2. Topic continuity

Projects in conceptual engineering (ameliorative projects) typically advocate a revisionary analysis of a key term. Thus Clark and Chalmers (Reference Clark and Chalmers1998) advocate a revisionary analysis of the term “belief”; Haslanger (Reference Haslanger1999) advocates a revisionary analysis of the term “knowledge”; Haslanger (Reference Haslanger2000) advocates revisionary analyses of the gender terms “man” and “woman” and of race terms such as “Latino” and “white”; Eklund (Reference Eklund2002) and Scharp (Reference Scharp2013) advocate a revisionary analysis of the term “truth”; Eklund (Reference Eklund2017) advocates a revisionary analysis of normative terms; and public debates over the terms “marriage,” “rape,” and “person,” for example, are naturally understood as debates between those who want to retain the historical use of the relevant term, along with its traditional analysis, and those who think the use of the term ought to change in accordance with a proposed revisionary analysis. Such revisionary analyses aim to provide a term with a new intension (a new meaning) as well as a new corresponding extension. Cappelen puts the point as follows:

Projects in conceptual engineering, then, aim to change the intension and extension of a key term; they aim for semantic change.Footnote ² But projects in conceptual engineering are not defined merely by the fact that, if successful, they would result in semantic change, since semantic change need not come about as the result of conceptual engineering and hence need not count as ameliorative. Think for example of the semantic change over time of terms such as “meat,” “wicked,” “ace,” and “awful,” which occur as a result of “mere semantic drift” rather than as a result of conceptual engineering. In cases of mere semantic drift, the intension and extension of a term changes, but so does the subject matter or topic that the term allows us to talk about. This is not typically so in cases of conceptual engineering. Those engaged in conceptual engineering typically propose a revisionary analysis as a better way of talking about the same topic. What marks out a semantic change as ameliorative, then, is topic continuity. In each of the examples of conceptual engineering mentioned above, we want to say that both those who advocate in favor of the proposed revisionary analysis and those who resist the proposed revisionary analysis are, despite their differences, nonetheless talking about a shared topic: belief, knowledge, men, women, Latinos, whites, truth, marriage, rape, and persons. Similarly, we want to say that if the semantic change takes effect, the relevant term is used to talk about the very same topic both preamelioration and postamelioration. In none of these cases has the topic, or subject matter, changed.Footnote ³

This gives rise to what Cappelen calls “Strawson’s Challenge”: How can a change in the intension and extension of a term be understood as preserving a topic rather than simply changing it? (Cf. Strawson Reference Strawson and Schilpp1963 in response to the account of explication in Carnap Reference Carnap1950.) To see the problem, note that if the topic marked out by a term were identified with the extension of that term, then a topic would necessarily change in line with a change in extension. This would, effectively, render conceptual engineering as described above a conceptual impossibility. A metasemantic framework within which to understand the general phenomenon of conceptual engineering must, then, have the resources to separate the extension of a term from the topic or subject matter that it concerns. Only then will it be able to accommodate the claim of the conceptual engineer that her revisionary analysis provides a better way of talking about the same topic.

Noting that expressions such as “content,” “intension,” and “extension” are theoretical terms, in contrast to expressions such as “what she said,” “what she was talking about,” and “talking about the same topic,” which Cappelen takes to be pretheoretic notions (108), he argues that “topics are more coarse-grained than extensions and intensions, and so expressions that differ with respect to extensions and intensions can be about the same topic” (101).Footnote ⁴ He takes this point to be captured in his principle: “Coarseness: Sameness of topic doesn’t track sameness of extension” (109). The truth of the principle would allow for continuity of topic through a change in the intension and extension of a term, which would in turn allow a revisionary analysis to provide, at least in principle, a better way of talking about the very same topic. Note that what the principle says is that sameness of topic doesn’t track sameness of extension. But by labeling the principle “Coarseness,” Cappelen makes reference to the further claim that this is because topics are more coarse-grained than extensions. Clearly the claims are distinct. It is possible to agree with the content of the principle—that sameness of topic doesn’t track sameness of extension—and to agree that in certain cases topics are more coarse-grained than extensions, but to reject the claim that the latter provides a proper explanation of the former. I return to this point below.

Cappelen aims to establish the point that topics are more coarse-grained than extensions, and hence that sameness of topic does not track sameness of extension, by appeal to two sets of linguistic data: the first concerns the phenomenon of context-sensitivity (cf. Cappelen and Lepore Reference Cappelen and Lepore1997; Cappelen and Hawthorne Reference Cappelen and Hawthorne2009; Cappelen and Dever Reference Cappelen and Dever2016); the second concerns the phenomenon of semantic plasticity (cf. Dorr and Hawthorne Reference Dorr and Hawthorne2014). Since I find Dorr and Hawthorne’s discussion of semantic plasticity problematic in a number of respects, and since Cappelen draws on it merely to reinforce a point he takes already to be established by the linguistic data on context-sensitivity, I leave it to one side here and focus exclusively on the case of context-sensitive terms.Footnote ⁵ According to Cappelen, then, gradable adjectives such as “cold,” “happy,” “expensive,” “smart,” and “interesting” are context-sensitive in the sense that their extensions are determined by fine-grained, contextually variant factors such as comparison classes or cut-off points on a scale. As such, there are infinitely many potential extensions for such adjectives because there are infinitely many possible comparison classes and infinitely many possible cut-off points on different scales. Nonetheless, we can often describe people who utter, for example, the sentence “T is an interesting theory” in different contexts as having said the same thing—namely, that T is an interesting theory. Cappelen says “it is a fact about our reporting practice that we don’t, in such cases, require the comparison class of interesting theories or the cut-off point on a scale of interestingness to be identical” (109). Since sameness of topic goes hand in hand with samesaying, and people can say the same thing despite differences in the intensions and extensions of their terms, people can also talk about the same topic despite differences in the intensions and extensions of their terms; and given the plethora of context-sensitive expressions in our language, this is, according to Cappelen, a general phenomenon.

For present purposes, I am willing to grant the data on context-sensitive terms and, hence, to grant that Cappelen’s examples establish that topic continuity is consistent with variations in intensions and extensions. As he says, “hardly anyone has advocated the view that identity of semantic content is required for samesaying” (110; original emphasis). This means that the very idea of conceptual engineering should not be rejected on the grounds that topic continuity is inconsistent with a change in the intension and extension of a term; it is not. But establishing this point falls short of explaining how topics and extensions can diverge in a way that would make sense of the phenomenon of conceptual engineering. This is for two related reasons. First, to show that topic continuity is consistent with variations in intensions and extensions is not yet to explain how this is possible. To identify the phenomenon is not yet to explain it. Cappelen’s metasemantic framework provides an account of the relation between a term and an extension: there are conditions that need to be satisfied in order to fall into the extension of a term, and it is these conditions that change in the process of conceptual engineering, thereby resulting in a change in the intension and extension of the relevant term. But Cappelen’s metasemantic framework does not provide an account of the relation between a term and a topic. What is needed is an account of how we connect representationally to a topic per se, and that account needs to explain why the representational connection between a term and a topic is not disrupted by intensional and extensional variation of the kind that occurs in conceptual engineering. I return to this point below.

Focusing on cases of context-sensitivity might lead one to believe that topics can be identified with contextually determined similarity classes of extensions. The relation between a term and a topic could then be constructed on this basis. For example, a term t could be thought of as related to, or as representing topic T in context C if and only if t has an extension that is a member of the similarity class of extensions that constitute T in C. We could even allow some vagueness. This understanding of topics would count as metaphysically lightweight, and, hence, prima facie acceptable within Cappelen’s austere framework because it wouldn’t introduce any new representational devices or new kinds of entities. But even if an account along these lines could be made to work in the case of context-sensitive terms, it is fundamentally flawed as an account of topic continuity in the ameliorative case. This is because there is a significant difference between the kind of variation that is tolerated within the bounds of topic continuity in cases of context-sensitivity and the kind of variation that is tolerated within the bounds of topic continuity in cases of amelioration. This is the second reason that Cappelen’s discussion of context-sensitive terms falls short of explaining how topics and extensions can diverge in a way that would make sense of the phenomenon of conceptual engineering. We can see this by looking at each kind of case in turn.

In cases of context-sensitivity, the pretheoretic intuition of samesaying, and hence of topic continuity, is mapped theoretically by a significant overlap between the different extensions of the term as used by each of the samesayers. It may be, as Cappelen says, a fact about our reporting practice that samesaying does not require the comparison class or the cut-off point to be identical, but it does require a significant amount of overlap. What is tolerated within the bounds of topic continuity in cases of context-sensitivity, then, is not variation across the board, but merely variation at the periphery. I do not deny, of course, that two uses of a context-sensitive term may have entirely different extensions. For example, the term “tall” can be applied on one occasion in relation to five year olds and on another occasion in relation to professional basketball players. The heights that fall into the extension of the term in the first context may be entirely different from the heights that fall into the extension of the term in the second context. But discussions of tallness in these different contexts would not be understood as discussions of the same topic. Rather, we would differentiate the topics by noting that the first discussion concerned tallness-for-a-five-year-old while the second concerned tallness-for-a-professional-basketball-player. Samesaying and topic continuity involving context-sensitive terms, then, are not constrained by identity of extension, but they are constrained by significant overlap in extension.

In cases of conceptually engineered terms, in contrast, the pretheoretic intuition of samesaying, and hence of topic continuity, is not mapped theoretically by a significant overlap between the different extensions of the term as used by each of the samesayers. Indeed, there may be no overlap whatsoever between the extension of a term preamelioration and the extension of that term postamelioration, and yet the intuition of samesaying, and hence of topic continuity, may persist. In cases of amelioration, then, what is tolerated is not merely variation at the periphery but a much more radical kind of variation that runs through to the core. For example, the amelioration of race terms can be understood as an attempt to replace a mistaken, biological understanding of race by an explicitly social understanding of race on the grounds that there are no such things as biological races. In this kind of case, there would be topic continuity in the absence of any overlap in extension because the topic of a term such as “Latino,” for example, would be Latinos both preamelioration and postamelioration, even though the conditions that determine the extension of the preameliorative term determine an extension which is empty—empty because there is nothing that satisfies the condition of being a person of biological race. Think also of conceptual engineering projects relating to the term “truth” (cf. Tarski [1933] Reference Tarski, Corcoran and Woodger1983; Eklund Reference Eklund2002; Scharp Reference Scharp2013). Such projects are typically motivated by the attempt to eliminate internal inconsistency that results in paradox. Since, if all goes well, there is internal inconsistency preamelioration but not postamelioration, the term has an empty extension preamelioration (nothing can satisfy inconsistent conditions) but has a nonempty extension postamelioration. Consequently, there is no overlap in extension, but there is nonetheless, as Cappelen is happy to accept, a single topic that persists throughout the process: truth. Although one might want to treat the cases differently, topic continuity through dramatic change in extension is also commonplace in science. For example, the definition of the term “atom” once included the condition that an atom be indivisible but now includes the condition that an atom be divisible. There is, of necessity, nothing which is both divisible and indivisible and, hence, there is, of necessity, no overlap between the extensions of the term pre- and postchange. Nonetheless, there is a single topic that persists throughout the theoretical change: atoms.Footnote ⁶

Many examples of successful conceptual engineering will, as it happens, involve an overlap between the extension of a term preamelioration and the extension of that term postamelioration, but the absence of any overlap is no impediment to topic continuity in the ameliorative case. The upshot is that extensional variation in the context-sensitive case, in which topic continuity presupposes significant extensional overlap, does not provide an adequate model for extensional variation in the ameliorative case, where significant extensional overlap cannot be presupposed. Cappelen’s appeal to the linguistic data on context-sensitivity, then, fails to explain how topics and extensions can diverge in a way that would make sense of the phenomenon of conceptual engineering.Footnote ⁷

I said earlier that a metasemantic framework within which to understand the general phenomenon of conceptual engineering must have the resources to separate the extension of a term from the topic it concerns. Only then, I said, will it be able to accommodate the claim of the conceptual engineer that her revisionary analysis provides a better way of talking about the same topic. Reflection on specific cases makes it clear that an adequate metasemantic framework must have the resources not only to separate the extension of a term from the topic it concerns, but to treat the two as ontologically independent rather than, for example, treating the former as constructed from the latter, or as being coarse-grained relative to the latter. Only then will it be possible for a term to concern the same topic at two times despite there being no overlap between the extensions of the term at those times. This is where Cappelen’s austere metasemantic framework runs into difficulty. In order for topics and extensions to be ontologically independent, there must be two ontologically independent representational relations: one to connect a term to an extension, the other to connect a term to a topic. Only then will it be possible for the topic of a term to remain stable through a radical shift in its extension. But two ontologically independent representation relations requires two representational devices.Footnote ⁸ And this is what Cappelen’s austere metasemantic framework does not have. As such, it cannot provide an adequate explanation of the phenomenon of conceptual engineering.

A metasemantic framework within which to understand the general phenomenon of conceptual engineering, then, requires two kinds of representational device: one to connect a term to an extension, the other to connect a term to a topic. We have been assuming that the intension, or meaning of a term, performs the former role. What performs the latter role, I suggest, is the concept expressed by a term. There is good reason not to conflate the two, and good reason to recognize both (cf. Burge Reference Burge1986; Sawyer Reference Sawyer, McLaughlin and Cohen2007; Sawyer Reference Sawyer2018; Sawyer Reference Sawyer, Burgess, Cappelen and Plunkett2020). I will offer a brief characterization of meanings and concepts here, and I return to the distinction between the two toward the end of the next section when I consider the role of externalism in Cappelen’s austere metasemantic framework.

The meaning, or intension, of a term is externally determined in the sense that it is determined by social practices—patterns of actual use and deference across the community—rather than by the dispositions or beliefs of an individual speaker. Since the social practices relating to a given term may change over time, the intension and extension of a term may change over time. This picture fits naturally with Cappelen’s account. Note that on this picture, linguistic norms are conventional because they are determined by social practice. It is their conventional nature that not only allows them to change but opens them up to explicit challenge. As Burge says:

This is precisely what motivates projects in conceptual engineering. Ameliorative projects challenge the current social practice surrounding the use of a term and, hence, challenge the conventional linguistic norms that govern the use of that term on the grounds that the nature of, for example, belief, knowledge, men, women, Latinos, whites, truth, marriage, rape, or persons has not been correctly characterized. A revisionary analysis is a proposal for how the key term ought to be used, and provides, in effect, what the ameliorators take to be the correct characterization of the relevant topic. If the ameliorative project is successful, the social practice surrounding the use of the key term changes in accordance with the proposed revisionary analysis, and, hence, the revisionary analysis comes to reflect the new conventional linguistic norms that derive from the new social practice.

Thus, one set of conventional linguistic norms can be replaced by another. A conventional linguistic norm can, of course, be challenged on the grounds that it conflicts with other existing linguistic norms, as well as on the grounds of mere preference for a different social practice, but to challenge a conventional linguistic norm on the grounds that it does not correctly characterize the nature of, for example, belief, knowledge, men, women, Latinos, whites, truth, marriage, rape, or persons is to appeal to a norm that goes beyond social practice and, hence, is not determined by convention. Social practice provides linguistic norms relative to which individual use can be assessed as correct or incorrect. But in order for social practice itself to be assessable as correct or incorrect, there must be norms that are not themselves determined by social practice. If all representational norms were conventional, semantic change would always be mere semantic drift. Topic continuity requires nonconventional representational norms. Burge calls such nonconventional, representational norms “intellectual norms,” and their importance is reflected by the title of his paper: “Intellectual Norms and Foundations of Mind” (1986). The sofa/safo example from the paper is often cited in favor of externalism. But the significance of the distinction between conventional linguistic norms and intellectual norms both to the specific example and to externalism more generally are typically overlooked. The protagonist in the example has a complete understanding of the linguistic meaning of the term “sofa,” but doubts the communally accepted definition of the term because he thinks the community as a whole is mistaken about the nature of sofas. He is aiming for truth. It is concepts—components of thought—that make his doubt possible. More generally, it is concepts that allow us to doubt, for example, that beliefs are internal states, that women are females, that marriage precludes rape, and so on. Such doubts are object-level doubts about topics that are represented conceptually. This is the sense in which, as Burge says, “thought can correct meaning” (1986, 714).

The distinction between linguistic norms and intellectual norms (norms of thought) brings with it a corresponding distinction between meanings and concepts. Concepts are externally determined, but not in the way that meanings are. The concept expressed by a term is determined neither by the dispositions or beliefs of an individual speaker nor by the dispositions, beliefs, or social practices of the community as a whole. Rather, it is determined by fundamental, nonconceptual relations to objective, nonlinguistic properties. Again, as Burge says, “even where social practices are deeply involved in individuating mental states, they are often not the final arbiter. This is because the sort of agreement that fixes communal meaning … is itself open to challenge” (1986, 707). Concepts provide norms of thought that are not conventional but are instead beholden to the nature of the objective properties they represent.

Cappelen is right to say that we can “talk about the same topic, e.g., knowledge, belief, freedom, or marriage, even though the extension and intension of ‘knowledge,’ ‘belief,’ ‘freedom,’ and ‘marriage’ change” (103; original emphasis), and he is right to say that such a change can constitute an improvement in the sense of providing a better way to talk about the same topic. But this is not because topics are more coarse-grained than extensions. Rather, it is because a concept represents a topic in a way that transcends social practice; it provides a representational anchor to a subject matter that is capable of remaining fixed as the social practice ebbs and flows around it. It is this distinction between thought and meaning that accounts for the distinction between topics and extensions and, hence, grounds an adequate explanation of the phenomenon of conceptual engineering.

3. The worldly construal of conceptual engineering

At the beginning of chapter 12, Cappelen identifies three possible responses to the question of what is engineered in the process of conceptual engineering:

Cappelen argues, to my mind successfully, against options 1 and 2. Instead, he defends what he takes to be the only remaining option — option 3: the worldly construal of conceptual engineering. But the worldly construal of conceptual engineering is problematic for at least two reasons. Moreover, it is the austerity of Cappelen’s metasemantic framework that forces him to choose between the three options he identifies. The alternative, less austere metasemantic framework outlined toward the end of the previous section brings a fourth, much more palatable option into view. I deal with each of these considerations in turn.

The first reason that option 3 is problematic is that it threatens to undermine the very possibility of topic continuity that we were trying to accommodate. If the process of conceptual engineering changes, for example, what a belief is, and does so, as Cappelen says, by changing the conditions that have to be satisfied in order for an object to fall into the extension of the term “belief,” then the property of being a belief is not a stable objective property that exists independently of the conditions that we associate with the meaning of the term “belief.” Aside from its being intuitively false that conceptual engineering changes the nature of belief, knowledge, men, women, Latinos, etc., the worldly construal of conceptual engineering undermines topic continuity by eradicating the possibility of a single topic—belief, knowledge, men, women, Latinos, etc.—that persists throughout the change.

The second reason that option 3 is problematic concerns statements such as “what a belief is has changed,” the truth of which are definitive of the worldly construal of conceptual engineering. The problem, as Cappelen recognizes, is that when we say “what a belief is has changed,” the semantic content of our utterance is false because on any given occasion of use, the term “belief” will have a stable meaning and, hence, a stable extension, and that stable extension could not, therefore, truly be said to have changed. As Cappelen says, “the conditions imposed on being in the extension are stable relative to all points of assessment” (139). Cappelen responds as follows:

He goes on to say:

I am happy to grant for the sake of the argument that each utterance expresses a large number of propositions only one of which is semantically expressed, but it is unclear how any of the additional propositions expressed by an utterance of “what a belief is has changed” could be true without being metalinguistic. What is needed, after all, is a kind of content that “reflect[s] semantic change” without mentioning the semantic entities (the words) that have undergone the change. Note that Cappelen’s own attempt to capture the salient proposition is manifestly metalinguistic. But unless an utterance of “what a belief is has changed” expresses a proposition which is true without being metalinguistic, the worldly construal of conceptual engineering is incoherent and, hence, untenable.

Finally, once we free ourselves from the constraints of Cappelen’s austere metasemantic framework and embrace the less austere metasemantic framework outlined toward the end of the previous section, a fourth, more palatable, option comes into view. The less austere metasemantic framework allows us to reject options 1 and 2: conceptual engineering is not about concepts, the constitutive principles for concepts, or the constitutive principles for concept possession, and it is not about the words of a specific language. It also allows us to agree with the first part of option 3: conceptual engineering is about the world in the sense of being about, for example, belief, marriage, rape, and persons. But it does so without committing us to the implausible and problematic claim that conceptual engineering involves changing what belief, marriage, rape, and persons are. On the contrary, conceptual engineering involves an individual or group of individuals purporting to point out that the community’s entrenched theory of belief, marriage, rape, and persons is mistaken, and trying to replace that theory with one which they take to be better (even if not ultimately correct).Footnote ¹⁰

The less austere metasemantic framework that I have suggested is based on the distinction between conventional linguistic norms and norms of thought that underpins externalism. It brings with it the corresponding distinction between meanings and concepts. I said earlier that these distinctions are not typically acknowledged by those who advocate externalism, despite the central role the distinctions play in Burge’s own discussion of externalism. Cappelen is typical in this regard. He cites both “experts in the community” and “complex patterns of use over time” as well as “what the world happens to be like” (where this means what the world is like beyond linguistic practice) as elements of the external environment that partly determine extensions and intensions (63). But there is a tension between the claim that the extension of a term is determined by use (social practice), and the claim that the extension of a term depends on what the world happens to be like beyond linguistic practice. If the extension of a term is determined by use (social practice), then the extension of a term will change over time as the beliefs and dispositions of the community change. But if the extension of a term is determined by what the world happens to be like beyond linguistic practice, then the extension of a term will not change even if there is a radical shift in belief across the community as a whole. The worldly construal of conceptual engineering depends on running the two distinct elements together.Footnote ¹¹

The less austere metasemantic framework allows us to distinguish two different ways in which an ameliorative project might be thought to be successful. On the one hand, an ameliorative project might be thought to be successful in the sense that the proposed revisionary analysis is accepted, resulting in a change in the social practice surrounding the use of the relevant term. This sense of success is connected to the ameliorator having “won her case.” On the other hand, an ameliorative project might be thought to be successful in the sense that the proposed revisionary analysis is true. This sense of success is connected to the ameliorator’s revisionary analysis correctly characterizing the relevant subject matter, or topic. Clearly there can be acceptance without truth and truth without acceptance. In Cappelen’s austere metasemantic framework, in contrast, these two ways in which an ameliorative project might be thought to be successful merge together; the worldly construal of conceptual engineering blurs acceptance and truth. According to the worldly construal, what it is to be a belief is determined by the conditions associated with the meaning of the term “belief,” but the conditions associated with the meaning of the term “belief” are determined by communal use. This creates a hint of antirealism that runs through the entire project and which goes against the very spirit of externalism properly understood.