and at the difference between skill in chess and skill in tennis. Think now of games like ring-a-ring-a-roses; here is the element of amusement, but how many other characteristic features have disappeared! And we can go through the many, many other groups of games in the same way; can see how similarities crop up and disappear.
And the result of this examination is: we see a complicated network of similarities overlapping and criss-crossing: sometimes overall similarities, sometimes similarities of details.
Nonetheless, the concept and the term remain useful to language users.
Empirical work presented further challenges to the classical model. Experiments in the 1970s demonstrated that language users agree that some category members are better examples of the concept than others, contrary to the classical model’s stipulation. The work began with natural kinds such as colour. It was later extended to artificial kinds and social practices. In particular, objects and practices that share more features with other members of the category are identified by subjects as more typical of the concept. Typical or central instances are recognised more quickly and more confidently than atypical or marginal instances. Moreover, and again contrary to the classical model’s postulate that category membership is binary, language users regard the boundaries of some categories as blurred.
What, then, can be said to constitute the definition of a concept if it is not a set of essential features? Wittgenstein described the similarities between instances of a concept as ‘family resemblances’. Although it was not possible to define the boundary of the concept, the concept defined by family resemblance was still useful to language users. A language user describes the concept by giving an example or examples (with the rider ‘and other like things’), not so that the listener might deduce the common essential elements that the speaker was unable to identify, but because that is what the concept is. Blurred boundaries are endemic but not fatal to communication and understanding.
Philosophers of language and cognitive scientists have followed Wittgenstein in developing ‘family resemblance’ models to explain the empirical findings about categorisation as a cognitive process. Although these models differ among themselves in significant ways, they all differ from the classical model in a similar respect. Whereas the classical model determines whether something is an instance of a concept by checking its features against a canonical list of essential features, these models determine whether something is an instance of the concept by considering whether a thing is ‘similar’ in relevant respects to some representation of the concept. In general terms, they ask whether there is a family resemblance between the thing and the known instances of the concept. Again following Wittgenstein, these models examine the ‘complicated network of similarities overlapping and criss-crossing’ and ask whether an instance is sufficiently like some representation of the concept to be regarded as an instance of it.
The first of these ‘family resemblance’ models is commonly referred to as the ‘prototype model’. The prototype is not necessarily a member of the category. Rather, in this model, a concept is represented by an idealised instance that is ‘most typical’ of the category. The features of this prototypical instance of the category are determined by considering the features of the already-encountered members of the category. They consist of an average of the features of the members of the category. (In fact, the features of the prototype are a weighted average of the features of the category members. This means that more weight is given to a feature in computing the average if more members of the category share the feature than if few do. So in calculating the prototypical bird, the features ‘has feathers’ and ‘flies’ will be weighted more highly than ‘swims’, because most birds have feathers and fly, while few swim. The prototypical bird is more likely to be a robin than a penguin.)
Whether a new instance is regarded as a member of the category is determined by its similarity – its family resemblance – to the prototype. The more characteristic features a particular instance has, the better that instance fits the category. (If the prototypical bird is somewhat like a robin, a small bird that flies (for example, a starling) is likely to be recognised as more typical of the category than a large bird that runs (for example, an emu).)
Notwithstanding that the prototype model does not depend on lists of essential features, it remains meaningful to speak of features or attributes. ‘Attributes are simply the dimensions along which different entities are regarded as similar.’ It is also consistent with the prototype model to insist that all members of a category possess a particular attribute. This does not lead to the collapse of the prototype model into the classical model because possession of that attribute itself can, on the prototype model, be a matter of degree. Prototypes reflect the fuzziness and gradedness of categories; they ‘contain a richness of sometimes culturally bound detail which, on a strictly Aristotelian view, would have to be regarded as accidental’.
The prototype model addresses some of the empirical deficiencies of the classical model noted above. But it has its own limitations.
First, and most conspicuously, the prototype model (by design) does not define sharp category boundaries. That may reflect how cognitive processes actually operate on concepts but it may be a weakness if the model is to be transplanted to the legal domain.
Second, the weighted averages required to identify the prototype and, hence, to determine whether something is a member of the category depend on already knowing the members of the category. A problem of circularity looms. Prototype theories, therefore, tend to identify categories as ‘natural’, ‘reflecting the natural partitioning of objects in the real world by our perceptual systems’. This may avoid the circularity in relation to natural kinds, but may limit the prototype model’s application to social institutions and practices.
Third, the assumption that graded typicality judgments correlate with graded category membership is problematic. This can be seen most clearly in the case of categories that do in fact have classical definitions. In one study, subjects rated three as being a more typical odd number than 57. It can also be seen in more complex categories. The authors of one study conjecture that ‘typicality judgments may be based on apparent similarity to a prototypical representation, but category membership may be based on an entirely different process’.
The exemplar model of concepts and categorisation developed alongside the prototype model. Like that model, it attempted to explain the typicality effects (that language users regarded some instances of a concept as more typical than others) that the classical approach could not accommodate. It differed from the prototype model in rejecting the idea that concepts were represented by an abstracted prototype that was distilled, or calculated, from the instances of the concept that had already been encountered. Rather, the concept was represented by exemplars – the actual collection of instances of the concept that had been encountered rather than a single averaged representative. Categorisation then proceeds not by comparison of a new instance with the abstracted prototype but with all the exemplars.
The exemplar model can clearly account for the empirical findings that support the prototype model. Its richer encoding of concepts assists it in explaining the empirical results that demonstrate that cognitive tasks involving concepts are context sensitive. In contrast, prototype theories have difficulties with concepts that are polycentric. Connell reports:
[F]or the concept spoon, subjects judge small spoons more typical than large ones, and metal spoons more typical than wooden ones. The family resemblance view would infer that the spoons with the greatest degree of family resemblance (and hence highest typicality) would be small metal ones, and those with the least family resemblance (lowest typicality) would be large wooden ones, with other combinations falling in between. In fact, what Medin and Shoben (1988) found was that [large wooden spoons were considered the most typical with small metal spoons being second most typical]. Family resemblance theory cannot explain this.
This is because the single prototype fails to reproduce information about correlations between features that appear to be relevant to the actual process of categorisation. A difficulty with the exemplar model is the finding that subjects use the central tendencies of categories in cognition after they have forgotten instances they have previously encountered, suggesting that (at least some) central tendencies are calculated as the prototype model suggests.
Following these and other empirical challenges to the predictions of family resemblance approaches, psychologists and cognitive scientists have proposed other models of categorisation. As seen above, a central question for categorisation theories is how they identify category members. On the classical approach, this is straightforward: a thing is a member of a category (an instance of a concept) if it possesses all the defining features of the concept. Prototype and exemplar models, on the other hand, depend on judgments about how similar an instance is to the prototype instance or the exemplars of the concept. Some writers contend that these judgments of similarity are too unconstrained to be considered the basis of categorisation.
One response is to posit facts about the ‘inherent structure of our cognitive mechanisms’:
We need … to recognize … that similarity is vacuous as an explanatory concept because any two objects are similar in infinitely many different ways. What causes our general agreement on the relative similarity of objects is more strongly a function of the inherent structure of our cognitive mechanisms, which more or less automatically selects those dimensions along which objects are compared. That is, it is an efect of categorization not its cause. Once again we must distinguish between the conceptualizations people have about how they classify and the actual cognitive mechanisms responsible for classification; mechanisms that can produce universals in how people classify the biological world.
Others, perhaps more persuasively, argue that categorisation is based on theories (or explanations) about the relationships between category members, their features and the world. Conceptual coherence on such explanation-based or theory-based models of categorisation derives not from possession of a set of necessary and essential features, or from family resemblance, but from the fact that category members ‘each fit our theory of what the category is about’. Common theory-based models hold that theories include beliefs about essences (for example, that birds are warm blooded biological organisms) and explanatory-causal relations between a thing’s features and other observations about how it behaves (for example, that birds use their wings for flying). Theory-based models enrich the prototype model by including explanatory-causal relations between features as part of the concept. Thus, as Komatsu summarises the literature:
[P]eople’s natural kind concepts are not limited to simple lists of attributes, lists of central tendencies on attributes, ranges of values on attributes or sets of instances. Rather, people’s natural kind concepts are (at least when fairly well developed) rather like theories, attempts to explain the distribution of attributes and instances. When these concepts are not well-developed, people’s concepts (reflecting a division of linguistic labor) may simply include the belief that such an explanation exists and that others in the linguistic community (eg, scientists or other experts) may know what that explanation is or are working on uncovering it. … With nonnatural kinds, people’s beliefs may include the idea that there are no experts and no to-be-discovered explanations, but only linguistic conventions.
What does this mean in the legal context?
It is important not to push any analogy between natural kinds and legal concepts too far. Natural kinds and legal concepts are fundamentally different. The structure of the former is revealed by empirical science; the latter is normative. It follows, as Dennis Patterson has recently argued, that while widely-accepted scientific methods can resolve disagreements about the essence of natural kinds, we cannot know whether a theory of a legal concept truly identifies the ‘deep structure’ of the concept that explains its concrete manifestations. There is simply no ‘methodological consensus in morals and politics’ that enables us to reach such a conclusion.
That is not to say that theories of legal and constitutional concepts are impossible. Rather, they are likely to be plural and the subject of disagreement – and of disagreement about how they are evaluated. Theories of legal concepts come in several kinds. Take just two instances. First, a positive doctrinal theory of a constitutional concept might attempt to identify the already-recognised instances of the concept (perhaps its central instances or exemplars) and to identify other instances. It might do so, for example, by identifying the attributes of the concept (which may be neither necessary or sufficient) and the weight to be given to them. Second, a justificatory theory of a constitutional concept might attempt to fit, and provide a normative justification for, the already-recognised instances of the concept. It might do so by reference to deeper (perhaps unwritten) principles of political morality. I provide some examples of such theories in the next Part. The important point is not that adopting a theory-based approach to categorisation will resolve controversies about the content of constitutional concepts. It will not. There will almost inevitably be disagreement both as to whether a theory of the first kind or the second kind is to be preferred and as to which of several theories of either kind is to be preferred. Rather, a theory-based approach can be used to identify theories, each of which fits (to a greater or lesser extent) the known instances of a concept and provides (again, to a greater or lesser extent) normative resources that may be used to identify other instances of the concept in the course of constitutional adjudication.
In Part II(A) above, I argued that the standard semantic model in Australian constitutional adjudication identified the meaning of constitutional terms through a set of necessary and sufficient or essential features. If, as I argued in Part III, that model does not reflect how we think or speak and is incapable of generating useful definitions for terms, it should not be surprising that it does not reflect how judges actually go about the business of constitutional categorisation. Judges use exemplars, prototypes and theories to identify and explain the meaning of constitutional concepts. They often do so with a nod to the classical model, as if to suggest that these approaches to categorisation are deviant. They should instead embrace the other approaches.
In this Part, I consider three instances where there has been some greater or lesser deviation from this element of the standard model. First, I consider taxation and excise, where the High Court has progressively moved away from an apparently classical approach exemplified by the set of essential characteristics of a tax that Latham CJ identified in Matthews v Chicory Marketing Board (Vic). The current approach – which emphasises a set of characteristics of variable weight – leaves a vacuum as to how the constitutional concept of a tax is to be stabilised. Second, I consider judicial power. Here, the High Court has been much more willing to acknowledge the difficulty in defining the concept with a set of essential characteristics. The problem is to give structure and stability to a concept that lacks such a classical structure. Finally, I consider trial by jury. Of the three instances that I consider, this is where the High Court has most clearly invoked the language of ‘essential characteristics’ and equivalent formulae. And yet its non-classical explanation-based or theory-based approach is the clearest (if not entirely satisfactory) demonstration of non-classical approaches to categorisation in constitutional adjudication. It demonstrates an approach that can inform the Court’s approach to other problems of constitutional categorisation.
In Matthews v Chicory Marketing Board (Vic), Latham CJ gave a statement of the essential features of a tax that, for a time, became the canonical definition. A tax is, according to this definition, ‘a compulsory exaction of money by a public authority for public purposes, enforceable by law, and ... not a payment for services rendered’.
In Air Caledonie International v Commonwealth, the High Court described this as ‘an acceptable general statement of positive and negative attributes which, if they all be present, will suffice to stamp an exaction of money with the character of a tax’. The judgment added two further negative criteria to the definition – that the exaction is not by way of penalty and that it is not arbitrary. The semantic model presumed by this statement remains plainly classical.
However, the remainder of the judgment in Air Caledonie suggests a shift away from the classical model. The Court held that the criteria identified by Latham CJ were not necessary and sufficient to establish that an exaction is a tax:
[Chief Justice Latham’s statement] should not be seen as providing an exhaustive definition of a tax. Thus, there is no reason in principle why a tax should not take a form other than the exaction of money or why the compulsory exaction of money under statutory powers could not be properly seen as taxation notwithstanding that it was by a non-public authority or for purposes which could not properly be described as public.
It could be said that it was a necessary (but not sufficient) feature of a tax that it be a compulsory exaction (not necessarily of money). The relationship between the other features was not clear but they were neither necessary nor sufficient. The Court went on:
[T]he negative attribute – ‘not a payment for services rendered’ – should be seen as intended to be but an example of various special types of exaction which may not be taxes even though the positive attributes mentioned by Latham CJ are all present. Thus, a charge for the acquisition or use of property, a fee for a privilege and a fine or penalty imposed for criminal conduct or breach of statutory obligation are other examples of special types of exactions of money which are unlikely to be properly characterised as a tax notwithstanding that they exhibit those positive attributes.
There are three non-classical aspects to this passage. First, the ‘special types of exaction’ category, which carves out a subset of exactions that are not taxes, is open-ended or fuzzy rather than binary. Second, it is defined by exemplar rather than classically (and the definition of tax which depends on it is, therefore, equally non-classical). Third, the fact that an exaction is one of the ‘special types of exaction’ does not mean that the exaction is not a tax; rather ‘it is unlikely to be properly characterised as a tax’. This last point is reinforced in the following passage:
On the other hand, a compulsory and enforceable exaction of money by a public authority for public purposes will not necessarily be precluded from being properly seen as a tax merely because it is described as a ‘fee for services’. If the person required to pay the exaction is given no choice about whether or not he acquires the services and the amount of the exaction has no discernible relationship with the value of what is acquired, the circumstances may be such that the exaction is, at least to the extent that it exceeds that value, properly to be seen as a tax.
Thus, although there are necessary features of a tax, they do not constitute a classical definition in terms of necessary and sufficient conditions. Chief Justice Gleeson emphasised this in Luton v Lessels:
The often-quoted words of Latham CJ in Matthews v Chicory Marketing Board (Vic) were not offered as a definition of a tax. They were an explanation of the features of the impost under consideration that justified the conclusion that it bore the character of a tax. … [The factors identified by Latham CJ had appeared in an earlier decision of the Privy Council but] were not put forward by the Privy Council as a definitive statement of the essence of a tax. In the result, revenue raising, a feature said to be indispensable in Australian Tape Manufacturers Association v Commonwealth, was said to be ‘while ... not a universal determinant ... often ... significant in deciding whether an exaction, or the imposition of a liability, bears the character of taxation’. The picture is equally complex in relation to duty of excise and shows a similar move away from the classical model. At one level, the positive features of an excise, ‘an inland tax on a step in production, manufacture, sale or distribution of goods’, are long established. In Bolton v Madsen, a unanimous court identified the ‘the directness of relation between the tax and the goods’ as ‘the essential characteristic of a duty of excise’. This ‘criterion of liability’ approach is classical in orientation, even if the ‘essential characteristic’ that is identified in Bolton v Madsen is not atomic or primitive. However, the move away from the ‘criterion of liability’ approach towards an approach that looked to the ‘substantial and practical operation approach’ initiated a progressive retreat from the classical model. So, for example, in Hematite Petroleum Pty Ltd v Victoria, it was held that a relationship to quantity or value of the good produced was not an essential feature of a duty of excise. Now, following Phillip Morris Ltd v Commissioner of Business Franchises (Victoria), ‘a range of factors must be taken into consideration when determining whether or not a particular impost amounts to an excise for the purposes of s[ection] 90’. The factors, which are described as ‘indicia’ (implying they have weight rather than being binary), may include:
the proximity of the relevant period to the licence period, the shortness of the licence period, the size of the tax imposed ad valorem and the fact that it is to be borne only once in the course of distribution …
The picture that appears is that tax and excise are not classically defined. In particular, the features which the Court has regard to in determining whether an exaction is a tax or an excise are now said to have ‘significance’ or weight, rather than being conclusive. But although the model is clearly not classical, it is harder to see it fitting within any of the other models described above. The judgments present the categorisation process as a modification of the classical model which still depends on the identification and cataloguing of features, albeit for the purpose of weighing them. Although they assume knowledge of previous cases in which exactions have been characterised as taxes or not-taxes, the recent judgments do not present the categorisation process as one of assessing the similarity of an exaction to (prototypes or exemplars of) previously recognised taxes in order to resolve an indeterminacy at the margins of the concept whose central instances (those prototypes or exemplars) are defined classically.
Perhaps, then, the categorisation process is based on a explanation or theory of the concept of tax or excise in the context of the Australian constitutional order? There is some evidence of a theory of the concept of tax in the decisions of the High Court, particularly in Tape Manufacturers, where the Court had regard to the constitutional provisions establishing parliamentary control over executive expenditure in determining whether an exaction ought to be regarded as a tax. The difficulty with a theory based on these provisions is that those provisions do not (and, hence, the theory does not) distinguish between taxes and other revenues received by the Commonwealth. The provisions are too thin to produce a fully fledged theory of what a tax is but the illustrative point about the theory is that it is a theory about what exactions ought to be regarded as a tax. The constitutional concept of tax is not an abstract economic concept but is rather a concept shaped by its normative constitutional context. A tax is an exaction that, within the normative order established by the Constitution, ought to be treated in a particular fashion (enacted according to certain formalities respecting the bicameral nature of the legislature and the principles of responsible government, collected and paid into the Consolidated Revenue Fund, appropriated by legislation and so on). It is even more clear that excise is a normatively laden concept. It reflects underlying (and disputed) theories about federalism and the intended financial autonomy or control of the constituent elements of the Australian federation. The sharp division of opinion between the majority and minority in Ha v New South Wales reflects a division of opinion about a normatively-shaped concept, Australian federalism, not about the essential elements of a classical concept.
The advantages of having such a theory are clear, both for the formation of the concept and its application. In Australia, constitutional adjudication is a specialised form of common law adjudication, ultimately resting on the constitutional text, but in which concrete determinations in earlier cases are used to formulate the rules that are applied in later cases. That process of regarding earlier cases as providing evidence of rules, that is, general propositions applicable in later cases on similar facts (rather than as providing simply standing for their individual results) depends on a theory of what cases, what fact patterns, are in fact similar. The objection made against family resemblance models of concepts applies equally here – similarity is not a perspicuous standard but one which requires an account, a theory, of what features are significant for assessing similarity. Such a theory might also stabilise the concept and make its application in later cases more predictable. A theory of ‘tax’ and ‘excise’ provides more information about the concept than the collection of previously recognised instances or a list of features with undifferentiated and variable weight.
There are many judicial statements to the effect that it is not possible to provide a set of necessary and sufficient conditions to establish that an exercise of governmental power is or is not an exercise of judicial power. As Windeyer J said in R v Trade Practices Tribunal; Ex parte Tasmanian Breweries Pty Ltd:
The concept seems to me to defy, perhaps it were better to say transcend, purely abstract conceptual analysis. It inevitably attracts consideration of predominant characteristics and also invites comparison with the historic functions and processes of courts of law.
In Brandy v Human Rights and Equal Opportunity Commission, Mason CJ, Brennan and Toohey JJ explicitly contrast their approach with the earlier classical approach. They commented:
[I]t has not been found possible to offer an exhaustive definition of judicial power. So, in R v Davison, Dixon CJ and McTiernan J observed:
[T]he enforcement of a judgment or judicial decree by the court itself cannot be a necessary attribute of a court exercising judicial power. The power to award execution might not belong to a tribunal, and yet its determinations might clearly amount to an exercise of the judicial power.
Their Honours cited a passage from the earlier unanimous decision in Precision Data Holdings Ltd v Wills that explicitly identified the source of the difficulty:
The acknowledged difficulty, if not impossibility, of framing a definition of judicial power that is at once exclusive and exhaustive arises from the circumstance that many positive features which are essential to the exercise of the power are not by themselves conclusive of it. Thus, although the finding of facts and the making of value judgments, even the formation of an opinion as to the legal rights and obligations of parties, are common ingredients in the exercise of judicial power, they may also be elements in the exercise of administrative and legislative power. Again, functions which are ordinary ingredients in the exercise of administrative or legislative power can, in some circumstances, be elements in the exercise of what is truly judicial power.
So, in Brandy, Mason CJ, Brennan and Toohey JJ rejected the earlier view that ‘[t]he exercise of power by a tribunal to enforce its own orders’ was ‘an essential element in the exercise of judicial power’. Instead, it was a feature having weight rather than dispositive force:
The fact that the Commission cannot enforce its own determinations is a strong factor weighing against the characterisation of its powers as judicial; though it must be recognised that this is not an exclusive test of the exercise of judicial power.
What constituted judicial power was not an Aristotelian abstract but depended on context: a power can be characterised as judicial when exercised by a court and as administrative when exercised by a member of the executive. Moreover, the structure of the concept rests in part on abstraction from prior exemplars:
Thus, it has always been accepted that the punishment of criminal offences and the trial of actions for breach of contract and for wrongs are inalienable exercises of judicial power. The validity of that proposition rests not only on history and precedent but also on the principle that the process of the trial results in a binding and authoritative judicial determination which ascertains the rights of the parties.
The approach of Justices Deane, Dawson, Gaudron and McHugh was broadly similar. Their Honours saw the Aristotelian semantic model as primary, but recognised the difficulty in applying it in the context of judicial power:
Difficulty arises in attempting to formulate a comprehensive definition of judicial power not so much because it consists of a number of factors as because the combination is not always the same. It is hard to point to any essential or constant characteristic. Moreover, there are functions which, when performed by a court, constitute the exercise of judicial power but, when performed by some other body, do not.
And yet they resisted an unfocussed appeal to exemplars to define the concept:
One is tempted to say that, in the end, judicial power is the power exercised by courts and can only be defined by reference to what courts do and the way in which they do it, rather than by recourse to any other classification of functions. But that would be to place reliance upon the elements of history and policy which, whilst they are legitimate considerations, cannot be conclusive.
Rather, like Mason CJ, Brennan and Toohey JJ, they attempted to abstract features and identify features that had significant or dispositive weight. In this case, those features were ‘indicative’ of, and ‘point[ed] in many respects to’, the exercise of non-judicial power. But those indications were outweighed by another attribute of the exercise of power by the Tribunal: ‘[h]owever, there is one aspect of judicial power which may serve to characterise a function as judicial when it is otherwise equivocal. That is the enforceability of decisions given in the exercise of judicial power’.
But, notwithstanding its weight, that feature was not essential to characterisation as judicial power.
This weighing or balancing approach was made explicit in Attorney-General (Commonwealth) v Breckler:
Given that it is impossible to point to any essential or constant characteristic of the judicial power, the task upon which a court must embark, when responding to submissions such as the foregoing, is rarely, if ever, a straightforward one. The court must examine the features of the legislation impugned to draw from them those elements which appear to involve an impermissible conferral of judicial power on a non-judicial body. At the same time it must weigh those features which are neutral and which may exist in both courts and non-judicial tribunals. It must also weigh those which are peculiarly characteristic of administrative decision-making.
An exemplar-based concept of judicial power seems intuitive. Judicial power can readily be regarded as an abstraction from historical instances. Those instances exemplify clearly-in members of the category. So, Brennan, Deane and Dawson JJ said in Chu Kheng Lim v Minister for Immigration:
There are some functions which, by reason of their nature or because of historical considerations, have become established as essentially and exclusively judicial in character. The most important of them is the adjudgment and punishment of criminal guilt under a law of the Commonwealth.
Equally, there are clearly-out (non-)members of the category, again defined by particular instances recognised in historical practice and tradition. Some of these instances can be abstracted into an exception for non-punitive detention; but others rest solely on tradition. As McHugh J wrote: ‘[h]istorical practice plays an important, sometimes decisive, part in determining whether the exercise of a particular power is legislative, executive or judicial in character’. The problem with the abstraction is that it may be regarded as thin and as incapable of coherent application in future instances.
If the exemplars and prototypes provided by history do not establish a stable concept, perhaps a theory-based account of judicial power can do better. What Sir Anthony Mason referred to as a ‘purposive functional approach’ to Chapter III might fit that bill. That function might be seen as directed to preservation of the federal structure (as Dixon J saw it in R v Kirby; Ex Parte Boilermakers’ Society of Australia) or as directed to protection of individual liberty particularly in disputes with government (an emerging theme in the Court’s recent jurisprudence). Even if agreement is reached on the function of judicial power (and its constitutional separation from other types of power) there is no single unique theory that explains the decided cases on whether particular powers are to be categorised as judicial or non-judicial or compatible or incompatible with the holding of federal judicial office and so on. The range of theories will make different predictions about whether future powers are to be categorised as judicial or non-judicial or compatible or incompatible. Once again, as was the case with tax and excise, the range of opinions within many of the decided cases makes this clear.
In Brownlee v The Queen, the question for the High Court was whether the trial process leading to Brownlee’s conviction was consistent with the requirements of section 80 of the Constitution. Brownlee argued that it was not because of two features of that process. These features were that the jury was allowed to separate overnight after it retired to consider its verdict and that the guilty verdict was returned by a jury of 10 after two jurors had been discharged during the trial. These features were authorised by the relevant legislation, but said by Brownlee to be inconsistent with the constitutional concept of ‘trial … by jury’. All members of the High Court agreed on the basic approach to that question. It was to consider whether those features were ‘essential characteristics’ of ‘trial ... by jury’ within the meaning of section 80. This was the language of the High Court in Cheatle v The Queen where it was held that jury unanimity was such an essential characteristic but that all-male juries (although equally a characteristic of juries at 1900) was not.
Accordingly, the ostensible point of disagreement between the Justices in Brownlee v The Queen did not concern whether identifying the ‘essential characteristics’ of ‘trial … by jury’ was relevant. Rather, it was how those characteristics were to be determined and, in particular, what place the understanding of ‘trial … by jury’ in 1900 was to play in answering that question. As I indicated at the outset, the originalism/non-originalism debate is beyond the scope of this article. Accordingly, I focus my attention on the structure of the semantic argument used by the members of the Court. The key questions are: were the members of the Court committed to an Aristotelian conception of the language of ‘essential characteristics’? And, whatever their position in the originalism/non-originalism debate, where did ‘essential characteristics’ come from?
Chief Justice Glee son and McHugh J noted that ‘the incidents of [trial by jury] never have been immutable; they are constantly changing’ and, for that reason, ‘it is impossible to contend that all of its characteristics at any given time ought to be regarded as essential. Its history demonstrates that they are not’.
History – instances of the concept – could demonstrate that a feature (in this case, that the jury had at least 12 members) was not essential because there were instances where this feature was not present. But history is not conclusive; it could be subordinated to the function of a concept in determining what were the concept’s essential features. ‘The function of jury trial is not such as to make it essential that the common law [feature] be preserved in its full rigour.’ A system which allowed for the absence of this particular feature ‘is not inconsistent with the objectives of independence, representativeness and randomness of selection, or with the need to maintain the prosecution’s obligation to prove its case beyond reasonable doubt’. Another feature (that the jury be sequestered while considering its verdict) was not an essential feature because it was a measure ‘taken to guard against the danger of jurors being subjected to improper outside influence’ and ‘the danger itself changes with varying social conditions and methods of communication’. Hence, the essential features were those that served the purposes of the concept in the particular context in which that concept was employed from time to time.
This emphasis on the purpose of a concept in determining its essential features was explicit in the judgment of Gaudron, Gummow and Hayne JJ: ‘[c]lassification as an essential feature or fundamental of the institution of trial by jury involves an appreciation of the objectives that institution advances or achieves’. That statement contains no explicit reference to history or understandings of the institution as at 1900. But Gaudron, Gummow and Hayne JJ also wrote:
As with the legal terms ‘a writ of … prohibition’ in s[ection] 75(v) of the Constitution and ‘patents of inventions’ in s[ection] 51(xviii), an appreciation of the essential characteristics of the legal institution identified as ‘trial by jury’ in prosecutions on indictment is assisted by an understanding of that legal institution at the time of the commencement of the Constitution. That understanding may assist in a perception of the ends sought to be advanced or achieved by the sequestering of the jury and the insistence upon a verdict returned by 12 jurors. The question then is whether these ends are such as to give rise to essential features of the trial by jury stipulated by s[ection] 80.
Thus, a historical feature of trial by jury (strict sequestration) promoted an objective (‘deliberation and attention to the evidence, without distraction of other material not in evidence and the threat of influence by outsiders upon that deliberation’); that in turn supported what was the ultimate objective of the institution itself: ‘the determination of guilt according to law, with the interposition between the accused and the prosecution of ‘the commonsense judgment of a group of laymen’. So although
it [was] an essential feature of the jury system that the jury should deliberate upon its verdict uninfluenced by an outsider to the trial process …
… [A]n understanding and construction should be given to the words in s[ection] 80 that the framers of the constitutional guarantee intended that a jury exercise its function without fear or favour and without undue influence in the context of community standards and expectations as current from time to time.
The reasoning on the requirement of a verdict of 12 was similar. The requirement that the jury initially empanelled have 12 members ‘is to be supported on utilitarian grounds’. This is because ‘[i]t ensures that the trial gets underway with fact-finding entrusted to a group of laymen which is large enough to promote measured deliberation and indicates to the community sufficient participation by its members to vindicate the outcome’. If a fresh jury must be empanelled because jurors are subsequently discharged this involves expense and delay for the state, the accused and witnesses. Legislation prior to federation
adapted the institution [of trial by jury] to what already in the late nineteenth century were perceived to be particular needs, whilst retaining the substantial character of the institution as an eficient instrument in the administration of justice.
It followed that a jury of at least 12 members was not an essential feature of the constitutional concept of ‘trial … by jury’.
Justice Kirby’s approach is harder to fathom. At one level, he rejects the connotation-denotation distinction, writing that the constitutional words are ‘set free’ from the understanding of their meaning at the time of their enactment and must now be interpreted ‘in their contemporary institutional setting and as they must operate in accordance with the “accepted standards of a modern democratic society”’. But his rejection of history is never absolute. The Constitution is a statutory text, albeit a special statute, and must be interpreted as such, inevitably, therefore, as an attempt to bind the future by the words of the past.
Words are not necessarily confined to the meaning that would subjectively have been ascribed to them by the Parliament that enacted them. … A recognition of this fact does not render wholly irrelevant the consideration of history – as in the debates that preceded adoption of the Constitution. But it does limit the utility of such searches
Most significantly for present purposes, he appears to attempt to reconcile these positions by accepting that concepts have ‘essential characteristics’. But even this attempt is not wholly convincing because he argues both that the essential characteristics of ‘trial … by jury’ are identified ‘from the perspective of contemporary considerations’ and that the relevant essential features ‘hav[e] enduring constitutional operation’. The tension between contemporary influences and enduring operation is not squarely resolved.
Thus, in this case, all members of the Court adopt a semantic model that uses the language and logic of Aristotelian essentialism: a thing is not a trial by jury in the sense of the constitutional term unless it has all the essential features of that concept. On this model, ostensibly at least, there is a bright line dividing the world into trials by jury and non-trials by jury. But when the apparently ‘essential features’ of the concept are examined more closely, another picture emerges. The essential features are determined by reference to a theory of the concept, in this case a teleological theory that identifies the purposes or objectives served by instances of the concept.
The judgments in Brownlee v The Queen demonstrate the difficulty in identifying where the theory comes from. None identify how the function of trial by jury is to be determined. Most simply adopt (and do not explain or justify) the statement made by White J in the United States Supreme Court in Williams v Florida. None identify how the constitutionally entrenched functions of ‘trial ... by jury’ in section 80 are to be distinguished from the functions that are not constitutionally entrenched.
Here, the explanation-based or theory-based semantic model must confront the same problem as the classical model and the (similarity-based) prototype and exemplar models: how are interpreters to determine the essential features, feature weights or theories applicable to define a concept?
We have seen that family resemblance and theory-based approaches can enrich our understanding of current approaches to constitutional interpretation. This is important because the difficulty of producing classical definitions of legal terms can produce a ‘defeatist scepticism’. James Penner quotes Jeremy Waldron:
When jurists express doubt about the usefulness of general terms … they imply that if we are unable to specify necessary and jointly sufficient conditions … then those terms are to be regarded as ambiguous or confused and certainly as analytically unhelpful.
Penner identifies this tendency in certain strands of property jurisprudence which leads those subject to it to the conclusion that ‘property is a flexible or malleable concept, with no definable essence, and no guidelines for definition which might in any way govern its application in particular circumstances’.
There is no need for this scepticism in constitutional adjudication. The argument here has shown that the absence of classical definitions that can be expressed in terms of necessary and sufficient characteristics does not mean that (in the domain of ordinary language and cognition) terms cannot be given a stable, workable meaning. That meaning may reflect an underlying conceptual structure based on family resemblance or on theories about concepts, and for ordinary language users the underlying conceptual structure may not matter much. For jurists, however, the means by which conceptual stability is achieved is as important as that stability itself. This is because jurists must expose their reasoning, and not just their conclusions, to public scrutiny.
The case studies – taxation and excise, judicial power, and trial by jury – demonstrate a superficial commitment to the classical model, qualified by a movement away from the classical model and its underlying essentialism, corresponding with a tendency to categorise on the basis of consideration of a ‘range of factors’. Family resemblance approaches are usually too thin in their predictive capacity to provide a suitable replacement. Should the Court embrace an explanation-driven or theory-driven approach to categorisation? Two points can be made, the first about judicial theorising in general, the second about the High Court’s attitude to theory.
There are good reasons to be cautious about some forms of judicial theorising. As Cass Sunstein has argued, it is not always practical or desirable that decisions be justified by a full theorised argument. He points to the merits of incompletely theorised agreements that allow multimember courts to reach agreements on outcomes without needing to secure agreement on every high level theoretical premise that might support those outcomes. Incomplete theorisation responds to the need for decision making within finite time constraints and in the face of incomplete information. It allows potentially divisive issues to be reserved for another day or postponed indefinitely. However, as Adrienne Stone has emphasised, ‘[f]or all its virtues, … incomplete theorization comes at a cost’, particularly when it is adopted by apex courts that not only must resolve concrete cases but can shape the development of the law. ‘[I]ncomplete theorization might undermine appropriate development of the law by discouraging the examination, and consequent understanding’ of the underlying normative concerns that influence agreement on low level theories and practical outcome s. It may conceal from other participants in the legal system the ways in which the law may develop in the future under the influence of those normative concerns. In other words, the benefits of incomplete (as opposed to complete theorisation) depend on institutional and substantive context. Theory-based or explanation-based approaches to categorisation can respond to this context-sensitivity. It does not involve a commitment either to comprehensive theories or to judicial minimalism.
The High Court itself is extraordinarily resistant to theory, at least of the top-down variety. In a common law context, Gummow J wrote in Roxborough v Rothmans of Pall Mall Australia Ltd:
To the lawyer whose mind has been moulded by civilian influences, the theory may come first, and the source of the theory may be the writing of jurists not the decisions of judges. However, that is not the way in which a system based on case law develops; over time, general principle is derived from judicial decisions upon particular instances, not the other way around.
In the constitutional context, the Court has insisted that, in order to remain within the legitimate bounds of constitutional interpretation, a foundation for constitutional doctrine must be found in the text and structure of the instrument and not in freestanding theories.
It would seem likely, therefore, that the Court’s embrace of an explanation-based or theory-based approach to categorisation is likely to be slow and cautious. But again note two things. First, the Court already uses an explanation-based or theory-based approach to categorisation in some contexts, even if it rhetorically prefers a classical approach. And second, an explanation-based or theory-based approach is consistent with the Court’s own methodological commitments to a modest role for theory in judicial decision making. The explanations or theories that stabilise the meaning of constitutional terms can be derived from ‘judicial decisions upon particular instances’; they can be derived from constitutional text and structure. They will, no doubt, be richer and constrain future decision making to a greater extent if they draw openly on a wider range of normative sources. But an explanation-based or theory-based approach to categorisation does not commit the Court to a substantive or methodological revolution. Rather, it promises a more candid approach to identifying the meaning of constitutional terms, that avoids the façade of a classical approach and provides a rational basis for conceptual stability and coherence.
[*] Director, Centre for Comparative Constitutional Studies, Faculty of Law, University of Melbourne. I began this project in 2002 as a Sabbatical Fellow in the Law Programme, Research School of Social Sciences, Australian National University. I am grateful to the Journal’s referees for their extremely helpful comments and to John Daley for piquing my interest in the questions discussed here.
 Constitution s 51(ii).
 Constitution s 51(xx).
 Constitution s 71.
 Constitution s 44(i).
 Cf William Labov, ‘The Boundaries of Words and Their Meanings’ in Charles-James N Bailey and Roger W Shuy (eds), New Ways of Analysing Variation in English (1973) 340, 342 in John Taylor, Linguistic Categorization: Prototypes in Linguistic Theory (2nd ed, 1995) 2.
 Although my focus is on constitutional interpretation, the argument and analysis are also applicable in other legal contexts.
 The terminology derives from John Stuart Mill, A System of Logic: Ratiocinative and Inductive (1875) 31–42 noted in Street v Queensland Bar Association  HCA 53; (1989) 168 CLR 461, 537 (Dawson J). For a philosophical treatment of Mill’s semantics, see Willem Remmelt de Jong, The Semantics of John Stuart Mill (Herbert Morton trans, 1982 ed) [trans of: De semantiek van John Stuart Mill]. This metaphysical distinction has fared better than Mill’s explanation of causation (see National Insurance Company of New Zealand Ltd v Espagne  HCA 15; (1961) 105 CLR 569, 591 (Windeyer J)) or his distinction between direct and indirect taxation (see Philip Morris Ltd v Commissioner of Business Franchises (Victoria)  HCA 38; (1989) 167 CLR 399, 429 (Mason CJ and Deane J)).
  HCA 53; (1989) 168 CLR 461.
 Ibid 537. Here, Dawson J uses ‘connotation’ and ‘denotation’ in the conventional legal sense, ‘doubt[ing] the wisdom of’ his having used them in their ‘popular’, opposite, sense in Commonwealth v Tasmania  HCA 21; (1983) 158 CLR 1, 3 02–3 (‘Tasmanian Dam Case’).
 In this model, connotation is identified with meaning: Re Wakim; Ex parte McNally (1999) 198 CLR 511, 551 (McHugh J); R v Commonwealth Conciliation and Arbitration Commission; Ex parte Association of Professional Engineers, Australia  HCA 47; (1959) 107 CLR 208, 267 (Windeyer J).
 R v Federal Court of Australia; Ex parte WA National Football League  HCA 6; (1979) 143 CLR 190, 234 (Mason J) (‘Adamson ’s Case’) in Re Wakim; Ex parte McNally (1999) 198 CLR 511, 552 (McHugh J). Justice McHugh noted ‘[t]he distinction between meaning and denotation is not without its difficulties’: at 552. See also Leslie Zines, The High Court and the Constitution (4th ed, 1996) 19; Eastman v The Queen  HCA 29; (2000) 203 CLR 1, 80–1 (Kirby J). Justice Kirby doubted the coherence of the distinction.
 See, eg, Jeffrey Goldsworthy, ‘Originalism in Constitutional Interpretation’  FedLawRw 1; (1997) 25 Federal Law Review 1; Christopher Birch, ‘Mill, Frege and the High Court: The Connotation/Denotation Distinction in Constitutional Interpretation’ (2003) 23 Australian Bar Review 296.
 Non-constitutional invocations include interpretation of the legal concepts of tenancy-in-common: Nullagine Investments Pty Ltd v Western Australian Club Inc  HCA 45; (1992) 177 CLR 635, 656 (Deane, Dawson and Gaudron JJ), identifying an ‘essential feature’ of the concept; debenture: Handevel Pty Ltd v Comptroller of Stamps (Victoria)  HCA 73; (1985) 157 CLR 177, 199 (Mason, Wilson, Deane and Dawson JJ), a fleeting reference to the denotation of the concept; banking: Commissioners of the State Savings Bank of Victoria v Permewan, Wright & Co Ltd  ArgusLawRp 139; (1914) 19 CLR 457, 470–1 (Isaacs J), where ‘[t]he essential characteristics’ of the concept were all that was required to make the statute applicable; desertion, in the context of marriage: Powell v Powell  HCA 48; (1948) 77 CLR 521, 535 (Dixon J); pawn or pledge: Palgo Holdings Pty Ltd v Gowans  HCA 28; (2005) 221 CLR 249, 282 (Kirby J).
  HCA 94; (1908) 6 CLR 469 (‘Union Label Case’).
 (2000) 202 CLR 479 (‘Grain Pool of WA’).
 Ibid 528.
 Ibid 527.
 Ibid 528 citing Union Label Case  HCA 94; (1908) 6 CLR 469, 606 (Higgins J, dissenting).
 Cheatle v The Queen  HCA 44; (1993) 177 CLR 541, 557–8 (Mason CJ, Brennan, Deane, Dawson, Toohey, Gaudron and McHugh JJ); Brownlee v The Queen (2001) 207 CLR 278.
 Bolton v Madsen  HCA 16; (1963) 110 CLR 264, 271 (Dixon CJ, Kitto, Taylor, Menzies, Windeyer and Owen JJ); Dennis Hotels Pty Ltd v Victoria  HCA 10; (1960) 104 CLR 529, 545 (Dixon CJ), 549 (McTiernan J), 55 1–8 (Fullagar J), 559 (Kitto J), 575 (Taylor J), 588–9 (Menzies J), 593, 598, 608 (Windeyer J); Philip Morris Ltd v Commissioner of Business Franchises (Victoria)  HCA 38; (1989) 167 CLR 399, 430–1, 435, 437 (Mason CJ and Deane J), 447, 458 (Brennan J), 488, 490 (McHugh J); Dickenson ’s Arcade Pty Ltd v Tasmania (1974) 130 CLR 177, 199, 203 (McTiernan J), 231, 235 (Stephen J), 241 (Mason J).
 Singh v Commonwealth  HCA 43; (2004) 222 CLR 322, 350–1 (McHugh J).
 Union Label Case  HCA 94; (1908) 6 CLR 469, 560, 572, 577, 581.
 Re Refugee Review Tribunal; Ex parte Aala  HCA 57; (2001) 204 CLR 82, 93 (Gaudron and Gummow JJ), 133, 135 (Kirby J); Re Minister for Immigration and Multicultural Afairs; Ex parte Miah (2001) 206 CLR 57, 123 (Kirby J).
 McGinty v Western Australia  HCA 48; (1995) 186 CLR 140, 200–1 (Toohey J); Cunlife v Commonwealth  HCA 44; (1994) 182 CLR 272, 365 (Dawson J).
 A-G (Cth) v Breckler (1999) 197 CLR 83, 125–8 (Kirby J); Polyukovich v Commonwealth  HCA 32; (1991) 172 CLR 501, 704 (Gaudron J); Leeth v Commonwealth (1992) 174 CLR 455, 50 1–2 (Gaudron J).
 Grannall v Marrickville Margarine Pty Ltd  HCA 6; (1955) 93 CLR 55, 78 (Dixon CJ, McTiernan, Webb and Kitto JJ); SOS (Mowbray) Pty Ltd v Mead (1972) 124 CLR 529, 550 (Barwick CJ), 565 (Menzies J); Hughes & Vale Pty Ltd v New South Wales [No 2]  HCA 28; (1955) 93 CLR 127, 162–3 (Dixon CJ, McTiernan and Webb JJ), 215 (Fullagar J); Hospital Provident Fund Pty Ltd v Victoria  HCA 8; (1953) 87 CLR 1, 17–18 (Dixon CJ).
 Re Finance Sector Union of Australia; Ex parte Financial Clinic (Vic) Pty Ltd  HCA 34; (1993) 178 CLR 352, 362, 367 (Mason CJ, Deane, Toohey and Gaudron JJ); R v Ludeke; Ex parte Queensland Electricity Commission  HCA 55; (1985) 159 CLR 178, 181 (Gibbs CJ, Mason, Wilson, Brennan, Deane and Dawson JJ).
 (2001) 207 CLR 278.
  HCA 2; (1942) 66 CLR 198.
 Ibid 223.
 See, eg, Sir Owen Dixon, ‘Concerning Judicial Method’ in Sir Owen Dixon, Jesting Pilate (1965) 152.
 I do not need to engage in the metaphysical and empirical debates about concepts and cognition. For those interested, a colourful (if highly partisan) account of one subset of the debates may be found in Jerry Fodor, Concepts: Where Cognitive Science Went Wrong (1998); a wider survey (on which I rely extensively in this article) appears in Lloyd Komatsu, ‘Recent Views of Conceptual Structure’ (1992) 112 Psychological Bulletin 500.
 Note that this preserves the meaning-application distinction assumed as part of the connotation-denotation model: see above Part II(A).
 Again, I do not need to buy into the metaphysical debates about properties, including whether they are universal or can only be individualised. For those interested in the debates, see Chris Swoyer, ‘Properties’ in Edward N Zalta (ed), The Stanford Encyclopedia of Philosophy (Winter 2000 ed) <http://plato.stanford.edu/archives/win2000/entries/properties> at 25 October 2006.
 How exactly categories are represented is a controversial question in cognitive science and other disciplines. The representation may include knowledge of exemplars (perhaps Phar Lap and Bucephalus); it may include features or attributes possessed by instances of the category (for example, that members of the category ‘horse’ are non-aquatic quadruped animals); it may include theories about what it is to be a horse.
 See Taylor, above n 5, 22–4.
 Ibid 23.
 It also follows that concepts are primitives with clear boundaries.
 Eleanor Rosch, ‘Principles of Categorization’ in Eleanor Rosch and Barbara B Lloyd (eds), Cognition and Categorization (1978) 27, 28; cf Komatsu, above n 32, 50 1–2.
 Cass R Sunstein, ‘Problems with Rules’ (1995) 83 Cal ifornia Law Review 953, 1022.
 Bruner, Goodnow and Austin, writing in 1956 in an important early study of categorisation, regarded concepts as particular kinds of rules:
[W]hen one learns to categorise a subset of events in a certain way, one is doing more than simply learning to recognise instances encountered. One is also learning a rule that may be applied to new instances. The concept or category is basically this ‘rule of grouping’ and it is such rules that one constructs in forming and attaining concepts: Jerome Bruner, Jacqueline Goodnow and George Austin, A Study of Thinking (1956) 45.
 Woo-Kyoung Ahn, ‘Why Are Different Features Central for Natural Kinds and Artefacts?: The Role of Causal Status in Determining Feature Centrality’ (1998) 69 Cognition 135, 162.
 Hilary Putnam, ‘Is Semantics Possible?’ in Stephen P Schwartz (ed), Naming, Necessity, and Natural Kinds (1977) 104 in Ahn, above n 42, 162.
 Douglas L Medin and Andrew Ortony, ‘Psychological Essentialism’ in Stella Vosniadou and Andrew Ortony (eds), Similarity and Analogical Reasoning (1989) 179, 183–4. Michael Strevens, ‘The Essentialist Aspect of Naive Theories’ (2000) 74 Cognition 149 discusses other variants of psychological essentialism.
 Morton Horwitz, The Transformation of American Law 1870–1960: The Crisis of Legal Orthodoxy (1992) 13.
 This is not to say that the classical model necessarily is a realist model (ie, one committed to the view that the things in the world exist independently of our thoughts or perceptions), or that only the classical model is realist. There is an association, no more.
 Of course, in practice, the criteria of membership for legal categories may be vague, and membership of those categories may be indeterminate. The classical semantic model may, nonetheless, be useful in rhetorically signalling a commitment to bright lines, legal definition and certainty, even if it cannot deliver on that commitment.
 One possible explanation for part of that failure is that some terms might define clusters of concepts rather than single concepts. (Perhaps ‘bird’ does not correspond to a single concept bird but to separate concepts, eg, flightless bird, raptor, penguin etc).
 John Gerring, ‘What Makes a Concept Good? A Criterial Framework for Understanding Concept Formation in the Social Sciences’ (1999) 31 Polity 357, 363.
 For example, on some accounts, the meaning of ‘human’ might be determined not by the conjunction of the features noted in the text but by some distinguishing attributes of human DNA. See, eg, Dennis Patterson’s account of Kripke and Putnam’s semantics of natural kinds: Dennis Patterson, ‘Dworkin on the Semantics of Legal and Political Concepts’ (2006) 26 Oxford Journal of Legal Studies 545, 553.
 Perhaps plausibly in relation to natural kinds.
 R v Brislan; Ex parte Williams  HCA 78; (1935) 54 CLR 262, 275 (Latham CJ).
 Constitution s 51(v). The very language of the section, with its extension to ‘other like services’, might more naturally suggest an approach that determines category membership by assessing the similarity between a putative category member and the central instances, telephonic and telegraphic services.
 R v Brislan; Ex parte Williams  HCA 78; (1935) 54 CLR 262, 275 (Latham CJ).
 Ibid 278–9 (Latham CJ).
  HCA 78; (1935) 54 CLR 262.
 Ibid 283 (Rich and Evatt JJ).
 Ludwig Wittgenstein, Philosophical Investigations (Gertrude Elizabeth Margaret Anscombe trans, 1958 ed) [trans of: Philosophische untersuchunge].
 Ibid. Rowe argues that ‘game’ can in fact be defined classically as having essential features that it is an abstract object, designed to lack instrumental value and intended to absorb the interest of its participants or spectators: M W Rowe, ‘The Definition of “Game”’ (1992) 67 Philosophy 467, 478.
 Eleanor Rosch, ‘Cognitive Reference Points’ (1975) 7 Cognitive Psychology 532–47.
 See generally Taylor, above n 5.
 Wittgenstein, above n 58, 32e–3e, 34e.
 Ibid 33e, 34e.
 The empirical discovery of common characteristics would not show that the concept in question was not a family resemblance concept; what is decisive is the existing practice of explaining the expression.
 Wittgenstein, above n 58, 55.
 Rosch, ‘Principles of Categorization’, above n 39, 29, emphasised this as a theory of categorisation rather than as a theory of mental representations (concepts). However, the prototype model of categorisation rapidly became a model of concepts and continues to be defended as such (see, eg, James Hampton, ‘Concepts and Prototypes’ (2000) 15 Mind & Language 299) though some remain unconvinced: see, eg, Fodor, above n 32. Komatsu distinguishes using ‘family resemblance’ to refer only to the prototype model; he brackets the prototype model with the exemplar model (discussed below) as ‘similarity-based views’: Komatsu, above n 32, 500.
 Taylor, above n 5, 63.
 Ibid 63–4.
 Ibid 42.
 Komatsu, above n 32, 505. As Komatsu notes, this is a problem for family resemblance models generally, not just the prototype model.
 Charles Kalish, ‘Essentialism and Graded Membership in Animal and Artefact Categories’ (1995) 23 Memory & Cognition 335, 336, 347–8; Komatsu, above n 32, 503.
 Sharon Lee Armstrong, Lila Gleitman and Henry Gleitman, ‘What Some Concepts Might Not Be’ (1983) 13 Cognition 263, 278–9, 284, 292 in Komatsu, above n 32, 506.
 See, eg, Kathleen Dahlgren, ‘The Cognitive Structure of Social Categories’ (1985) 14 Cognitive Science 379 in James Worthen et al, ‘The Internal Structure of Social Categories: Components and Predictors’ (1998) 111 American Journal of Psychology 581, 583–4.
 Komatsu, above n 32, 506.
 Komatsu refers to an alternative reading of the exemplar model which treats a concept as a set of family resemblance concepts (multiple prototypes in his description) rather than as previously encountered instances: ibid 507.
 Ibid 508.
 Douglas L Medin and Edward Shoben, ‘Context and Structure in Conceptual Combination’ (1988) 20 Cognitive Psychology 158, 166–7. Connell misreports Medin and Shoben and inverts the order of typicality of these two types of spoon. This does not affect the larger point that she is making.
 Louise Connell, Categories, Concepts and Co-occurrence: Modelling Categorisation Efects with LSA (MSc in Cognitive Science and Natural Language Thesis, Division of Informatics, University of Edinburgh, 2000) 14 <http://www.nl-webspace.co.uk/~unn_ppht6/papers/Connell-2000-MSc.pdf> at 28 September 2006.
 These include hybrid models, like the schema model, not discussed here but outlined in Komatsu, above n 32, 510–11.
 Medin and Ortony, above n 44, 179; Edward E Smith and Douglas L Medin, Categories and Concepts (1981).
 John Vervaeke and Christopher Green, ‘Women, Fire, and Dangerous Theories: A Critique of Lakoff’s Theory of Categorization’ (1997) 12 Metaphor and Symbol 59 <http://www.yorku.ca/christo/papers/wfdt8.htm> at 28 September 2006.
 Connell, above n 79, 27.
 Komatsu, above n 32, 514.
 Patterson, above n 50, 553.
 See below n 168 and associated text.
 They may incidentally provide resources for resolving those disagreements to the extent that they expose their normative underpinnings.
 Leslie Zines has argued:
Clearly the dichotomies of essence and incidence or connotation and denotation do not resolve the question [of the meaning of constitutional terms over time]. They merely restate the issue and perhaps mislead some to the view that there must be some fixed ‘nature’ that a term has which can be ascertained by examination and pure thought. In reality, we often develop our concepts with their connotations from familiarity with specific objects or instances. A child might be told that a particular object is a ‘chair’ and that another is also a chair, even though it is not exactly the same as the first; its colour, size and texture are different: Zines, above n 11, 19.
Constitutional concepts are not the products of pure thought. They are the products of social institutions and practices. It is not reason that provides the starting point for identifying their meaning but experience. Where I depart from Zines is the next step. He argues that the child who has encountered a number of chairs ‘thus learns the meaning of the concept and to distinguish features that are essential to the notion and those (such as colour and size in this case) that are not’: at 19. The retreat into the classical model is unnecessary and suggests that the concept does indeed have a fixed ‘nature’.
 Not because the non-classical models are true in any sense demonstrated by the psychology and cognitive science literature. Whatever that literature establishes about cognition in relation to natural and nominal kinds cannot apply directly to the ‘artificial reason’ that judges and lawyers engage in.
  HCA 38; (1938) 60 CLR 263, 267.
 Another example of a non-classical approach can be found in The Church of the New Faith v Commissioner of Pay-Roll Tax (Vic)  HCA 40; (1982) 154 CLR 120 (‘Scientology Case’), where Wilson and Deane JJ held that ‘religion’ was not defined by a set of necessary and sufficient features: at 171–2. Rather, they gave a list of indicia or guiding features of variable weight: at 174–5. See also Street v Queensland Bar Association  HCA 53; (1989) 168 CLR 461. In considering the meaning of s 117, Deane J stated:
[T]he notion of residence is a complex one, it is not one which can be subdivided, in the abstract, into a number of discrete, necessary and sufficient elements or factors. Regardless of the precise meaning which one gives to the word ‘resident’ in s 117, the relative importance, and even the identity, of the factors which are determinative of whether a particular person is or is not resident in a particular State are likely to vary from case to case: at 526.
 Matthews v Chicory Marketing Board (Vic)  HCA 38; (1938) 60 CLR 263, 276. As to whether Latham CJ intended his remarks as a ‘definition’, see below n 99 ff.
 (1988) 165 CLR 462 (‘Air Caledonie’).
 Ibid 466–7 (Mason CJ, Wilson, Brennan, Deane, Dawson, Toohey and Gaudron JJ).
 Ibid 467.
 Ibid (emphasis added).
 Ibid (emphasis added).
  HCA 13; (2002) 210 CLR 333.
 Ibid 342 (emphasis added); see also 352 (Gaudron and Hayne JJ) (‘features ... typical of a tax’); cf 365 (Kirby J) (referring to ‘the definition ... offered by Latham CJ’ and ‘his Honour’s description’).
 Ibid 342 referring to Lower Mainland Dairy Products Sales Adjustment Committee v Crystal Dairy Ltd  AC 168, 175–8.
  HCA 10; (1993) 176 CLR 480, 522 (Dawson and Toohey JJ, dissenting) (‘Tape Manufacturers’). Justice McHugh, concurring generally with their Honours, appeared to agree as to the importance of an impost being ‘revenue’ as well (at 529–40); whilst the majority, Mason CJ, Brennan, Deane and Gaudron JJ, finding that it did raise revenue, did not need to consider whether this was essential: at 505–8.
 Luton v Lessels  HCA 13; (2002) 210 CLR 333, 343 (Gleeson CJ) (emphasis added); cf 372 (Kirby J). Both Gleeson CJ and Kirby J cited Airservices Australia v Canadian Airlines International Ltd (1999) 202 CLR 133, 178 (Gleeson CJ and Kirby J). In that case, Gleeson CJ and Kirby J stated:
Not all taxation has as its primary purpose the raising of revenue; and some forms of taxation are [a] notoriously inefficient means to that end. An objective of raising revenue is not, therefore, a universal determinant. Even so, the presence or absence of such an objective will often be significant: at 178.
 Ha v New South Wales  HCA 34; (1997) 189 CLR 465, 490 (Brennan CJ, McHugh, Gummow and Kirby JJ).
  HCA 16; (1963) 110 CLR 264.
 Ibid 271.
 Under the criterion of liability approach, an exaction constituted an excise only if the person was rendered liable to taxation because he or she had taken a step in the production or distribution of goods. It was not enough that the practical effect of the exaction was the same as an excise: see, eg, Dickenson ’s Arcade Pty Ltd v Tasmania (1974) 130 CLR 177. ‘The directness of the relation between the tax and the goods’ is not atomic or primitive (as the strict classical model would insist) because it can be decomposed into sub-features or sub-criteria: at 223 (Gibbs J).
  HCA 23; (1983) 151 CLR 599.
  HCA 38; (1989) 167 CLR 399.
 Ibid 434–5 (Mason CJ and Deane J); see also 484–5 (Toohey and Gaudron JJ).
 Ha v New South Wales  HCA 34; (1997) 189 CLR 465, 501 (Brennan CJ, McHugh, Gummow and Kirby JJ).
 Tape Manufacturers  HCA 10; (1991) 176 CLR 480, 503, 506–8.
 See Ha v New South Wales  HCA 34; (1997) 189 CLR 465, 491, 497 (Brennan CJ, McHugh, Gummow and Kirby JJ); cf 506, 508 (Dawson, Toohey and Gaudron JJ).
  HCA 34; (1997) 189 CLR 465.
 One focus of that disagreement is whether s 90 ‘was intended to give the Parliament a real control of the taxation of commodities and to ensure that the execution of whatever policy it adopted should not be hampered or defeated by State action’: Parton v Milk Board (Vic)  HCA 67; (1949) 80 CLR 229, 260 (Dixon J).
 None of this is to suggest that constitutional adjudication should follow the top-down approach anathematised in the context of common law adjudication in Roxborough v Rothmans of Pall Mall Australia Ltd  HCA 68; (2001) 208 CLR 516, 544–5 (Gummow J).
 See, eg, Julius Stone, ‘The Ratio of the Ratio Decidendi’ (1959) 22 Modern Law Review 597.
 See Sue v Hill  HCA 30; (1999) 199 CLR 462, 515 fn 187 ff.
  HCA 8; (1970) 123 CLR 361.
 Ibid 394.
  HCA 10; (1995) 183 CLR 245 (‘Brandy’).
  HCA 46; (1954) 90 CLR 353.
 Ibid 368 in Brandy  HCA 10; (1995) 183 CLR 245, 257.
 (1991) 173 CLR 167.
 Ibid 188–9 in Brandy  HCA 10; (1995) 183 CLR 245, 267.
 Brandy  HCA 10; (1995) 183 CLR 245, 256.
 Ibid 256.
 Ibid 257.
 Ibid 258. See also Re Ranger Uranium Mines; Ex parte Federated Miscellaneous Workers’ Union of Australia  HCA 63; (1987) 163 CLR 656, 666 (the character of a power may be determined by ‘the purpose for which it is undertaken’); Precision Data Holdings Limited v Wills (1991) 173 CLR 167, 189–90, 191.
 Brandy  HCA 10; (1995) 183 CLR 245, 258.
 Ibid 267 (footnote omitted).
 Ibid 269.
 Ibid 268.
 Ibid 268–9 citing Waterside Workers’ Federation of Australia v J W Alexander Ltd  HCA 56; (1918) 25 CLR 434, 451 (Barton J).
 (1999) 197 CLR 83.
 Ibid 124–5 (Kirby J). For the analysis of the majority, along these lines without an explicit statement to that effect, see 109–12 (Gleeson CJ, Gaudron, McHugh, Gummow, Hayne and Callinan JJ).
 (1992) 176 CLR 1.
 Ibid 27.
 Ibid 27–8, 32.
 Ibid 28; cf 55 (Gaudron J) (referring to ‘presently accepted categories’). See also Sue v Hill  HCA 30; (1999) 199 CLR 462, 518 (Gaudron J).
 Chu Kheng Lim v Minister for Immigration (1992) 176 CLR 1, 67. As the cases cited by McHugh J indicate, this is not a new phenomenon, but see the joint reasons in which his Honour participated in Brandy.
 Sir Anthony Mason, ‘A New Perspective on Separation of Powers’ (1996) 86 Canberra Bulletin of Public Administration 1, 2.
  HCA 10; (1956) 94 CLR 254 (‘Boilermakers’ Case’).
 See, eg, Kable v DPP (NSW)  HCA 24; (1996) 189 CLR 51; Al-Kateb v Godwin  HCA 37; (2004) 219 CLR 562, 573 (Gleeson CJ) (‘[it] does not involve an invalid attempt to confer on the Executive a power to punish people who, being in Australia, are subject to, and entitled to the protection of, the law’).
 See, eg, Grollo v Palmer, Commissioner of the Australian Federal Police  HCA 26; (1995) 184 CLR 348; Wilson v Minister for Aboriginal and Torres Strait Islander Afairs  HCA 18; (1996) 189 CLR 1; Kable v DPP (NSW)  HCA 24; (1996) 189 CLR 51.
 See Jury Act 1977 (NSW) ss 22(a)(i), 54(b). These are applicable because of s 68 of the Judiciary Act 1903 (Cth).
 Constitution s 80.
 Brownlee v The Queen (2001) 207 CLR 278, 284 (Gleeson CJ and McHugh J), 298 (Gaudron, Gummow and Hayne JJ), 327 (Kirby J), 342 (Callinan J).
  HCA 44; (1993) 177 CLR 541.
 Ibid 562.
 Brownlee v The Queen (2001) 207 CLR 278, 286.
 Ibid 287.
 Ibid 288.
 Ibid 289.
 Ibid. See also Brandy  HCA 10; (1995) 183 CLR 245, 267 (Deane, Dawson, Gaudron and McHugh JJ).
 Brownlee v The Queen (2001) 207 CLR 278, 290.
 Ibid 298. Justice Callinan too used functional arguments to identify the essential characteristics of ‘trial … by jury’: at 341, 342–3.
 Ibid 299 (emphasis added, footnote omitted).
 Ibid 302.
 Ibid (footnote omitted).
 R v Brownlee (1997) 41 NSWLR 139, 145–6 in Brownlee v The Queen (2001) 207 CLR 278, 302.
 Brownlee v The Queen (2001) 207 CLR 278, 303.
 Ibid 302.
 Ibid 302–3.
 Ibid 303.
 The conclusion may have been different so far as s 22 of the Jury Act 1977 (NSW) authorised a jury of fewer than 10: see Jury Act 1977 (NSW) s 22(a)(iii); Brownlee v The Queen (2001) 207 CLR 278, 303–4.
 Brownlee v The Queen (2001) 207 CLR 278, 32 1–2 (footnotes omitted); see also 314. See also Cheatle v The Queen  HCA 44; (1993) 177 CLR 541, 560.
 Brownlee v The Queen (2001) 207 CLR 278, 321–2 (footnotes omitted).
 Ibid (footnotes omitted, emphasis added).
 Ibid 327.
 Ibid; cf 317–18.
 Or purposes (ibid 288 (Gleeson CJ and McHugh J)), or objectives (at 288 (Gleeson CJ and McHugh J)) or ends (at 299–300 (Gaudron, Gummow and Hayne JJ)).
  USSC 150; 399 US 78, 100 (1970):
The purpose of the jury trial … is to prevent oppression by the Government … Given this purpose, the essential feature of a jury obviously lies in the interposition between the accused and his accuser of the commonsense judgment of a group of laymen, and in the community participation and shared responsibility that results from that group’s determination of guilt or innocence: cited in Brownlee (2001) 207 CLR 278, 288, 302, 329–30.
 James E Penner, ‘The “Bundle of Rights” Picture of Property’ (1996) 43 University of Cal ifornia Los Angeles Law Review 711, 768. Of course, it is possible to produce a classical legal definition of any term provided that one accepts that the legal definition may have a denotation that differs from the ordinary denotation of the term and one accepts the normative dissonance that this produces. The more that one is committed to a normative (rather than formal) vision of the Constitution, the more likely one is to relax the requirements of the classical approach and the more necessary it becomes to find an alternative basis for stabilising constitutional concepts.
 Jeremy Waldron, The Right to Private Property (1988) 30 in Penner, above n 180, 769. For (most) judges, it is no answer to the difficulty of discerning the criteria for membership of constitutional categories to suggest that they stipulate what those criteria shall be. That avenue is open to experts and scientists, as the recent redefinition of ‘planet’ and the relegation of Pluto to the status of minor planet shows. There are at least two reasons why judges cannot redefine categories in this way. Most judges are committed to a view of their role that sees them as interpreters of the Constitution rather than makers of constitutional meaning (otherwise than in exceptional cases). And, as the public reaction to the redefinition of ‘planet’ shows (see, eg, Tim Thwaites, ‘Goodbye to Pluto from Good, But Painfully Slow, Science’, The Age (Melbourne), 8 September 2006, 15), intuitive or cultural understandings of the content of (even apparently non-normatively determined) categories can exert a strong pull that render definitions that are arrived at by pure stipulation quite unstable. (The planetary example suggests that ‘planet’ is a cultural concept as well as a scientific one.)
 Penner, above n 180, 770.
 Cass R Sunstein, Legal Reasoning and Political Conflict (1996).
 Ibid 39–40.
 Ibid 45.
 Adrienne Stone, ‘Incomplete Theorizing in the High Court: Cass R Sunstein, Legal Reasoning and Political Conflict’  FedLawRw 8; (1998) 26 Federal Law Review 195, 198.
 Ibid 199.
 Ibid 198–9.
  HCA 68; (2001) 208 CLR 516.
 Ibid 544.
 Lange v Australian Broadcasting Corporation  HCA 25; (1987) 189 CLR 520, 566–7. Its members’ attempts to explain how this is to be done have not been notably successful: see, eg, Justice McHugh’s attempted rejoinder to Adrienne Stone’s arguments in Coleman v Power  HCA 39; (2004) 220 CLR 1, 46–8.