1. Introduction
Old versions of expressivism—the view that ethical thoughts are practical attitudes that get expressed by ethical sentences—said that ethical sentences don’t have truth-conditions, don’t describe or represent things, and don’t express beliefs. Recently, expressivists have dropped these claims, influenced by minimalist theories of truth and related concepts, which make them inconsistent with expressivists’ prior commitments. This threatens to make expressivism indistinguishable from rival views; James Dreier (Reference Dreier2004) calls this the problem of creeping minimalism.
Dreier (Reference Dreier2018) defends what he calls an explanationist solution to the problem, which focuses not on whether ethical language and thought has the features mentioned above, like truth-conditions, but whether we need to appeal to those features to explain what gives ethical sentences and beliefs their content. Dreier’s ontological explanationism says that expressivism is distinctive in avoiding ethical ontology (facts and properties) when explaining ethical content. Dreier faces problems accounting for false beliefs which do not correspond to facts, and error theories which reject ethical ontology but nevertheless are not expressivist. Representational explanationism instead distinguishes expressivism by its evasion of concepts like truth, representation, and belief. However, Dreier (Reference Dreier2018) argues that this is too open-ended and threatens to make minimalism entail expressivism not just about ethics but about all language, an unwelcome consequence for many.
In this paper, I defend an explanationist solution to the problem of creeping minimalism that captures the advantages of both ontological and representational accounts. Expressivism is distinct from its rivals because it does not mention the subject matter of ethical language and thought, where subject matter is read as not ontologically committing. I will argue that this subject matter view emerges from and captures the advantages of both ontological and representational explanationism but avoids their drawbacks. I will also argue that it accounts for other non-representationalist views like expressivism about modality and causation, and that it is the best way to understand the broader non-representationalist movement of which expressivism is a part.
2. Minimalism
In the literature on creeping minimalism, it is not always clear what minimalism involves. To understand it, let’s start with Paul Horwich’s (Reference Horwich1998) minimalism about truth, the best-known minimalist view. Horwich’s view is also known as deflationism; I will use these two names interchangeably, though some philosophers do not.
Horwich’s minimalism centres on a negative claim and a positive one. The negative claim is that the property of truth—considered for now as a property of propositions—has no constitution thesis (Horwich, Reference Horwich1998, 120–22). What this means is that there is no claim of the form:
$$ \mathrm{For}\ \mathrm{all}\ \mathrm{propositions}\kern0.5em x,\ x\kern0.5em \mathrm{is}\ \mathrm{true}\ \mathrm{iff}\kern0.5em x\kern0.5em \mathrm{is}\kern0.5em F $$that is necessarily true, and which explains the facts about truth. The property F-ness that such a claim would specify would be a property that constitutes truth; Horwich thinks there is no such property, no property we can identify as truth in a way that would explain the facts about truth. Any proposed candidate will either fail to be necessarily coextensive with truth or fail to explain the facts about truth. The positive claim makes use of a schema:
$$ (T)<p>\mathrm{is}\ \mathrm{true}\ \mathrm{iff}\kern0.5em p $$Horwich argues that grasp of (T) constitutes understanding of truth: to understand the truth predicate ’… is true’ just is to accept (T) in general (1998, 35, 126).Footnote 1
This view has two important consequences for our purposes. First, given other facts, (T) tells us which propositions are true. Once we know whether grass is green or whether dogs bark, we know whether <grass is green> and <dogs bark> are true. Whenever we know a worldly fact of this kind, we know that a corresponding proposition is true.
Second, truth loses its explanatory power. (T) is intended not only as a biconditional, but as an explanatory claim; when a proposition <p> is true, it is true because p. Footnote 2 There is nothing more to <grass is green>’s being true than grass’s being green, and <grass is green> is true entirely because grass is green. Whether this proposition has the property of truth is completely settled by the colour of grass; the truth of the proposition is a trivial consequence of grass’s being green. This means that there will be no explanations in terms of the truth of <grass is green> that couldn’t be replaced by an explanation in terms of grass’s being green.
Minimalism “creeps” in two directions. The first covers certain entities that have a close relationship with truth: facts and properties. Minimalism about facts and properties says that claims about them trivially follow from claims about truth (Horwich, Reference Horwich2008, 273). Given what (T) says about truth, we get the following schemas:
$$ (F)\kern0.5em \mathrm{The}\ \mathrm{fact}\ \mathrm{that}\kern0.50em p\kern0.5em \mathrm{exists}\kern0.50em \mathrm{iff}\kern0.5em p\kern0.5em (P)\kern0.5em O\kern0.50em \mathrm{has}\kern0.50em \mathrm{the}\ \mathrm{property}\kern0.5em F-\mathrm{ness}\kern0.5em \mathrm{iff}\kern0.50em O\kern0.50em \mathrm{is}\kern0.5em F $$Minimalism about facts and properties says that grasp of (F) and (P) constitutes our understanding of facts and properties. It has the same two results as minimalism about truth. It tells us which facts exist and which things have which properties given other things—if we know whether grass is green, we know whether the fact that grass is green exists, and we know whether grass has the property of being green. It also tells us that facts and properties have no explanatory role here—neither the existence of the fact that grass is green nor grass’s having the property of being green can explain why grass is green (Horwich, Reference Horwich2008, 273).
Minimalism about truth, facts, and properties defines the truth of propositions, the existence of facts, and the instantiation of properties as trivial consequences of some worldly state of affairs. The next kind of minimalism instead defines a property of a sentence, belief, or other representation of the world as a trivial consequence of a different property of that representation. To see how this works, consider truth again. Horwich’s schema tells us what it takes for a proposition to have the property of truth. Horwich extends this to sentences and beliefs by way of the idea that sentences and beliefs can express propositions, or mean that p, for some p (1998, 100–2). What we get is the following schema:
$$ ({T}^{*})\kern1.00em \mathrm{A}\ \mathrm{sentence}\ \mathrm{or}\ \mathrm{belief}\kern0.5em x\kern0.5em \mathrm{is}\ \mathrm{true}\ \mathrm{iff}\ \mathrm{for}\ \mathrm{some}\kern0.5em p,\ x\kern0.5em \mathrm{means}\kern0.5em p\kern0.5em \mathrm{and}\kern0.5em p $$The minimalist claim is that grasp of (T*) constitutes understanding of truth as applied to sentences and beliefs.
This in turn yields a broader principle concerning truth-conditions—for a sentence or belief to have truth-conditions is for there to be some p such that it is true iff p, which given (T*) just means that there is some p such that it means that p:
$$ (TC)\ \mathrm{x}\kern0.5em \mathrm{has}\kern0.5em \mathrm{truth}\hbox{-}\mathrm{conditions}\ \mathrm{iff}\ \mathrm{for}\ \mathrm{some}\ \mathrm{p},\ x\kern0.5em \mathrm{means}\ \mathrm{that}\kern0.5em p. $$The minimalist idea is that (TC) gives us a definition of the property having truth-conditions in terms of another property, namely meaning p for some p or, in other words, having propositional content. This view has the same results as before. First, via (TC), minimalism tells us which things have truth-conditions given other facts—about which things have propositional content. Second, it prevents explanations of meaning in terms of truth-conditions. For (TC) is intended to say that x has truth-conditions because it has propositional content, not the other way round; minimalism about truth-conditions tells us that having truth-conditions is a trivial consequence of having propositional content.
So one property of beliefs, sentences, and other representations—having truth-conditions—has been explained as a trivial consequence of another property they have, namely having propositional content. This manoeuvre also applies to certain other concepts, including representation, belief, and description. We can give the following schemas:
$$ (RD)\kern0.5em x\kern0.5em \mathrm{represents}\ \mathrm{and}\ \mathrm{describes}\ \mathrm{grass}\ \mathrm{as}\ \mathrm{green}\ \mathrm{iff}\kern0.5em x\kern.15em \mathrm{means}\ \mathrm{that}\ \mathrm{grass}\ \mathrm{is}\ \mathrm{green}\kern0ex \hspace{0.5em}(B)\kern0.5em x\kern0.5em \mathrm{expresses}\ \mathrm{the}\ \mathrm{belief}\ \mathrm{that}\ \mathrm{grass}\ \mathrm{is}\ \mathrm{green}\ \mathrm{iff}\kern0.5em x\kern0.5em \mathrm{means}\ \mathrm{that}\ \mathrm{grass}\ \mathrm{is}\ \mathrm{green} $$The minimalist claim is that grasp of these principles constitutes understanding description, representation, and expressing belief. These schemas, like the others, are intended to have an explanatory consequence: x represents and describes grass as green and expresses the belief that grass is green because it means that grass is green, not the other way round. (RD) and (B) can be generalised to yield a definition of being representational, descriptive, and belief-expressing in terms of having some propositional content or other.
As before, we get two consequences. These claims tell us which things are representational, descriptive, and belief-expressing given other facts—about which things have propositional content. And they undermine explanations of meaning in terms of these concepts. We can’t say that the sentence ‘grass is green,’ or someone’s belief that grass is green, means that grass is green because it describes or represents grass as green, since minimalism tells us that it’s the other way around. This is part of the idea that representation and the other concepts here are just trivial add-ons to the notion of meaning.
A further question here is what a minimalist might say about meaning itself. However, we do not need to answer this question for the purposes of understanding the problem of creeping minimalism, because it does not affect the two crucial consequences of the views described here. The first consequence is that they tell us when these concepts apply in virtue of other facts, namely facts about what has propositional content. We’ll see that expressivists all agree about these latter facts whatever they might ultimately say about meaning, and this is the source of the problem of creeping minimalism. The second is that they block explanations of meaning in terms of these concepts. This is true whether or not ultimately meaning gets a similar treatment which blocks explanations of other phenomena in terms of it. Moreover, it is this kind of view we see in the literature on and around the problem of creeping minimalism.Footnote 3 All this makes this last view worth calling “minimalism,” as I shall continue to do.
We’ve seen how minimalism applies to three kinds of phenomena. The first is truth, a property of some but not all representations, which minimalism says applies to representations in virtue of (i) their meaning and (ii) some other state of affairs. The second is facts and properties, which are things in their own right, and which minimalism says exist and are instantiated in virtue of the same states of affairs that explain when truth applies. The third is what we might call representational phenomena like description, representation, and belief-aptitude, which minimalism says are trivially had by all representations with propositional content. Despite the differences, minimalism has the same two consequences in each case. It tells us which representations have these features, which facts exist, and which properties are instantiated, in terms of other facts. And it undermines certain explanations in terms of these phenomena.
3. The problem
We can now move on to the problem of creeping minimalism. Old versions of expressivism made the following claim:
(N) (a) There are no ethical truths, facts, or properties.
(b) Ethical thought and language is not representational: it does not represent or describe things in the world, or express propositions or beliefs, or aim to state facts or truths. Ethical sentences do not have truth-conditions.
This distinguishes expressivism from its rivals. Realism rejects (Na) and (Nb); error theory and other anti-realist views reject (Nb).
Earlier we saw that minimalism tells us when concepts like truth and representation apply to beliefs and sentences given other facts. The problem is that expressivists nowadays accept both minimalism and these other facts, and so we get this claim:
(¬N) (a) There are (or may be) ethical truths, facts, or properties.
(b) Ethical thought and language is representational: it does represent and describe things in the world, it expresses propositions and beliefs, and aim to state facts or truths. Ethical sentences have truth-conditions.
(¬Na) follows because expressivists are happy to say that things can be good or bad, wrong or right—given minimalism, they must be happy to say that there are ethical truths, facts, properties. (¬Nb) follows because expressivists agree that ethical sentences have propositional content.Footnote 4 So expressivism plus minimalism yields (¬N) rather than (N), but now it’s hard to see how expressivism is distinct from its rivals since (N) marked that distinction.
To appreciate how this is a problem of creeping minimalism, consider how we might react if expressivists didn’t make the wide-ranging claim (Nb), but instead focused on just one concept, say description:
$$ \mathrm{Ethical}\ \mathrm{language}\ \mathrm{is}\ \mathrm{not}\ \mathrm{descriptive}. $$If expressivists accept minimalism about description, they must agree with their rivals that ethical language is descriptive. There’s a simple way to distinguish expressivism again—find another feature that minimalism doesn’t apply to, that expressivists will say does not apply to ethics but its rivals will say does apply. Though expressivists must accept that ethical language and thought is descriptive, they will still be distinctive in denying something else, for instance that it is representational.Footnote 5
The problem of creeping minimalism concerns the fact that any relevant features seem to be good candidates for a minimalist treatment. They are conceptually tied either to truth (in the case of facts and properties) or to propositional content (in the case of representation and belief). This seems true of all of the notions mentioned in (Na) and (Nb). As we saw in the previous section, this means minimalism can easily creep to them, and so we cannot distinguish expressivism using them.
The problem is a combination of two factors, then. Together with prior expressivist commitments, minimalism about any of the concepts mentioned in (N) entails that those concepts apply in the ethical case. And minimalism creeps to all of the concepts mentioned in (N) and seems likely to creep to any notion that might appear in (N) or an extension of it.
4. Explanationism
There are several strategies for solving the problem. We could find some concept minimalism doesn’t creep to which expressivists think does not apply to ethics but which non-expressivists think does apply, or vice versa. The difference between expressivism and its rivals will lie either in what ethical ontology it posits, or what features it ascribes to ethical thought and language. A different strategy, which Dreier (Reference Dreier2018) calls explanationism, focuses on what expressivists use in their explanations of certain facts about ethical thought and language. The facts in question concern the content of ethical thought and language, for instance that the English sentence ‘stealing is wrong’ means that stealing is wrong, or that Ella’s belief that charity is good is a belief that charity is good. The kind of explanation at stake is a constitutive one, which tells us what it is in virtue of which these things have ethical content—what it is to have ethical content, to make an ethical assertion, and to have a belief about ethics rather than something else.Footnote 6
Explanationism, then, is the view that what distinguishes expressivism is how it completes claims like:
(1) The English sentence ‘stealing is wrong’ means stealing is wrong in virtue of …
(2) Ella’s belief has the content stealing is wrong in virtue of …
Before we explore this further, note how explanationism avoids the problem of creeping minimalism. While minimalism forces expressivists to accept that concepts like truth, fact, representation, and belief do apply in the ethical case, it does not force them to give these concepts any role in explanations of the content of ethical language and thought.Footnote 7 So the problem is solved if we can find something that expressivists alone think has no role in explanations of ethical content. I think explanationism therefore solves the problem of creeping minimalism at least by giving a general way to neutralise the main threat minimalism poses.
Which things do non-expressivists think explain ethical content, but expressivists think do not? At this point we can distinguish two versions of explanationism. Ontological explanationism is the view that what distinguishes expressivists is their evasion of ethical ontology—the things mentioned in (Na), such as ethical facts, properties, truths—in their explanations of ethical content. Representational explanationism finds the distinction in expressivists’ evasion of what we might call representational concepts, the things mentioned in (Nb): within its explanations of ethical content, expressivism does not say that ethical thoughts are beliefs, that ethical language describes or represents things, and so on. The two views are related but we’ll keep them distinct for now.Footnote 8
4.a Ontological explanationism
Ontological explanationism has been defended most prominently by Dreier (Reference Dreier2004; Reference Dreier2018), drawing on Fine (Reference Fine2001) and O’Leary-Hawthorne and Price (Reference Price and O’Leary-Hawthorne1996). Dreier claims that realists explain facts like (1) and (2) by citing normative properties, while expressivists do not. For the expressivist, “There is no explanatory weight borne by these normative properties in the account of what we are doing when we use normative concepts” (Dreier, Reference Dreier2018, 533).
Matthew Chrisman (Reference Chrisman2008) raises a difficulty for ontological explanationism: there are theories that are clearly not expressivist but that Dreier’s view would count as expressivist. First, suppose we say that an expressivist explanation of a belief is one that doesn’t appeal to the corresponding fact when explaining its content. Then every false but representational belief must get an “expressivist” explanation since it has no corresponding fact (348). Perhaps we should instead appeal to properties: the belief that the moon is made of cheese is not explained using the fact that the moon is made of cheese but in terms of the moon and the property being made of cheese. This is why the explanation is non-expressivist.
However, this idea falters when we consider beliefs about non-existent entities, which won’t be explained in terms of relations with entities yet can still be representational. Chrisman’s example is the belief that the planet Krypton is made of kryptonite. Here there are no suitable relata, yet the belief is still representational. Beliefs about properties which no real objects have (like being made of kryptonite) can be given a representationalist account. Dreier’s view would require such an account to commit to the existence of properties that are uninstantiated, in this case being made of kryptonite. But not all representationalists would accept this commitment, and so Dreier’s view will (incorrectly) not count them as representationalists.
In an earlier paper (Simpson, Reference Simpson2018, 515–16), I argued that in the ethical case, ontological explanationism incorrectly classifies error theory. Error theorists cannot appeal to ethical truths, facts, or properties since they deny their existence. But then ontological explanationism says error theory is expressivist, and this cannot be right. Dreier replies that it nevertheless captures a distinction between realism and its rivals (2018, 534–35), for error theory is anti-realist in this sense and is rightly classified as such by ontological explanationism.
Whether or not this is true, ontological explanationism still misses the distinction between expressivism and non-expressivism. The primary difference between expressivism and its rivals concerns whether we need to treat ethical language and thought as representational in order to explain it. This is both the difference between expressivism and error theory and the difference between expressivism and realism. Ontological explanationism cannot capture this distinction. The deeper reason for this is that the very idea of representation—of representing things as being a certain way in language or in thought—does not by itself carry a particular ontology (Simpson, Reference Simpson2018, 516–17). Ontology is not the central marker of representationality: philosophers can disagree about the ontology involved in a belief that everyone will agree is representational.
4.b Representational explanationism
Those who find these criticisms plausible may therefore be attracted to representational explanationism, which aims to identify the distinction between expressivism on the one hand and both realism and error theory on the other, a difference over whether ethical language and thought is representational. The view is that this distinction concerns whether explanations of ethical content involve concepts like representation, belief, description, and so on.Footnote 9 Expressivists accept that concepts like these apply to ethical language and thought, but deny that we need them when we say what gives ethical sentences and beliefs their distinctive ethical content. While rivals will explain Ella’s belief that stealing is wrong as having its content in virtue of being a representation of stealing as wrong, or a belief in the proposition <stealing is wrong>, expressivists will not do this. Instead they will appeal to something else, like Ella’s disapproval of stealing. Since, as we’ve seen, concepts like representation and belief do not carry any particular ontology, representational explanationism will not necessarily count error theory as expressivist. Yet it will correctly categorise realist theories too, since they appeal to ethical facts and properties and the representational relations between them and ethical language and thought.
Representational explanationism is appealing because it focuses on the specific element of expressivism that distinguishes it from both realism and error theory. Expressivism does not just differ from realism by avoiding ethical ontology in its explanations of ethical content—error theory is the same in this regard—but by avoiding representational concepts in those explanations too. However, representational explanationism faces its own problems. First, it needs to say what things count as “representational.” Without a principled account of this, it is too vague, and risks classifying theories incorrectly. However, giving such an account is not an easy task. One particular threat, which Dreier discusses (Reference Dreier2018, 539), comes from reductionist accounts of concepts we would naturally count as representational, like a causal tracking theory of reference (see e.g., Boyd Reference Boyd and Sayre-McCord1988). Such views will not use terms like “representation” or “belief” in their accounts, yet nevertheless should clearly count as representationalist. The representational explanationist risks only classifying as representationalist what we might call primitivist views which appeal to concepts like belief and representation but decline to explain them any further; this classification is simply too narrow.
In earlier work, I responded to this problem by arguing that representationalist accounts of ethics explain ethical content either in terms of representation, belief, description, and so on, or things they could plausibly be reduced to (Simpson, Reference Simpson2018, 518). This would capture a causal tracking view of ethical predicates since reference plausibly reduces to causal tracking. However, this faces further problems. First, it does not tell us what should be in our list of the concepts or phenomena whose potential reduction bases we might consider. It does not tell us how to informatively replace the “and so on” in the list. Second, it is not obvious that one has to think that causal tracking is a plausible reduction base for reference in order to take accounts which appeal to it as representationalist. Suppose a Quinean eliminativist about reference was convinced by a causal tracking view of ethics. She would naturally count as a representationalist, and certainly not an expressivist, yet she does not believe in reference nor any reduction of it (indeed she might be an eliminativist partly because she thinks reduction is impossible). So it’s not clear that the notion of reduction does any useful work here. In any case, it only works given a more informative characterisation of the phenomena whose reduction bases we might also include.
The second major problem concerns minimalism itself. We saw how minimalism undermines the explanatory role of everything it applies to. In particular, the minimalism about representational concepts from section 2 implied that those concepts cannot explain the meaning of any language or thought because they are just trivial add-ons that follow from meaning itself, as set out in schemas like (RD) and (B).Footnote 10 If ‘good’ means good then it represents things as good—but it does not mean good because it represents things in this way. However, now we have a simple path from minimalism about representational concepts to a completely general extension of expressivism. If minimalism is true of representational concepts, then they never explain the content of any language or thought, and given the way representational explanationism defines expressivism, that just means that expressivism is true for every kind of language and thought. So the second major consequence of minimalism about representational concepts—undermining explanations of meaning in terms of them—causes trouble for representational explanationism.Footnote 11
Dreier (Reference Dreier2018, 539) offers an ontological version of this problem. He argues that minimalism about representational concepts precludes any robust ontological view in any domain—it demands the same ontological view that expressivists take about ethics. A robust ontological view seems to invoke representational relations and properties in an explanatory context and minimalism blocks that. But surely our solution to the problem of creeping minimalism should not do this—it should not prevent us from being robust realists about, say, mass.
The general worry is that minimalism about representational concepts—which is where the problem of creeping minimalism starts—entails both a global expressivism and a global rejection of robust realism. Suddenly we have waded into very deep waters! While not everyone dislikes this consequence (see e.g., Price Reference Price, Johnson and Smith2015a), it is a prima facie problem for representational explanationism.
The second problem for representational explanationism is closely linked to the first. The argument here is this:
(3) Minimalism about X entails that X never explains the content of any language or thought.
(4) Metaethical expressivists accept minimalism about all representational concepts.
(5) So, metaethical expressivists accept that representational concepts never explain any language or thought.
Then the idea is that to accept that representational concepts never explain any language or thought is to accept global expressivism. I have been assuming premise (3) throughout this paper. So the rest of the argument entirely depends on which concepts are representational; without knowing this, we don’t know whether expressivists must accept minimalism for all of them. So now, representational explanationism depends on what counts as a representational concept. This is just the first problem, concerning open-endedness.
On the basis of the considerations discussed above, I suggest the following desiderata for a solution to the problem of creeping minimalism:
(a) It is as informative and non-arbitrary as possible.
(b) It characterises expressivism, error theory, and realism correctly.
(c) It classifies both error theory and realism as representationalist.
(d) It characterises intuitively representationalist but reductionist views correctly.
(e) It allows non-expressivist views to be compatible with minimalism.
I’m going to argue that there’s a view which meets these conditions and has a further advantage besides.
5. Subject matter
Ontological explanationism distinguished realism by its commitment to the actual existence and explanatory role of properties or entities corresponding to ethical language and thought. The problem was that representationalism in general needn’t involve this commitment. Representational explanationism avoided this problem by focusing on representation itself but faced the problem that the category of representation—or of representational concepts, as I’ve been putting it—is too open-ended. Perhaps ontological explanationism can inform a good way to define that category.
5.a Representation and subject matter
What is representation? One idea is that it is a relation between language and thought on the one hand and things in the world on the other; a belief or assertion involves a relation with the stuff the belief and assertion are about. The belief that the table is brown, for instance, is a representation because it involves a relation between the believer, the table, and perhaps brownness. This conception of representation informed Dreier’s ontological explanationism.
Ultimately, ontological explanationism fails because representation does not require a genuine relation with what is represented: a belief can represent O as being F even if neither O nor F-ness exists.Footnote 12 But we can maintain that representation involves aboutness, yet say that this doesn’t necessarily require a genuine relation with existing things.Footnote 13 The resulting idea is that a representationalist account treats the relevant language and content in terms of what it is about—its distinctive subject matter. While the idea of subject matter includes things like the entities and properties the language or thought is about, it does not stop there. First, it might go beyond just entities and properties. We might want to think about the subject matter of logical constants and yet deny that this concerns special logical entities or properties. Second, and crucially, we must not require that the subject matter exists; for instance, if the subject matter is properties, we must not require that these properties exist at all, whether instantiated or uninstantiated, or considered as universals or something else. Otherwise the view would hardly be different from Dreier’s ontological explanationism. But as long as we don’t require that subject matter exists, we can avoid the problems with ontological explanationism while capturing its insights along with those of representational explanationism, as we’ll see later.
The subject matter view is that expressivism diverges from its rivals by not mentioning the subject matter of ethical language and thought when explaining it. More precisely it does not mention the subject matter on the right-hand side of claims like (1) and (2). It does not mention goodness, values, or duties when saying what it is to have ethical language and thought. Its rivals, however, do mention those things. Here, mentioning subject matter must not be taken to be ontologically committing, at least in the sense that mentioning goodness does not commit us to the existence of goodness (instantiated or uninstantiated) or good things. A non-expressivist (representationalist) account is one that mentions ethical properties (or other ethical entities) in its explanations of ethical content, whether or not it is thereby committed to the existence of such things. This is subject matter explanationism.
Subject matter explanationism can be viewed as its own distinctive view, or we can use it to define the category of representational concepts used by representational explanationism. In that case representational concepts are just those that involve mentioning the subject matter of words and thoughts. While it is helpful to read subject matter explanationism as a version of representational explanationism in order to appreciate its advantages, I will mostly treat it as a distinct view.
The subject matter view will classify the following as representationalist accounts:
(6) ‘Stealing is wrong’ means stealing is wrong because it represents/describes/pictures stealing as wrong.
(7) ‘Stealing is wrong’ means stealing is wrong because it is true iff stealing is wrong.
(8) ‘… is wrong’ means … is wrong because ‘wrong’ refers to/predicates the property wrongness.
(9) ‘… is wrong’ means … is wrong because (the relevant) utterances of ‘wrong’ are (typically) caused by something being wrong.
(10) Ella’s belief has the content stealing is wrong because it involves a concept C whose teleological function is to track wrongness.
On the right of every “because” in this list—in the explanans of each explanatory claim—we find the word ‘wrong.’ These accounts all mention the subject matter of the predicate ‘… is wrong’ (though they are not all committed to the existence of wrongness) so they are all representationalist views. Expressivism on the other hand does not mention subject matter, as we can see in the following (crude) formulations:
(11) ‘Stealing is wrong’ means stealing is wrong because it expresses disapproval of stealing.
(12) Ella’s belief has the content stealing is wrong because it is a plan not to steal.
Wrongness is not mentioned on the right-hand side of these explanations.
The subject matter view captures what was right about representational explanationism. For belief, representation and description all involve mentioning subject matter. Saying that a mental state is a belief that O is F, or a representation or description of O as F, or a belief relation to the proposition <O is F> means mentioning F-ness. So the subject matter view captures everything representational explanationism does, and later we will see how this means both views avoid several objections in similar ways. But there are other things whose description involves mentioning subject matter: invoking a representational concept may be sufficient for mentioning subject matter, but it is not necessary for it. Causal tracking relations do as well, so the subject matter view, like ontological explanationism, classifies these as representationalist. However, tracking relations only automatically count as representational concepts if we have an informative characterisation of those concepts, which is one of the major problems with representational concepts. So the subject matter view correctly classifies views that representational explanationism is not guaranteed to get right.
5.b Classifying theories correctly
The first challenge the subject matter view faces is an analogue of the reductionism challenge facing representational explanationism: What does it say about views that are reductionist not about representational concepts, but the subject matter in question? For instance, what about a view which says that ‘good’ means good because it causally tracks the property of pleasure maximisation? This does not mention goodness on the right-hand side, but is a classic reductive naturalist view which clearly counts as representationalist.Footnote 14
The answer is that reductive views will accept that the thing they mention is the subject matter of the language and thought in question: the pleasure maximisation view will involve the claim that pleasure maximisation is goodness. More broadly, the stuff mentioned on the right-hand side gives us truth-conditions for the beliefs and assertions in question, in terms of pleasure maximisation. To the extent that the truth-conditions for an assertion or a belief mention its subject matter, this view invokes subject matter too.
The subject matter view satisfies the desiderata we’ve set out so far. It correctly classifies expressivism as expressivist, and realism and error theory as representationalist; it counts the latter two as representationalist for the same reasons, so doesn’t miss the general distinction between representationalism and expressivism like ontological explanationism did. It correctly classifies views that are reductionist about representational phenomena as well as those that are reductionist about ethical properties and facts. And unlike the representational explanationist view I defended before, it’s not arbitrary either: it is not a list, but an informative characterisation using the intuitive notion of subject matter.
The subject matter view has further advantages. It correctly characterises non-representationalist views in other areas. Consider the following:
(13) To believe that Xs cause Ys is to expect a Y given an X.
(14) To believe that necessarily, P is to be disposed to use <P> in reasoning from any premise.
(15) To believe that if P then Q is to be disposed to believe that Q on coming to believe that P.
(16) To believe that pr(A) = n is to be disposed to bet on A at certain odds.Footnote 15
These are all non-representationalist views. They are all non-representationalist because there is no mention of the relevant subject matter on the right-hand side of these claims: there’s no mention of causation, necessity, conditional facts, or probability, or anything that can be described in other terms but is intended to provide truth-conditions for the beliefs on the left.
Contrast these views with some rivals:
(17) To believe that Xs cause Ys is to believe that an X raises the chance of a Y.
(18) To believe that necessarily, P is to believe that <P> is true in all possible worlds.
(19) To believe that if P then Q is to believe that not-P or Q.
(20) To believe that pr(A) = n is to believe that in an infinite series of relevant tests, the frequency of As will tend to n. Footnote 16
Why are all these representationalist views? Because their right-hand sides specify a proposition that is meant to have the same truth-conditions as the target, or things that supposedly constitute the subject matter of the left-hand side. These reductive views therefore count as representationalist on the subject matter view, just like their non-reductive counterparts which appeal directly to representational concepts.
So the subject matter view nicely distinguishes not only metaethical expressivism but other kinds of non-representationalist views too. This is a significant advantage for the account.Footnote 17 This view is also superior to the strategy of finding something expressivists will say applies to ethical language and thought but non-expressivists won’t. The likely candidate is something about expressing attitudes, but not all non-representationalist accounts will give the same positive story (for causal and probabilistic language for instance), so this strategy will not yield a generalisable account of non-representationalism.
A further advantage emerges when we consider hybrid expressivist views, such as Michael Ridge’s (Reference Ridge2014) ecumenical expressivism. Ridge’s view is roughly that an ethical judgement consists in a belief and an attitude: someone believes that an object has a property that is the subject of some relevant norm, and they accept that norm. Ridge’s view should count as non-representationalist, but since it appeals to belief—a representational concept—in its explanation of ethical judgements, representational explanationism would count it as representationalist. Subject matter explanationism does not, because the belief Ridge appeals to is not about the subject matter of ethics; it is not about goodness or anything that is meant to constitute it.Footnote 18
Finally, let’s consider non-naturalist views, which posit causally inert normative properties. We cannot distinguish these from expressivism using the notion of tracking, as we did with other cases. However, the subject matter view does not need to appeal to tracking exclusively. There are other ways of mentioning subject matter, and to the extent that non-naturalist views mention goodness or other normative subject matter in their explanations—for instance, by saying that ethical beliefs are constituted by primitive concepts of goodness, or by primitive acquaintance relations with goodness, and so on—subject matter explanationism will rightly count them as representationalist. Crucially, there’s nothing about explanationism that requires representationalism to give a causal role to normative subject matter: it just needs to mention that subject matter. A similar judgement can be made by the representational explanationist too (Simpson, Reference Simpson2018, 524), as well as Dreier’s ontological view which does not require normative properties to play a causal role in the constitutive explanation of normative content. In this regard, subject matter explanationism is on equal footing.Footnote 19
A more challenging case is Ralph Wedgwood’s view (Reference Wedgwood2001) which appeals to conceptual roles to explain the meaning of moral language.Footnote 20 Wedgwood’s main claim about the conceptual role of moral language does not appear to mention moral subject matter, just a link between accepting normative claims and preferring certain courses of action over others (15). The subject matter view might therefore count Wedgwood’s view as non-representationalist, which would jar with Wedgwood’s own view of himself as a non-naturalist realist.
However, I think Wedgwood’s view does count as representationalist—or at least does not obviously count as non-representationalist. Wedgwood’s account uses the notion of a correct preference and the (correct) purpose of practical reasoning to explain why moral terms mean what they do. Both of these are normative notions which play an explanatory role in his account. In this sense, Wedgwood mentions the subject matter of moral terms—for him this involves the correctness of preferences—when explaining why they mean what they do. Expressivists do not mention the correctness of the attitudes they think constitute moral judgements. Moreover, since Wedgwood does not also apply his view to the normative term ‘correct’ here, which he agrees is a normative term, arguably the account he offers for moral terms is not complete, and therefore we cannot tell whether he will appeal to the subject matter of ‘correct’ when explaining it. Wedgwood seems to suggest that he will have to do so, since he recognises that his project involves giving an account of correctness itself (Reference Wedgwood2001, 20).
Perhaps Wedgwood’s account does not need to appeal to the notion of correctness to explain why moral terms mean what they do. If that is so, then Wedgwood’s theory may well count as non-representationalist according to the subject matter view. In such a case, however, I do not think this is a problem. For Wedgwood’s view itself is clearly interestingly different from other realist views, relying as it does on a conceptual role approach which is usually the domain of non-representationalists, including inferentialists (see for instance Chrisman Reference Chrisman2016, Brandom Reference Brandom1994). So while I think the subject matter view will likely count Wedgwood as a representationalist, it is not a major problem if it does not.Footnote 21
6. Global expressivism?
We’ve seen that the subject matter view is informative; correctly characterises expressivism, error theory, and realism; and counts error theory and realism as distinct from expressivism in the same way. The final desideratum is that it avoids collapsing metaethical expressivism into global expressivism. Earlier we saw two versions of this idea. There was the argument that since minimalism blocks representational concepts from ever explaining meaning, to accept minimalism just is to accept global expressivism if the latter is defined according to representational explanationism. There was also Dreier’s worry that minimalism blocks any realist view, for similar reasons, though since realism and representationalism are distinct, we should distinguish these two arguments.
Note that the worry is not that minimalism rules out some representationalist accounts. Not all representationalist accounts will be compatible with minimalism—claims (6) to (8) above are clearly incompatible with minimalism about representational concepts. This is not a problem by itself: we should not expect minimalism to be compatible with every theory of how language and thought gets its meaning.Footnote 22 The problem is whether minimalism blocks every representationalist account because, if it does, metaethical expressivism must collapse into global expressivism.
Let’s start with the first argument:
(3) Minimalism about X entails that X never explains the content of any language or thought.
(4) Metaethical expressivists accept minimalism about all representational concepts.
(5) So metaethical expressivists accept that representational concepts never explain any language or thought.
This argument does not threaten the subject matter view. We can see this point in two different ways.
I mentioned earlier that a representational explanationist could read the subject matter view as a version of her approach: expressivism avoids representational concepts, but representational concepts just are things which invoke subject matter. On this view, premise (4) is false because minimalism does not creep to all properties and relations which invoke subject matter and, hence, does not creep to all representational concepts as currently defined. Earlier we saw how minimalism spreads to any property that is trivially entailed by a representation’s having propositional content: the fact that such a property is so entailed actually constitutes the kind of minimalism I set out in section 2.Footnote 23
However, there’s no reason to think that any property or relation that involves subject matter will be trivially entailed by propositional content.Footnote 24 For instance, consider a causal tracking relation. There’s no reason to believe the claim:
$$ x\kern0.5em \mathrm{tracks}\ F\hbox{-}\mathrm{ness}\ \mathrm{because}\kern0.5em x\kern0.5em \mathrm{means}\kern.5em F $$which would constitute minimalism for tracking along the same lines as minimalism about representation, description, and belief. More broadly, there’s no reason to think that any property or relation whose use mentions subject matter will succumb to minimalism. There’s therefore no reason to think that expressivists must accept minimalism for everything they might want to say doesn’t explain ethical thought and language but does explain other kinds. So premise (4) of the argument fails: metaethical expressivism does not collapse into global expressivism.Footnote 25
The other reading of the subject matter view doesn’t combine it with representational explanationism, but instead just asserts directly that representationalist views are ones that mention the subject matter of the language and thought in question, and expressivism does not. This means that the arguments (3) to (5) have no effect because, on this view, global expressivism is not the view that representational concepts never explain any language or thought! Rather it is the view that subject matter never explains any language or thought. So we can accept (5) without accepting that metaethical expressivism entails global expressivism. Moreover, the view is not vulnerable to an alternative argument couched in terms of subject matter, whereby we assert that expressivists must accept minimalism for all properties and relations that invoke subject matter, for that is false, as we saw with the causal tracking example.
So subject matter explanationism avoids the global expressivism argument by blocking minimalism from creeping over everything we might use to distinguish expressivism from its rivals. Let’s now turn to Dreier’s argument:
Why shouldn’t we be able to combine deflationism about representation with an inflated theory of wrongness itself? Can’t deflationists about truth and representation have robust theories of anything? Of mass, for example? But then in their theories, they will not be able to make their representational relations and properties carry explanatory weight, since the mark of deflated properties is their explanatory inefficacy. And then [representational explanationism] will count these theories as non-representational. So that seems wrong.
(Dreier, Reference Dreier2018, 539)The idea is that being a minimalist (deflationist) about representational concepts seems to undermine any robust realist theory at all because, according to representational explanationism, a robust theory is one that uses such concepts. Yet it seems, in principle, possible for a minimalist about representational concepts to be as robust a realist as is possible about some things.
Dreier asks why we shouldn’t be able to combine minimalism about representation with an inflated theory of wrongness itself; this is presumably the view of someone who likes minimalism about representation, yet thinks that ‘wrong’ needs the kind of treatment that we might want to apply to a term like ‘mass.’ But the subject matter view does not block this, for we’ve seen that minimalism does not creep to everything that involves subject matter. So it is possible to mention the subject matter of a term like ‘mass’ or ‘wrong’ without appealing to the kind of property or relation which minimalism creeps to, i.e., one which is trivially entailed by propositional content. Consequently, being a minimalist about representation and similar notions does not rule out mentioning wrongness or mass in explanations of ‘wrong’ and ‘mass.’ But if the mark of believing in such things in a robust sense is, as Dreier’s ontological explanationism suggests, believing that they have an explanatory role, then it is possible to believe in them in the robust sense while being a minimalist about representation and other representational concepts more generally. Mentioning the subject matter does not mean invoking the representational concepts minimalism covers, because minimalism simply does not creep to all the properties and relations we might appeal to when mentioning subject matter.
There’s a small wrinkle in the argument here. Dreier talks about using ethical facts and properties in explanations of ethical language and thought as the mark of realism. If we accept minimalism about facts and properties—which, recall, takes those things to be trivial consequences of the way the world is—then no facts and properties ever have an explanatory role to play. However, we shouldn’t misunderstand this consequence. It just means that we cannot make an explanatory appeal to facts and properties considered as genuine, distinctive entities in the world. It does not mean that we cannot appeal to what we might ordinarily call facts and properties, considered broadly as ways the world is, and ways things are. A minimalist about facts and properties might still appeal to the facts about water and the properties of water to explain our use of the word ‘water,’ she just would not be appealing to distinctive entities as characterised and debated by metaphysicians in order to do this. Instead she would be appealing to ways water is. The issue of minimalism about facts and properties is a red herring when it comes to the question Dreier raises.Footnote 26
So the subject matter view does well here. It meets the five desiderata I set out earlier: it correctly classifies expressivism, error theory, and realism, and counts realism and error theory together as representationalist. It correctly classifies views which are reductionist about either representational concepts, or the subject matter in question, or both. It is informative and non-arbitrary: we can see this in how it neatly extends to (non-)representationalist views about other kinds of language and thought. And it does not collapse metaethical expressivism (or any other local form of expressivism) into global expressivism or anti-realism.
7. Conclusion
The core of the solution to the problem of creeping minimalism was, in my view, given by Dreier in the same paper in which he introduced the problem (Reference Dreier2004). Minimalism forces expressivists to agree that certain features apply in the case of ethics; Dreier’s focus on explanation makes this consequence irrelevant to the definition and distinctiveness of expressivism. However Dreier’s specific solution is unsatisfactory, as is its main rival, representational explanationism, at least as it has been defended in the recent literature. I have argued that the closely related subject matter explanationism captures the advantages of both accounts while avoiding their drawbacks, and gives us a correct, satisfying, generally applicable account of what it is to hold a non-representationalist view about a certain kind of language and thought, why expressivism counts as such a view, and why it is in that respect genuinely different from its rivals even if expressivists accept minimalism for truth, properties, facts, and representational concepts in general.
Acknowledgements.
Thanks to Jamie Dreier and Tim Crane for helpful discussion, and to an editor and two anonymous reviewers for the Canadian Journal of Philosophy for their constructive and helpful comments.
Matthew Simpson is a teaching fellow in philosophy at University College London. Starting October 2020, he will be a junior research fellow at Robinson College, Cambridge.
