1. Reference and practice

One of the most salient features of the later Wittgenstein's conception of language is its emphasis on the diversity of functions linguistic expressions can discharge. They are like the tools in a toolbox: ‘The functions of words are as diverse as the functions of these objects’.Footnote ¹ Or like the handles in the cabin of a locomotive: they look more or less alike, but they work in radically different ways.Footnote ² This is one respect in which he saw his position as departing from his earlier approach, which had concentrated exclusively on the representational function of language – on how sentences can represent things as being a certain way:

The later Wittgenstein's position on this question is incontestable – we clearly do lots of different things with linguistic expressions. Representing things as being a certain way is only one of them.Footnote ⁴ If anyone ever thought that this is all we do with language, they were clearly wrong. The rest of us realise that some linguistic episodes represent the world and others do other things – or nothing at all. Sometimes instances of linguistic representation can be distinguished from other linguistic activities by grammar. Imperatives and interrogatives, for example, don't tend to have a representational function. Linguistic representation is usually effected by declarative sentences. However not all declarative sentences succeed in representing the world, or even try to do so. How do we distinguish the representational from the non-representational in declarative discourse?

It will simplify matters to focus on predicates, and the contrast between those that discharge representational duties and those that don't, either because they can't or because they are not even trying. We think that many of our predicates are fit for representational duties: by ascribing them to an object, we represent the object as being a certain way. The object may or may not be as represented by the predicate, and this determines whether the object satisfies the predicate, and whether a sentence that ascribes the predicate to the object is true or false. Let me refer to predicates with this feature as representational. The notion can be interchangeably applied to concepts, conceived as the mental correlates of predicates, and everything I say about the linguistic case is meant to be equally applicable in the mental realm.

Our conviction that a predicate is representational is not infallible. By ascribing a predicate that we take to be representational we might be doing something entirely different from representing objects as being a certain way. This is the situation that would be faced by those who regard ethical predicates as representational if expressivists were right about their true function – if the sentences in which they are ascribed were, as A.J. Ayer maintained, ‘simply expressions of emotion which can be neither true nor false’.Footnote ⁵ It's also possible that by ascribing a predicate that we take to be representational we are not achieving anything of note – neither representing the object nor the performance of any other function. Those who regard ethical predicates as representational would find themselves in this situation if error theorists were right about them.Footnote ⁶

Fallible as our verdicts may be, we have procedures that we employ for deciding which of our predicates are representational – tests that some predicates pass and some fail. Some of these procedures can be broadly described as pragmatic: they concern the rules that govern our use of predicates and the non-representational purposes (e.g. explanatory) served by the theories in which they figure.Footnote ⁷ But we also seem to have at our disposal a different, non-pragmatic procedure for deciding on the representational status of a predicate.

In order for a predicate to discharge representational duties, it has to be connected with the world in the right sort of way. It has to bear a relation to a property as a result of which ascribing the predicate represents the property as present in the object of predication, and the satisfaction of the predicate by an object is determined by whether the property is present in the object. Let me refer to this predicate-property relation as reference, and to the property to which a representational predicate bears the reference relation as its referent. The non-pragmatic procedure I have in mind for determining the representational status of a predicate consists in ascertaining whether the predicate has a referent – whether there is a property in the world to which the predicate bears the reference relation, where this is achieved without the help of pragmatic procedures. The predicate will be deemed to be representational if and only if there is a property playing this role with respect to it. Let me refer to this as the referential procedure for establishing the representational status of a predicate.

How are pragmatic procedures and the referential procedure related to one another? A tempting conception of the relationship goes as follows. Pragmatic procedures are used in our day-to-day lay-person determinations of the representational status of predicates. The features they focus on are only contingently related to the representational status of predicates, so the procedures can produce false results. The referential procedure, by contrast, focuses on the facts that determine whether a predicate is representational. Hence, when correctly applied, the referential procedure produces infallible results, and can be used to validate the verdicts reached by pragmatic procedures. Applying the referential procedure is a job for the philosophers, who specialise in the language-world connections that the procedure focuses on. They will be able to determine whether the verdicts reached by lay people using pragmatic procedures are correct or incorrect. Let me refer to this conception of the relationship between the two types of procedure as the primacy-of-reference picture.

2. Against the primacy of reference

I want to argue that the primacy-of-reference picture is incorrect. The referential procedure cannot play the validating role with respect to pragmatic procedures that the picture ascribes to it. To appreciate the problems I want to raise, we need to consider the kind of access that we enjoy to the realm of properties to which predicates are connected by the reference relation. In order to designate a property as the referent of a predicate, we need to single out the property somehow. But the only procedure we seem to have at our disposal for singling out a property is through the use of a predicate (or concept) that refers to it, or a nominalization of it, as in ‘“flies” refers to the property of floating in the air’. But this exercise will succeed in designating the referent of the target predicate (‘flies’) only on the assumption that the predicate that we use to single out the referent (‘floats in the air’) is itself representational. If ‘floats in the air’ weren't representational, ‘the property of floating in the air’ would not denote anything, and ‘“flies” refers to the property of floating in the air’ would fail to pair ‘flies’ with a referent. The representational character of ‘floats in the air’ can be established in turn using again the referential procedure, but once more we will need to use predicates to single out the property that we are designating as the referent of the new target predicate, and we face exactly the same problem.

This suggests that the referential procedure can be applied only in one of two scenarios. The first scenario would require that we are able to single out at least some properties directly, without the mediation of predicates or concepts. Then, on the assumption that we are capable of ascertaining when a predicate bears the reference relation to one of the properties we can single out in this way, we would be able to establish the representational status of a predicate in a way that would not be conditional on the representational status of other predicates. The second scenario would involve having at our disposal a stock of predicates whose representational status has been established by means other than the referential procedure. We could then use these predicates to single out the properties that we want to designate as the referents of other predicates, in applications of the referential procedure.

The first of these scenarios strikes me as wildly implausible and I'm not even sure whether it has any contemporary advocates. In any case I'm not going to argue against it here, but simply assume, for the sake of the argument, that the kind of unmediated access to properties that it contemplates is not to be had. It follows from this assumption that the referential procedure can be applied only in the second scenario, starting from a pool of predicates whose representational status has been established by non-referential means. Provided that all the alternatives to the referential procedure can be classified as pragmatic, in applying the referential procedure to establish the representational status of predicates we would single out the properties that we designate as the referents of predicates using other predicates whose representational character has been established using pragmatic procedures.

3. The primacy of practice

These thoughts result in an alternative picture of the relationship between pragmatic procedures and the referential procedure, to which I shall refer as the primacy-of-practice picture (PoP). According to PoP, our most fundamental procedures for establishing the representational status of predicates are pragmatic. Pragmatic procedures are employed for establishing the representational character of a basic pool of predicates, with these verdicts not being subsequently validated by the referential procedure. The referential procedure plays a subsidiary role in PoP. It can be used in cases where we haven't been able to establish the representational status of a predicate by applying pragmatic procedures directly to it. In this situation, if we were able to ‘identify the referent’ of the target predicate using other predicates that have been deemed representational by pragmatic procedures, we would be able to extend the sanction of the pragmatic procedures on these predicates to the target predicate. If, e.g., pragmatic procedures designated the predicate ‘floats in the air’ as representational, establishing the truth of ‘“flies” refers to the property of floating in the air’ would enable us to extend this verdict to the predicate ‘flies’. The referential procedure would be a tool for establishing indirectly that a predicate satisfies the pragmatic procedures. According to PoP, there's generally no advantage in reducing the number of predicates whose representational character is established directly with pragmatic procedures, without the intervention of the referential procedure.

On this view, when we identify, say, the referent of the predicate ‘is gold’ as the property of having atomic number 79, our vindication of the representational status of ‘is gold’ is conditional on the representational status of the predicate ‘has atomic number 79’ that we use to single out the referent of ‘is gold’. But the representational status of ‘has atomic number 79’ itself is unlikely to be established with another application of the referential procedure. More likely it will be established in terms of pragmatic criteria, including the role that it plays in theories to which we ascribe a certain explanatory role. Even if we insisted, implausibly, on treating as representational only those predicates whose referent could be identified in terms of the language of basic science, we wouldn't bypass pragmatic procedures, as it's hard to see how the referential procedure could be used to establish the representational status of the predicates of the language of basic science.

Similar remarks apply to negative verdicts on the representational status of a predicate. According to PoP, when we have established the representational character of a predicate using pragmatic procedures, no purpose is served in trying to establish it again using the referential procedure. The referential procedure comes into play when we haven't been able to establish the representational character of a predicate using pragmatic procedures. In this situation, as we've just seen, the referential procedure might enable us to establish the representational character of the target predicate by linking it to the representational status of the predicates in terms of which we specify its referent, with the representational character of these predicates established by pragmatic procedures. If we fail to achieve this, it might seem in the first instance that the conclusion that the target predicate is not representational is the result of our failure to identify its referent. However, in reality, the verdict will have been reached through our failure to establish its representational character using pragmatic procedures, either directly or indirectly, through the mediation of the referential procedure.

In light of the role that PoP assigns to the referential procedure, there is no justification for maintaining that a predicate should not be treated as representational unless this treatment can be supported with the referential procedure. Let me use the label reference fixer to designate sentences of the form ‘“A” refers to the property of B-ing’, and let me say that a reference fixer is homophonic when ‘A’ = ‘B’. It follows from PoP that the unavailability of a non-homophonic reference fixer for a predicate should have no tendency to undermine our decision to treat it as representational, if that is the verdict supported by pragmatic procedures. We should have no problem treating predicates as representational when the only reference fixers at our disposal for them are homophonic, or even when we are convinced that no non-homophonic reference fixers will ever be available. It is sometimes assumed that in the absence of a non-homophonic reference fixer for a predicate we have only two options at our disposal: on the one hand, we can adopt a non-cognitivist approach, according to which ascriptions of the predicate serve a purpose other than representing the world; on the other hand, we can endorse an error theory, according to which, although we aim to represent the world with our ascriptions of the predicate, we fail to achieve this because the predicate is not connected with the world as representation requires. If PoP is along the right lines, this is a false dilemma.

4. Tu quoque?

Advocates of the primacy-of-reference picture might try to undermine PoP with a version of the reasoning I deployed against that view. I argued that the referential procedure for establishing the representational status of a predicate can be applied only if there are some predicates whose representational status can be established by other means. What I want to consider now is whether the reasoning I used in support of this conclusion can be adapted to establish a parallel conclusion for pragmatic procedures – that they can be applied only if there are some predicates whose representational status can be established by other means.Footnote ⁸

Let C be a pragmatic criterion that we employ for establishing the representational character of predicates. If a predicate satisfies C, we conclude that it is representational. Let's assume, for the sake of simplicity, that C is the only criterion that we employ in this way. Suppose we establish that ‘flies’ satisfies C, and conclude on this basis that ‘flies’ is representational.

The advocate of the primacy-of-reference picture could argue as follows: ‘“flies” satisfies C’ would not establish the representational character of ‘flies’ unless it represented ‘flies’ as being a certain way.Footnote ⁹ This requires, in turn, that ‘satisfies C’ is a representational predicate. But on the assumption that C is our only pragmatic criterion for representational character, establishing the representational character of ‘satisfies C’ with our pragmatic procedures would consist in establishing that it satisfies C. And this, the argument goes, would be unacceptably circular. Hence establishing the representational character of ‘flies’ on the basis of whether it satisfies C would require an alternative, non-pragmatic procedure for establishing the representational character of predicates.

I want to take issue with the claim that using C to establish the representational character of ‘satisfies C’ would involve vicious circularity. C is a criterion that a predicate may or may not satisfy. When a predicate satisfies it, we say it is representational. Now, what goes for predicates in general goes in particular for ‘satisfies C’. This predicate may or may not satisfy C. In fact, since we haven't specified which criterion ‘C’ stands for, we have no way of knowing whether ‘satisfies C’ satisfies it or not. If ‘satisfies C’ satisfied C, we would conclude, as with any other predicate, that it is representational, and hence that by ascribing it to a predicate, whether this be ‘flies’ or ‘satisfies C’, we represent it as being a certain way. Our ability to use C to establish the representational character of predicates doesn't depend on the availability of an alternative, non-pragmatic criterion.

Couldn't the advocate of the primacy-of-reference picture use a parallel move to save her view from my attack? The idea would be that if the representational character of ‘satisfies C’ can be established on the grounds that it satisfies C, similarly the representational character of, say, ‘floats in the air’ could be established with a homophonic reference fixer, on the grounds that ‘floats in the air’ refers to the property of floating in the air.

But the two cases are not parallel. There is at least the following difference between them. The procedures that we use for applying criterion C put predicates to a test that they might fail independently of whether ‘satisfies C’ is representational. Furthermore, the test is just as substantial when applied to ‘satisfies C’ as when applied to any other predicate. I could come to the conclusion that ‘satisfies C’ doesn't satisfy C without drawing on independent information to the effect that ‘satisfies C’ is not representational.

Now, we certainly have substantive procedures for determining whether a predicate refers to the property of floating in the air – they are the procedures that we employ for deciding how to interpret predicates and, in particular, for determining when to interpret a predicate as referring to the property of floating in the air. A predicate could fail the relevant tests independently of whether or not ‘floats in the air’ and ‘refers to the property of floating in the air’ are representational. This is the situation for predicates other than ‘floats in the air’. I can determine that, say, ‘“sings” refers to the property of floating in the air’ is false even if ‘floats in the air’ is representational. However, for ‘floats in the air’ itself (as meant by me now), interpretation procedures do not apply. The procedures that I use for deciding on the ascription of ‘refers to the property of ϕ-ing’ dictate that there's only one circumstance in which I shouldn't ascribe the predicate to ‘ϕ’, namely if ‘ϕ’ is not representational. I could only come to the conclusion that ‘floats in the air’ doesn't refer to the property of floating in the air if I had independent information to the effect that ‘floats in the air’ is not representational. There is no other reason for rejecting a homophonic reference fixer. If homophonic reference fixers could be used to establish the representational character of predicates, every predicate would have to count as representational.

5. Knowledge of reference

Representing objects with predicates for which we have no non-homophonic reference fixers might seem to pose a problem. It might seem that in the absence of a non-homophonic reference fixer we won't know how the predicate represents objects as being. Hence when we represent objects with a predicate in these circumstances, the thought goes, we won't know how we are representing the object as being – representation would happen, so to speak, behind our backs.

But this can't be a genuine problem. It is true that sometimes we don't know how a predicate represents objects as being until we find a non-homophonic reference fixer for it, and, when we do, the reference fixer expresses our knowledge of how the predicate represents objects as being. Nevertheless, it can't be true in general that we can't know how a predicate represents objects as being unless we have a non-homophonic reference fixer for it. The reasons for this are familiar and related to the argument offered above in support of PoP. A reference fixer will enable us to grasp the referent of the target predicate only if it is specified in terms of a predicate whose referent we know. This knowledge, in turn, can be given by another reference fixer, but sooner or later the resulting chain will have to bottom out in a predicate whose referent we know by means other than reference fixers. It follows that, if knowledge of reference is possible at all, there can't be a problem with knowing the referent of a predicate for which we don't have a non-homophonic reference fixer.Footnote ¹⁰

What else could knowledge of reference consist in? Once again, we can appeal to features of the practice. On this model, our knowledge of how a predicate represents things as being could consist in the cluster of procedures that we employ for deciding on its ascription and the role that it plays in wider practices. I'm proposing that there is no problem with representing objects with a predicate when we know in this way how the predicate represents objects as being.

It's important to see that this practice-based model of knowledge of reference does not carry a commitment to some similar-sounding questionable views. First the practice-based account of knowledge of reference doesn't entail a parallel practice-based account of reference itself. On this account of reference, whether an object satisfies a predicate would be determined by whether ascribing the predicate to the object would satisfy a cluster of pragmatic criteria. This would give us, in effect, the material for a non-homophonic reference fixer for the predicate: ‘“P” refers to the property of being the object of predication in an ascription of “P” that satisfies such-and-such pragmatic criteria’.

Practice-based accounts of reference along these lines are highly implausible. A major obstacle they face is the revisable character that our practice attributes to the verdicts currently endorsed by the ascription procedures we associate with our predicates. Revisions can come about as a result of what we see as improvements in our state of information or of changes that we take to be for the better in the procedures we employ for deciding on the ascription of predicates. We simply don't see the procedures we currently use, or the verdicts we take these to support, as defining the reference or satisfaction conditions of our predicates. What I'm proposing is to apply the practice-based model of knowledge of reference to predicates for which we have no non-homophonic reference fixer, including the practice-based proposal outlined here. Ascription procedures would explain knowledge of reference, but they wouldn't also single out the referent of the predicate.

Second, the practice-based model of knowledge of reference does not entail either the possibility of specifying in terms of pragmatic factors conditions under which the ascription of a predicate would be warranted or would enjoy some other positive epistemic status.Footnote ¹¹ Ascribing the predicate using the procedures invoked in our explanation of knowledge of reference would not make the predicate ascription warranted or justified or, even if it were true, an instance of knowledge. Whether a predicate ascription deserves any of these accolades would be determined by the procedures associated with the epistemic predicates that express them.Footnote ¹²

Finally, explaining knowledge of reference in terms of the practice doesn't in principle force us to endorse practice-based necessary and sufficient conditions for sameness of reference. Our verdicts on when someone refers with a predicate to the same property to which we refer with one of ours will be based on the kinds of aspects of their practice in terms of which knowledge of reference is explained on this model. However, these verdicts won't be reached by checking whether their practice meets predefined sets of criteria associated with our predicates. Instead, we will use general interpretation procedures to determine whether their practice should be characterised as attaching to their predicate a referent that we could single out with one of ours.

This last point is compatible with the observation that there are cases in which we regard certain features of a predicate-ascription practice as necessary and even sufficient for a predicate to have the reference it has. What this means is that for one of our predicates our interpretative procedures might involve never treating a predicate as coreferential with ours unless its associated ascription practices exhibit a certain feature (necessity), or always treating a predicate as coreferential with ours if its associated ascription practices exhibit a certain feature (sufficiency).Footnote ¹³ These aspects of our interpretative procedures involving certain predicates sometimes sustain promising solutions to philosophical difficulties involving these predicates.Footnote ¹⁴

Notice that, even if we thought that a feature of our procedures for ascribing a predicate was necessary and sufficient for the predicate to have the reference it has, we shouldn't expect to be able to obtain from this fact necessary and sufficient conditions for the satisfaction of the predicate.Footnote ¹⁵ And, of course, our view that a feature of our predicate-ascription procedures has this status is open to revision. We might come to reject it without thereby seeing ourselves as changing the reference of the predicate.

I'm suggesting, then, that a predicate can be representational even if we don't have or expect to have at our disposal a non-homophonic reference fixer for it. Having no non-homophonic reference fixers forces us to provide an alternative account of our knowledge of the referent of the predicate, but this can be constructed in terms of the procedures that we employ for deciding on the ascription of the predicate and the practices in which it is embedded. This account of knowledge of reference doesn't force us to obtain from the practice an identification of the referent of the predicate, conditions under which its ascriptions have some positive epistemic status or conditions under which someone else's predicate has the same referent as ours. Predicates that answer to this description are not peripheral cases of representation. The arguments offered above for practice-based accounts of representation and of knowledge of reference suggest that predicates of this type form the core of our representational apparatus.

It should be clear also that there are no legitimate obstacles to describing these predicates as referring to properties, whose presence in or absence from an object determines whether the object satisfies the predicate. Restricting this treatment to predicates for which we have non-homophonic reference fixers would betray a lack of awareness of the subsidiary role played by reference fixers in establishing the representational status of predicates and in explaining knowledge of reference.

6. Representation and practice

Now, the template I've just outlined for a representational predicate can be applied, specifically, to the predicate ‘is representational’, or ‘represents’. Suppose we don't have at our disposal a non-homophonic reference fixer for ‘represents’ – a substantive account of the conditions that a predicate would have to satisfy in order to count as representational. We could still treat the predicate ‘represents’ as representational if our pragmatic procedures for deciding on the representational status of a predicate supported an affirmative verdict in this case. Then we would be entitled to maintain that ascribing ‘represents’ to a predicate represents this predicate as being a certain way (that ‘represents’ represents). We would also be entitled to speak of the way a predicate is represented as being when we ascribe ‘represents’ to it as a property of predicates – the referent of ‘represents’ – whose presence in or absence from a predicate determines whether the predicate satisfies the predicate ‘represents’. In the absence of a non-homophonic reference fixer for ‘represents’, we would need an alternative account of our knowledge of its referent, but this could be constructed in terms of the (ultimately pragmatic) procedures that we employ for deciding whether a predicate represents. In this case, exceptionally, the procedures in terms of which we explain knowledge of reference would be the same procedures that we use to support ascribing representational status to the predicate. As in the general case, explaining knowledge of reference in this way wouldn't force us to endorse practice-based necessary and sufficient conditions for the satisfaction of the predicate, for ascriptions of the predicate to enjoy any positive epistemic status, or for someone else's predicate to have the same reference as ours.

The relational predicate ‘refers’ could receive the same treatment. Suppose we don't have at our disposal a non-homophonic reference fixer for it. We could still treat the predicate as representational if this verdict was supported by our pragmatic procedures for establishing representational status. Then we would be entitled to speak of the way we represent a predicate and a property as related when we ascribe ‘refers’ to them, and to describe this way of being related as the relation the predicate refers to, whose exemplification by a predicate-property pair will determine whether they satisfy the predicate ‘refers’. As before, our knowledge of the reference of this predicate could be explained in terms of the procedures that we employ for deciding on its application and of the practices in which it figures. These procedures will sanction, specifically, homophonic ascriptions of the predicate, in all instances of ‘“ϕ” refers to the property/relation of ϕ-ing’, where ‘ϕ’ is a predicate to which we ascribe representational status, including, of course, ‘“refers” refers to the relation of referring’.

7. Naturalizing reference

The treatment of ‘represents’ and ‘refers’ outlined in the preceding section might have no obvious competitor on the assumption that no non-homophonic reference fixers are available for these predicates. But this assumption is emphatically rejected by those engaged in the project of naturalizing semantic notions. A naturalistic account of predicate reference would identify, in non-semantic terms, the relation that pairs each representational predicate with the property whose instantiation conditions determine the satisfaction conditions of the predicate – i.e. the property, in our terms, to which the predicate refers. This would give us a non-homophonic reference fixer for ‘refers’. Thus, for example, a familiar template for naturalistic accounts takes the following form:

where C-conditions are specified non-circularly.Footnote ¹⁶ This can be easily reformulated as a reference fixer for ‘refers’:

This can then be used to provide a non-homophonic reference fixer for ‘represents’:

If reference fixers of this kind are available, the alternative model outlined in the preceding section would be entirely redundant.

Some supporters of this approach might see it as a vindication of the primacy-of-reference picture, but this is by no means compulsory. The naturalistic approach is in principle compatible with PoP. The naturalist could concede that our procedures for determining the representational status of predicates are ultimately pragmatic, as PoP dictates, but contend that the representational character of ‘refers’ and ‘represents’ can be established with the referential procedure, using predicates, e.g. ‘causes’, whose referential status is ultimately established by pragmatic procedures.

The prospects of this kind of naturalization project for semantic concepts have been the subject of extensive debate. The issue is often characterised as involving two separate stages.Footnote ¹⁷ First, there is the question whether naturalists can provide an extensionally accurate account of reference. Second, there is the question whether an extensionally accurate naturalistic account of reference could be treated as identifying the relation that ‘refers’ refers to – the relation pairing each representational predicate with the property that determines its satisfaction conditions.

Opponents of the naturalization project have argued that a naturalistic account of reference would fail to identify the reference relation even if it were extensionally accurate, on the grounds that no natural relation can capture the normative character of reference and other semantic notions.Footnote ¹⁸ Naturalists have resisted this, maintaining that achieving extensional accuracy is the only substantive challenge that their project faces.Footnote ¹⁹ They contend that if they could provide an extensionally accurate account of reference there would be no legitimate obstacle to treating this account as identifying the reference relation.

I want to suggest that this way of framing the debate might mislocate the difficulties that the opponents of naturalism are trying to articulate. Consider the criterion of extensional accuracy that naturalists recognise as the only challenge they have to meet. An extensionally accurate account of reference is one that gets the predicate-property pairings right – pairing each predicate with the property, if any, that actually determines its satisfaction conditions. Or, slightly more modestly, an extensionally accurate account will pair each representational predicate with the objects that actually satisfy it. It is unquestionable that this is what naturalists aspire to, but this formulation of the criterion doesn't provide a complete description the procedure that we would need to follow for assessing their proposal. We don't enjoy any form of direct access to the predicate-property links that naturalists are trying to chart, in terms of which we could assess the extensional accuracy of their proposals. All we have are the procedures that we employ for deciding on the ascription of predicates and the verdicts that we take these to support. These are the only standards against which we can assess the extensional accuracy of naturalist accounts.

If we consider the extensional-accuracy requirement in this light, we soon encounter some problems. One source of difficulties is a phenomenon mentioned above: the verdicts supported by our ascription procedures are typically treated as revisable, in light not only of additional information, but also of the adoption of new procedures that we regard as improving on the existing ones. Hence suppose that a naturalistic account of reference were to provide a perfect match with the ascription verdicts currently supported by our procedures. At that point we would consider it extensionally accurate. But suppose that we came to revise some of our ascription verdicts, as a result either of new information or of new procedures that seemed to us to do a better job at detecting the presence in an object of the referent of the predicate. Accepting the naturalistic account would require sticking to our old verdicts and discounting the new ones. But what would be the justification for this? The naturalist might recommend this attitude on the grounds that the account is extensionally accurate, but this reply would clearly beg the question. The only reason we had for thinking that the account was extensionally accurate was that it provided a perfect match for the verdicts previously endorsed by our procedures. Since we now think that in light of what we regard as better information or better procedures some of those verdicts were incorrect, surely the right conclusion to draw is that we were wrong in thinking that the naturalistic account was extensionally accurate, and to reject it on these grounds.

If the naturalistic account offers only an imperfect match for the verdicts supported by our current ascription procedures we face the same problem immediately. For the cases in which there is a disagreement between the naturalistic account and what our current verdicts recommend, accepting the naturalistic account would require rejecting these verdicts as incorrect. Granted that we accept the fallibility of the verdicts supported by our current procedures, but what adopting the naturalist account requires of us is not simply accepting that our verdicts might be wrong, but that they are actually wrong. But, once again, what would be the justification for this? As we've just seen, invoking the extensional accuracy of the naturalistic account at this point would beg the question. The reason we've been offered for thinking that the account is extensionally accurate is the fact that it matches many of the verdicts currently supported by our ascription procedures. But if this is a reason for thinking that the account is extensionally accurate, then the kind of disagreement that we are now considering should count as a reason for thinking that the account is extensionally inaccurate, not for abandoning the verdicts currently supported by our procedures.

The advocate of the naturalist account could defend her proposal on the grounds that it offers the best possible overall match for our verdicts. She could argue that, although accepting her proposal will require rejecting some of our verdicts as incorrect, all the alternatives would require an even greater departure from the verdicts that we now consider correct. Now, this argument would carry some weight on the assumption that some non-homophonic reference fixer for ‘refers’ has to be right.Footnote ²⁰ Then the need to reject some of our current verdicts as incorrect wouldn't by itself invalidate a proposal, as the level of rejection required by any plausible alternative might be greater still. But we have seen that this assumption can be challenged. We can represent the world with predicates for which no non-homophonic reference fixer is to be had, and we have the option of treating ‘refers’ in this way. Unless a reason can be given why a non-homophonic reference fixer is needed in this particular case, the fact that the naturalist's proposal is more compatible than the alternatives with the verdicts currently endorsed by our ascription procedures doesn't provide legitimate support for rejecting the verdicts that disagree with the proposal.

In sum, adopting a naturalistic account of reference requires treating it as extensionally accurate. This requires, in turn, being prepared to reject predicate-satisfaction verdicts endorsed by our ascription procedures when they disagree with the naturalistic account. However, this move would be justified only to the extent that we had reason to believe that the account is extensionally accurate. But our reason to believe that the account is extensionally accurate consists in its ability to match predicate-satisfaction verdicts endorsed by our ascription procedures. This suggests that a conflict between the naturalistic account and the verdicts endorsed by our procedures should be taken as undermining the extensional accuracy of the account, not as a reason for abandoning the verdicts endorsed by our procedures. If the representational character of ‘refers’ depended on the availability of a non-homophonic reference fixer, rejecting the verdicts endorsed by our procedures when they clash with the naturalistic account could be justified if any alternative account required an even greater departure from these verdicts. However, we've seen that, according to PoP, a predicate can be representational even if there is no non-homophonic reference fixer in the offing for it. In the absence of a specific reason for thinking that ‘refers’ can't be treated in this way, we have no reason for treating a naturalist proposal as extensionally correct when it ascribes to predicates satisfaction conditions that clash with the verdicts supported by our ascription procedures. And even if the proposal provides a perfect match for the verdicts we currently endorse, subsequent revisions to our satisfaction verdicts should be taken as undermining its extensional accuracy.