The problem of things that don’t exist

Matthew Chrisman raises an obvious worry, one that I am a bit ashamed of having failed to notice (Chrisman Reference Chrisman2008). Chrisman points out that when somebody asserts that the moon is made of cheese, we can be confident that their words and belief are not made what they are by the fact that the moon is made of cheese, since there is no such fact. But we are not thereby anti-realists, or expressivists, or anything but realists about the moon and cheese. To address this rather glaring counterexample, though, an Explanationist need only adjust the theory a bit. The explanation of the naif's belief is a matter of standing in certain relations to the moon, and to cheese; there need be no fact that verifies the belief, of course. But, as Chrisman notes, a second problem is standing in the wings. Suppose, having been disabused of the mistake about the moon, our dupe offers the opinion that Superman was born on Krypton. Now our trick doesn’t help. Krypton plays no role in our explanation of the dupe's belief, for it is handicapped by non-existence. Chrisman takes it to be obvious that we should not conclude that realism about the domain of Superman talk is hopeless.

Well, in fact, I would have said that it is an open question which side of the divide is home to Superman talk. We are error theorists about such talk. And typically philosophers have placed error theory on the anti-realist side of the divide. I am personally ambivalent (Dreier Reference Dreier, Joyce, Kirchin and Hetherington2010). Let me table this question for a moment to discuss why.

In unpublished work, Matthew Simpson argues that the right way to think of the distinction that we seem to have lost to deflationary forces is not the distinction between realism and anti-realism, but the distinction between Representationalism and Non-Representationalism. His case in point is the error theory, which he takes to be paradigmatically anti-realist and Representationalist. For the same reason, we might say that an atheist is an anti-realist about the domain of the divine but also a Representationalist, since she thinks of the language of divinity to be an attempt to represent real properties, only a failed attempt since she thinks there is nothing divine. John Mackie insisted that on matters of conceptual analysis his view matched Moore's or Plato’s, that in saying that there are no objective values he was denying exactly what Moore and Plato were asserting, that his error theory was a theory of ontology and not meaning. If we agree with Simpson that the lost distinction is the boundary between the Representational and the Non-Representational, then it is easy to see Chrisman's point. Even if we are happy to let the Explanation criterion classify Mackie as an anti-realist, we should want him to count as a Representationalist. But then, what is that supposed to amount to? It's no good saying Representationalism understands normative talk as representing the world. ‘Representing’ is too easily deflated!

The point is not that Explanationism has failed to map the true distinction it was after. The real point is that there was more than one distinction, and Explanationism has (so far) articulated only one. So one theoretically important distinction between non-naturalist realism and Quasi-realism is, indeed, the one Kit Fine explicated in ‘The Question of Realism’; but there is another we haven’t yet managed to analyze. I now turn to Chrisman's own solution to the problem. While it has a philosophically satisfying feel to it, Chrisman's view faces a problem, spelled out by (Tiefensee Reference Tiefensee2016). Addressing that problem is complicated, and it will take us through waters mapped out by Huw Price and Michael Williams, and again some obstacles discovered by Tiefensee.

Chrisman's inferentialism

Chrisman suggests an Inferentialist approach to understanding how language works, in the tradition of Brandom and Sellars, and recently of Horwich, Price and Michael Williams. As Chrisman puts it, Inferentialism ‘seeks to explain meaning (even of uncontroversially descriptive statements) in terms of inference rather than representation’ [‘Saving the Debate’, 349]. Here an inference is a transition from sentence to sentence in a public language, rather than a bit of reasoning that takes place in someone's head. While Chrisman stresses this difference by way of contrasting Inferentialism with, e.g. Allan Gibbard's way of understanding the meaning of normative language, I suspect it turns out not to be a very important contrast for current purposes. In any case, the main point is that an Inferentialist has, according to Chrisman, a particularly nice way of characterizing the difference between the way a Quasi-realist expressivist thinks language works and the way a realist thinks it works. For the two will have quite different stories about the position that a normative sentence typically occupies within the web of inference.

Chrisman suggests that a Quasi-realist expressivist should be contrasted with realists by their insistence that normative language has a special role in inference. ‘Ought’ sentences, a Quasi-realist will say, and other pieces of normative language, are inferred from others with the support of practical reasons, and they license inferences to actions. Footnote ³ By contrast, theoretical sentences in descriptive language are inferred with the support of theoretical reasons, of evidence, and they license inferences to beliefs and theoretical knowledge about the world. Whereas, one must suppose, a realist will draw no such contrast. According to realism, normative language seeks to describe a special section of reality, apart from the natural world, Footnote ⁴ and so it must be characterized by the same patterns of inference as other descriptive language.

Chrisman, then, offers a method of distinguishing the Representationalism of realist theories from the Non-Representationalism of Quasi-realist expressivist theories from within the theoretic framework of Inferentialism. Instead of Explanationalism he rests his taxonomy on the distinction between practical and theoretical inference.

Now, there are some technical issues that will have to be dealt with rigorously before this distinction can bear weight. One of these issues is the sorting of sentences into the kind that occupy the practical inference position and those that fit into the theoretical inference slots. To illustrate the complication I have in mind, consider an example of Arthur Prior's Reference Prior1960: ‘Whatever all church officers ought to do, undertakers ought to do.’ This statement appears to be fraught with ought, it is paradigmatically normative language, and thus might be expected to occupy one of the practical reasoning slots in our inferential web if Chrisman's expressivist is to prevail. But it follows deductively, as Prior notes, from ‘Undertakers are church officers’. Perhaps it can be marked off by its downstream inferential consequences as practical? But any such logical consequences it has must a fortiori be shared by everything to the left of its turnstile, including, of course, ‘Undertakers are church officers’, which is not supposed to have ‘immediate’ practical consequences. Is expressivism thus refuted? Surely it can’t be so easy!

When I call these technical problems I mean that they call for technical fixes, and I will assume some such fix is available. A promising strategy, to my mind, would be to start with (Gibbard Reference Gibbard2003)'s fact-plan world sets, which are designed to capture the mix of descriptive and (as Gibbard puts it) plan-laden content that is ubiquitous in real life normative talk, with apparatus for extracting the distinctively practical inferential capacity of sentences from the mixture when the theory calls for it. Although Chrisman sees the semantic formalities in Gibbard's work as straying too close to a truth-conditional semantics and too far from the Inferentialist program, in fact I think the fact-plan world sets can be thought of as serving precisely the function of coding inference information, so that the semantic values of sentences turn out to be, seen from the right perspective, nothing more than inferential roles. They are like Library of Congress numbers, but more systematic, indexing positions in an algebra, with the individual fact-plan worlds as atoms. In any case, I will assume, without undue optimism I think, that the technicalities can be worked out.

There is, though, a deeper problem, one explicated by (Tiefensee Reference Tiefensee2016). The problem is that upon closer examination, it looks like Chrisman has articulated the wrong divide. It's a perfectly good divide, and of certain theoretical interest, but it is not the one that separates Quasi-realism from non-naturalist realism. For take a typical explication of the practical role of an ought judgment. We can use Chrisman's own example: ‘We ought to put the quiche in the oven now’. In his story, a couple is reasoning about their dinner party. From facts about when the guests will arrive, they reach the conclusion about the quiche, employing obviously practical reasons, and then having reached the conclusion they go ahead and put it in the oven, a step that is on its face licensed by their acceptance of the normative sentence. But that's just the problem: it is just obvious that ordinary normative judgments come with practical licenses and are licensed by practical reasons. It is not a distinctive component of expressivism, but a commitment common to all sorts of metaethical views.Footnote ⁵ For example, there is no doubt that Scanlon accepts this practical role for judgments about what reasons people have.Footnote ⁶ Indeed, it would be a grievous weakness in a metaethical theory if it could not make out a distinctive practical role for normative judgment, in both thought and language. It's true that the tradition of expressivism has taken this role to be utterly central, and built its theory of meaning around it, while realists merely attempt to accommodate; but many agree that accommodation is necessary. Again, if Chrisman's point is that the expressivist approach is very much more promising than any realist theory (especially non-naturalist realism), I couldn’t agree more. But that is to point to an attractive feature of expressivism rather than to define it.

Eleatic explanationism

The Eleatic view of ontology (dubbed by Price Reference Price, Blackburn and Brandom2013) says that our official ontology admits only those objects (including properties, facts, ‘objects’ broadly speaking) that pull their weight in causal explanations. It is a version of Explanationism, since causal explanation is a kind of explanation. In the present context it is particularly relevant whether our explanation of how words come to denote their contents is causal explanation: when it is, we’ll have a tidy test for whether the vocabulary deserves a realist construal. And if we wed Inferentialism to Eleaticism, we have another scheme of classification.

To fix ideas, let's take Michael Williams's clear explication (Williams Reference Williams2013). In Williams's Inferentialist account of meaning, the meaning of an expression is given by what he calls ‘Explanations of Meaning in terms of Use’, or EMUs. Each EMU has three components: a summary of the inferential connections exhibited by sentences in which the expression occurs; a specification of the epistemic conditions in which the expression may be competently applied; and a story about the function of the expression, what it is used to do. When Williams outlines an expressivist EMU for the term ‘good’ and a realist EMU for the same term, he marks out the difference in two ways. First, perhaps most obviously, he builds into the expressivist EMU an action-guiding function; in the realist one he puts a ‘tracking’ function:

Similarly, in the realist EMU's epistemic component Williams includes this:

These EMU elements give ‘good’ a role structurally similar (though of course with entirely different inferential connections) to that of ‘platinum’ or ‘crimson’. The point of the expression is to reliably indicate the presence of a certain property, and competence with the expression requires reliability in the user in registering the presence of the property. Here registration, keeping track, indicating are all naturalistic, causal relations: they are like the relations of a thermometer to the temperature. But the expressivist EMUs for ‘good’ include no such causal elements. So, in Williams's story, according to the Eleatic Explanationist, the expressivist doesn’t get goodness in his official ontology, while the realist does.

This is a nice distinction. It captures a theoretically important difference between the way expressivists understand the meaning of normative expressions and the way most of us think of naturalistic vocabulary as getting its content. But (as Tiefensee explains), it cannot capture the distinction between realism and Quasi-realism. For many normative realists, and indeed the ones we’ve been most concerned to distinguish from Quasi-realists, are nonnaturalists, and thus do not take our normative thought to be caused by normative properties; they may not think of normative properties as having any causal powers whatsoever.Footnote ⁷ It is unsurprising that a distinction drawn by Williams to capture the difference between what normative expressions do and what natural kind terms and naturalistic property-denoting predicates do should leave a nonnaturalist realist view of normative expressions on the wrong side of the divide. Useful as this Eleatic Explanationism is, it won’t play the role we are looking to fill.

Simpson's explanationism

Matthew Simpson (Reference Simpson2017) offers a version of Explanationism designed to shore up the problems in the kind I have defended. Simpson's innovation is to focus on the explanatory weight borne in a theory of meaning by what he calls representational properties and relations. In this category he includes the usual deflationary suspects: representation, reference, truth, expressing belief, being about the world & Simpson knows well that these properties and relations are ripe for deflation, and that in their deflationary guises they are perfectly acceptable to Quasi-realists: nobody will deny that ‘Slavery is unjust’ is true, or that it represents slavery as unjust, and so on. The point is that these relations will not show up in a Quasi-realist theory in any explanatory role. Quasi-realists do not explain meanings of predicates by trying to say which properties they represent, or in a Gricean mode by saying (merely) which beliefs they express. Their explanations are in quite other terms, as we know. Whereas a realist account will, Simpson argues, inevitably rest its explanation of meanings on these representational features. So, Simpson's idea is thoroughly Explanationist, in that it finds a distinction by the mark of what bears explanatory weight; but instead of the target properties and facts as explanans, Simpson's version looks to the very representational properties that deflationists are fondest of deflating to classify theories.

This is a lovely move. On the old view, what it is to be an ‘inflated’ or ‘robust’ property or fact is to be load-bearing in a certain kind of explanation, an explanation of representation. Deflated properties can’t carry this weight. Simpson's insight is to distinguish representation itself into inflated and deflated versions according to the load they bear or shirk in a theory. I have two worries about this strategy.

The first is that in abstract terms it looks to be a way of distinguishing deflationist and inflationist conceptions of the semantic notions in question, and not robust vs. quasi-version conceptions of, say, moral facts. 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 Simpson's criterion will count these theories as non-representational. So that seems wrong.

The second problem is that it might be quite difficult to tell whether a given theory satisfies Simpson's test for being Representational. For a theory may appeal to certain relations between words and the world, say causal relations (as in Richard Boyd's theory of natural kind terms), and state its account of meanings in those terms. The use of the word ‘platinum’ in our linguistic community is causally regulated by the metal with atomic number 78, and in virtue of that causal relation, a naturalistic scientific realist might say, the word has the meaning it does. Now, has our theorist relied on a Representational relation? He hasn’t named his relation in Representational language, but surely the idea is that the causal regulation in question is what constitutes reference (for natural kind terms). This case is a relatively easy one: the spirit of Simpson's classification scheme calls for sorting a Boydian view into the Representationalist bin, and we can just say, ‘Well, Boyd is talking about reference, only without using the vocabulary’. But there are sure to be trickier cases.Footnote ⁸

I have now reached the end of my supply of positive suggestions for dividing Quasi-realism from non-naturalist realism. Let me pause to take stock before moving on to some Big Picture issues. My Explanationist way of distinguishing realism from Quasi-realism was by asking whether the properties and facts in the target domain explain what it is to talk and think about them. But Chrisman points out that this distinction has the error theory on the Quasi-realist side, so it does not manage to say what is distinctive about Representationalist theories.

Matthew Simpson's alternative Explanationism asks whether the representational properties and relations explain our talk and thought about the domain. Both of these versions have some fuzzy regions in their boundaries, especially in the neighborhood of error theories, rather than drawing bright lines everywhere. And Chrisman's own distinction, drawn from the perspective of Inferentialism, classifies many self-styled realists with expressivists on the grounds that they recognize the intrinsically practical role of normative concepts. Finally, Eleatic Explanationism seems to pick out something important about what we might call naturalistic realist vocabulary, but like Chrisman's distinction it leaves nonnaturalist realism in the camp with expressivism.

The failure to find any single boundary criterion that classifies all the theories we know into the intuitively right categories should not, I think, worry us too much. Let us embrace the pragmatic spirit: there are many lines to draw, and each of the ones we’ve looked at does appear to capture a theoretically important distinction. And if I’m drawing my bullseyes around the holes my bullets pierced in the side of the barn, like the farmer in the joke, that's not cheating: articulating some theoretical distinctions can help us see post hoc what was important about the groupings we were taking for granted.

The problem of the accommodating expressivist

Camil Golub (Reference Golub2017) thinks the effort of redrawing the old distinction has amounted to wasted ink. The paper takes apart a number of attempts to partition the realist from the Quasi-realist accounts, using different useful tools to do the dismantling. Golub particularly wants us to remember that certain features of theories like Gibbard's and Blackburn's are hallmarks not of expressivism but of deflationism itself (which Golub calls ‘minimalism’); he is clearly quite right about this and it bears remembering. We can’t distinguish Quasi-realism from realism by noting that the former and not the latter take the list of instances of schema T to be trivial tautologies; even if that is an accurate description of all expressivist theories and some realist ones, it only shows that expressivists have tended to embrace deflationism about truth more than realists have. And like me, he declines to assume that there is a single watershed awaiting our discovery. On both these issues, Golub's reminders are salutary.

When he discusses the Explanationist criterion for drawing a line, though, Golub argues that no Quasi-realist will want to deny what according to Explanationism is the defining claim of realist Representationalism. I will call this, the Problem of the Accommodating Expressivist. I want to move slowly and carefully through this part of his argument, since I am not sure I am understanding it correctly. To start with, here is what Golub thinks an Explanationist will take to be the defining claim of Representationalism:

Whether Realism-Con accurately captures Explanationism's criterion depends on what's meant by an ‘account of the semantic content’. For example, in recent years there has been an explosion of interesting work in the formal semantics of deontic modals, written both by semanticists in linguistics departments as by philosophers of language. As far as I know, not a single theory in this recent corpus mentions wrongness, a primitive ought relation, or a property of permissibility (They tend to be about ordering sources, accessibility relations, restrictors of modal quantification, and functions from contexts to sets of possible worlds). I assume Golub does not think of this work in formal semantics as giving any ‘account of the semantic content of normative discourse’; a set of possible worlds would not count as ‘semantic content’ in the relevant sense. So what does Golub mean?

The Explanationists Golub cites are Fine, Hawthorne, Price, me, and more tentatively Gibbard and Blackburn. So, an account of the semantic content of normative discourse ought to mean: a story about which facts are the ones in virtue of which a sentence counts as meaning what ‘Abortion is morally permissible’ or ‘We ought to put the quiche in the oven’ in fact mean in our language. Quasi-realists say that these sentences express certain attitudes (or, to bring Chrisman aboard, they are attached in certain ways to other sentences and to actions and evidence in a network of inference). But Golub seems to disagree:

It is not easy to understand what Golub is getting at. Representationalist talk is internal to normative theorizing in this way: that when someone asks how Shakespeare represents Iago in Othello, we may say straightforwardly that Iago is represented as thoroughly evil. In so doing we are, as Simon Blackburn says, moralizing, since we are committing to a moral view according to which the deception and conniving of the character is bad; a Machiavellian moralist might say instead that Iago is represented as powerful and realistic. And there are all the familiar points to make about the truth of Julia Child's instructions about how long we ought to keep the quiche in the oven, and so on. But none of this seems to be on point. How can it follow, from these familiar thoughts about the deflated conception of truth and representation in ethics, that there is no tension between expressivism and Realism-Con? For expressivists say that normative facts and properties play no substantive role in explaining what it is to mean that abortion is permissible.

I think Golub wants to say that there is a tautological conception of meaning, according to minimalism, the kind given maybe by the list of schema T instances, which can be used by expressivists and realists alike to explain the semantic content of normative sentences. And he thinks this is a substantive explanation. So, he concludes, expressivists can agree with realists that normative properties play a substantive role. There they are, on the right-hand side of the bi conditionals, and the left side is about the truth of the sentences, so the normative properties are entering into the truth-conditions. Golub thinks that no expressivist need deny this, so long as she is a deflationist.

I think this is badly wrong. So although it will be a detour from the main track of this paper, I will explain why it is wrong.Footnote ⁹ The explanation may help some understand what is going on in some contemporary debate over how to give the meanings of sentences.

In his landmark of deflationism, Truth, (Horwich Reference Horwich1998), Paul Horwich offers a theory of truth in the form of the instances, restricted somehow to avoid the contradictions generated by self-reference, of the schema: The proposition that p is true iff p. A variant of Horwich's theory could be given for a truth predicate whose primary bearers are sentences rather than propositions: ‘S’ is true iff S (This is again a schema, generating instances by replacing ‘S’ with a sentence). From this collection, Horwich claims, everything one needs to know about the meaning of the truth predicate, everything there is to know about how to use it, can be deduced. Someone who understood our language except for the word ‘true’ and its cognates (and synonyms) could become fully competent with the predicate by attending to these axioms, taking them to be ‘free’, meaning-giving postulates. Now, there is also a tradition, exemplified by Davidson, of understanding what it is to know a language as knowing all of the truth conditions of the sentences of that language; more realistically, knowing how, by means of one's knowledge of words and syntax, to calculate truth conditions for any given sentence. So if one knew, for each sentence, what the world had to be like in order for that sentence to be true, one would know what all the sentences mean. And we can typically (after accounting for indexicals and other expressions whose content varies with context) write out the truth conditions for a sentence just by writing the sentence: the truth conditions for ‘Everest is the tallest mountain’ are that Everest is the tallest mountain. So it is perhaps tempting to think that the same list that the variant of Horwich's theory uses to define truth (of sentences) can be used also to give the meaning of the sentences of the language in which the theory is written down.

But even if the collection of bi conditionals can be used in either way, it definitely cannot be used for both purposes together. Merging the two theories is absurd. For in Horwich's theory, each bi conditional is a tautology, serving to specify a bit of the meaning of ‘true’. And on the other hand, in the Davidsonian theory the bi conditionals together capture what the sentences of the language mean, which is an empirical fact. You cannot learn German by writing down each sentence of German in quotation marks, followed by ‘ist wahr’, followed by ‘↔’, followed by the sentence again, and then studying that list every night. You would have a list of sentences of German, but you would not know what any of the sentences means. And if you could learn what each of those sentences means, then of course you would not need the list!Footnote ¹⁰

No doubt all of this is obvious, but I wanted to show as pedantically as necessary that the way in which truth conditions give semantic content of a sentence is not at home in the deflationary conception of truth (for sentences). Maybe an easier way of clarifying is to suppose that all parties agree that truth is explained in a satisfactory deflationary way by the schema for propositions, and then to ask what makes a given sentence express a given proposition. What has to be true of a linguistic community, we may wonder, in order for a string of words in their language to mean that abortion is permissible? We know what sort of answer expressivists give: the community has to use the string of words to express a certain type of attitude toward abortion. Realists, on the other hand, even if they agree that communities that use ‘permissible’ in a way that does in fact express some attitude, will answer differently the question of what makes the string of words mean that abortion is permissible. They (I hypothesize) will mention a non-natural property in giving their answer.

As I understand him, Golub thinks this part of the theory is not semantics, but meta-semantics.Footnote ¹¹ Against Explanationism applied to metasemantics, though, he complains that the version of deflationism we are taking for granted in this discussion ‘entails that many semantic facts need no deeper explanation: for example, the fact that “Genocide is wrong” is true just in case genocide is wrong. On their view, such tautologies simply follow from the grammar of the word “true”.’ But that is just the mistake I warned against. The semantic fact that ‘Abortion is permissible’ means what it does needs some explanation, indeed, one which expressivists give in a distinctive way. Think of all the trained EMUs Michael Williams needs to explain what it is for a sentence to mean that (even just for the simplified, toy version of the language!).

As I said, I am not sure I have understood Golub's point, but as things stand I believe the Problem of the Accommodating Expressivist is not a real problem. Serious expressivists should indeed deny that the normative subject matter itself bears any weight in semantic or metasemantic explanation.

The problem of conceptual role semantics

I now turn to the final problem. And it is to my mind the most worrisome; I honestly don’t know what to think about it. The problem shows up when we think about a way of specifying meanings that is sometimes called ‘Conceptual Role Semantics’ (Harman Reference Harman and Harman1987). It is similar in most ways to Chrisman's Inferentialism, but is more naturalistic and psychologistic. Meanings are assigned first to speakers’ assertions by way of the meanings of their psychological states, and only then to sentences in a language, and the ‘roles’ in question are causal, psychological roles rather than normative inferential ones. Suppose we wondered whether someone in an alien linguistic community meant ‘platinum’ by a certain word. We would want to see how the sentences in which the word occurred were linked to other words and sentences, a word, perhaps, for ‘shiny’, or ‘metal’, or ‘electricity’. Probably we would try to identify a network of inferences the speaker was inclined to follow, and then assign a bunch of meanings all at once, by Ramsification (Lewis Reference Lewis1970) or the like (Jackson Reference Jackson1998). We would also have to check to see whether the speaker or her community were acquainted with platinum, whether they had any causal contact with it, whether they could collectively recognize it. We might construct an EMU for our word ‘platinum’ and see whether any word in the speaker's language followed its pattern of use. Now at some point we would decide that yes, indeed, the speaker had a word we could translate as ‘platinum’, and then we would count her as talking about platinum. The pattern of use, the conceptual role, we would think, qualified the word as being about some stuff in the world around us. There would be no further question of whether, having filled the conceptual role, the word really is about that stuff: reference happens, as it were, automatically.

Now imagine two philosophers who start off with a straightforward inferentialist story about normative expressions. One approaches metaethics in the manner of (Gibbard Reference Gibbard2003), the other along the lines of (Wedgwood Reference Wedgwood2007), but their starting strategy is the same. They work out the same pattern of inference, including what inferentialists call ‘output-transitions’ to action or intention or motivation. When an expression fits one of the patterns, one of the EMUs maybe, it will count as meaning what ‘ought’ means, they agree.

But when they’re done with that part, the Gibbard-like theorist says, ‘That's the whole story, except we can add some deflationary talk about properties and truth and the like.’ Whereas the Wedgwoodish theorist says, ‘So, when people speak in this way, they manage to latch on to the real, non-natural normative properties and they will count as thinking and talking about those properties.’Footnote ¹² It's not that my Wedgwoodish character takes the pattern of use to be excellent evidence of some other connection people must be making to the non-natural properties. The extra bit of the theory, he thinks, comes for free, emergent. We achieve reference to the non-natural properties precisely by having expressions in our language that fit the patterns. It works analogously to the way we manage to refer to things by descriptions: we just have the description in our minds, and then it refers to whatever it is in the world that satisfies the description, automatically.

These two philosophers certainly seem to have different ontologies and different accounts of normative language, the one anti-realist and the other realist, but their explanations of what it is in virtue of which our language and thought counts as normative language and thought are the same. The extra bit the Wedgwood character tacks on is not explaining, at least not explaining the things that are supposed to matter in my taxonomy. So I am worried that I classify them together when they should be separated.

I find this rather mysterious. Maybe it's just an illusion, and there really isn’t any difference – the bit that gets added on by the ‘realist’ story is just a kind of flourish. But it doesn’t feel that way to me, so I still want to figure out what to say about the difference.

Recap

In recent literature on the problem of drawing the boundary between realism and Quasi-realism, between Representationalism and Non-representationalism, some philosophers raise objections to the criteria deployed by others. But we should be accommodating: there are many lines to draw, and many distinctions of theoretical interest, so the criteria do not conflict with one another.

Some philosophers have raised doubts about the possibility of distinguishing realist domains from Quasi-realist ones, at least under broadly deflationist presuppositions; but so far no good reason has been offered for why Explanationist criteria couldn’t draw useful distinctions. One problem still worries me: whether having offered a pragmatist or expressivist style account of how, for example, the normative expressions in a language come to have the meanings they have, we could still wonder whether expressions that have those meanings also denote properties; that is, whether the account might still be thought incomplete by those with more realistic bent. The problem, as I see it, is not so much how we would tell, but rather whether there is anything at all to wonder about.