Some Properties are More Essential Than Others: How the Meaning of Kind Terms Changes in Scientific Discourse

Abstract 

I argue that a modified version of Thomas Kuhn’s philosophy of language can accommodate scientific realism and the changes in the meaning of scientific kind terms. Kuhn’s philosophy of language draws upon the work of Ludwig Wittgenstein so as to bolster his magnum opus, The Structure of Scientific Revolutions (SSR) (Pirozelli 345). The basic Wittgensteinian structure of Kuhn’s account of kind terms is correct: scientific kind terms correspond to family resemblance concepts, and family resemblance concepts must hang together within a taxonomy. Kuhn hastily concludes from this that it is impossible to compare one taxonomy to another because kind terms cannot be translated from one taxonomy to another. I agree with Sankey that once we make a distinction between sense and reference, Kuhn’s semantic theory does not imply antirealism. Kuhn focuses too narrowly on physics and chemistry (Mayr 333) and Sankey’s critique inherits this myopia (“Incommensurability” 11). The goal of scientific kind terms is to correspond to natural kinds as closely as possible, as opposed to role-bearers within natural laws. I introduce a concept of natural kinds where objects are grouped according to their fundamental resemblances. To make sense of this in terms of the philosophy of language, I suggest that the relevance of a property for the purposes of classifying an object in a taxonomy comes in degrees. 

I. Introduction

The philosopher Thomas Kuhn spent the latter portion of his career outlining a philosophy of language that clarifies and undergirds his magnum opus, The Structure of Scientific Revolutions (SSR). In SSR, Kuhn argues that scientific revolutions are paradigm shifts—a shift from an earlier paradigm, corresponding to a vocabulary that makes contextual sense of scientific terms, to an entirely new paradigm with an entirely new vocabulary. Even if some terms survive the paradigm shift, the understanding and concepts behind them change (Pirozelli 368). The result is Kuhn’s “incommensurability thesis”: two theories or paradigms are incompatible if and only if

1. “The meaning of the vocabulary employed by theories varies between theories. 

2. Translation is impossible from the vocabulary of one theory into the vocabulary of another.

3. As a result of (1) and (2), the content of such theories may not be compared” (Sankey, “language of science” 127). 

The incommensurability thesis challenges a key common assumption about scientific terms: that the meanings of scientific terms are “theory-neutral” as opposed to “theory-laden” (Sankey, “language of science” 126). 

I agree with Kuhn that family resemblance concepts undergird the meaning of scientific kind terms, and that the meaning of scientific terms depend on their taxonomy; however, I disagree with Kuhn that a change of taxonomy, or corresponding paradigm, must be arbitrary. I agree with Sankey that once we make a distinction between sense and reference, there will be a non-arbitrary way of comparing and selecting taxonomies (“Incommensurability” 12). I offer some critiques of Sankey and conclude that we can select a taxonomy based on the fundamentality of members’ properties so as to more closely cut nature at its joints. An important consequence of my thesis is that Kuhnian semantic theory does not entail scientific anti-realism. 

II. Overview of Kuhn’s Semantic Theory 

To understand how Kuhn arrives at the incommensurability thesis, we need to understand his semantic theory. A kind term refers to a family resemblance concept. Two things resemble each other if and only if they share at least one property. A family resemblance concept, by extension, is a loose collection of things, or members, that resemble each other and whose resemblance relations often overlap (Pirozelli 352). Kuhn recognized, however, that a concept is only coherent if it is closed, that is, if it has definite boundaries. On their own, family resemblance concepts are open by the very nature of the resemblance relation: only one shared property with any existing member of a family is necessary for a new member to join a resemblance family. If this were the end of the story, then family resemblance concepts would result in nothing more than nominalism, i.e., the claim that the meaning of kind terms is just arbitrary (Pirozelli 351). According to Kuhn, the boundaries of a family resemblance concept are defined by other surrounding family resemblance concepts in the same taxonomy. So, a taxonomy is a collection of family resemblance concepts whose boundaries are mutually defining (Pirozelli 352). 

Taxonomies, according to Kuhn, follow the “no-overlap principle”; that is, there are no intermediate members that could be placed in more than one family within a given taxonomy (Pirozelli 356). So, although the concepts in a taxonomy cannot stand on their own, each taxonomy is perfectly self-consistent and independent. For any given taxonomy, therefore, there is a fact of the matter as to which kind the member belongs in the taxonomy. However, there is no fact of the matter as to which taxonomy is the most appropriate or true according to Kuhn. This has the immediate consequence that scientific kind terms only have meaning in the context of specifiable taxonomies, respectively, each taxonomy corresponding to a paradigm. Statements that use kind terms are only true or false with respect to a certain taxonomy. Because taxonomies do not have truth values, scientific discourse does not refer to the external world according to Kuhn (Pirozelli 361). 

I find Kuhn’s analysis compelling up to a point: there is good reason to believe that the concepts corresponding to scientific kind terms are family resemblances, and there is good reason to believe that these family resemblances delimit each other within a taxonomy. However, the incommensurability thesis and the resulting anti-realism are hastily drawn inferences. 

III. Mechanics of Kuhn’s Semantic Theory

Andersen gives an example of the set theory behind Kuhn’s family resemblance concepts and taxonomies (qtd. in Pirozelli 351). Suppose we have some taxonomy with families K, J, and Q. We use these families to categorize members that may have any number of properties. We call the relevant Properties α, β, γ, δ, ε, ζ, η, θ, λ, and μ, respectively. Table 1 catalogs and compares the properties of six different objects, and based on that comparison, Table 1 categorizes the objects.

Resemblance is a binary reflexive relation between two objects: two objects either resemble each other or they do not; they either share at least one property or they share no properties. It’s intuitive to think, however, that resemblance comes in degrees rather than in binaries: that is, even though j1 and k2  resemble each other in the binary sense because they share Property ζ, one could not place k2 in J because it has more properties in common with k1 and presumably with the other members of K not listed in Table 1. This is how K and J define each other’s boundaries and how the no-overlap principle is preserved (Pirozell 351). 

There is another way in which the binary, mathematical resemblance relation may be too primitive for scientific kinds: the relevance of properties need not be binary. We can filter which properties of objects are of interest based on our taxonomy. For example, the property ‘weighs on average 1,000 pounds’ is not relevant for placing the tiger in the category ‘mammal.’ In Table 1, the Properties θ, λ, and μ must be, likewise, considered totally irrelevant properties. If we considered Properties θ, λ, and μ relevant, then q2 would be an intermediate member; that is, we could reasonably place q2 in J or Q, and so there would be no way to determine the boundary between J and Q. 

Suppose, however, that relevance itself were not binary as I suspect is the case in real scientific taxonomies. Instead, the relevance of a property could have some score between 0 (completely irrelevant) and 1 (completely relevant). For example, the luminosity and relatively large size of a star are relevant properties for classifying that star as a red giant, but the age and mass of the star are even more relevant for classifying the star as a red giant. Both propositions are still true: (1) the approximate age and mass of a star are more relevant for taxonomizing it with other stars than luminosity and size and (2) the luminosity and size are still relevant in the grouping process.  

IV. Blocking Kuhn’s Inference to Anti-Realism.

My intention is to block the Kuhnian argument for anti-realism. We can express Kuhn’s argument as a syllogism: 

P1: There is no epistemically objective way to select one taxonomy over another.

P2: If scientific realism is true, there is an epistemically objective way to select one taxonomy over another. 

C: Therefore, scientific realism is not true. 

I have assented to P2 and the conclusion. So we must dissect P1, which is itself a conclusion of the incommensurability thesis. Recall that the incommensurability thesis states the following: 

P1.1) “The meaning of the vocabulary employed by theories varies between theories. 

P1.2) Translation is impossible from the vocabulary of one theory into the vocabulary of the other.

P1) As a result of (P1.1) and (P1.2), the content of such theories may not be compared” (Sankey, “language of science” 127). 

This argument, however, is invalid; P1 does not follow conclusively from P1.1 and P1.2. Sankey argues that it does not matter whether one taxonomy can be translated to another because we can compare how many true statements can be formulated within each taxonomy and select the taxonomy that allows us to make more true statements. To justify how a statement can be true independent of its taxonomy, Sankey draws a sharp distinction between sense and reference, and he states that truth depends on reference alone. Sense is roughly whatever associations are in the head of the scientist who utters a particular term whereas reference is the external thing or set of things to which the term refers. Although changes in taxonomy alter the sense of scientific kind terms, the things to which the taxonomy refers remain the same (Sankey, “Incommensurability” 12). 

V. Critique of Sankey

  So far, I agree with Sankey’s reply to Kuhn; however, Sankey does not flesh out his account enough. Take the following example: with respect to Taxonomy A, “The duck-billed platypus is a mammal” is true, and with respect to Taxonomy B, “The duck-billed platypus is a reptile” is true. Both Taxonomy A and Taxonomy B contain the kind terms “mammal” and “reptile.” Sankey would say that if Taxonomy A allows us to make more true generalizations about animals, we should conclude that Taxonomy A is more likely to be True (with a capital ‘T’) and therefore, duck-billed platypus is more likely a mammal species. But if the real test of a taxonomy were just that one could assert more true statements about the objects themselves, the best taxonomy would be the one where every object occupies its own kind. The duck-billed platypus, for example, is such a unique creature that there are many generalizations about reptiles and mammals that do not apply to it. Consequently, if we follow Sankey’s advice, the duck-billed platypus should really occupy its own animal group. I doubt that biologists will take this advice, however, because the very purpose of grouping animals is to make interesting generalizations about multiple species. Therefore, the real test of a taxonomy is both to maximize the number of true generalizations one can make and to make the most interesting generalizations. 

Furthermore, interesting generalizations must be relevant for the purposes of the taxonomy. Sankey agrees with Kuhn that the purpose of scientific taxonomies is to delimit groups of objects following different natural laws such that the natural laws are built into the meaning of the kind terms (“Incommensurability” 11). This has the peculiar consequence that a statement like “Planets revolve around the Sun” is an a priori statement under the Copernican taxonomy because when we analyze Copernicus’ concept ‘planet,’ we will find that ‘revolving around the Sun’ is just part of what it means to be a planet (Sankey, “Incommensurability” 11; Pirozelli 366). Peculiar though it may seem, Kuhn is not far from the truth. To bring Sankey’s commentary closer to the truth, however, we should make three critiques: first, the emphasis on natural laws unduly privileges sciences like chemistry and physics where the concept of natural law is most applicable (Carroll); second, natural laws are distinct from true generalizations (Carroll); and third, different natural kinds may follow the same regularities or laws. If we are to make Kuhn’s semantic theory compatible with scientific realism, we have to move away from Kuhn’s obsession with natural laws. As Mayr points out, Kuhn focuses almost exclusively on the history of physics where nearly exceptionless generalizations are possible (333). But philosophers have long questioned whether there are natural laws at all in the special sciences—branches of science besides physics (Carroll). 

Sankey commits himself to two distinct ideas that are in tension: that we should select one taxonomy over another by comparing the number of assertable true generalizations and that the better taxonomy latches onto the natural laws more accurately. Carroll notes that natural laws, in addition to being true generalizations, have an element of necessity to them. So, although the statement “All gold spheres are less than a mile in diameter” may in fact be true, there is no fact about the universe that makes it necessarily true (Carroll). The statement “All uranium spheres are less than a mile in diameter,” by contrast, is necessarily true due to the critical mass of uranium (Carroll). 

It’s clear why Sankey prefers the criterion of assertable true generalizations over natural laws: metaphysical necessity, a necessary and missing ingredient in natural laws, is difficult to detect. We only have straightforward epistemic access to true generalizations; however, for Sankey’s account to work, we would really need epistemic access to the much more elusive natural laws. 

Finally, being classified in a certain kind in a special science taxonomy does not entail following a set of laws unique to that kind. For example, the differences between mammals’ and reptiles’ behaviors and looks are not a result of their different roles in natural laws. Reptiles and mammals follow the same fundamental Darwinian and genetic regularities. 

VI. Fundamentality and Natural Kinds

What realists expect of scientific taxonomies is not only that they make more true statements assertable than their predecessor taxonomies, but also that the kind terms themselves refer more accurately to natural kinds. Consider the difference between statements like “This bear is a mammal” and “Mammals are furry or hairy.” The latter statement is a true generalization, the type of statement that Sankey and Kuhn overly fixate on. The former statement, however, implies that mammal is a natural kind because the statement employs the ontological form of the verb “to be.” Therefore, because realists should say that both statements are approximately True (with a capital ‘T’), the realist goal of scientific taxonomies is not only to make true generalizations but also to arrive at the natural kinds. Without affirming statements within scientific taxonomies that employ the ontological form of the verb “to be,” a realist’s ontology cannot be constituted by scientific taxonomies. 

We want to choose the taxonomy that allows us to make more interesting and true generalizations. For a claim to be interesting, it has to make claims about multiple members and to fulfill the purposes of the taxonomy. Although Kuhn and Sankey think that the purposes of the taxonomy are to describe the workings of natural laws, we have shown that this is not necessarily applicable to special sciences or readily accessible epistemically. Instead, we can arrive at a natural kind by way of the fundamentality measure, as a family resemblance within a Kuhnian taxonomy where the objects fundamentally resemble each other. 

VII. Fundamentality as A Measure of Relevance

The more fundamental a property is, the more relevant it is in the taxonomy. An understandable worry may arise at this suggestion. We are trying to demonstrate how Kuhn’s semantic theory is compatible with scientific realism, and so we cannot subtly assume scientific realism in our discussion of the fundamentality of properties. 

We have to provide an epistemic account of how scientists access the relative fundamentality of properties. Once scientists come to form justified beliefs about the relative fundamentality of properties, they can weigh those properties accordingly when building a taxonomy. A justified attribution of fundamentality does not have to be necessarily true. This preserves the meaning of scientific kind terms of an outmoded paradigm or taxonomy. Although propositions that include categories of an outmoded paradigm are presumably not true (to the scientific realist according to whom truth is a concern of science), they still retain meaning within their historical context or paradigm. So by providing an epistemic account of fundamentality as a measure of relevance, we leave space for false but meaningful scientific statements, an important implication of scientific realism. 

Bradley argues that given a “sufficiently rich description of the world,” we can tell whether one property is more fundamental than another property a priori (55). This is not to say that we can know what fundamental reality is a priori; rather, we can infer from our prior knowledge which properties are fundamental and which are derivative (Bradley 55). Bradley imagines a bookshelf of physics textbooks, each containing just as much information on the laws of physics in one unique possible universe as all the others. From this bookshelf, we cannot infer a priori which book describes the actual laws of physics, but we can infer for each possible universe which properties are fundamental as opposed to derivative (Bradley 56). The claim that we can tell which properties are fundamental a priori is crucial for my thesis because as we noted earlier, scientific realists are implicitly committed to the idea that there are meaningful but false statements in science. So we can say of each physics textbook on Bradley’s imaginary bookshelf that it is meaningful and coherent precisely because the fundamentality of properties in each textbook does not rely on that textbook’s correspondence with the actual world.  

Bradley states that fundamentality has three qualifications that can be evaluated a priori: (i) “similarity,” (ii) “causality,” and (iii) “minimality” (51). We observe objective (i) similarities across objects, and we rightly infer that a more fundamental property probably explains these similarities. We observe that (ii) objects have similar causal powers, and we rightly infer that a more fundamental property probably explains these causal powers. And finally, (iii) the minimality qualification consists in not positing more fundamental properties than are necessary. An account of fundamental properties is more likely to be correct if it minimizes fundamental ontological commitments and maximizes explanatory power (Bradley 52-53).

One might worry that we are subtly begging the question by including in that sufficiently rich description of the world the very taxonomies whose meaning we are trying to explain. In other words, one might worry that fundamentality is not paradigm independent. But we must show fundamentality to be paradigm independent if we want to make Kuhn’s semantic theory (along with my modification) compatible with scientific realism. Thanks to Bradley, we can see that a property’s fundamentality can indeed be determined independent of any paradigm or taxonomy (55-56). None of the three qualifications of fundamentality– similarity, causality, or minimality–requires the kind terms whose meaning we are trying to tether to the real world. 

Bradley notes that we have to assume that there is an “ontological base” in order to use (iii) minimality as a criterion for fundamentality (54). Bradley admits that we must assume that there are rock-bottom fundamental properties before using the minimality criterion, although metaphysicians like Schaffer have controverted that claim (qtd. in Bradley 54). For my purposes though, it is enough to clarify three things: (1) there are good a priori reasons to think that there are rock-bottom fundamental properties though I won’t explore those here, (2) by assuming that objects have fundamental properties, I am not assuming scientific realism, and (3) were we to reject rock-bottom fundamental properties and (iii) the minimality criterion, that would not call into question the (i) similarity and (ii) causality criteria. Scientists don’t need to know the rock-bottom fundamental properties in order to discern that some properties are more fundamental than others (what I call relative fundamentality). And the justified belief of relative fundamentality is all that I need to make Kuhn’s semantic theory compatible with Darwinian epistemology and scientific realism. The (i) similarity and (ii) causality criteria alone provide us with such justifications. So, while I am confident in (iii) the minimality criterion and the existence of rock-bottom fundamental properties, that’s not a hill that I have to die on.   

Lastly, I should note a limitation that my thesis inherits from Bradley’s, that his account strictly applies to properties of objects. It does not attempt to explain how we can know that some objects themselves, for example, are more fundamental than others. Bradley talks of “objects” and “properties” in the strict metaphysical sense (63-64), but Kuhn talks of “objects” and “properties” in the more expanded linguistic sense such that object is synonymous with member (Pirozelli 349). Kuhn, for example, might call a supernova an object in the sense that it is something that scientists taxonomize. But, in fact, a supernova is not an object but an event in the strict metaphysical sense. Nonetheless, my modified version of Kuhn’s semantic theory will apply to most scientific taxonomies: those that concern metaphysical objects with metaphysical properties. I can easily imagine that my modified version of Kuhn’s semantic theory could in principle expand to include events and other referents of scientific taxonomies. Bradley has opened the floodgates for philosophers in the future to offer accounts of how we can justify a belief in the fundamentality of other referents besides objects (64-65). These future accounts will be compatible with my thesis so long as those accounts are likewise a priori, those accounts don’t require the taxonomies in question to get off the ground, and they don’t assume scientific realism. 

VIII. Conclusion: Consequences for epistemic objectivism and the notion of fundamentality

Although Kuhn is essentially correct in his formulation of family resemblance concepts and taxonomies, Kuhn’s fundamental error is in assuming that sense determines reference. Sankey’s error is in conceding to Kuhn’s analysis that kinds necessarily play a role in natural laws. The purpose of scientific taxonomies is not necessarily to delimit the objects that play certain roles in natural laws, but rather, the purpose of scientific taxonomies is to pick out natural kinds. We can get closer to natural kinds by selecting the taxonomy that groups objects by the most fundamental class of properties of which we are aware. Other less fundamental properties may still be relevant because relevance is not binary. Our epistemic access to the relative fundamentality of these properties does not depend on the taxonomy in question. Hence, we can objectively select one taxonomy over another. Therefore, scientific realism is preserved even as we embrace this modified Wittgensteinian view of the meaning of kind terms.  

As for the implications in the philosophy of language, the relationship between this modified Wittgensteinian view and essentialism is intriguing and worthwhile to assess in future research. At first glance, essentialism and Wittgensteinianism share this assumption that properties are either completely relevant or irrelevant, and as I have argued, this assumption does not hold up for scientific kind terms. 

Suppose for a moment that essentialism were true. All the relevant properties for an essentialist taxonomy would be the union of the sets of essential properties (where each kind has one set of essential properties). So, a property is essential if and only if it is relevant. So, were I to explain my modified Wittgensteinian view to an essentialist (e.g., a layperson), I would say, “Some properties are more essential than others.” 

Table 1

Members (k1-q2) and their Properties (α - μ) in a Generic Taxonomy

Note:  This table is based on figures and examples from Pirozelli (2020). The table represents a single taxonomy with three kinds, or categories: K, J, and Q. The six objects in the first column are named according to the kind by which they are classified. (Objects k1 and k2 belong to K; Objects j1  and j2 belong to J; and Objects q1 and q2 belong to Q.) A checkmark  indicates that the object corresponding to that row possesses the property corresponding to that column. For example, the checkmark on Row “j1” Column “Property ε” indicates that Object j1 possesses Property ε. With this table format, two objects resemble each other if and only if their rows have checkmarks in at least one and the same column.