Explication or Amelioration? Carnapian Clarification as the Normative Basis for Conceptual Engineering

Author: Matthieu Queloz
Published in: Forthcoming in The Monist. Special issue on Explication and Conceptual Engineering.
Canonical entry: https://www.matthieuqueloz.com/entries/explication-or-amelioration-carnapian-clarification-as-the-normative-basis-for-conceptual-engineering/
Published PDF: https://philpapers.org/archive/QUEEOA.pdf

Machine-readable text companion generated from the PDF. Page markers follow the printed pagination.

[p. 1]

ABSTRACT

As conceptual engineering fractures into explication pursuing exactness and amelioration pursuing justice, the field risks losing its focus. I argue that unifying these projects requires retrieving a crucial insight from Rudolf Carnap: that attempts to improve concepts must start with the preliminary stage of practical clarification. However, Carnap’s account of clarification in terms of predictive proficiency remains normatively inert and biased towards exactness. I expand it into a normative diagnosis of the needs underpinning a concept’s inferential structure. This reveals whether properties like vagueness are flaws that need fixing or features worth preserving. Ultimately, practical clarification can orient a division of conceptual labour, determining when concepts call for scientific exactness or social utility and accommodating both aspirations within one methodology.

Keywords: Carnap, clarification, normativity, explication, amelioration, conceptual engineering

1. A FIELD DIVIDED: SCIENTIFIC EXACTNESS VS. SOCIAL JUSTICE

As philosophers increasingly embrace an image of themselves as conceptual engineers whose contribution lies in rewriting the mindware of inquiry and society, the field is becoming riven by a methodological divide. On one side are those channeling the spirit of Rudolf Carnap, who championed the method of explication as a way of replacing the messy tangle of vague, inconsistent, and insight-foreclosing concepts we inherited with the crystalline exactness, consistency, and fruitful simplicity of scientific instruments (Brigandt 2010, 2012; Scharp 2013; Cappelen 2018, 2020; Scharp 2020). On the other side is a growing cadre of conceptual engineers whose aim is not explication, but amelioration: they seek to recast concepts to rectify social injustices (Haslanger

[p. 2]

2012, 2020a, b). Where the former chase theoretical virtues like exactness and consistency, the latter prioritise practical virtues like justice, moral progressiveness, and political expedience.

Caught between these diverging aims, conceptual engineering risks losing its focus. Is it a form of scientific theory-building or a form of political activism? Both aspirations are present and to varying degrees intertwined in the conceptual engineering literature – individual conceptual engineers themselves move back and forth across this divide (Haslanger 2000; Brigandt and Rosario 2020; Cappelen 2023), and some conceptual engineering projects combine both aspects (Dutilh Novaes 2020). From a historical perspective, that is not surprising: Enlightenment philosophers already sought to reform tradition in pursuit of the twin ideals of science and social progress; the Vienna Circle’s reaffirmation of Enlightenment ideals against the neo-Romantic fervour that gripped Vienna in the early twentieth century likewise reacted to anti-scientistic streaks on the one hand and to social, anti-modernist streaks on the other (Uebel 2004; Carus 2007, 3; Romizi 2012; Sigmund 2017; Edmonds 2020); Robert Musil keenly captured both aspects by portraying the cultural crisis in The Man Without Qualities as precipitated by the interplay between anti-scientistic intellectuals and anti-modernist aristocrats. In reviving the spirit of the Vienna Circle, present-day conceptual engineering inherits this longstanding duality.1

But the coexistence of these divergent tendencies threatens to produce a tension within the field. There will be cases where the scientific ideals of explication pull in one direction while the social ideals of amelioration pull in another direction. Does this mean that we should see conceptual engineering less as one philosophical method and more as a fractured field on the brink of bifurcation, or is there a way to accommodate both aspirations within a single methodology? And

See Dutilh Novaes (2020) and especially Yap (2022), who recover the respects in which the Vienna Circle anticipated socio-politically engaged, ameliorative engineering as much as science-minded, explicative engineering.

[p. 3]

if we retain both aspirations, how do we decide, for any given concept, which set of conceptual virtues to prioritise? In this paper, I argue that the key to seeing what holds these two kinds of project together lies in retrieving a crucial insight from Carnap: that attempts to improve concepts must start with the preliminary stage of clarification. Before we can determine how a concept should be engineered, we must clarify what the concept is and why we use it to begin with. While the need for clarification has been comparatively neglected by recent work on conceptual engineering, Carnap was acutely aware of how the very problem we set out to address by transforming our conceptual repertoire remains underdetermined unless we first put in the work of clarification. In Logical Foundations of Probability, he warned against the “temptation to think that […] it does not matter much how we formulate the problem” (1962, 4). On the contrary, he insisted, before the work of explication can begin, one must make the concept one is concerned with “at least practically clear enough to serve as a basis for an investigation” (1962, 4). For Carnap, however, this “practical” clarity was to be understood in purely descriptive and predictive terms: person Y achieves practical clarity regarding person X’s use of a term if Y can accurately predict how X will apply it in ordinary cases. While Carnap was right to insist on clarity, I argue that his conception of practical clarity as predictive proficiency is in itself not normative enough to guide attempts to improve concepts, while being at the same too normative in its implicit presuppositions about what would count as an improvement. I propose to remedy this by developing an account of clarification as a normative diagnosis that offers more guidance while presupposing less, thereby enabling it to act as the orienting preliminary to both explication and amelioration. Clarification must involve more than disambiguating terms and gaining a sense of how they are used; it must involve grasping what role the inferential articulation of the concept plays in our reasoning and diagnosing what practical needs this meets. For it is only by understanding what needs underpin the distinctive reasoning

[p. 4]

pattern encoded in a concept that we can know in which direction to re-engineer it. We cannot know whether a concept should be more or less exact until we know whether its vagueness is a help or a hindrance. If clarification reveals a need for greater exactness, then it should indeed be followed by explication in Carnap’s sense. As we shall see, however, clarification will sometimes reveal conceptual needs pulling in a different direction, including towards greater vagueness. Then the pursuit of exactness may prove a case of sharpening a conceptual tool to uselessness.

2. CARNAP’S “PRACTICAL CLARITY” AND ITS LIMITS

If Carnap’s work is well worth revisiting for present-day conceptual engineers, it is notably because of the attention he pays to the easily overlooked, but crucial stage of clarification that must precede any attempt at explication. For Carnap, explication is not an act of stipulation ex nihilo, of the sort that David Chalmers envisages under the heading of de novo conceptual engineering (Chalmers 2025). Explication is the transformation of concepts we already possess. It is essential to this enterprise that it should start out from an explicandum – a term from everyday language or some prescientific or pretheoretical stage of inquiry – that is “inexact” (Carnap 1962, 3). The aim is to replace this explicandum with an explicatum: a new concept that is exact, fruitful, and preferably simple while retaining some similarity to the explicandum.2

When set up in these terms, however, explication immediately runs into a difficulty. Because the explicandum will, by definition, never be given in exact terms – for “if it were, no explication would be necessary” (1962, 4) – the problem to which explication is to be a solution will itself never be “stated in exact terms” (1962, 4). How can one improve what one only has a wobbly grasp on to begin with?

For discussions of these criteria guiding explication, see Brun (2016), Dutilh Novaes and Reck (2017), Dutilh Novaes (2020), and Reck (2024).

[p. 5]

Carnap’s solution is to insist on a preliminary stage of clarification. Before we can construct the explicatum, we must make the explicandum “at least practically clear enough to serve as a basis for an investigation” (1962, 4). Carnap is not demanding exactness at this stage – that is the aim of the ensuing explication stage, not its prerequisite. He is demanding practical clarity. What does practical clarity consist in? Judging by Carnap’s own examples (1962, 4–5; 1963, 933), it notably involves being clear about what contexts of use one has in mind: are we concerned with the concept expressed by the term “salt” as used in chemistry or as used in household parlance? He also takes the achievement of practical clarity to involve disambiguating between different concepts that a term expresses in ordinary usage: before replacing the concept warm with the quantitative concept of temperature, one must first informally distinguish between two different concepts that coexist in ordinary usage: (1) “the thing x feels warm to person y” and (2) “the thing x is warm” (Carnap 1963, 933). But clarification cannot just be a matter of disambiguating terms. One might think it must be a matter of giving an analysis of the concept expressed by a term before attempting to improve it – in line with recent injunctions to treat conceptual analysis as a necessary preliminary to good conceptual engineering (Glock 2025). This cannot be what Carnap had in mind, however, because he explicitly says that he takes conceptual analysis to suffer from precisely the same difficulty: those who rush to answer questions like “What is justice?” or “What is mind?” without first clarifying the basis of their investigation set themselves up for failure (1962, 4). Carnap thus takes practical clarification to be something that must precede analysis as much as explication. The closest that Carnap comes to spelling out what he means by “practically clear” is the following passage:

[p. 6]

What X means by a certain term in contexts of a certain kind is at least practically clear to Y if Y is able to predict correctly X’s interpretation from most of the simple, ordinary cases of the use of the term in those contexts. (Carnap 1962, 4) A first feature of this criterion worth noting is that instead of treating practical clarity as a property of a concept or term, it treats it as a relation between two people: what X means by a certain term is practically clear to Y. This gives the criterion a quasi-ethnographic character and moves it from sphere of pure semantics into the sphere of descriptive semantics (Ebbs 2026). Carnap’s motivation for framing the criterion this way is, I think, best understood in light of his “Meaning and Synonymy in Natural Languages” (1955). There, Carnap makes explicit his verificationist ambition to give the notion of the intension of a predicate “a clear, empirically testable sense” by specifying an “empirical procedure for testing, by observations of linguistic behavior, a hypothesis concerning the intension” (1955, 41).3 His criterion of practical clarity embodies the same ambition: it renders one’s achievement of practical clarity empirically testable by specifying a procedure by which one can put it to the test.

Second, Y’s practical clarity in relation to X manifests as an ability on Y’s part, namely the ability to accurately predict X’s “interpretation” of the term in question (in contexts of a certain type). What does it mean to accurately predict X’s “interpretation”? To retain the dispositionalist spirit of spelling out clarity in terms of predictive ability, X’s “interpretation” of a certain term is best read as a reference to what X treats the term as applying or not applying to. This dispositionalist reading is, again, supported by “Meaning and Synonymy in Natural Languages,” where Carnap writes:

For a detailed account of Carnap’s descriptive semantics,

[p. 7]

That a predicate “Q” in a language L has the property F as its intension for X, means that among the dispositions of X constituting the language L there is the disposition of ascribing the predicate “Q” to any object y if and only if y has the property F. (1955, 42) In other words, X’s “interpretation” of a predicate “Q” is to be captured in terms of the conditions an object y must fulfill in order for X to be disposed to apply “Q” to y (though it should be noted that Carnap is talking here mainly about “observable properties,” meaning properties that are either directly observable or explicitly definable in terms of directly observable properties). Third, we can address the looming worry that it would be unreasonable to expect Y to accurately predict X’s application of some term “Q” across all situations that arise in some type of context C by recasting Carnap’s criterion of clarity in gradualised terms:

X’s use of term “Q” in context C will be practically clear to Y to the extent that Y can successfully predict, for any object y presenting itself in that context, whether X will treat “Q” as applicable to y or not. Finally, the inductive basis for this prediction, Carnap tells us, is given by “most of the simple, ordinary cases of the use of the term in those contexts” (1962, 4). This must refer not just to X’s past uses, but to the term’s ordinary use in such contexts more broadly, for it would be implausible to suggest that Y draws on nothing but X’s own past use of “Q” (as far as Y observed it). We do not approach every interaction as an exercise in radical translation. Rather, we approach someone’s use of a term with our expectations initially calibrated to the term’s ordinary use, and then adjust for systematic deviations from that use as far as necessary to interpret someone’s idiolect (as we are now doing in interpreting Carnap’s distinctive use of “clear”). Practical clarity is achieved once we are no longer surprised by X’s applications. Where X’s use aligns with ordinary use, practical clarity is thus secured from the outset.

[p. 8]

There is a valuable insight here that conceptual engineers ignore at their peril: before we can improve a concept, we must know what work it is currently doing. However, Carnap’s conception of clarification also suffers from two limitations.

The first limitation is the lack of normative guidance provided by his notion of practical clarity. How much does one really understand about a concept when one knows only how a speaker will apply (or withhold) a term in a given situation? It is not obvious that one understands enough to explicate or otherwise re-engineer it well.4

What exactly would someone who had achieved practical clarity still be missing? One worry is that being able to “predict correctly,” as Carnap puts it, is not the same as understanding why the application of a term is fitting. We can sharpen this objection by considering the striking alignment between Carnap’s notion of practical clarity and the kind of competence exhibited by transformerbased Large Language Models (LLMs). The latter are prediction machines par excellence, trained to anticipate exactly how a human language user in a certain role and context would apply a term. By Carnap’s definition, therefore, LLMs achieve “practical clarity,” yet seemingly without understanding why a term might be fitting.

One might counter on Carnap’s behalf that accurately predicting a term’s application often itself requires understanding why the term is or is not fitting. Granted, the application of “bachelor,” “even number,” or “palindrome” follows relatively simple rules. But one cannot become a good predictor of X’s use of the term “sarcastic praise” without learning to recognise the relevant features of an utterance and its context that might justify the term’s application. This shows that the task of predicting a term’s application is more demanding than it first appears. Just as predicting the resolution of a whodunnit requires one to track who emerges as the likely culprit and why (simply guessing “the butler” every time is not going to cut it), predicting how a term will be applied

My use of the term “re-engineer” follows Brun (2016), who uses it as a hypernym of explication.

[p. 9]

requires one to track what features of a situation render it applicable or inapplicable. If LLMs can accurately predict what utterances we would apply the term “sarcastic praise” to, which they seem able to do (Templeton et al. 2024), this should be taken as evidence that they develop some sensitivity to the features of utterances and their context that warrant the application of the term. After all, what a system is trained to do tells us little about how it learns to do it (Levinstein and Herrmann 2024; Herrmann and Levinstein 2025), and there is increasing evidence that optimizing for the deceptively simple objective of next-word prediction can entrain the emergence of strikingly sophisticated mechanisms (Beckmann and Queloz 2025). Whatever the eventual verdict on LLMs, however, the comparison brings out that Carnap’s notion of practical clarity is less thin and more demanding than it seems. If a cognitive system fulfills the criterion of predictive accuracy for a term whose applicability cannot be determined by simple rules, this implies that it has some grasp of what makes the term fitting. This defense of Carnap’s criterion is insufficient, however. For even if accurate prediction implies a sensitivity to the criteria that govern a term’s application, it does not guarantee insight into the point of thinking in those terms. What is gained by organising one’s thoughts and practices around the concept that the term expresses? What is the background of human concerns that renders this particular concept important? One can predict a term’s application and even engage in conceptual analysis without addressing these questions. Yet this ignores something crucial. As one witness to the heyday of conceptual analysis later reflected:

What we tended to do was to pick up some distinction or opposition, and go very carefully into it and into the various nuances that might be attached to it, […] without enough reflection on what background made this set of distinctions, rather than some other, interesting or important. (Williams 1982, 119).

[p. 10]

This retrospective critique of conceptual analysis vindicates Carnap’s insistence that even analysis should be rooted in practical clarity; but it also points beyond Carnap’s own elaboration of the relevant form of clarity, underscoring that conceptual analysis runs the risk of aimlessly and endlessly subdividing ever finer threads in the tissue of our conceptual repertoire unless it is guided by a sense of what renders a particular concept or distinction important. Even a descriptive analysis of a concept should take its bearings from a clarifying grasp of why we have the concept to begin with. To truly understand the what, we cannot ignore the why.

This same point applies even more forcefully once we turn from analysis to explication. Predictive proficiency alone is normatively silent, providing no normative guidance as to how to improve the concept. We cannot ignore the needs a concept answers to if we are going to improve it. If, as Carnap himself declares, “language, whether natural or artificial, is an instrument that may be replaced or modified according to our needs” (1963, 937), we need to understand what those needs are in order to replace or modify concepts.5 And notice that our needs – our conceptual needs in this case, i.e. our needs for certain conceptual instruments – are not transparent to us the way our purposes are. In Timothy Williamson’s (2000, 13) terminology, purposes are luminous: having them implies knowing that one has them. But conceptual needs are not luminous. They only reveal themselves to us once we understand what demands our luminous purposes – along with our physiological and psychological needs, values, projects, and commitments – place on our conceptual repertoire as a result of being pursued with certain capacities and limitations in a certain setting. Conceptual needs are the opaque pressures that result from the interaction between agential concerns, capacities, and circumstances. Uncovering those pressures is notably more demanding than getting clear about one’s purposes in using a concept or about a concept’s intended

See also Carnap (1963, 912).

[p. 11]

use, since purposes and intentions are luminous while conceptual needs are not.6 Yet getting clear about one’s conceptual needs seems crucial to determining in what respects a concept should be improved.

This brings us to the second limitation, which is Carnap’s assumption that the direction of improvement will always be towards greater exactness. While he acknowledges that ordinary language serves “a hundred different purposes,” like a “crude, primitive pocketknife” (1963, 938), he invariably frames explication as a movement towards greater precision. “The only essential requirement,” he writes in reply to P. F. Strawson, “is that the explicatum be more precise than the explicandum” (1963, 936).7 Carnap’s guiding ideal is the production of the conceptual equivalents of a “microtome” (1963, 938) – a scientific instrument that can carve tissue and materials into slices so thin that they become translucent and can be stained to reveal their innermost structure under a microscope.

Yet this creates a Procrustean bias in the clarification phase. If one assumes from the outset that the aim is to replace inexact with exact concepts – “pocketknives” with “microtomes” – then the clarification phase fails to ask whether the “crudeness” of the pocketknife might be exactly what makes it useful (try whittling a stick by the campfire with a microtome). The crucial insight that what look like theoretical vices might in fact be practical virtues is one of the principal ways in which the later Wittgenstein breaks with the aspiration to crystalline purity that had enthused the Tractatus, and to some extent also the Vienna Circle. As Wittgenstein put the point in a note from

Erich Reck has argued that “the initial clarification involved in an explication is guided by the explicatum's ‘intended use’, thus pragmatically” (Reck 2012, 99). While I agree with Reck’s emphasis on the “practical” aspect of practical clarification, getting clear about what a concept’s “intended” use should be will, at least sometimes, require getting clear about more than concept-users’ intentions prior to the clarification.

See also Carnap (1962, 5, 7).

[p. 12]

the late 1930s: “If I ask him for a bread knife, and he gives me a razor blade because it is sharper, I shall not thank him” (2000, MS 120, 142v). In the clarification phase, conceptual engineers need to keep an open mind about whether vagueness, superficiality, or inconsistency might not be features rather than flaws. A method of clarification that does not prejudge these questions will be capable of acting as the unifying underpinning of both explication and amelioration. We must therefore elaborate Carnap’s account of “practical clarity” to offer normative guidance and avoid being biased towards exactness from the outset.

3. CLARIFICATION AS NORMATIVE DIAGNOSIS

In order to offer normative guidance to conceptual engineering, clarification must do more than remedy ambiguities and predict what someone takes to fall under a term. It must be diagnostic and normative rather than merely descriptive and predictive; it must uncover the conceptual needs underpinning the use of a concept – the practical needs to which it answers. Only then can we determine whether inexactness is a defect to be corrected or a feature to be preserved. Clarification, in short, must remedy a lack of clarity over what needs, if any, give a concept its point. One lesson we can draw from Carnap’s writings on clarification, however, is that such an inquiry into the needs underpinning a concept, much like analysis or explication, presupposes a working understanding of the concept itself. We must achieve a firm grip on the basis of our investigation, however vague or inexact it may be, before we can inquire further into its relation to human purposes and the conceptual needs they engender. Perhaps the most common way to achieve such a grip is to offer an informal definition. In answer to the question “What do you mean by ‘F’?,” one crams into a definition all the characteristic features of F one can think of. This differs from a strict definition in that the characteristics need

[p. 13]

not be individually necessary. They might include information about the kind of context in which the concept would be applied (e.g. in chemistry or household parlance). Yet this obscures an important difference between two kinds of characteristics associated with a concept: those licensing the application or introduction of the concept and those licensed by the application or introduction of the concept (“introduction” fits certain concepts such as sentential connectives better, though “application” is the more natural and well-worn choice when talking about concepts whose introduction is licensed by the observation of certain worldly conditions). Some features associated with a concept give us reasons to introduce the concept into our thought and talk in a given situation. Other features give us reasons to move on from that concept to other concepts whose introduction it licensed in turn. In this respect, concepts can be thought of as inferentially articulated in terms of a bowtie structure, with several conditions (which may or may not be necessary) licensing the application of the concept as the central node, which in turn licences the drawing of several inferential consequences. This two-sided articulation of concepts was first emphasised by Michael Dummett (1973, 434) when he generalised Gerhard Gentzen’s work on sentential connectives, and many conceptual role theorists and inferentialists have stressed it since (Peacocke 1992; Brandom 1994, 2000; Boghossian 2003; Wedgwood 2007; Kukla and Lance 2009; Peregrin 2014). More recently, Jorem and Löhr (2022) have stressed the importance for conceptual engineering of focusing not only on truth and application conditions, but also on what follows from a concept’s applicability. This is consonant with Carnap’s own trail-blazing inferentialist work in the early 1930s (Peregrin 2020). What this bowtie structure brings out is that concepts are not merely labels we stick on objects; they form inferential bridges, licensing transitions from observation to judgment and from judgment to action. Thus, Ludwig Wittgenstein thinks of a concept as “the technique of our use of

[p. 14]

an expression: as it were, the railway network that we have built for it” (1988, 50). Clarifying a concept should notably involve mapping out this network. Yet definitions of the form “x is F iff it has properties P1 … Pn”, just like informal enumerations of characteristic features of F, tend to obscure the bowtie structure of concepts. They simply pull together the grounds for applying a concept and the implications of applying it into a single, static description. This conceptual “pincer movement” fails to differentiate between the conditions under which the concept is correctly applied and the consequences that follow from its applicability. While such comprehensive definitions have their uses, they obscure the dynamic role that concepts play in reasoning. Our minds use concepts to move from certain conditions to certain conditions. Indeed, we discover new concepts notably by realizing that we are justified in moving from certain conditions to certain consequences. And crucially, we discover that a concept is in need of re-engineering precisely when this inferential bridge begins to buckle: e.g. when the conditions for applicability are met, but the consequences of applicability no longer hold as reliably as the concept presupposes. To better capture a concept’s dynamic role and connect it to practical needs, we can operationalize the framework of “Explaining Meaning in terms of Use” (EMU) developed by Michael Williams (2013), which I shall liberally adapt here. On the conception I propose, clarification should not cram as much as possible into a single comprehensive definition, but should disentangle the inferential and instrumental connections a concept stands in by characterising the concept in terms of three components with distinct roles: entry clauses, exit clauses, and need clauses. First, entry clauses specify the conditions under which we are entitled to apply a concept. These clauses identify the features of the world, the agent, or the context that serve as criteria or reliable indicators for the concept’s applicability. For observational concepts, these might be perceptual

[p. 15]

cues; for theoretical or institutional concepts, they might be specific experimental results or procedural facts. Entry clauses map the upstream input triggering the applicability of the concept. Second, exit clauses specify the inferential consequences of applying a concept. They answer the question: what follows from the fact that this concept is applicable here? These clauses map the concept’s downstream output – both theoretical (what else must be true) and practical (what must be done or felt). By separating entry from exit conditions, we gain critical leverage over a concept. We can ask whether the criteria we use to apply the concept (as specified in the entry clauses) actually warrant the consequences we draw from it (as specified in the exit clauses). Conceptual defects often manifest not simply as vagueness or a lack of exactness, but as a misalignment in these two sets of clauses. The history of the medical concept of preeclampsia – a condition sometimes arising during pregnancy – offers a vivid illustration. Historically, the entry clauses for this concept consisted of a rigid triad of necessary and jointly sufficient conditions (Tanner et al. 2022): a patient had to exhibit (1) hypertension, (2) edema (swelling), and (3) proteinuria (excess protein in the urine). The exit clauses, meanwhile, consisted of dire predictions and drastic prescriptions: the applicability of the concept predicted a severe risk of seizures (eclampsia), stroke, and organ failure, and the prescription was the induction of labour or a C-section to deliver the placenta. As medical understanding advanced, however, a lethal misalignment emerged. It became clear that the risk of seizures and organ dysfunction was also frequently present in women who did not meet the full triad of criteria. A patient might present with high blood pressure and impending liver dysfunction but lack protein in her urine; another might lack significant swelling, but still be at risk of stroke. Under the strict regime of the classic triad of entry clauses, these women were effectively rendered conceptually invisible. Because they did not register as satisfying the concept, life-saving interventions were not triggered, leading to preventable deaths.

[p. 16]

Though the old concept of preeclampsia was defective, this was not due to its being vague or inexact, however. If anything, it was too exact, treating the classical triad of hypertension, edema, and proteinuria as necessary and sufficient conditions when in fact, the seizures and organ failures associated with preeclampsia often occurred in women who did not meet all three criteria. The concept was defective because of a misalignment between the inferential connections encoded in its entry and exit clauses and the connections that in fact obtained between the various symptoms of the disease and its causal consequences. Doctors thought: “If no proteinuria, then it is inappropriate to expect eclampsia.” The reality, however, was that the physiological state of the patient was sometimes physically capable of producing eclampsia even without proteinuria. The conceptual map, in articulating which inferential transitions were required, permitted, or illicit, represented as closed off a transition that the causal terrain actually left open. To fix this, the aperture of the concept had to be widened to match the variety of ways in which the disease could manifest itself: today, preeclampsia is diagnosed by high blood pressure plus any one of several signs of organ damage or neurological symptoms. Proteinuria is no longer strictly required, and edema was removed entirely because it is too common in normal pregnancy. This allows the concept of preeclampsia to better track the population actually at risk from eclampsia. Recognising the bowtie structure of entry and exit conditions reveals this change to be an act of conceptual repair in pursuit of a certain kind of isomorphism: the isomorphism between the inferential connections encoded in our concepts and the causal connections embodied in the world. The guiding ideal here, to use Spinoza’s phrase, is that “the order and connection of ideas” should be “the same as the order and connection of things” (2002, Ethics II, prop. vii) – a conception elaborated by Brandom (2019, 61) and Hlobil and Brandom (2025, 11) in terms of an alignment between deontic and alethic modality, i.e. between what is rationally necessary, possible, or impossible and what is causally necessary, possible, or impossible.

[p. 17]

However, to clarify why the concept is structured the way it is, and whether that structure is worth preserving, a third component of clarification is required: the need clauses. Need clauses specify what practical needs are met by a concept governed by certain entry and exit clauses. While the entry and exit clauses describe the use of the concept in terms of the rules or inferential proprieties governing its application, the need clauses describe the point of using a concept with those rules of application. The need clauses state what the concept does for us by functioning as it does in our reasoning. In the case of preeclampsia, the need clause is relatively straightforward: by tracking the causal precursors of a specific physiological malfunction, the concept serves the practical need to identify pregnancies at imminent risk of seizures and organ failure in order to trigger life-saving delivery. This adds a diagnostic and normative dimension to clarification by revealing that the concept is defective if its entry conditions are too narrow to capture the patients who effectively require the drastic interventions prescribed by its exit conditions. The entry clauses must be sufficiently sensitive and accommodating of variation to capture all patients facing the relevant risks and requiring the relevant interventions. This tripartite account of clarification in terms of entry, exit, and need clauses thus transforms clarification from a merely descriptive and predictive exercise into a normative diagnosis. It moves us from the observation and prediction of how a term is applied (“X applies “Q” in context C”) to a functionalist hypothesis explaining the widespread cultivation of the very disposition to think in those terms (“Group G cultivates the disposition to think in terms of “Q” because it meets need This allows us to clarify a concept not just semantically, by specifying the entry and exit clauses governing its use, but – in a deeper sense than the one Carnap articulates – genuinely practically. Clarification elucidates the role a concept plays – its role in our reasoning, in the first instance, but also its role in our lives, by revealing what needs it meets by playing this role in our reasoning.

[p. 18]

4. NEED MATRICES: MAPPING THE LANDSCAPE OF CONCEPTUAL NEEDS

If the clarification phase is to culminate in the articulation of need clauses – statements of the practical exigencies a concept answers to – more has to be said about how one arrives at a need clause, and what such clauses are themselves grounded in.8

The basic idea is that practical clarification in the expanded sense encompassing need clauses must identify a specific combination of practical pressures rendering a particular concept needful. A concept does not inherently meet a need. Nor can conceptual needs arise in a vacuum. Rather, conceptual needs are generated by the interaction of three factors:

1. Concerns: What do the concept-users care about? What values, purposes, or projects are they trying to realize? 2. Capacities: What are concept-users already capable of, physically, perceptually, cognitively, and technologically? And what are the corresponding limitations of their powers? 3. Circumstances: What is the world like? What are the causal laws, environmental constraints, and social institutions within which concept-users operate? Tying a concept back to the concerns, capacities, and circumstances of those who use it can radically change the way we look at it. To take a classic example: once we tie the concept of knowledge back to a need matrix, as Edward Craig has done (1990), we shall no longer regard it simply as a way to label “justified true belief,” but as being, at root, a tool for finding reliable

While the approach to the formulation of need clauses offered in this section follows Queloz (2025), it is also possible to recast philosophical methodologies aiming to reverse-engineer the functions of concepts as attempts to formulate need clauses; see Queloz (2021, 2022b).

[p. 19]

informants in situations where a great deal depends on the information they have to offer. This already intimates that the concept needs to track typical indicators of reliability, such as being able to offer justifications or standing in the right causal relation to the facts of interest. Yet our need for a concept is also a need for an instrument that fits our capacities, just as a tool must fit our hand. This introduces conceptual ergonomics into the equation: the requirement that a concept be suitable for use by agents like us (Nimtz 2024; Queloz 2025). In the example of the concept of knowledge, this means that the presence of the tracked properties must be recognizable to users of the concept, imposing further constraints on what the concept must track (the property of having true beliefs, for instance, would make a fine indicator of being a reliable informant, were it not for the fact that the truth of a belief is precisely what we cannot immediately assess when seeking information we lack).9 Finally, circumstances matter as well: a concept tracking indicators that are generally reliable in a high-trust, homogeneous society may become dangerously misleading in a polarized, low-trust society, because the correlations that used to make for reliable indicators no longer hold – a problem that may well be amplified when the potential informants are no longer human agents, but artificial ones.

To clarify what needs are being met by a concept, we can construct need matrices: minimal sets of concerns, capacities, and circumstances that are sufficient to engender a conceptual need for something like this concept. If a “need clause” is a statement of what the concept does for us, a “need matrix” explains why it must do that given the world we live in. The term “matrix” (from the Latin for “womb”) helpfully connotes not just the natural environment in which an organism develops, but also the social, cultural, and institutional environment in which an idea develops (“Oxbridge was the matrix of this ideology”). Specifying exactly what conjunction of conditions in

I elaborate on this point in Queloz (2021, 135–36).

[p. 20]

that environment generates which conceptual need clarifies a concept’s relation to its formative environment and ultimately to the human concerns that fundamentally animate its use. Achieving practical clarity about a concept by tying it back to a need matrix gives us not just a retrospective diagnosis of what it might have done for us in the past, but a normative, actionguiding diagnosis, because it tells us what concept would be most apt within this need matrix. As I have argued elsewhere (Queloz 2026b), conceptual needs are best understood as possessing a distinctively aptic kind of normativity. The term goes back to Selim Berker (2022, 2024), who identifies the aptic as a distinct domain of normativity pertaining to fittingness, distinguishable from both deontic normativity (the normativity of requirement, permission, and prohibition) and evaluative normativity (the normativity of goodness and badness). To say that we have a conceptual need is not to say that we are required to possess a concept, nor merely that it would be valuable to have it; rather, it is to say that the concept is called for by the situation – that it fits the conceptuser’s predicament much as a key fits a lock. However, the diagnostic picture is rarely as simple as a single key fitting a single lock. Concepts have complex histories; they accumulate functional roles over time, often answering simultaneously to multiple, disparate, and potentially conflicting needs. A single concept may be the focal point of several needs that are all being met at once (less a crude pocketknife than a Swiss army knife). Consequently, the task of clarification is not necessarily concluded after identifying one need. Another source of complexity is that a concept might prove apt relative to a need matrix, but the matrix itself might turn out to be built around an objectionable concern. Far from automatically endorsing the status quo, clarifying what needs underpin a concept thus allows us to distinguish between concepts that are defective in that they fail to serve legitimate concerns as well as they could and concepts that are all too functional, serving concerns we have no wish to see satisfied.

[p. 21]

And lastly, when we find that a concept is a focal point of multiple needs rooted in concerns we endorse, clarification can also become a diagnosis of potentially irreducible tensions. It may reveal shortcomings of the explicandum, including its vagueness or inconsistency, to be irremediable. For if these features are a reflection of the conflicting demands we place upon the concept, we may not be able to improve the concept in one respect without losing something in another respect. Any inadequacies rooted in conflicting yet ineliminable needs will themselves prove hard to eliminate, or be eliminable only at the cost of leaving one of the needs unmet. In some cases, we may be able to meet both needs more satisfactorily through “conceptual fission” (Maund 1981), splitting the concept into two more specialised concepts (as Carnap (1962) himself suggests already happened with the concept of probability). But if the needs really do need to be met simultaneously and by the same concept, the normative upshot may be that we just have to live with the inadequacies.

5. THE DIVISION OF CONCEPTUAL LABOUR

This account of practical clarification as a normative diagnosis of the conceptual needs underpinning a concept provides the resolution to the tension with which we began: the apparent conflict between explication’s drive towards exactness and other theoretical virtues and amelioration’s drive towards justice and other practical virtues. Once we view clarification as a mapping of the landscape of conceptual needs, we can see that these are not rival methodologies so much as responses to different kinds of need matrices. In scientific contexts, one kind of need matrix tends to dominate. When the chief concern is theory-building, prediction, and control, the capacities are extended by scientific instruments, and the circumstances are controlled experimental settings, this generates a strong need for theoretical virtues like exactness and consistency. In such a setting, the theoretical virtues are practical virtues,

[p. 22]

and Carnapian explication will often be the apt response. We should replace fish with piscis, because the need matrix demands it. In many sociopolitical contexts, however, the need matrices will often look radically different, and the Carnapian ambition to tidy things up may end up frustrating our conceptual needs, because the relationship between epistemic accuracy and conceptual precision will be more complicated. Let me give three examples. First, as I have argued in detail elsewhere (Queloz 2025, 205-6), the very concern for accuracy can sometimes demand vaguer concepts. Consider the case of the U.S. Social Security Administration’s disability program (Mashaw 1983, 52–53; Zacka 2017, 53–55). At one point, legislators realised that a precise, sharp-edged definition of “disability” ended up systematically excluding cases that, on the ground, could clearly be seen to merit support. Legislators then intentionally introduced a vaguer conception of disability in recognition of the fact that street-level administrators were epistemically better placed to appreciate and evaluate the complexities of individual cases. The very concern to more accurately pick out the set of people deserving support thus created a need for a vaguer concept, because the resulting play or leeway in how the concept was to be applied gave lower-ranking administrators just the discretion they needed to leverage their superior epistemic position. Here, the decentralised nature of the state apparatus generates a conceptual need for vagueness. Secondly, inconsistency in our conceptual scheme can be a way of more accurately reflecting real and ineliminable conflicts between our values. Consider the concept of liberty. A Carnapian explication might seek to define liberty in terms of a bundle of rights consistent with the rights bestowed by other political concepts such as equality, thereby creating a perfectly consistent conceptual system where liberty and equality never clash. This would be theoretically tidy, but it would fail to meet our conceptual needs in practice. As Bernard Williams argued against Ronald Dworkin, we need a concept of liberty that registers losses in liberty even when those losses are

[p. 23]

justified by the demands of equality.10 The tension between the concepts of liberty and equality is not a flaw to be corrected, but a faithful reflection of a real conflict between two fundamental human concerns. A concept of liberty that was perfectly consistent with the concept of equality would be less truthful, because it would conceal real and irreducible trade-offs inherent in political life.

And finally, even conceptual superficiality – understood, following Michael Strevens (2008), as a disregard for the causal processes underpinning something – can be a way to calibrate a concept to the level of description at which it can best serve our concerns. Consider the concept of voluntary action. The history of philosophy is replete with attempts to “deepen” the concept in terms of metaphysical accounts of the will as something unconditioned by contingency. Yet, as Williams also suggested, conceptual depth can be a liability.11 The concept of the voluntary plausibly answers to a concern to allocate responsibility in a way that is fair and respects individual freedom, but it must arguably do so in a world where every action is conditioned by contingency. A deepened concept that required total freedom from contingency would find no application, rendering us unable to hold anyone responsible. What we need, therefore, is a superficial concept of the voluntary: one that is sensitive not to the deeper causal details of neurophysiology and biography, but to surfacelevel features of behaviour, like the absence of coercion or the presence of deliberative impairments. To deepen the concept would be to break its aptic fit with our limited capacities and our need for the allocation of responsibility on a fair basis.

See especially Williams (2001) and Dworkin (2001), and see Queloz (2024) and Cueni (2024) for detailed interpretations of this Dworkin–Williams debate, which extended over decades.

See the highly compressed argument in Williams (1993, 68–69). I expand on this suggestion in Queloz (2022a, 2026a).

[p. 24]

The normativity of clarification, on this account, derives from its ability to reveal the practical needs that give a concept its point – needs that vary across different contexts of inquiry. In this respect, the account I propose echoes the reading of Carnap offered by Georg Brun (2016). Brun argues that “the requirements an adequate explication has to meet cannot be specified in general but only with respect to a specific task of explication,” and consequently, that “choosing an adequate explicatum is a practical decision which has to be taken in view of the specific problems the explicatum is expected to solve” (2016, 1225). On the interpretation I propose, the “specific problems” can themselves be further specified by constructing need matrices. This amounts to the piecemeal modelling of the landscape of our conceptual needs. If the clarification phase reveals a matrix dominated by concerns for prediction, generalization, and systematicity (as in physics or formal logic), those needs will indeed be for theoretical virtues such as exactness, consistency, fruitfulness, and simplicity. In such cases, Carnap is right: inexactness is a defect, and explication is the fix. However, our concepts help us to solve an enormous variety of problems in an enormous variety of settings. Practical clarification may equally reveal conceptual needs for what tidy-minded theoreticians would consider vices in a concept: vagueness, inconsistency, or superficiality. Then the engineer’s task is not to sharpen the concept, but to adapt it, if need be, to the complex challenges of social reality.

6. CONCLUSION

Carnap was right to warn that explication without clarification risks “becoming entirely futile” (1962, 4). But the futility does not fundamentally arise, as he seemed to imply, from a lack of exactness in the explicandum. It ultimately arises from a lack of aptness in the explicatum. To

[p. 25]

sharpen a concept that needs to be vague, or to deepen a concept that needs to remain superficial, is to misunderstand what tool one is dealing with. Reconceiving clarification as normative diagnosis – as unfurling a concept’s inferential structure and tying it back to the needs it meets – promises to provide conceptual engineering with the orientation it can seem to lack. It bridges the divide between explication and amelioration by giving them a shared basis in an essential preliminary step: the clarifying inquiry into what we need our concepts to do for us. Such practical clarification is crucial for both explication and amelioration. It puts conceptual engineers in a position to distinguish between concepts that really do suffer from a lack of theoretical virtues and concepts whose apparent defects are actually adaptations to a complex world: they are as exact as they can be while remaining as vague as they need to be. Ultimately, to clarify a concept is to determine how the distinctive reasoning pattern it articulates fits into the wider tapestry of human concerns. Thus opening up the clarification phase to a wider variety of practical considerations sits well with the increasingly open-minded pragmatism of Carnap’s later thought: “The choice of a method for the solution of a given philosophical problem,” he wrote, “should be decided in each case by practical considerations” (1963, 938). The practical clarification of concepts, I have argued, is the method by which we identify those practical considerations. And the method itself is best conceived as remaining open about whether these considerations will pull us towards explication or amelioration. The mature Carnapian project, as A. W. Carus observes in his study of Carnap’s intellectual development, was not a conquest of ordinary language by scientific terminology, but a refinement of our conceptual repertoire on a case-by-case basis, grounded in the full range of practical concerns and values that animate human affairs:

Unlike rational reconstruction, explication no longer envisaged one-way replacement of the ordinary, intuitive world view by a scientific one, but a dialectical interchange between the

[p. 26]

two kinds of system. Our practices and our values reside within an intuitive Lebenswelt that can be progressively improved, whose quality can be raised piecemeal through explicative replacement of its concepts by constructed ones, but we decide what replacements to undertake from the overall standpoint of the Lebenswelt, our practical concerns and our values. (Carus 2007, xi) Clarifying a concept’s relation to our conceptual needs is a key step in this “dialectical interchange.” It allows us to decide, from the broader standpoint of the Lebenswelt, whether a concept requires the precision of the laboratory or the strategic vagueness of social life. And there is a spectrum here, as Carnap himself recognised: “between everyday concepts and scientific concepts […] I see [… ] no sharp boundary line but a continuous transition” (1963, 634). Similarly, there need be no sharp boundary between explication and amelioration. Together, they span a continuous spectrum of reengineering strategies, mirroring the continuity of the conceptual needs to which they answer.

8045 words incl. notes

[p. 27]

Bibliography Beckmann, Pierre, and Matthieu Queloz. 2025. ‘Mechanistic Indicators of Understanding in Large Language Models’. arXiv arXiv:2507.08017: 1–32. Berker, Selim. 2022. ‘The Deontic, the Evaluative, and the Fitting’. In Fittingness: Essays in the Philosophy of Normativity. Edited by Christopher Howard and R. A. Rowland, 23–57. Oxford: Oxford University Press. Berker, Selim. 2024. ‘Is There Anti-Fittingness?’. Ergo 11 (39): 1051–1082. Boghossian, Paul. 2003. ‘Blind Reasoning’. Aristotelian Society Supplementary Volume 77 (1): 225– 248. Brandom, Robert. 1994. Making It Explicit. Reasoning, Representing, and Discursive Commitment. Cambridge, MA: Harvard University Press. Brandom, Robert. 2000. Articulating Reasons. Cambridge, MA: Harvard University Press. Brandom, Robert. 2019. A Spirit of Trust: A Reading of Hegel’s Phenomenology. Cambridge, MA: Harvard University Press. Brigandt, Ingo. 2010. ‘The Epistemic Goal of a Concept: Accounting for the Rationality of Semantic Change and Variation’. Synthese 177 (1): 19–40. Brigandt, Ingo. 2012. ‘The Dynamics of Scientific Concepts: The Relevance of Epistemic Aims and Values’. In Scientific Concepts and Investigative Practice. Edited by Uljana Feest and Friedrich Steinle, 75–103. Berlin: De Gruyter. Brigandt, Ingo, and Esther Rosario. 2020. ‘Strategic Conceptual Engineering for Epistemic and Social Aims’. In Conceptual Engineering and Conceptual Ethics. Edited by Alexis Burgess, Herman Cappelen and David Plunkett, 100–124. Oxford: Oxford University Press. Brun, Georg. 2016. ‘Explication as a Method of Conceptual Re-Engineering’. Erkenntnis 81 (6): Cappelen, Herman. 2018. Fixing Language: An Essay on Conceptual Engineering. Oxford: Oxford University Press. Cappelen, Herman. 2020. ‘Conceptual Engineering: The Master Argument’. In Conceptual Engineering and Conceptual Ethics. Edited by Alexis Burgess, Herman Cappelen and David Plunkett, 132–151. Oxford: Oxford University Press. Cappelen, Herman. 2023. The Concept of Democracy: An Essay on Conceptual Amelioration and Abandonment. Oxford: Oxford University Press. Carnap, Rudolf. 1955. ‘Meaning and Synonymy in Natural Languages’. Philosophical Studies 6 (3): 33–47. Carnap, Rudolf. 1962. Logical Foundations of Probability. 2nd ed. Chicago: The University of Chicago Press.

[p. 28]

Carnap, Rudolf. 1963. ‘Replies and Systematic Expositions’. In The Philosophy of Rudolf Carnap. Edited by P. A. Schilpp, 859–1013. La Salle: Open Court. Carus, André. 2007. Carnap and Twentieth-Century Thought: Explication as Enlightenment. Cambridge: Cambridge University Press. Chalmers, David. 2025. ‘What Is Conceptual Engineering and What Should It Be?’. Inquiry 68 (9): Craig, Edward. 1990. Knowledge and the State of Nature: An Essay in Conceptual Synthesis. Oxford: Clarendon Press. Cueni, Damian. 2024. ‘Constructing Liberty and Equality – Political, Not Juridical’. Jurisprudence Dummett, Michael. 1973. Frege: Philosophy of Language. New York: Harper and Row. Dutilh Novaes, Catarina. 2020. ‘Carnapian Explication and Ameliorative Analysis: A Systematic Comparison’. Synthese 197 (3): 1001–34. Dutilh Novaes, Catarina, and Erich Reck. 2017. ‘Carnapian Explication, Formalisms as Cognitive Tools, and the Paradox of Adequate Formalization’. Synthese 194 (1): 195–215. Dworkin, Ronald, Bernard Williams, Mark Lilla, Thomas Nagel, Richard Wollheim, Frances Kamm, and Steven Lukes. 2001. ‘Pluralism’. In The Legacy of Isaiah Berlin. Edited by Mark Lilla, Ronald Dworkin and Robert Silvers, 121–139. New York: New York Review of Books. Ebbs, Gary. 2026. ‘Carnap, Quine, Putnam and Burge on Concepts’. In A Philosophical History of the Concept. Edited by Stephan Schmid and Hamid Taieb, 418–438. Cambridge: Cambridge University Press. Edmonds, David. 2020. The Murder of Professor Schlick: The Rise and Fall of the Vienna Circle. Princeton: Princeton University Press. Glock, Hans-Johann. 2025. ‘Conceptual Engineering and Conceptual Analysis: Historical and Conceptual Connections’. In New Perspectives on Conceptual Engineering - Volume 1: Foundational Issues. Edited by Manuel Gustavo Isaac, Steffen Koch and Kevin Scharp, 1–23. Cham: Springer Nature Switzerland. Haslanger, Sally. 2000. ‘Gender and Race: (What) Are They? (What) Do We Want Them to Be?’. Haslanger, Sally. 2012. Resisting Reality: Social Construction and Social Critique. Oxford: Oxford University Press. Haslanger, Sally. 2020a. ‘Going On, Not in the Same Way’. In Conceptual Engineering and Conceptual Ethics. Edited by Alexis Burgess, Herman Cappelen and David Plunkett, 230–260. Oxford: Oxford University Press. Haslanger, Sally. 2020b. ‘How Not to Change the Subject’. In Shifting Concepts: The Philosophy and Psychology of Conceptual Variation. Edited by Teresa Marques and Åsa Wikforss, 235–259. Oxford: Oxford University Press.

[p. 29]

Herrmann, Daniel A., and Benjamin A. Levinstein. 2025. ‘Standards for Belief Representations in LLMs’. Minds and Machines 35 (1): 1–25. Hlobil, Ulf, and Robert Brandom. 2025. Reasons for Logic, Logic for Reasons: Pragmatics, Semantics, and Conceptual Roles. New York: Routledge. Jorem, Sigurd, and Guido Löhr. 2022. ‘Inferentialist Conceptual Engineering’. Inquiry: 1–22. Kukla, Rebecca, and Mark Lance. 2009. ‘Yo!’ and ‘Lo!’: The Pragmatic Topography of the Space of Reasons. Cambridge, MA: Harvard University Press. Levinstein, Benjamin A., and Daniel A. Herrmann. 2024. ‘Still No Lie Detector for Language Models: Probing Empirical and Conceptual Roadblocks’. Philosophical Studies: 1–27. Mashaw, Jerry L. 1983. Bureaucratic Justice: Managing Social Security Disability Claims. New Haven: Yale University Press. Maund, J. B. 1981. ‘Colour: A Case for Conceptual Fission’. Australasian Journal of Philosophy 59 Nimtz, Christian. 2024. ‘The Power of Social Norms: Why Conceptual Engineers Should Care about Implementation’. Synthese 203 (6): 215. Peacocke, Christopher. 1992. A Study of Concepts. Cambridge, MA: MIT Press. Peregrin, Jaroslav. 2014. Inferentialism: Why Rules Matter. New York: Palgrave. Peregrin, Jaroslav. 2020. ‘Rudolf Carnap’s Inferentialism’. In The Vienna Circle in Czechoslovakia. Edited by Radek Schuster, 97–109. Cham: Springer. Queloz, Matthieu. 2021. The Practical Origins of Ideas: Genealogy as Conceptual Reverse- Engineering. Oxford: Oxford University Press. Queloz, Matthieu. 2022a. ‘The Essential Superficiality of the Voluntary and the Moralization of Psychology’. Philosophical Studies 179 (5): 1591–1620. Queloz, Matthieu. 2022b. ‘Function-Based Conceptual Engineering and the Authority Problem’. Queloz, Matthieu. 2024. ‘The Dworkin–Williams Debate: Liberty, Conceptual Integrity, and Tragic Conflict in Politics’. Philosophy and Phenomenological Research 109 (1): 3–29. Queloz, Matthieu. 2025. The Ethics of Conceptualization: Tailoring Thought and Language to Need. Oxford: Oxford University Press. Queloz, Matthieu. 2026a. ‘Law as a Test of Conceptual Strength’. In Bernard Williams on Law and Jurisprudence: From Agency and Responsibility to Methodology. Edited by Veronica Rodriguez- Blanco, Daniel Peixoto Murata and Julieta Rabanos. Oxford: Hart. Queloz, Matthieu. 2026b. ‘Needs of the Mind: How Aptic Normativity Can Guide Conceptual Adaptation’. manuscript. Reck, Erich. 2012. ‘Carnapian Explication: A Case Study and Critique’. In Carnap’s Ideal of Explication and Naturalism. Edited by P. Wagner, 96–116. New York: Palgrave.

[p. 30]

Reck, Erich. 2024. ‘Carnapian Explication: Origins and Shifting Goals’. In Interpreting Carnap: Critical Essays. Edited by A. Richardson and A. T. Tuboly. Cambridge: Cambridge University Press. Romizi, Donata. 2012. ‘The Vienna Circle’s “Scientific World-Conception”: Philosophy of Science in the Political Arena’. HOPOS: The Journal of the International Society for the History of Philosophy of Science 2 (2): 205-242. Scharp, Kevin. 2013. Replacing Truth. Oxford: Oxford University Press. Scharp, Kevin. 2020. ‘Philosophy as the Study of Defective Concepts’. In Conceptual Engineering and Conceptual Ethics. Edited by Alexis Burgess, Herman Cappelen and David Plunkett, 396– 416. Oxford: Oxford University Press. Sigmund, Karl. 2017. Exact Thinking in Demented Times: The Vienna Circle and the Epic Quest for the Foundations of Science. New York: Basic Books. Spinoza, Baruch de. 2002. Complete Works. Edited by Michael L. Morgan. Indianapolis: Hackett. Strevens, Michael. 2008. Depth: An Account of Scientific Explanation. Cambridge, MA: Harvard University Press. Tanner, Michael S., Mary-Ann Davey, Ben W. Mol, and Daniel L. Rolnik. 2022. ‘The Evolution of the Diagnostic Criteria of Preeclampsia-Eclampsia’. American Journal of Obstetrics and Gynecology 226 (2): S835–S843. Templeton, Adly, Tom Conerly, Jonathan Marcus, Jack Lindsey, Trenton Bricken, Brian Chen, Adam Pearce, Craig Citro, Emmanuel Ameisen, Andy Jones, Hoagy Cunningham, Nicholas L. Turner, Callum McDougall, Monte MacDiarmid, C. Daniel Freeman, Theodore R. Sumers, Edward Rees, Joshua Batson, Adam Jermyn, Shan Carter, Chris Olah, and Tom Henighan. 2024. ‘Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet’. Transformer Circuits Thread. Uebel, Thomas. 2004. ‘Carnap, the Left Vienna Circle, and Neopositivist Antimetaphysics’. In Carnap Brought Home: The View from Jena. Edited by Steve Awodey and Carsten Klein, 247– 78. Chicago: Open Court Publishing. Wedgwood, Ralph. 2007. The Nature of Normativity. New York: Oxford University Press. Williams, Bernard. 1982. ‘The Spell of Linguistic Philosophy: Dialogue with Bernard Williams’. In Men of Ideas: Some Creators of Contemporary Philosophy. Edited by Bryan Magee, 110–124. Oxford: Oxford University Press. Williams, Bernard. 1993. Shame and Necessity. Berkeley: University of California Press. Williams, Bernard. 2001. ‘Liberalism and Loss’. In The Legacy of Isaiah Berlin. Edited by Mark Lilla, Ronald Dworkin and Robert Silvers, 91–103. New York: New York Review of Books. Williams, Michael. 2013. ‘How Pragmatists can be Local Expressivists’. In Expressivism, Pragmatism and Representationalism. Edited by Huw Price, 128–144. Cambridge: Cambridge University Press.

[p. 31]

Williamson, Timothy. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Wittgenstein, Ludwig. 1988. Wittgenstein’s Lectures on Philosophical Psychology 1946–47. Edited by P. T. Geach. Hemel Hempstead: Harvester Wheatsheaf. Wittgenstein, Ludwig. 2000. Wittgenstein’s Nachlass. The Bergen Electronic Edition. Edited by The Wittgenstein Archives at the University of Bergen. Oxford: Oxford University Press. Yap, Audrey. 2022. ‘Conceptual Engineering and Neurath’s Boat: A Return to the Political Roots of Logical Empiricism’. In The Political Turn in Analytic Philosophy: Reflections on Social Injustice and Oppression. Edited by David Bordonaba Plou, Víctor Fernández Castro and José Ramón Torices, 31–52. Berlin: De Gruyter. Zacka, Bernardo. 2017. When the State Meets the Street: Public Service and Moral Agency. Cambridge, MA: Harvard University Press.