Gibbard on Quasi-realism and Global Expressivism

Abstract

In recent work Allan Gibbard claims to be both a local quasi-realist, in Blackburn’s sense, and a global expressivist. His local quasi-realism rests on an argument that for naturalistic discourse but not ethical discourse, the semantic relation of denotation and the causal relation of tracking can and should be identified; that denoting simply is tracking, for naturalistic vocabulary. I argue that Gibbard’s case for this conclusion is unconvincing, and poorly motivated by his own expressivist standards. I also argue that even if it were successful, it is doubtful whether the resulting position would count as global expressivism, as Gibbard and I both use that term.

1 Expressivism and Quasi-realism, Local and Global

Quasi-realism is a well-known view in meta-ethics. There, as Simon Blackburn says, it is ‘a position holding that an expressivist account of ethics can explain and make legitimate sense of the realist-sounding discourse within which we promote and debate moral views.’ (Blackburn 1994, 127) In one familiar idiom, the quasi-realist seeks to explain why ethical claims ‘look’ representational, if their true function is expressive.

As Blackburn emphasises, quasi-realism is a naturalistic view, in Hume’s sense. If successful, it explains the role of ethical thought and language in the lives of natural creatures like us, without invoking materials such as moral properties, that are themselves difficult to ‘place’ in the kind of world described by science. Blackburn takes quasi-realism to have similar benefits in other areas. He recommends it for causal necessity and probability, for example, arguing that the former case, at least, is another one in which Hume is our guide to a fruitful naturalistic stance (Blackburn 1993). Again, the quasi-realist’s goal is to explain the (apparently) representational character of causal and probabilistic claims, on expressive foundations.

But why stop there? Taking a step further than Blackburn,¹ I have argued that quasi-realism should aspire to be a global programme. That is, I claim, there is no area of ‘realist-sounding discourse’ for which the quasi-realist’s explanatory stance is not the one we need. In the ethical case, for example, the quasi-realist is interested in explaining why ethical claims take declarative form, why we call them true and false, and so on. Early expressivists tended to take for granted that there was some unproblematic answer to the corresponding questions about what were assumed to be non-expressive cases. From an early point in my own work, I was inclined to challenge this assumption. In Facts and the Function of Truth (Price 1988, 2023), for example, I recommended a pragmatic, expressive account of truth in all its uses. I now call this generalised quasi-realism global expressivism. Among other things, I take it to extend both naturalism in the Hume–Blackburn sense–subject naturalism, as I have called it² –and the project of explaining the representational character of language to the vocabulary of natural science itself.

In support of global expressivism, I have argued that Blackburn’s local version of quasi-realism is unstable, a victim of pressures both internal and external (Macarthur and Price 2007; Price 2015). Internally, it is implausible that the quasi-realist’s explanatory programme could be successful locally–successfully explaining the representational face of ethics and modality, for example–without the same techniques being generalisable to other cases, including the language of natural science. It thus renders redundant other approaches to these cases. We’ll meet an example of this argument below, in application to Gibbard’s proposals.

Externally, Blackburn’s Ramseyan semantic minimalism deprives him of the theoretical vocabulary needed to mark a meaningful distinction, or ‘bifurcation’, between genuinely and merely quasi-realist cases. It is implausible to hold both that scientific claims (say) are apt for some non-quasi kind of truth, and that truth itself is to be understood in Ramsey’s minimal terms, or anything like them.³

In recent work (Gibbard 2015), Allan Gibbard defends a position he claims to be intermediate between my global expressivism and Blackburn’s local quasi-realism. Gibbard characterises his own expressivist views since (Gibbard 2003) as quasi-realist, in Blackburn’s sense. And he argues that in at least one good sense of the term, he is a global expressivist. Yet, contrasting his view with mine, he says that he is not a global quasi-realist. On the contrary, he wants to retain a distinction between cases in which quasi-realism is appropriate, such as ethics, and cases in which he thinks that it is not, such as naturalistic discourse. As he says, ‘I stress a sense in which naturalistic thinking is especially representational.’ (2015, 210).

In this piece I explore this difference between my views and Gibbard’s. As Gibbard notes, one possibility is that it is partly terminological. Perhaps we mean different things by expressivism, so that local quasi-realism is compatible with global expressivism in Gibbard’s sense of the term but not mine? In my terms, that would put Gibbard in the same camp as Blackburn, and the question to ask would be whether he manages to block what I regard as the internal and external pressures to globalisation. If Gibbard and I mean the same thing by expressivism, however, then there are additional questions. Is he right that there is a possibility I have missed, of combining local quasi-realism with global expressivism? And if so, is it a plausible position?

I’ll begin below with the terminological issue, and argue that while there are differences of emphasis in how we understand expressivism, that isn’t the main source of our disagreement. But the differences of emphasis help to point us toward the true difference: Gibbard wants to ask, and answer, a theoretical question that I regard as ill-posed–as resting on a kind of category mistake.

To look ahead, the main difference amounts to this. Gibbard and I agree, more or less, that there are two important representational notions in play, notions that following Gibbard I will call denoting and tracking. Briefly, denoting is a generic notion of reference applicable as much in ethical as in naturalistic discourse; tracking is a naturalistic word–world relation, applicable only in naturalistic vocabularies. Gibbard’s view is that for naturalistic discourse, these two notions coincide: denoting just is tracking, in these cases.

I think that Gibbard’s argument for this conclusion is unconvincing, and more importantly, that as expressivists about denoting (as Gibbard also claims to be), we simply shouldn’t be asking what denoting is. For us expressivists, that question involves a methodological category mistake. Our proper questions are about the use and function of the term, not the nature of the thing.

I will thus be criticising Gibbard’s version of local quasi-realism, relying at some points on expressivist commitments that we share. I will touch only briefly on the question whether if Gibbard were right about naturalistic discourse, the position would really be compatible with his professed global expressivism. In §7 I’ll propose a reason for scepticism about this claim, but won’t have space to develop it in detail.

As I’ll explain, Gibbard and I agree on much of the landscape. This is what makes the remaining disagreement so helpful, of course. I hope that this discussion will clarify for others, as it has for me, the core commitments of my own views about naturalism and representation. In particular, I hope that it will explain my reasons for maintaining, following Sellars, that we need a clear distinction between two kinds of representational notions: those that call for expressivist treatment (subject naturalist, but not object naturalist), and those that may be regarded as matters for first-order scientific investigation in their own right.

2 What is Expressivism?

First, then, to what we mean by ‘expressivism’. Explaining his own use of the term, Gibbard notes an important difference between his own views, early and late.

As some writers use the term, being an expressivist regarding ethics involves denying that ethical judgments can be true, and denying that they are beliefs. In my first book, Wise Choices, Apt Feelings (1990), I did issue such denials. Eventually, though, I was convinced—especially by Horwich—that deflationary truth was the only kind of truth I understood, and that I didn’t understand what I was denying when I excluded ethical judgments from being ‘beliefs’. (2015, 211)

Accordingly, Gibbard suggests, early expressivists such as Ayer (1936) should have phrased their view differently. As a deflationist about truth, Ayer should have had no trouble applying ‘That’s true’ to ethical claims.

On a deflationary understanding of truth … ‘It’s true that stealing is wrong’ should mean nothing more than that stealing is wrong—and thus, according to Ayer, should mean the same as ‘Boo for stealing!’ If Ayer opposed stealing, then he should have been willing to say ‘That stealing is wrong is true’. (2015, 211)

This brings Gibbard to what he takes to be the defining characteristic of expressivism.

Ayer explains the meaning of ‘wrong’ by saying what sort of state of mind the word serves to express. ‘Stealing is wrong’ expresses feelings against stealing, and saying how it does so is a way to explain the meaning of the term ‘wrong’. The pattern to note here is that the meaning of a term is elucidated by saying how it figures in expressing states of mind, by explaining the states of mind the term serves to express. (2015, 212; here and below I use underlining to mark my own emphasis in quoted passages)

Gibbard notes that this account requires an important qualification.

[I]f the account is to qualify as expressivistic, it can’t specify the state of mind expressed by citing its content. The state of mind expressed by ‘Stealing is wrong’ can’t be specified, in the explanation, as anything like ‘believing that stealing is wrong’. The idea is to explain the meaning of the word via explaining the states of mind that constitute believing things couched with the term. We can’t, then, legitimately start out relying on an understanding of what it is to believe things so couched. We can’t just say that there’s a general relation of believing a proposition, and that believing that stealing is wrong is standing in this relation to the proposition that stealing is wrong. (2015, 212)

Gibbard says that he now regards himself as a global expressivist in this sense – more on that below. He raises the question whether he means the same by global expressivism as I do. As he puts it, ‘my sense of the term “global expressivist” may not be Price’s.’ (2015, 211) We’ll come to that issue, but first to what I mean by expressivism, global or not.

I now think of expressivism as a recipe with about five ingredients.⁴ One key ingredient is what we can call a use-first approach to meaning. Expressivism focusses on how words are used, rather than what they are about. What does ‘use’ mean, in this context? I have a broader conception of the factors that the relevant accounts of use are allowed to involve than many expressivists, including Gibbard. I think it is unhelpful to restrict them to psychological states, as opposed to more general aspects of speakers’ circumstances.⁵

An attractive alternative model is that of Michael Williams (2010, 2013). Developing ideas he finds in Sellars, Williams calls the central component of his model ‘an explanation of meaning in terms of use (an EMU)’ (2010, 321). Williams’ EMUs involve three kinds of clauses: intra-linguistic inferential (I) clauses, epistemological (E) clauses (roughly, language entry and exit rules for the vocabulary in question), and functional (F) clauses. Williams illustrates the model with several examples, including a deflationary theory of truth of Horwich’s sort, and an account of colour judgements–the latter example will be useful below.

Returning to my expressivist recipe, the second key ingredient is an attitude that presents itself as an alternative to metaphysics, or ontology. It may be motivated in the same way by so-called ‘placement problems’–that is, in their typical form, questions about the place of some seemingly problematic subject matter (e.g., morality, modality, meaning, or the mental), in the kind of world revealed to us by science. But expressivism combines an insistence that these be regarded as primarily linguistic or psychological issues–Why do we talk or think this way?–with a renunciation of the ‘representational’ moves that lead from there back to metaphysics (e.g., that of seeking ‘referents’, or ‘truthmakers’, in some non-deflationary sense).

The third ingredient, closely linked to the first and second, is an explanatory programme. It aims, roughly speaking, to account for the existence and practical relevance of the vocabularies in question; typically the former in terms of the latter, in some way. Why do creatures like us employ these terms and concepts? And why do these terms and concepts exhibit distinctive links to various aspects of our practical lives?⁶

The fourth ingredient, closely linked again to the third and first, rests on an identification of features of speakers–typically features of practical or ‘pragmatic’ significance–that play characteristic roles in expressivist accounts of particular vocabularies. I have called these features the pragmatic grounds of the vocabularies in question (Price 2019, 146).

The fifth ingredient, finally, is a kind of perspectivalism, with the pragmatic grounds of a vocabulary playing the role of the perspective from which the users of that vocabulary speak. I link this ingredient to the Copernican metaphor familiar from Kant, noting how well it characterises the sense in which expressivism provides an alternative to metaphysics. What we took to be in need of metaphysical investigation is instead explained as a perspectival matter, in which features of our own situation carry the main explanatory burden.

With the possible exception of my broader conception of use, I don’t think that any of these ingredients marks any major difference from Gibbard’s understanding of expressivism – most of them could easily be illustrated with selected passages from Gibbard’s own work. As we’ll see, however, I’m inclined to be more stringent in my insistence that expressivism is in a different game from metaphysics, or ontology. The main difference between Gibbard and me will turn out to be that by my lights, he blurs this line, in the case of the representational notions with which we will be concerned.

Coming back now to global expressivism, do Gibbard and I have the same thing in mind? I think we do, at least in all respects that will be relevant here. For Gibbard, as we saw, it is the view that there are no cases in which the illuminating way to specify a use-rule for a sentence is to say that it is used to express a belief with a certain propositional content, thought of as available in advance.⁷ This amounts to endorsing globally what Brandom calls ‘the pragmatist direction of explanation’:

The pragmatist direction of explanation ... seeks to explain how the use of linguistic expressions, or the functional role of intentional states, confers conceptual content on them. (2000, 4)

Starting with an account of what one is doing in making a claim, [the pragmatist direction of explanation] seeks to elaborate from it an account of what is said, the content or proposition—something that can be thought of in terms of truth conditions—to which one commits oneself by making a speech act. (2000, 12)

Elsewhere (Price 2017, 2023) I have characterised this as the view that content is ‘downstream’ rather than ‘upstream’ of use, and noted that it accords very closely with the crucial first ingredient of a global expressivism. In this respect, then, I think that Gibbard, Brandom and I are clearly on the same page.

3 Two Notions of Representation

Now to the distinction marked in Gibbard’s terminology by the contrast between denoting and tracking. I’ll begin by describing how I came to draw what Gibbard agrees to be a similar distinction, and how I see it as relating to global expressivism.

In early work (Price 1988, 1992, 1994) I proposed a two-level picture of assertoric language. At the top level, I suggested, the assertoric language game as a whole has an interesting global function–roughly, that of encouraging useful convergence between the commitments of the members of a speech community.⁸ But this global function sits on top of much functional diversity among various kinds of commitments–the sorts of diversity to which expressivists typically appeal.

In (Price 1994) I characterised this view by proposing that we could usefully distinguish two notions of description.⁹ The first was an inclusive, top-level notion, encompassing the entire assertoric domain, and marked by aptness for deflationary truth. The second was a lower-level functional notion, applicable only to a subclass within the domain.

I shall use the term “minimal description” for any utterance which is capable of being minimally true or false. The suggestion is thus that within the class of minimal descriptions, we may find sub-classes of utterances serving a range of different linguistic functions. ... Let us now suppose that one of the functions served by some minimal descriptions is ... to signal the presence of certain conditions in the physical environment of a speaker ... . [W]e thus have a distinction between the semantic ... notion of minimal description, and the functional notion of natural description (or physical signalling). (Price 1994, 67–8)

I then went on to propose that traditional expressivist claims could be charitably reconstructed within this framework:

My suggestion is … that the [expressivist’s] central thesis may be thought of as the claim that in certain cases we systematically confuse minimal descriptions for natural descriptions. Moral judgements (or whatever) are minimal descriptions, but are not natural descriptions. (1994, 68)

In more recent work (Price 2011a, 2015; Price et al 2013), I have used different terminology for a similar distinction. I have proposed that we distinguish two ‘nodes’ in work on representation in philosophy, cognitive science, and related fields. Each of these nodes is itself a cluster of notions, at least in the sense that the defining features of the node may themselves be developed in a number of different ways. Bracketing the latter kind of diversity, my proposal is that we see the nodes themselves as distinct notions, rather than different aspects of the same single concept of representation.

The first node–e-representation, as I call it–involves the world-tracking conception of representation, associated, in biological cases, with the idea that the function of an evolved representation is to co-vary with some (typically) external environmental condition. (In some cases the relevant piece of the world is within the skin, as for pain or thirst.) There are also familiar non-biological examples, of course: the needle in the fuel gauge and the level of fuel in the tank, the barometer reading and air pressure, and so on. What unites such cases, biological and non-biological, is that some feature of the representing system either does or (in some sense) is intended to vary in parallel with some feature of its environment.

Thus e-representation emphasises the system–world links. The second node–i-representation–gives priority to internal connections between one representation (such as a word) and another. By this criterion, a token counts as a representation in virtue of its position in some sort of cognitive, inferential or functional architecture–its links, within a network, to other items of the same general kind. The notion is flexible enough to accommodate several different kinds of network: causal–functional, inferential, even computational.¹⁰ The important thing is that the notion be divorced from any external notion of representation, thought of as a system–world relation.

These two notions of representation may have quite different uses, for various theoretical purposes. In particular, by the time we get to human language (from simpler forms of biological representation, in either sense), there may be no useful external notion of a semantic kind–no useful general notion of a relationship that words and sentences bear to the external world, that we might identify with truth and reference. This is a conclusion that a semantic minimalist has already come to from another direction, of course. The impression that there are such external relations can be regarded as a kind of trick of language–e.g., a misunderstanding of the nature of the disquotational platitudes. But we can hold this without rejecting the internal notion–without thinking that there is no interesting sense in which mental and linguistic representations can be characterised and identified in terms of their roles in networks of various kinds.

Once again, this e/i distinction allows us to draw a bifurcation in e-representational terms, while preserving homogeneity in i-representational terms. We can say that all declarative utterances are i-representations, but that only a subset of them are also e-representations. This captures something of the intuitions that originally motivated local expressivists such as Blackburn and (early) Gibbard, while remaining compatible with global expressivism, or global pragmatism. In my preferred way of developing things, we again have pragmatism at two levels: at the top level, a story (e.g., Brandom’s story) about assertoric discourse as a whole; and at the lower level, a range of pragmatic stories about the distinctive functions of particular discourses, e-representation being just one pragmatic function among others.¹¹

What is the relation between e-representation and i-representation, in this picture? I have explained this by noting an analogous distinction in the work of Wilfrid Sellars–in particular, his distinction between two notions of truth. On one hand, in Sellars’ view, there is a ‘generic’ notion of truth, defined in terms of what Sellars calls semantic assertibliity (‘S-assertibility’). This is the notion we find in ordinary use, and it is applicable, as Jim O’Shea puts it, to ‘all kinds of propositions, whether they are empirical, mathematical, or moral claims’ (O’Shea 2007, 144). Sellars’ account of it is expressivist and deflationary, in contemporary terms.

On the other hand, there is the notion of truth that Sellars characterises in terms of ‘picturing’, and relates to the representations animals possess of their environments. This notion is much narrower in scope. Sellars takes it to be applicable only to what he calls ‘matter-of-factual’ language. Sellars stresses that the two notions are quite distinct: ‘Picturing is a complex matter-of-factual relation and, as such, belongs in quite a different box from the concepts of denotation and truth.’ (Sellars 1968: 136). He draws similar distinctions with respect to related notions, such as ‘descriptive’, ‘fact’, and ‘proposition’. For these, too, he distinguishes their ordinary use from a narrower theoretical use.

In effect, then, Sellars is dealing with the fact that there is a cluster of notions that are being pulled in two directions, one inclusive and one exclusive. For all of these notions, Sellars ends up saying, there’s a generic notion application to declarative statements of all kinds and a local notion applicable much more narrowly–to the matter-of-factual, as Sellars himself puts it.

My diagnosis of this terminological tension is that all these notions are trying to serve two quite different masters. I think that we get a much clearer view of the landscape by making this explicit, by recognising that we have two quite distinct notions or clusters of notions in play, and by modifying our terminology accordingly. The terms e-representation and i-representation were my attempt to mark this distinction.

What does Sellars mean when he says that the two notions of truth belong in different boxes? He puts the difference by saying that picturing is a natural relation between objects–linguistic items considered as objects, on one side, and objects in our environment, on the other–whereas truth in the generic sense is a ‘pseudo-relation’, to be understood in terms of its inference-supporting role within the language game. I think we can emphasise this distinction even further by noting the different theoretical stance we employ in each case.

In the case of the generic notion, we are interested in a notion we find in use in ordinary language. The natural stance, at least for someone who approaches these topics from the expressivist direction, is the explanatory one. We will be asking, in effect, ‘What are creatures like us doing when they use this notion? Why do they have it in their language in the first place?’.¹²

In the case of picturing, our focus is quite different, as Sellars himself stresses. We are interested not in the use of a folk term, but in a complex, first-order relation that we postulate to be required in an adequate theory of (certain parts of) our language. There is no reason whatsoever to suppose that the notion we find ourselves investigating will be already in play in folk usage. And our theoretical interest is in the relation in the world, not in the use of certain terms in ordinary language.¹³

Deflation of the ‘folk’ semantic notions need not imply that there can be no role for theoretical notions that, at least at some superficial level, seem to be working the same territory, like Sellars’ broad and narrow uses of terms such as ‘descriptive’, ‘fact’, and ‘proposition’. But if we want to avail ourselves of this option then we need to keep a sharp eye on the distinction, and not confuse one for the other.¹⁴

Now to Gibbard. Gibbard responds to my remark that e-representation and i-representation ‘serve two quite different masters’ like this:

I think that there is a paradigm of representation that answers to both masters. The paradigm is keeping track of the layout of discrete objects and their manifest properties. We can contemplate, for example, the checkers arranged on a checkerboard, and when we succeed, our thinking corresponds to their layout in a way that gives the thinking content. An image of representation proceeds from this paradigm. Some of our thinking fits it and some, misleadingly, doesn’t. Good science extends this paradigm, whereas good ethics doesn’t. (2015, 213)

To explain what Gibbard has in mind, let’s turn to his remarks about the limits of semantic minimalism.

4 Gibbard on Co-reference

As I have already noted, Gibbard regards himself as a minimalist about truth and reference, in Horwich’s sense. But he observes that such a view requires an important qualification.

A cogent ‘minimalism’ about truth … can’t be the claim that the deflationary schema says all there is to say about truth. It clearly doesn’t, and that it does isn’t something that a minimalist like Horwich maintains in the end. We need synonymy too, or something to do the job that Horwich envisions for synonymy. I, as a speaker of English, can’t get by deflation alone and the whiteness of snow that Pierre’s ‘La neige est blanche’ is true. I need that his sentence means that snow is white. I need that it is synonymous with my own ‘Snow is white’. (2015, 218)

‘Or actually’, Gibbard goes on to say, ‘I need something different. I need a relation of co-reference.’ (2015, 218).

Pierre’s sentence ‘Je suis français’ is synonymous with my sentence ‘I’m French’, but Pierre’s ‘Je suis français’ is true whereas my ‘I’m French’ is false. What I need to get that Pierre’s sentence is true is, among other things, that his word ‘je’ is co-referential with my word ‘Pierre’. (2015, 218)

Gibbard concludes that ‘[f]or minimal truth, this relation of co-reference is crucial’:

In order to explain reasoning about truth, we must explain co-reference. Truth itself is not a substantial property, according to a coherent minimalism, but the account must help itself to a substantial relation, the relation of co-reference.¹⁵ Without co-reference, there’s no such thing as deflationary truth—except in the special case of one’s own present language. (2015, 218)

So how do we ‘explain co-reference’? This is Gibbard’s proposal, which tells quite different stories for the naturalistic and ethical cases. Gibbard begins with the naturalistic case.

In some cases, explanation [can start] with a substantial relation between a word and what it denotes. I’ll call this relation tracking. I won’t try to explain tracking in general, but suppose that we have such a relation and examine what follows. I’ll assume that some explanations can begin with a relation of tracking, which is word-to-world. Co-reference can then be explained as tracking the same thing. We thus don’t need to explain talk of denotation [i.e., in Gibbard’s terminology, a more general notion, also applicable in non-naturalistic vocabularies] as an upshot of co-reference plus deflation; we begin with a substantial explanation of denotation as tracking. Co-reference, if we need it, is a trivial upshot. (2015, 218)

Gibbard then explains why this won’t work in all cases, by expressivist lights.

In various other cases, though, co-reference is explained in some way other than as tracking the same thing. Pierre’s word ‘bon’ co-refers with my word ‘good’, but if ethical expressivists are right, it would be pseudo-explanatory to say that this is because both words track good. They co-refer in that they figure correspondingly in expressions of favoring. To disagree with claims couched with either term is to disagree, in corresponding ways, with favoring. (I disagree, for example, both with ‘Hitler is good’ and with ‘Hitler est bon’, in that I disagree with favoring Hitler.) No purported denotation figures as playing a genuinely explanatory role; any correct talk of denotation emerges from explanations that don’t purport to cite a substantial word-to-world relation of tracking. (2015, 218–219)

‘So in short’, Gibbard concludes,

we can have correct talk of co-reference that is rooted in recognition of a substantial word-to-world relation, and we can have talk of co-reference that isn’t. In the latter case, talk of denotation emerges, but such talk is not explained by citing a substantial word-to-world relation. (2015, 219)

Gibbard then relates this proposal to my i/e distinction.

In some cases, on this account, we can treat denotation both as an i-representational and as an e-representational notion. Do the relations then coincide? Or are they such different beasts that the same word might e-represent one thing and i-represent not it but something else?¹⁶ We can see that when both are in play, they indeed must coincide. Start with the general schema for denotation that figures in i-representational accounts. Pierre’s expression ‘la bonté’ denotes goodness, in that it is co-referential with my word ‘goodness’, and by deflation my term ‘goodness’ denotes goodness. That’s a case without e-representation, but the same schema applies to Pierre’s word ‘Londres’, which is e-representational: his ‘Londres’ co-refers with my ‘London’, which by deflation denotes London. The schema again is one of co-reference plus deflation. In this case, however, e-representation comes in too: we explain co-representation via a substantial relation of tracking. We start with the finding that both my ‘London’ and his ‘Londres’ track London, and if we apply the general i-representational schema, the upshot is as we would expect, that his ‘Londres’ denotes London. This follows so long as co-reference is explained as tracking the same thing. (2015, 219)

We now have the denoting/tracking distinction on the table. ‘Denoting’ is the term Gibbard is using for the generic, i-representational notion of reference, applicable in all assertoric discourses; ‘tracking’ is his term for the (or perhaps a) e-representational notion that he and I are agreeing that we can expect to find in naturalist cases.¹⁷

As we shall see, Gibbard himself goes on to provide the materials to undermine the claim that co-reference itself needs to be understood as a substantial relation. He outlines an expressivist account of taking to be co-referential, and this is apparently all we need to supplement minimalism in the way he rightly takes to be necessary. But setting that aside for the moment,¹⁸ his own account of co-reference in the ethical case permits an easy generalisation to the naturalistic language case, one that doesn’t identify denoting with tracking.

What Gibbard actually says about ‘goodness’ and ‘bonté’ is that they co-refer because ‘they figure correspondingly in expressions of favoring.’ But it is obvious how he would generalise this to other non-tracking cases–to Humean expressions of expectation, say, in the case of claims of causal necessity. Here favouring is not in the picture, but there is another psychological state, expectation, that plays exactly the same role.

The pattern here is immediate. What matters for two terms to be co-referential is simply that they be governed by the same use-rule.¹⁹ However, if this account of co-reference is sufficient for the ethical goose, it is sufficient for the naturalistic gander. We can say quite generally that co-reference is a matter of being governed by the same use-rule. In whatever sense this fact about the use of two terms is a substantial, explanatory, word–word relation in the ethical case, it is so in all cases.

The dialectic here is that of my internal challenge to Blackburn’s local quasi-realism. Gibbard’s success in finding a substantial relation to explain co-reference in the ethical case simply renders unnecessary his appeal to tracking to do the job in the naturalistic case. Instead, we can say–and it is clearly simpler to say–that the same substantial word–word relation does the job everywhere.

It is easy to confirm that this does the explanatory work Gibbard needs, in the naturalistic case as in others. To return to the question whether Pierre’s word ‘Londres’ denotes London, for example, Gibbard says in the passage above that ‘[t]his follows so long as co-reference is explained as tracking the same thing.’ (2015, 219) That’s true, but it cites more than the explanation actually needs. It also follows, as in the ethical case, simply from the generic fact that Pierre’s term ‘Londres’ and our term ‘London’ have the same use-rule.

It might be objected that the use-rule in the naturalistic case involves tracking, in a way in which that for ‘goodness’ does not. Perhaps this provides a justification for identifying co-reference with co-tracking, in this case? Well, let’s see. Michael Williams’ EMUs (see §2) are helpful at this point. Like Gibbard and me, Williams wants to say that some vocabularies are used to track elements of our environment and some are not. He explains how EMUs can mark this distinction. But he insists, as we’ll see, that this does not mean that in such cases, the semantic notion (denoting) and the causal notion (tracking) amount to the same thing.

Williams’ example of a tracking term is ‘red’ (‘in its reporting use’, as he puts it). He gives us a sketch of an EMU for ‘red’, ‘extracted from Sellars’ analysis’ (Williams 2013, 139). Extracting still further, let’s focus on the elements relevant to our present concern. First, the (E) (‘epistemological’) clause of the EMU includes the following sub-clauses:

(b) To master ‘red’ in its reporting use, the speaker must have a reliable discriminative reporting disposition (RDRD), a disposition, given appropriate motivation and conditions, to report ‘x is red’ only in the presence of a red thing in his field of vision.

(c) For a speaker fulfilling (b), a reporting move of ‘x is red’ is generally free but open to challenge, hence requiring justification, in special circumstances. (Williams 2013, 140)

The (F) (‘function’) clause then glosses the effect of these stipulations as follows:

In a reporting use, tokens of ‘x is red’ express reliable discriminative reactions to an environmental circumstance. (2013, 140)

Turning to the questions that concern us here, Williams then gives this characterisation of the case.

By [sub-clause (b)], observation reports are bound up with reliable discriminative reporting dispositions: this is what allows them to function as language entry transitions, making possible their distinctive role in the regulation of theory. But because of its appeal to such dispositions, the EMU for ‘red’, or any word with a reporting use, involves world–word relations essentially. To be sure, the world–word relations on which entry transitions depend are causal, not semantic. ‘Red’ does not refer to red things by virtue of this causal relation: the causal relation resides in the E-clause. Since the EMU attempts no reduction of reference to causal relatedness, it implies no representationalist backsliding. (2013, 141)

In our present terminology, then, Williams takes the view that denotation and tracking need to be kept apart–that they are, as Sellars insists, different beasts–even when an EMU appeals to a version of the latter. Far from the occurrence of a notion of tracking in the EMU providing a reason to identify denotation and tracking in the case in question, it shows what a mess we would get into by trying to do so. Imagine trying to replace the talk of ‘reliable discriminative reporting dispositions’ with talk of ‘reliably using “red” to denote red things’–we immediately lose what’s distinctive about the use-rule for a term with this reporting use.²⁰

There’s another danger close by, from an expressivist’s point of view. Think of the emotivist’s use-rule for ‘That’s good.’ It has the consequence that utterances of ‘That’s good’ are causally related to the presence of pro-attitudes on the speaker’s part, in normal circumstances. This is a word–world relationship, even though the relevant part of the world in this case is within the speaker’s head–let’s call it a word–head relation. Unless an emotivist can resist the pressure to identify this word–head relation with a semantic relation, she will fall victim to an objection raised long ago by (Jackson and Pettit 1998). Jackson and Pettit note that the emotivist agrees that in normal circumstances ‘X is good’ is a conventional indication that the speaker approves of X. They argue that such an utterance is therefore equivalent to a self-report: ‘I approve of X’. Emotivism thus reduces to subjectivism, a (not very plausible) form of cognitivism.

This objection is not specific to emotivism, of course. A version of it can be raised to any kind of expressivism of Gibbard’s sort, which takes the target utterances to be expressions of a distinctive kind of psychological state. In such an account there is always a word–head relation, at the core of the use-rule. Quite generally, then, expressivists need to resist pressure to identify denotation with this word–head relation. That way lies subjectivism–in fact, global subjectivism, if one claims, like Gibbard, to be a global expressivist.

It is true that as critics of Jackson and Pettit pointed out, it is not just expressivists who would be caught in their net. If it worked, the objection would apply equally to the anodyne view that an ordinary assertion ‘P’ is normally an expression of a belief that P. This, too, would have to be given a subjectivist reading, thus leading, apparently, to an unstoppable regress.²¹

The lesson is that everyone, not just expressivists, needs a distinction between what we say and when we say it–between saying when and saying that, as I put it in early work (Price 1983).²² As noted above, we global expressivists have a view about the proper explanatory relationship between these two things: the latter is always downstream of the former. We explain what is said by saying when it is appropriately said, or what one is doing in saying it–as Brandom puts it, ‘the content by the act, rather than the other way around.’ (2000, 4).

Similarly, we explain saying the same thing in terms of sameness of use, and Gibbard’s general explanation of denotation fits this model: ‘goodness’ co-denotes what any term with the same use-rule denotes. But having shown this necessary discipline with respect to one sort of word–world relation–having kept saying-when clearly distinct from saying-that–why should we lower our standards with respect to another, by giving a different account of co-denoting in the naturalistic case?²³

5 Tracking, Denoting and Intersubstitutability

Have I missed something crucial in Gibbard’s argument? At the end of his paper he summarises like this:

I asked why, if there is such a thing as ‘tracking’, it counts as denoting. My answer was that terms that track the same thing are intersubstitutable. For tracking to be denoting, substitutability must depend on co-tracking. What’s basic to the explanation, then, is substitutability. Substitutability is a normative quality: roughly, what’s substitutable is what it’s all right to substitute. The view, then, explains meanings via the oughts that govern accepting sentences in one’s language. This is a form of normatively couched expressivism. Even if, according to this view, denoting is tracking and thus plays a genuinely explanatory role, to say that tracking is denoting is to say how to use one’s words. Explaining the meanings of words as a matter of how to use those words is expressivism—of the normatively couched variety. (2015, 222–223)

This requires some unpacking. By this point, Gibbard has proposed that what’s central to an account of reference (in the generic, ‘denoting’ sense) is the observation that we use reference claims to licence substitutability. To say that ‘X’ refers to Y is to express the normative view that the term ‘X’ is substitutable for the term ‘Y’ (in appropriate contexts–let’s leave aside the issue as to which those are). Moreover, as he says here, and as the talk of normativity suggests, Gibbard takes this to be an expressivist account of reference.

This means, as Gibbard himself elsewhere emphasises, that the appropriate question is not what it is for two terms to denote the same thing, but what it is for a speaker to take two terms to denote the same thing. What the expressivist seeks is not an account of what denotation or reference is, but of what ascriptions of reference or denotation are–what role they play in the language game. In his book Meaning and Normativity (Gibbard 2012), Gibbard is admirably clear about this point:

What matters in the story is not which terms do designate what, but what terms the hearer implicitly believes designate what. Or if the story is in terms not directly of designating but of codesignating, what matters is not which terms in fact do codesignate, but which ones [a speaker] implicitly believes codesignate. To explain what is going on, then, we won't have to say what it is for terms genuinely to codesignate; we'll have to say what it is to believe that terms codesignate. Note too this contrast: Thoughts about meaning are, according to Horwich, thoughts about what explains the deployment of words apart from communication from standpoint to standpoint. In these vignettes, in contrast, the conversants' thoughts about reference are not directly thoughts about how to explain something. They are, rather, ways of relying on the beliefs of others. A causal explanation of the deployment of words in beliefs will invoke not what does denote what, but what, in effect, the hearer takes to denote what. I won't try to specify here what these ways consist in. Rather, I now turn to trying to develop normative concepts of reference and truth. (2012, 149)

Against this background, let’s return to the passage above. My question for Gibbard is this. What is someone who claims to be an expressivist about denoting doing, in telling us that (in some cases) denoting is tracking? My phrasing here is deliberately ambiguous. I mean the question both in the idiomatic sense of ‘What do you think you are doing?’ that expresses an attitude that the hearer is acting inappropriately, and in the more serious sense: ‘What do you take the point, use, or function of the question to be?’ Putting these meanings together, I am asking what theoretical role there could be for the question ‘What is denoting?’, even in this special naturalistic case; and expressing my scepticism about whether it has any legitimate role, by the expressivist lights that Gibbard and I share.

Gibbard’s answer seems to be that, as he puts it, ‘to say that tracking is denoting is to say how to use one’s words’–i.e., if I read him correctly, that the identity claim is part of the account of the use-rule for terms such as ‘denote’ and ‘co-refer’. However, that suggestion seems to me simply to perpetuate the category confusion. In saying how the term ‘T’ is normally or properly used, we are precisely not in the business of saying what T is.

To say how terms such as ‘denote’ and ‘refer’ are properly used, on Gibbard’s account, we need to put in the foreground that they signal intersubstitutability. What substitutability depends on varies from case to case, depending on the use-rules of the lower-level terms whose substitutability is in question. Terms are intersubstitutable, and hence are properly said to co-denote, when they have the same use-rules.

This means that for anything at all that figures in a use-rule for a term ‘T’, the substitutability of ‘T’ for other terms, and hence its denotation, may be said to depend on that thing, in some sense. For an emotivist, for example, co-denotation of ethical terms thus depends on co-expression of pro-attitudes. For Williams’ Sellarsian expressivist, co-denotation of colour terms in reporting uses depends on some co-relation to reliable discriminative or ‘tracking’ dispositions. But we have found no more reason to say that denoting is tracking in the naturalistic cases, than the emotivist account means that denoting is expressing approval, in the ethical case.

To return to the passage from Gibbard at the beginning of this section, let us imagine replacing ‘tracking’ with ‘expressing’ in the first three sentences.

I asked why, if there is such a thing as ‘tracking’ [‘expressing a psychological state’], it counts as denoting. My answer was that terms that track the same thing [express the same state] are intersubstitutable. For tracking [expressing] to be denoting, substitutability must depend on co-tracking [co-expression]. What’s basic to the explanation, then, is substitutability.

Gibbard doesn’t want to endorse the revised first sentence, obviously–that way lies the Jackson-Pettit objection. Yet the revised second sentence is true, and in the case of the third, we have been given no reason to prefer the original to the revised version. The appeal to substitutability doesn’t distinguish the original version (with ‘tracking’) from the revised version (with ‘expressing’); and yet such an appeal is what Gibbard takes to be basic to his explanation as to why ‘tracking … counts as denoting.’

6 A Prescriptive Interpretation?

Suppose we accept that the identity of denoting and tracking in naturalistic cases doesn’t fall out directly from an expressivist account of ‘denote’, and indeed that expressivists simply aren’t in the business of asking ‘what is’ questions (such as ‘What is denotation?’) in the first place. Does that prevent the issue being raised as in a different key, in two senses: first, as a question not about denotation but about the use of the term ‘denote’; and second, as a prescriptive matter, rather than a straightforwardly descriptive claim about use as we actually find it?

This suggestion is prompted by another possible reading of Gibbard’s remark that ‘to say that tracking is denoting is to say how to use one’s words.’ Taken this way, we would unpack the remark as a recommendation, or prescription, something like this:

(Treat D_nat=T) To say that tracking is denoting is to say how we should use the words ‘denoting’ and ‘tracking’: namely, that we should regard them as inter-substitutable in naturalistic cases.

If implemented, this recommendation would lead us, according to Gibbard’s own expressivist account of denoting, to say that the term ‘denote’ denotes tracking (in suitable cases). Wouldn’t that count as accepting that denoting and tracking ‘coincide’, in these cases?

What can we say for and against this prescription? Let’s distinguish two possibilities. The first is that the recommendation might seem mandated by some discovery about denoting and/or tracking, the second that it be treated purely as a recommendation, perhaps in a spirit of conceptual engineering.

As for discovery, we have seen that such a recommendation doesn’t emerge from an expressivist account of ‘denote’, for two reasons. First, the availability of a general explanation of co-denoting in terms of sameness of use-rule shows that it is unnecessary. Second, the internal details of the expressivist account of tracking terms (e.g., Williams’ Sellarsian account of ‘red’) make it clear that (Treat D_nat = T) is unhelpful–better in that context to distinguish denoting and tracking, than to treat them as co-substitutable.

What other sort of discovery could be relevant? Perhaps a discovery that the term ‘denote’ tracks tracking, in naturalistic cases? We might be doubtful about this suggestion on the grounds that it is unclear whether even the term ‘track’ could track tracking. Could the tracking relation take itself as one of its own relata in this way? (By analogy, think of the question whether the causal relation itself can be a cause or effect.) Leaving that aside, it would seem to be question-begging, in this context, to infer from the supposition that ‘denote’ tracks tracking to the conclusion that ‘denote’ denotes tracking.²⁴

Moreover, I think it is helpful at this point to stop pretending that we have a well-worked-out notion of tracking already in place, a notion that could possibly give us determinate answers to these kinds of questions. We should bear in mind the points about such naturalistic notions of reference that were made by (Stich 1996). Stich was concerned with eliminativism about the notions of folk psychology–belief, desire, and so on. He notes that many philosophers at the time took the eliminativist thesis to be that the terms ‘belief’ and ‘desire’ do not refer. In other words, they took the thesis to be couched in these semantic terms. But if that’s what eliminativism is, Stich argues, then in order to assess it we need a theory of reference–a theory capable of guiding our judgement about whether these terms succeed in referring to anything. Stich argues that this leaves the metaphysics of folk psychology in an unenviable position, hostage to the almost inevitable indeterminacies in a scientific theory of reference. In other words, it means that we can’t decide whether eliminativism is true until we sort out the issue between competing theories of reference–and that’s likely to mean ‘never’, given the nature of scientific theory.²⁵

For Stich there was an alternative. He recommended that in place of the question whether the terms ‘belief’ and ‘desire’ refer to anything, we should simply ask whether there are beliefs and desires. But in the present case, where the (supposed) naturalistic relation of tracking is centre stage, that’s not an option. The advisable course seems to be very cautious about what we do with the notion of tracking. Among other things, that recommends not seeking to identify it with another notion, denoting, that has other allegiances–and especially so in virtue of the apparently self-referential character of any such identification, by its own lights.

In other words, Stich’s argument militates strongly against the recommendation (Treat D_nat = T). It suggests that it is unlikely that there is anything that we could discover about tracking that would support the recommendation, and that it would be little short of reckless as a piece of conceptual engineering. Why should we tie our conceptual hands together in that way, when there is so much uncertainty about where the notion of tracking will find itself, in the light of the science of natural representation?

In my Sellarsian proposal, the e/i distinction separates denoting and its siblings on one side, from tracking and its cousins on the other. One of the great advantages of keeping these notions in different boxes is that it lessens the tendency for the family pathologies of one to seem to infect the other. Does ‘gavagai’ denote rabbits or undetached rabbit parts? Does the frog’s visual system track flies or small black objects? These familiar issues may look like the same kind of concern, but they have very different homes–on the one hand within the i-representationalist business of radical interpretation, on the other in the e-representational business of keeping track of one’s environment. By rejecting the recommendation (Treat D_nat = T), we can avoid confusing one for the other.

7 Conclusion: Quasi-realism and Global Expressivism

Gibbard professes to be a global expressivist. We saw that what he means by the term is much what I mean by it, and what Brandom means by a global insistence on ‘the pragmatist direction of explanation.’ Yet Gibbard also claims to be a local rather than a global quasi-realist. This raises two questions: (i) is the case Gibbard makes for local quasi-realism a good one; and (ii) if so, is the position really compatible with global expressivism?

We have been focussing on question (i), and have considered two possible routes to a defence of local quasi-realism. The first route rested on Gibbard’s claim that the explanation of talk of denoting needs to be fundamentally different in the case of naturalistic and ethical vocabularies. For the former but not the latter, Gibbard argues, such an explanation needs to appeal to tracking, in a way that implies that co-denoting is co-tracking, in those cases.

If successful, this argument would indeed show–contrary to my view, and the way I have interpreted Sellars–that tracking and the semantic notion of denotation belong in the same box, in the naturalistic case. I think Gibbard’s characterisation of the view would be fair, in the sense that it would amount to a local quasi-realism. But, turning briefly to question (ii), would it really be compatible with global expressivism?

Gibbard himself touches on this question only briefly. At the very end of his paper the passage I quoted at the beginning of §5 concludes like this:

Explaining the meanings of words as a matter of how to use those words is expressivism—of the normatively couched variety. (2015, 223)

Gibbard then remarks, as the final sentence of the paper, ‘I am thus both a representationalist for some vocabulary and a global expressivist.’ (2015, 223).

This argument seems to me a little swift. After all, it is uncontroversial that non-expressivists, too, may want to say something ‘of the normatively couched variety’ about how words are properly used. In the simplest case, they may want to say that a descriptive sentence ‘P’ is properly used to express a belief that P. Where they differ from the expressivist, or from the pragmatist in Brandom’s sense, is in thinking of the propositional content P as ‘already available’, upstream of this proposed account of the use of the sentence ‘P’. This means that we don’t qualify as global expressivists simply by virtue of insisting that meaning is always a matter of use, even normatively couched use. It also depends on what theoretical resources we take to be available to, as opposed to explained by, specifications of use.

In the light of this, my concern about Gibbard’s claim to be a global expressivist is the following. Identifying tracking and denotation in the naturalistic case seems likely to open the door to a naturalistic account of content, too, in those cases. Ruth Millikan’s work (e.g., Millikan 1989, 2009) would be an obvious place to start; Millikan herself would take her view to have this implication, presumably. For naturalistic discourse, then, content would turn out to be upstream of use, contrary to the understanding of global expressivism that Gibbard and I share with Brandom.²⁶ So far as I can see, Gibbard says little or nothing to exclude this possibility. Accordingly, I think he has not made a convincing case that his version of local quasi-realism would be compatible with global expressivism.

However–returning now to question (i)–I argued that Gibbard’s argument for identifying tracking and naturalistic denotation does not succeed. I made two main points. First, there is a pattern of explanation of co-denotation that invokes the same explanatory property, sameness of use-rule, in both the naturalistic and ethical cases. This allows us to confine tracking where it belongs–e.g., in Williams’ terms, within the EMU for naturalistic claims. Second, Gibbard’s argument for identifying co-denoting and co-tracking in naturalistic cases is no stronger than a parallel argument, disastrous for an expressivist, for identifying co-denoting and co-expressing in expressive cases.

So much for the first route to a defence of local quasi-realism. The second route relies on the prescription (Treat D_nat = T). In naturalistic cases, (Treat D_nat = T) recommends, ‘denoting’ and ‘tracking’ should be treated as inter-substitutable. In a sense, this gives the local quasi-realist two options. At a meta-level, they can point out that no such recommendation makes sense for non-naturalistic vocabularies, so that there is already a distinction where they want one. Or, taking the recommendation on board, they can simply say something like this:

“I concede your argument concerning the explanation of talk of denoting. That does generalise to the naturalistic case, in the way that you claim. Nevertheless, there’s a bifurcation in the notion of denoting, right where I want it. For science but not for ethics, denoting just is tracking (that is to say, the two terms are inter-substitutable).”

However, we found no grounds for endorsing the recommendation (Treat D_nat = T), and several grounds for rejecting it. The second route seems no more promising than the first. I conclude that Gibbard has not made a good case for local quasi-realism; or, though we have not discussed this in comparable detail, for its compatibility with global expressivism.

To close in an irenic spirit, I want to emphasise again that there is much on which Gibbard and I agree. In particular, I am with him in hoping to account for some of the structural features of i-representation in terms of analogous features of e-representations. Gibbard proposes this for disjunctive reasoning, arguing that we might find what amounts to it, in the behaviour of non-human, non-linguistic creatures. This is certainly an interesting suggestion. I think it has some affinities with my own early proposals (Price 1990) about the source of the kind of incompatibility that needs to underlie an expressivist account of negation. But it doesn’t depend on identification of the e and i levels.

Indeed, I agree with Gibbard even more than he thinks, in some respects. At one point, he offers the following assessment of some similarities and differences between our views.

Price, to be sure, allows something like [my view] when he allows e-representation, and he allows that some i-representations are e-representational as well. But on his picture, the general case is one of i-representation, where genuine explanations cite only features that are internal to thinking and language. My own picture is significantly different: I picture a paradigm of representation where genuine e-representation explains i-representational features like the validity of disjunctive syllogism. The ethical case is non-paradigm: it fits the paradigm in its internal workings, but without e-representation to account for why. We have, then, cases that fit the paradigm, as well as cases with some like internal features that don’t fit it. That is the bifurcation that I claim, and it is part of the global expressivist story. (2015, 216)

It is true, in one sense, that in my view ‘the general case is one of i-representation’. In other words, this is where we should look, in my view, if our interest is in the general features of assertoric discourse, those that are in common between the ethical case, the scientific case, and others. However, it is not true that all the ‘genuine explanation’ I admit happens at this level. On the contrary, I want the lower level part of the explanation, too, the part that distinguishes the various discourses. And here not only is there room, and need, for e-representation, in the lower-level component for naturalistic vocabulary; but I can also allow, as just noted, that it may play a fundamental role elsewhere. Putting it crudely, assertoric language might begin as a tool for coordinating e-representational commitments, and inherit some of its general i-representational features from this role.²⁷ But the investigation of this possibility will go much more smoothly, in my view, if we do not saddle ourselves with the attempt to identify i-representational and e-representational notions, in naturalistic cases.²⁸