Quasirealism as semantic dispensability

Abstract

I argue that standard explanationist solutions to the problem of creeping minimalism are largely on the right track, but they fail to correctly specify the kind of explanation that is relevant to distinguishing realism from quasirealism. Quasirealism should not be distinguished from realism in terms of the explanations it gives of why a normative judgment—a normative sentence or attitude—has the semantic content that it has. Rather, it should be distinguished in terms of the explanations it offers of what the semantic content of a normative judgment is.

1 Introduction

The predicate ‘true’ is the source of the problem, that and the verb ‘believe’. Initially, expressivism seemed like an irrealist doctrine. Despite having the surface grammar of an ordinary descriptive sentence, ‘murder is wrong’ did not describe. It expressed a noncognitive attitude of, say, disapproval towards murder. Neither normative claims nor attitudes were supposed to be truth-apt (e.g., Ayer 1936/1952).

Unfortunately, expressivism is a claim about our normative language, the one we grow up speaking. But it is perfectly good English to say, ‘It’s true that murder is wrong’. Similarly, we talk about people’s moral beliefs. As a theory of our normative language, irrealist expressivism makes poor predictions.

So expressivists add sophistication to their story. There is a minimalist theory of truth, and by helping themselves to it, they can better accommodate the linguistic data. We can extend this minimalism to beliefs, as well. But as expressivism becomes more sophisticated, it risks losing its distinctiveness. This same minimalism looks like it can be extended to every component of the characteristically realist ideology. But then expressivism seems to collapse into realism. Dreier has named this the problem of creeping minimalism (2004).

Dreier also offers a solution: the explanation explanation (ibid). The expressivist may agree that there are normative properties, in an appropriately minimal sense of property. But, ‘to explain what it is to make a moral judgment, we need not mention any normative properties’ (39). The realist’s explanation, on the other hand, must invoke such properties.

I will argue that Dreier’s explanation explanation gets something right. Normative properties are different for the quasirealist in that they are in an important sense dispensable. But Dreier assumes the relevant kind of dispensability is metasemantic—that normative properties are dispensable to explanations of “what it is to make a … judgment” with a certain semantic content (or at least the explanation has been understood in the subsequent literature as a metasemantic one). This general strategy of carving out the difference between quasirealism and realism in terms of metasemantic explanations has, moreover, become the standard one (e.g., Chrisman 2008; Dunaway 2016; Simpson 2018).¹

This overlooks one of the most striking features of traditional expressivism. It is, or traditionally has been at least, a novel semantic thesis, explaining what normative utterances mean in terms of the attitudes they express.² On this basis, I will argue that the quasirealist is not offering a special type of metasemantic explanation. Rather, the quasirealist is committed to a distinctive kind of semantic explanation.

Admittedly, the exact borders between semantics and metasemantics are themselves disputed. As I will use these phrases, a metasemantic explanation explains why something means what it does, while a semantic explanation simply tells us what it means. It is dispensability to this latter kind of explanation, I will argue, that is most relevant to the expressivist metaethical project. The proposal will also allow us to develop both robustly realist and quasirealist versions of inferentialist and conceptual role semantics, the value of which will emerge in the discussion. Finally, I will discuss the extent to which this solution is available to the recent program of reinterpreting expressivism (or some version of quasirealism) as a metasemantic thesis (e.g., Ridge 2014; Chrisman 2016; Köhler 2017, 2018). Any apparent incompatibility is probably mostly verbal.

I will start with a brief explanation of why minimalism creeps.

2 Why minimalism creeps

Expressivism risks being an overly revisionary theory of normative language.³ So we adopt deflationism about truth. According to the deflationist, the predicate ‘true’ does not denote a substantive property. To attribute truth to a sentence or proposition is not to attribute correspondence, coherence, or possession of a primitive property of being true. Rather, what truth is, is exhausted by the instances of the T-schema:

p is true if and only if p.⁴

Notice that even if ‘true’ does not denote a substantive property, the predicate is still very useful. Given the T-schema, attribution of truth can serve as a device for disquotation, reassertion, and quantification. I can say, for example, “Everything Kant said is true,” and thereby reassert everything Kant said. I can do this, even if I am not totally sure of everything Kant said.

With truth sufficiently domesticated, the expressivist is ready to tame belief. What is it to believe that p? The minimalist answer is: to regard p as true (e.g., Chrisman 2008: 344).⁵ But calling p true is simply another way of saying that p. So, to have a belief, then, is simply to be in a state that could be expressed in a declarative sentence (Lenman 2003: 38; Dreier 2004: 27–28).

The expressivist is thereby able to account for realistic aspects of our language. But quasirealism comes with a challenge. We’ve agreed that some normative claims are true and others false, and that these claims express beliefs. How is that not just realism? In what sense is the realism merely quasi?

Perhaps it is in the denial of more arcane theoretical commitments that the quasirealist will stake out her position. There might be normative beliefs and truth-apt normative sentences, but we can at least deny that there are normative properties and propositions. The problem is, our minimalist account of truth seems to lead to minimalist accounts of philosophical arcana as well. There is a plausible deflationary schema for ‘property’:

x has property F if and only if x is F.

The term ‘property’ would seem to make a useful contribution to communication very similar to that proposed for ‘true’. We can say things such as, ‘These two objects share many important properties’, without predicating each of the similarities individually. The generalization would otherwise be impossible, since English grammar does not allow for second-order quantification.⁶

Propositions can be accounted for with a deflationary schema as well:

‘S’ expresses the proposition that S.

Talk of propositions and the use of that-clauses is another convenient way of making generalizations, of expressing that one statement can be translated into another, and relating our mental states to what is linguistically communicated.⁷

So there is a natural way of giving deflationary accounts of the arcana. Still, one may wonder whether the deflationist about truth must accept them, whether she can’t keep her deflationism specific to truth alone. The difficulty is, properties and propositions are, on standard understandings, already connected with the realistic terminology the expressivist has taken on board. What are propositions? They are the fundamental bearers of truth-values, the things that get disquoted, that figure as the contents of beliefs (Schroeder 2008: 160). So it looks like we have normative propositions. These normative propositions predicate properties of objects (e.g., murder is wrong), and so if they are true, the relevant property (being wrong) must be instantiated. And so on.⁸

This is the problem of creeping minimalism. It leaves the expressivist with two connected problems. First, she must find a way of differentiating her account from other metaethical theories. Second, she must preserve the idea that there is some asymmetry between the normative and the nonnormative, that properties like being green or being a star are somehow more real that properties like being wrong. Otherwise the fundamental motivation for the view slips away.

3 Dreier’s explanation explanation

Enter Dreier’s explanation explanation. Even after we embrace minimalist notions of truth, belief, propositions, and properties, there is still, he holds, a distinctively expressivist thesis.

…[E]xpressivists are distinguished by their claim that there is nothing to making a normative judgment over and above being in a state that plays a certain ‘non-cognitive’ psychological role, a role more like desire than it is like factual belief. In particular, to explain what it is to make a moral judgment, we need not mention any normative properties.

(2004: 39)

The expressivist can agree that there are normative properties and propositions. But these properties and propositions are dispensable. We can explain what it is to make a normative judgment without invoking them. One believes murder is wrong in virtue of, for example, disapproving of murder.⁹ That explanation makes no reference to being wrong. This provides a satisfying sense in which the realism is merely quasi. Normative properties do not even play the extremely minimal role of explaining what it is to believe a normative theory. That’s how dispensable they are.

One interpretive complication is that Dreier does not specify the kind of explanation he has in mind (Asay 2013; Taylor 2020). Nonetheless, it is most natural to understand his solution as a claim about the kind of metasemantic explanation a quasirealist will make—where a metasemantic explanation is a metaphysical explanation of the nature (or of the grounds) of content and other intentional properties. Dreier’s locutions here, that the quasirealist offers ‘to explain what it is to make a moral judgment” [emphasis added], or that the normative judgment is “nothing over and above” some noncognitive state, are both characteristic ways of indicating that the explanation in question is a metaphysical one. What’s more, in subsequent discussion Dreier worries that the explanation explanation mistakenly classifies Ralph Wedgwood as a quasirealist; and Wedgwood’s explanation of normative judgments is explicitly metasemantic.

4 Objections to the explanation explanation

Due to its complexity, I am unable to discuss in detail Taylor’s (2020) challenge to the explanation explanation. Taylor’s argument, if successful, shows that the explanation explanation is incompatible with deflationism. I will simply note that Taylor’s objection is to solving the problem of creeping minimalism with some substantive account of the nature of reference and predicate-satisfaction. Given that I draw the distinction between realism and quasirealism in terms of explanations of what something’s content is, rather than in terms of the metaphysical explanation of that content, Taylor’s objection will not apply to my solution.

Unfortunately, theories that seem straightforwardly nonexpressivist meet Dreier’s criterion.

First, an error theorist must think that the property of being wrong does not play any role in explaining what it is to believe that murder is wrong, since there is no such property and explanation is factive (Chrisman 2008; Dreier 2018: 534; Simpson 2018: 516). But this means that if we use Dreier’s explanation explanation to distinguish the quasirealist from the realist, an error theorist will count as a quasirealist.

Second, the explanation explanation seems to misclassify Ralph Wedgwood as a quasirealist (Eklund 2017: 156ff; Dreier 2018: 544–545). Wedgwood accepts a conceptual role semantics, according to which our concepts are individuated by their inferential or conceptual role (2001, 2007). For example, concept C is identical with the concept and in virtue of regulating beliefs in accordance with the introduction and elimination rules for conjunction. Believing p and believing q commits one to believing C(p, q); and believing C(p, q) commits one to believing p and to believing q (2001: 6ff.; 2007: 83ff.). Wedgwood extends this account of concept individuation to normative concepts. Concept B is identical to the concept of better than in virtue of B’s having the role of regulating an agent’s preferences. Believing better than (p, q) commits the agent to preferring p to q (2001). The concept ought similarly regulates intention, though in a more complex way (2007).

As Wedgwood describes the strategy: ‘we must avoid … using these [analysed] terms inside the scope of propositional attitude ascriptions’ in order to explain ‘what it is for a thought to be a thought involving the concept in question’ (2007: 83). So Wedgwood is explicitly committed to explaining what it is to believe that dogs are better than cats, for example, without appealing to a property of being better than. What is it to believe that? It is to be in the state that commits you to preferring dogs to cats.

Dreier acknowledges that his explanation explanation seems to identify Wedgwood as a quasirealist. Regarding Wedgwood’s normative semantics and Alan Gibbard’s expressivist semantics, he writes, “I am worried that I classify them together when they should be separated” (2018: 544–545).

5 Interlude on quasirealist inferentialism

The question of whether Wedgwood can be classified as a realist is made more pressing when we consider Chrisman’s inferentialism about normative expressions (2008, 2010, 2016). In his (2008, 2010), Chrisman proposes a rival solution to creeping minimalism. Rather than identifying the difference between the normative and the more genuinely real in terms of the role normative properties do or don’t play in explaining content, Chrisman advocates a turn to inferentialist metasemantics.¹¹ Inferentialism is similar to conceptual role semantics,¹² but rather than accounting for meaning primarily in terms of the functional properties of an individual’s mental states, it explains meaning in terms of the public social commitments that linguistic utterances create to reason (and act) in certain ways. As Chrisman puts it:

…[T]he inferentialist approach encourages us to explain the content of mental states in terms of the content of statements that conventionally express these mental states, which in turn is explained in terms of their role in the socially-embodied inferential practice of making statements and giving and asking for reasons for these statements.

(2008: 349–350)

Descriptive claims should be thought of as conclusions of theoretical reason, normative claims conclusions of practical reason. This means that the reasons one offers to support a descriptive claim will be evidential reasons, whereas those supporting a normative claim will be practical reasons. Practical conclusions rationally license actions in ways descriptive conclusions do not (2008: 350–351). When we consider these differences in the inferential roles between theoretical and practical claims, we have a sense in which the former but not the latter describe reality (352).

It is unclear, however, that Chrisman’s inferentialism is sufficient to rule out robust realism on its own.¹³ To see why, notice how similar it is Wedgewood’s conceptual role semantics. We could, presumably, “translate” Wedgwood’s position into an inferentialist one. Rather than talk about concepts, we will say the public term ‘F’ means better than just in case asserting ‘F (x, y)’ conversationally commits the speaker to choosing x over y. But if Wedgwood’s conceptual version of the doctrine was compatible with realism, it is unclear what happened in our translation to rule realism out. Or, if Chrisman’s inferentialism leads to quasirealism, how does conceptual role semantics avoid it? Both views say that a judgment is normative not in virtue of how it is guided by “upstream” inputs, but how it leads to motivation or licenses choice.

This suggests an additional criterion for quasirealism—ideally our theory would explain how a conceptual role theorist or an inferentialist could be either a realist or a quasirealist, depending on further commitments.

6 What the expressivist is thinking

My proposal is that Dreier’s explanation explanation is to a large extent correct, but that it fails to specify the kind of explanation the quasirealist is offering. The quasirealist can give a special kind of semantic explanation, whereas the explanation explanation runs into problems because it looks to metasemantic explanations.

This will require that we specify more clearly the distinction between semantics and metasemantics, along with characterizing the relationship between formal (or compositional) semantics and semantics more generally. As I will use the phrase, the paradigm case of a semantic explanation explains what a word or expression means (e.g., Achinstein 1983: 33 and fn. 21). This is a familiar kind of explanation. It is the kind of explanation I give when my son asks me “What is a reptile?” or “What is an election?” and I try my best to offer him a definition serviceable to a young child. To explain what a word or sentence means is usually just to offer a definition, or a translation, or provide examples of how the expression is used. A metasemantic explanation, in contrast, explains why these sentences have those meanings; it is an explanation in terms of the metaphysical nature or grounds of semantic properties.

On this taxonomy, formal or compositional semantics is only a portion of semantics, because it does not completely explain what the expressions in a natural language mean, nor does it aim to. Consider, for example, explaining the meaning of ‘necessarily’ with ‘Necessarily p is true at w iff p is true in all worlds accessible from w’. We have not fully explained the meaning of ‘necessarily’ until we say, for example, what worlds are. If talk of possible worlds is not intended literally—if they are not real objects in our ontology—but simply as a device for modeling aspects of meaning, the explanation is incomplete until we are told what is being modeled.

This is not to deny that there is much that formal semantics does explain about meaning. For example, among the semantic properties of a linguistic expression are its logical properties and its compositional properties, and these are properties which a formal metalanguage is uniquely suited to illuminate. In answer to our question what ‘necessarily’ means, it tells us that whatever it means must stand in certain logical relations to whatever certain other expressions mean, and whatever it means must be a function of a certain sort that takes certain other types of meanings—whatever they are—as arguments.

To avoid confusion, it is important to emphasize that mine is not the only available taxonomy.¹⁴ Chrisman, for example, classifies those questions about what an expression means beyond its compositional properties as metasemantic (2016: 15, fn. 17, and 17). Still, at this point in the paper, any disagreement is merely verbal. Chrisman agrees that compositional semantics does not provide a complete account of what natural language expressions mean (2016: 10–18). He is also clear that the use of truth-conditional semantics should not be thought to commit us to a realistic or representationalist picture of meaning (10–12); rather, truth-conditional semantics are a “convenient” device “to keep track of the commonality of content across… embeddings…” (11).

What is the motivation for my taxonomy? First, the claims of formal semantics are partial answers to what-questions, and so they make up a natural class with other theses that answer (partially or completely) the same what-questions, and this class contrasts naturally with theses that answer other questions, such as the why-question. Second, to identify semantics with formal or compositional semantics obscures the fact that nonlinguistic entities can have semantic properties as well. Maps, diagrams, and mental states can all have content; they can be true or false, satisfied or unsatisfied, veridical or nonveridical.

Finally, my own solution to the problem of creeping minimalism is distinguished from traditional explanationist answers in that it focuses on answers to what-rather than why-questions. This taxonomy gives a convenient handle for identifying the kind of solution I propose.

As noted, the issue at this point is verbal. Those opposed to my taxonomy are welcome to think of the distinction I have invoked in terms of explanations of what something’s content is, versus explanations of why it has that content.

Two final points should be made about semantic explanations. First, I said that the paradigm case of a semantic explanation is an explanation of what an expression means. But given my definition, it is natural to identify any explanation of what something’s content is as semantic. Thus, I will treat explaining what someone is thinking about, or explaining what a map or diagram tells us, as semantic explanations as well. The key idea here is that semantic explanations tell us what something’s content is, along with the intentional relations (e.g., illocutionary force, direction of fit, attitudinal relations) it bears to that content. Those who dislike this taxonomy could instead call these intentional explanations, to be contrasted with meta-intentional explanations of the why something has its content and other intentional properties. It is enough for my purposes if we are aware of what they have in common with paradigmatically semantic explanations.

Second, explanations of what an expression means are often indirect: we do not provide a translation or definition, but rather state or provide examples of how the expression is used. Importantly for the expressivist, an indirect semantic explanation of this sort may still be a complete one (though often it will be incomplete). The T-schema is an indirect explanation of the meaning of ‘true’, but one that, according to deflationists, exhaustively accounts for the predicate’s meaning. Expressivists should be understood as similarly giving an indirect explanation of what terms mean—explaining how ‘wrong’ is used to express various noncognitive attitudes is a way of exhaustively explaining what the predicate means (Dreier 2015: 282ff.).

While these indirect explanations do not identify for us what the content of an expression is, they are still of a class with direct explanations because they explain the meaning of an expression by showing how the expression modifies the way other expressions are related to their content. According to the deflationist, we can’t directly say anything informative about the content of ‘true’—because it doesn’t designate a substantive property. But we can say how ‘true’ contributes to the content of larger expressions. The content of the sentence ‘p is true’ is p, and ‘true’ makes the sentence into a way of indirectly asserting that content. The content of ‘Everything Kant said is true’ is the propositional content of each of Kant’s utterances, and again ‘true’ makes the sentence into an indirect assertion of these contents. In short, an indirect explanation of the meaning of ‘true’ still explains, at least partly, what the word contributes to the meaning of the sentence in which it occurs. For deflationists, it fully explains the word’s contribution.

With all this in place we can say that a striking feature of traditional expressivism is its unorthodox semantics: what a normative sentence means is explained in terms of the state of mind expressed. The meaning of ‘murder is wrong’ is an expression of disapproval of murder, or of planning on getting angry at murder, or of being for blaming for murder, etc.¹⁵

Quasirealism risks obscuring this unique expressivist semantic commitment. It does so because it also commits the expressivist to saying ‘Murder is wrong’ means that murder is wrong. The first meaning—in terms of disappointment—is supposed to, in some sense, be more fundamental. And so it is natural to understand the quasirealist expressivist as offering a distinctive metasemantic thesis: ‘Murder is wrong’ means that murder is wrong in virtue of expressing disapproval of murder.

This isn’t wrong. But we should be aware that insofar as she retains her unique semantic commitments, the kind of metasemantic explanation she is offering is unusual. The disapproval of murder is itself a meaning of the sentence. The expressivist is committed to saying that ‘Murder is wrong’ means one thing (that murder is wrong) in virtue of meaning another (disapproval of murder).

Metasemantic explanations are not always like this. If someone says the word ‘dog’ refers to dogs because of causal chains connecting us to an original baptism, they are not thereby committed to saying that the word ‘dog’ is about baptisms or causal chains.¹⁶

Quasirealism is unique, then, in saying that normative utterances have two complete meanings, one of which is given in terms of normative properties, and one of which is given in terms of expressed attitude. This points to a way in which normative properties are dispensable. We could, in principle, say what every possible normative sentence in our language means without mentioning normative properties.

Dreier’s explanation explanation, then, is in some sense right. But it has been developed in terms of the wrong kind of explanation, an explanation of why words mean what they do, or why thoughts have the content they do. The proposal works much better if we interpret it instead as a claim about semantic explanations. Quasirealism is distinctive for holding that there is an alternative way of explaining completely what the content of a normative thought or sentences is, without mentioning normative properties (Lenman 2003).

Dreier, it should be noted, puts things not primarily in terms of explaining linguistic meaning, but of explaining what it is to make a judgment with normative content. As noted, this has been interpreted in the subsequent literature as a kind of metasemantic (or meta-intentional) explanation, a metaphysical explanation of the nature or grounds of content. We can avoid standard objections, however, if we instead characterize quasirealism in terms of a special kind of semantic (or intentional) explanation, a special explanation of what a normative thought is about.

My proposal is this: the quasirealist thinks that we can fully explain what the person who believes that murder is wrong is thinking about without mentioning being wrong, as long as we can indicate the way in which she is thinking about it. What is she thinking about? Murder. How is she thinking about it? Disapprovingly. The quasirealist also thinks we can fully explain what ‘murder is wrong’ means, without mentioning wrongness, as long as we can say which attitude the sentence expresses.

So, as a first pass, the quasirealist is committed to the following theses:

INTENTIONAL. We can completely explain what any normative belief is about by giving a characterization of the belief (i) which does not mention normative properties in describing the attitude’s content, by (ii) describing the belief as a desire-like intentional attitude.

SEMANTIC. We can provide a complete explanation of what any normative assertion means in a metalanguage (i) which lacks any terms or expressions ascribing normative properties, by (ii) indicating which desire-like intentional attitude the assertion expresses.

I will identify both as semantic dispensability theses, as they identify certain contents as dispensable to explanations of what a sentence means or a thought is about.

It is important to note what the theses do not say. They do not claim that we cannot also explain what normative beliefs are about or what normative utterances mean by invoking normative properties. The sophisticated expressivist will say that ‘murder is wrong’ means murder is wrong and that it expresses the belief that murder is wrong (Timmons 1999). The distinctive commitment of the expressivist is that there is an alternative way of explaining what the sentence means or the thought is about, which does not invoke normative properties. This makes the more standard explanation dispensable, at least in principle.

The theses do not misclassify error theory. The error theorist must think we can say why normative sentences mean what they mean without invoking normative properties, since there aren’t any. But she also thinks that we cannot say what the sentence means in terms that do not ascribe normative properties—otherwise there would be no error.

What does the thesis say about Wedgwood? It says that Wedgwood need not be a quasirealist. Wedgwood can explain why a belief is about cats being better than dogs, without invoking the better than relation, explaining instead in terms of the judgment’s functional role. But this does not mean that he can say what the belief is about without mentioning the better than relation. The belief is not about a functional role, any more than a conjunctive belief is about introduction and elimination rules. Admittedly, we may be able to use the functional role to offer an indirect explanation of what the belief is about. But this will only lead us to classify Wedgwood as a quasirealist if this indirect semantic explanation is complete (and Wedgwood has grounds for saying it is not). This will be addressed in Sects. 6.2 and 6.3. For now, the point is simply that conceptual role semantics, as a metasemantic (or meta-intentional) thesis, does not by itself entail quasirealism.

Do the theses misclassify normative reductionism as a form of quasirealism? No, because they are given in terms of properties rather than vocabulary. Let’s say that being right reduces to maximizes happiness. Now either this means that being right is identical to maximizing happiness, or it does not (views about reduction and the fine-grainedness of properties differ). Let’s further say we try to explain that ‘Pushing the button is right’ means that pushing the button maximizes happiness. If the properties are identical, this explanation does not vindicate SEMANTIC, because we are explaining the meaning of the normative claim by using a complex predicate ‘maximizes happiness’ that ascribes a normative property. If the properties are not identical, the explanation of what the normative sentence means is at best incomplete. Mutatis mutandis for explanations of what normative beliefs are about.

Both theses need refinement. We must expand them to allow for inferentialist quasirealists, or expressivists about, say, epistemic modals. Before all of that, though, I want to explain why this kind of dispensability—dispensability to explanations of content—is relevant to questions of quasirealism. After all, we have already seen that metasemantic dispensability is not.

6.1 What are we arguing about?

Dreier’s proposal was that our beliefs about the quasireal are different, because we can say what it is to have such a belief without invoking the property the belief is about. This seemed to get at something right, but it proved too broad for a variety of reasons. There are many ways of characterizing normative thoughts which don’t invoke the property the thought is about (Eklund, ibid.). If physicalism is correct, there may be some day when we can say, ‘He’s in brain state 284-C again’, and thereby say what he’s thinking without mention of normative properties. Less hypothetically, there are accounts such as Wedgwood’s.

Metasemantic dispensability turned out to be irrelevant. But then why should semantic dispensability be the relevant sort?¹⁷ There are two reasons. First, expressivism is a metaethical theory, and therefore a project in metatheorizing. An essential part of metatheorizing must be determining what the first-order theories and debates of the relevant subject matter are about. But for that project, telling us what brain states the theorists are in is a non-starter. If philosophers ask what the subject matter of mathematics is, or what mathematicians are doing, they would not accept an answer in terms of how certain electrical signals lit up in the brains of these weird, ambulatory, fluid-filled bags, or an answer in terms that characterized the mathematicians as extremely complex input–output devices. A metatheoretical account is illuminating in virtue of helping us better understand what kind of reasoning the theories involve, and what aspects of the world the theorists are reasoning about.

Semantic dispensability theses are special because they identify what it is to hold an opinion on some matter of first-order normative theory in terms of what such a statement or judgment is about (e.g., its content), and they do so in a vocabulary that is more immediately bound up with our conception of ourselves as agents and reasoners. That normative properties may be dispensable from other kinds of explanations of what it is to make a normative judgment is simply not relevant to the expressivist’s metatheoretical project. But dispensability for semantic explanations or explanations of what someone is thinking are relevant, because those characterizations are compatible with continuing to understand those making the judgments in question as theorists or reasoners.

What does the expressivist think normative discourse is about? As noted above, the person who believes that murder is wrong is thinking about murder (disapprovingly). More generally, according to the quasirealist, normative discourse is about what to approve and disapprove of, or what to plan on doing,¹⁸ or what to favor (i.e., be for).¹⁹ On the most general level, it is about what to do and how to feel. According to the expressivist, we could, in principle, explain normative discourse to aliens unfamiliar with it, without ever making mention of normative properties—whether as genuine pieces of reality, or even merely as something to which those who engage in normative discourse are committed. To explain normative discourse, it would suffice to explain how those engaged in it were discussing what to do, think, and feel.²⁰

The motivation for the dispensability theses also helps us to determine whether a quasirealist must commit to both theses, or whether endorsing either one is sufficient for quasirealism.²¹ The default position is that a quasirealist must satisfy both. We want to explain what normative theorizing and reasoning are about. For that project, it seems we need to explain what our both normative thought and our normative talk are about. After all, normative theories are presented, and publicly debated about, with words. If we don’t satisfy both INTENTIONAL and SEMANTIC, there is some aspect of normative theorizing that we can only make sense of when understanding it as about normative properties.

But what if one can give principled reasons why, for example, the way that theorizing is linguistically presented is relatively unimportant for understanding such theorizing? Wouldn’t we then have reasons for thinking that a quasirealist only needs to satisfy INTENTIONAL?²² Yes, but then such arguments need to be given; and even in this case, I suspect any quasirealist must satisfy at least a weaker version of SEMANTIC to be worthy of the name.

Let me explain why. Suppose the aliens land, and they want to know what normative discourse and theories are about. So we explain what the participants are thinking about, and we do so without mentioning normative properties. Of course, we do have to mention normative properties when explaining what they are saying or writing, but that’s just because there is no other way to account for some of the complex grammar of English (or Russian or Japanese). And the aliens weren’t asking about grammar. Their eyes glaze over.

This may be enough to qualify as quasirealist. Talk normative properties is dispensable for understanding our normative reasoning, it seems. It’s just natural-language grammar that requires us to mention them.

There’s a question about the story, however. Have we told the aliens how the normative reasoners’ thoughts are logically related to one another? That is, can we explain what the first-order logical relations between normative beliefs are, while only characterizing those beliefs in noncognitive terms?

If the answer is yes, we have a solution to the Frege-Geach problem, at least in its simpler iterations. We can explain normative negation and modus ponens, for example. We should also be able to give a partial explanation of what ‘Murder is wrong’ and ‘If murder is not wrong, then encouraging murder is wrong’ mean in terms of the attitudes they express. After all, they are used, in part, to express attitudes. (It’s hard to see how a theory could claim to be expressivist if it doesn’t allow this much.) But notice that this partial explanation is, from what we have said, by itself sufficient to explain the logical relations between the two object sentences. That ‘wrong’ predicates a property is not needed to explain logical relations between sentences containing the term: the role of ‘wrong’ in expressing condemnation is already enough for that.

In short, we would have sufficient resources to satisfy:

SEMANTIC-MINIMAL. We can provide a partial explanation of what any normative assertion means in a metalanguage (i) which lacks any terms or expressions ascribing normative properties, by (ii) indicating which desire-like intentional attitude the assertion expresses, and (iii) which explains all first-order logical properties of sentences in the object-language.

Let’s say, on the other hand, the answer is no. We can’t explain how normative beliefs are logically related to each other insofar as we characterize those beliefs noncognitively. Then we either have to say that relations of entailment, inconsistency, implication, and so on are marginal to understanding normative reasoning and theorizing—which is false—or we have to accept that how normative theories and reasoning are linguistically represented is an essential part of metatheoretical understanding. In that case, to qualify as quasirealists, we need to show that we can at least account for the first-order logical relations in normative language without talk of normative properties—it’s only wonkier compositional properties for which normative properties must be mentioned. But that means we still need to satisfy SEMANTIC-MINIMAL.

6.2 More generally

Our two theses may provide a sense in which normative properties are dispensable, but at present they are too narrow. They assume a quasirealist is a metaethical expressivist. But expressivism has been proposed for other aspects of language, notably epistemic modals (Yalcin 2007; Incurvati and Schlöder 2019). An expressivist of this sort will likely want to be a quasirealist about epistemic possibilities (e.g., epistemically possible worlds). What’s more, a conceptual role theorist or an inferentialist may wish to embrace quasirealism as well.

Expressivism about a nonnormative domain is easy enough. We can simply broaden the state of mind that can be expressed and identified with the relevant kind of belief. For example, to judge that might p is to be in a state in which one’s beliefs fail to rule out p. Alternately, one may follow Incurvati and Schlöder (ibid.) in putting public speech acts first. To assert ‘It might be that p’ is to commit to ruling out adding ~ p to the conversational context.²³

As for conceptual role and inferentialist semantics, as already pointed out, these are standardly metasemantic theses, not semantic ones. Nonetheless, it is easy enough to combine them with quasirealist semantic commitments. They are both versions of meaning is use theories, and as noted, to provide information about how to use an expression is often a way of providing an indirect explanation of what something means. The expressivist can say that the belief that murder is wrong is a thought about murder, as long as she adds it is a disapproving thought about murder. But a conceptual role theorist could say something similar. What is the person who believes that cats are better than dogs thinking about? The ordered pair, [cats, dogs]. How is she thinking about that pair? In a way that commits her to preferring the first member to the second. The inferentialist quasirealist story will be similar, but it will put public speech acts first. What does the assertion, ‘Cats are better than dogs’ mean? It means cats are better than dogs, obviously, but we can say that it also is a commissive speech act, and the content of the commitment is the speaker choosing cats over dogs.

I am not saying that such an alternate but complete account of meaning is entailed by conceptual role or inferentialist semantics. Rather, such an alternative semantics is compatible with these metasemantic theses, and can be taken on should one wish to combine the metasemantic picture with quasirealism. To avoid quasirealism, as Wedgwood would, one must reject that the alternative meanings are complete.

We can now state the general form of quasirealism. To be a quasirealist about property F is to accept:

INTENTIONAL-GEN. We can completely explain what the belief that F(x₁, x₂, …, x_n) is about by giving a characterization of the belief (i) which does not mention property F in describing the state’s content, by (ii) describing the belief as a nonrepresentational intentional state.

SEMANTIC-GEN. We can provide a complete explanation of what assertion ‘F(x₁, x₂, …, x_n)’ means in a metalanguage (i) that lacks any terms or expressions for ascribing property F, by (ii) saying which nonassertoric speech act the assertion performs (e.g., expressive, commissive, imperative).

Mutatis mutandis for quasirealism about object o.

6.3 Two strategies for preserving realism

My proposal allows for a conceptual role theorist or inferentialist to accept or reject quasirealism, depending on whether they go on to accept or reject the semantic dispensability theses. But more needs to be said. Even if denial of INTENIONAL-GEN and SEMANTIC-GEN is logically consistent with either of these metasemantic views, one may worry that it is unmotivated. Wedgwood seems committed to holding that to believe that cats are better than dogs is equivalent to thinking about cats and dogs, in a way that commits us to preferring the one to the other. On what grounds could he reject INTENTIONAL-GEN?

It isn’t enough to say that Wedgwood intends his theory as a metasemantic thesis, rather than a semantic one. The problem follows from the nature of meaning is use metasemantic theories. They explain why an expression or concept means what it means in terms of how the expression or concept is used—but this means that their metasemantic explanations can also serve as indirect semantic explanations. (This means one and the same answer is an answer to both the what- and why-question about content, something we already saw was possible with expressivism.)

To see how an inferentialist or conceptual role theorist could reject the dispensability theses for some F, an analogy will be helpful. An inflationist about truth, a correspondence theorist, for example, can agree that the predicate ‘true’ is a device for disquotation, quantification, and indirect assertion. She can even say that describing these roles provides important information about the meaning of the term. She will deny, however, that any such indirect explanation is complete. The deflationist, by contrast, holds that an explanation in such terms would be complete, and so talk of truth is in principle dispensable, though it may require a language that allows for forms of quantification ruled out by natural language grammar. For the inflationist, we cannot dispense with truth talk, even in principle, without losing something.

Similarly, to accept the two semantic theses for some F is stronger than simply saying that the standard semantic value is determined by, say, expressed attitude, or inferential commitments, or conceptual role. It is also to say that the traditional semantic value is dispensable (in principle) in favor of talk about expressed attitude, inferential commitments, or conceptual role. One can embrace the determination claim but deny the dispensability claim.

This is what Wedgwood should do. Given his metasemantics, one could explain what someone was thinking about by saying, “She’s thinking about cats and dogs, in a way that commits her to preferring the former to the latter.” This explanation may be helpful in some contexts, it may provide important information about the content of the person’s thought. But, for Wedgwood, it is an incomplete explanation. A complete explanation must also make reference to being better than.

Why reject dispensability? Two obvious reasons come to mind. The first is simple conservatism. Both sides, the realist and the quasirealist, agree that wrong is one semantic value of ‘wrong’. Why would our language contain lots of semantic clutter? The default should be that semantic values are not dispensable for an explanation of what a given utterance means. We should only be quasirealists or deflationists about those properties or objects for which special positive reasons for accepting dispensability can be given.

This fits with how expressivists have traditionally presented their view, identifying aspects of normative thought and talk that are, well, weird when compared with more quotidian descriptive discourse. These idiosyncrasies are supposed to motivate a “beneath the surface” understanding of such thought and talk—with the implication that the surface understanding is fine for most parts of language. These oddities are numerous and well-known: the empirical inaccessibility of normative properties, widespread cross-cultural disagreement about their distribution, the tie with motivation, atypical intuitions about disagreement and translation, and so on. It may be that none of these is a good reason for treating normative properties as dispensable. But that is simply to say quasirealism might be wrong, not that there isn’t a distinctive view here, or no prima facie motivation for it. The local quasirealist thinks indispensability is the default, but thinks it is overcome in special circumstances.

The assumption that realism is the default fits moreover with a certain ‘just so’ story it is natural to see as working in the background. We start off using language to communicate what things in the world are like. We do so with sentences with a subject-predicate structure. What are we communicating with these sentences? Given how basic this form of communication is, we have no way of saying, other than to simply use the same subject-predicate structure. But after a while our communicative needs grow beyond simply saying what the world is like. We find we need to express our attitudes, or to indirectly assert and to quantify over the claims of others. In order to meet these communicative needs, we end up exploiting the existing subject-predicate structure of our language. We deploy predicates as devices for expressing our feelings or for quantifying. The properties ascribed by this mode of communication are dispensable, because the relation between what is communicated and subject-predicate structure is only one of convenience.

The second reason to reject dispensability has to do with the connection, already noted, between giving a semantic interpretation to some discourse and metatheorizing. A metatheoretical interpretation, if it is not to be a debunking account, needs to show that the target discourse is a more or less rational activity, that those involved in it can be understood as reasoning. A given F (or o) may be semantically indispensable, because we cannot make sense of first-order theorizing as a rational activity, unless we understand the theorists as talking about F.

This is Wedgwood’s reason, incidentally, for rejecting expressivism. For the expressivist, reasoning about the normative is a matter of establishing greater consistency among the relevant desire-like states. But, Wedgewood asks, “Why should the agent strive so hard to achieve … intrapersonal and interpersonal consistency…?” (2007: 50). If this were all that was at stake, normative reasoning would seem to be ‘a bizarre fetish for logical consistency’ (50). Wedgwood argues that we can only make sense of this striving for consistency because consistency is a tool for getting things right, for making correct judgments; and these judgments are correct in virtue of being true (51–57).

But here we have a difference. The quasirealist thinks that we can exhaustively explain what first-order normative discourse is about without mentioning normative properties. Wedgwood may think we can explain why discourse is normative without mentioning such properties, but as an explanation of what the discourse is about this would be incomplete: it would make normative discourse into an elaborate verbal and intellectual game, or a strange psychological compulsion. To fully explain what participants are doing and thinking, we need to add that they aim to get things right, to think something true.

Consequently, the full explanation of what the belief that cats are better than dogs would have to be: thinking about cats and dogs in a way that commits you to preferring cats to dogs in virtue of aiming to be true of cats and dogs. But to be true of cats and dogs the thought must predicate something of them. So mentioning the better than relation is essential to explaining—saying—what the thought is about.

Of course, this way of stating Wedgwood’s motivation for realism has it that our beliefs aim at truth. Have we smuggled inflationism back in? It seems unlikely. Defenders of deflationism have thought their view is compatible with the claim that we or our theories, aim at truth (Field 1994: 264–265; Horwich 1998a: 63ff.).

For example, in Wedgwood’s case, we could also say that we can only make sense of normative reasoning as reasoning insofar as we interpret normative reasoners as aiming to believe p if and only if p, for all normative p. Talk of truth is simply a way of getting around the fact that natural language doesn’t allow such forms of quantification.

6.4 Metasemantic expressivism

Is the solution here compatible with the recent project of interpreting expressivism as a metasemantic doctrine? I suspect it is. As noted earlier, metasemantic expressivists, such as Chrisman, draw the boundaries between semantics and metasemantics differently than I do—identifying semantics with compositional semantics of the sort practiced by linguists. But, as noted, they also maintain that truth-conditional semantics can be given a non-descriptivist interpretation (Ridge 2014; Chrisman 2016). For that to be true, compositional semantics must offer incomplete explanations of what expressions mean, a point Chrisman explicitly endorses.²⁴

From what I can tell, then, nothing rules out providing an interpretation of what our formal semantics tells us about the meaning of normative expressions which does not explain what they mean in terms of normative properties, but nonetheless suffices to capture the logical relations between those expressions. That is, metasemantic expressivists should be able to satisfy at least SEMANTIC-MINIMAL (or a generalized version of it in Chrisman’s case).

We should remember that the metalanguage in which SEMANTIC-MINIMAL is satisfied need not account for all of the compositional properties of English or other natural languages. An analogy with deflationism about truth is helpful once again. One way of capturing the way in which the property of truth is dispensable to the deflationist is to say that we could in principle cease talking about truth without losing anything, if it were replaced with forms of quantification natural language does not allow for. In short, it is enough to show that we could replace our current language with an artificial one that made no reference to truth without losing communicative power. One way to think about the challenge of SEMANTIC-MINIMAL, then, is that a quasirealist should think we could in principle present our normative theories and arguments in an artificial language which made no reference to normative properties, but which preserved the narrowly logical relations between those theories and premises (Lenman 2003). This seems compatible to me with the metasemantic expressivist project.

But if I am wrong, then to defend their right to quasirealism, metasemantic expressivists (or non-descriptivists) need to show why my arguments in Sect. 6.1 that SEMANTIC-MINIMAL really is a minimal commitment for quasirealist expressivists fail. The basic ideas behind SEMANTIC-MINIMAL are, first, that a quasirealist must be able to explain what normative theorizing and reasoning are about without mentioning normative properties. Second, this requires at the very least that they be able to account for the logical relations between normative thoughts without mentioning normative properties. Finally, if they can account for logical relations between normative thoughts without mentioning normative content, they should be able to account for strictly logical relations between normative sentences without mentioning normative content. It should be remembered that I am already assuming that arguments are available for thinking most aspects of how normative theorizing is linguistically represented are relatively unimportant to understanding the activity. If such arguments fail, metasemantic expressivists will need to satisfy something even stronger than SEMANTIC-MINIMAL.

7 Conclusion

Creeping minimalism forces the expressivist to explain how she can preserve what is distinctive of her view without embracing a revisionary picture of normative language. I have argued expressivism remains distinctive, even if it allows for normative properties, because we can say what those engaged in normative discourse are saying and thinking without making reference to such properties. This points to a way in which her realism is attenuated: talk and thought about the properties is dispensable. This is in the spirit of Dreier’s explanation explanation, but with a novel reinterpretation of the kind of explanation we want. If this is right, quasirealists should not look to distinguish their view from realism in terms of explanations why expressions or thoughts have their content, but rather should say that they are committed to a certain kind of explanation of what the content is. This way of drawing the distinction avoids the standard objections, and furthermore is revealing of the kind of metatheoretical project in which the quasirealist is engaged.