1.1 The theory

The central claim of ecumenical expressivism is that normative claims express hybrid, relational states of mind (see also Schroeder Reference Schroeder and Shafer-Landau2013b and Toppinen Reference Toppinen and Shafer-Landau2013 for other ‘relational’ versions of expressivism). These are complex states comprising a representational and nonrepresentational component. The nonrepresentational component is a kind of noncognitive practical stance (a ‘normative perspective’) that provides the agent with a set of policies about which standards of practical reasoning to reject and accept (Ridge Reference Ridge2014: 115). A standard is a rule that can be used in an action-guiding way as the basis of a practical judgment or decision (40). The standards that make up an agent's normative perspective are ‘ultimate’, in that they are fundamental (not derived from other standards) and provide a complete guide to action (116f.). However, standards are ultimately explained in terms of what it is for an agent to accept a standard, where this is for that agent to be disposed to issue the relevant prescriptions and intuitively endorse them (111f.).

On this view, the representational component of the state expressed by a normative claim is a robustly representational belief. The belief is indexed to the agent's normative perspective such that, for any normative judgment, the object of evaluation is evaluated or ‘ranked’ by the standards not ruled out by the agent's normative perspective (Reference Ridge2014: 119). The exact nature of the ranking will depend on the normative predicate employed in the claim (e.g., very roughly, ‘good’ will mean ‘ranked high’, while ‘bad’ will mean ‘ranked low’). While the content of the representational component is robustly representational, the overall content of the normative claim itself is not identified with any representational content. Rather, it is irreducibly normative.

To illustrate the above, consider the following example:

1. (1) ‘Giving to charity is good’.

According to ecumenical expressivism, (1) conventionally expresses the relational state of mind comprising:

1. (1_a) A normative perspective

2. (1_b) The belief that giving to charity is highly ranked by any admissible ultimate standard of practical reasoning

The state expressed is relational in virtue of the concept of an admissible standard, which refers to the standards not rejected by the normative perspective of the speaker (Ridge Reference Ridge2014: 119). As different speakers will have different normative perspectives, (1_a) and (1_b) together make up the relational state type that is multiply realized by any agent that tokens the state expressed by (1). Logical complexity is then ‘off-loaded’ to the representational component of the state, which sets the stage for Ridge's attempt to solve the Frege-Geach problem (144ff.). For example, the claim ‘If giving to charity is good, then Socrates would approve of it’ would express the multiply realizable relational state comprising (1_a) together with the belief that if giving to charity is highly ranked by any admissible ultimate standard of practical reasoning, then Socrates would approve of it.

Thus far, this account has explained normative thought and discourse without recourse to normative propositions. However, as I argued above, there are lots of good reasons for everyone to recognize normative thought and discourse as propositional. To accommodate normative propositions within ecumenical expressivism, Ridge appeals to Scott Soames's (Reference Soames2010, Reference Soames, King, Speaks and Soames2014, Reference Soames2015) cognitive act theory of propositions. For Soames, the nature of propositional content is explained by concrete cognitive activity. On this view, propositions are certain types of cognitive acts or events. (I will follow Soames in assuming that acts are events rather than, say, processes.) More specifically, propositions are acts that are tokened in representational activity: ‘Propositions are repeatable, purely representational, cognitive acts or operations the performance of which results in concrete cognitive events; to entertain a proposition is to perform it’ (Soames Reference Soames2015: 16, emphasis in the original). Identity conditions for propositions are specified in terms of the essential representational properties of such acts.

For example, consider the proposition that the sea is blue. According to Soames, this is identified with the cognitive act type of representing the sea as being blue. Judging, asserting, imagining, and so on, that the sea is blue are all ways of entertaining this proposition. To perform any of these acts is to token the act type of representing the sea as being blue. Thus, the identity conditions of the proposition that the sea is blue are explained in terms of representational properties of concrete acts of representing the sea as being blue. With this basic notion of entertaining a proposition in place, other propositional attitudes and acts can be defined in terms of it. For example, to judge that the sea is blue is to perform that predication in the affirmative manner, where this is cashed out in terms of its role in one's cognitive economy (Soames Reference Soames2015: 18). To believe that the sea is blue is to be disposed to judge that the sea is blue, and so on (Soames Reference Soames, King, Speaks and Soames2014: 97; Reference Soames2015: 18f.).

For Soames, as for most others, propositional thought is eo ipso representational. Hence, in order to accommodate normative propositions, expressivists need to broaden the relevant class of cognitive event types that constitute propositions. This is analogous to the more familiar expressivist move of broadening the relevant class of mental states that are ‘beliefs’ to include not only robustly representational beliefs, but any mental state conventionally expressed by sincere utterances of declarative sentences (Ridge Reference Ridge2014: 128). Whereas Soames appeals to an intuitive antecedent understanding of ‘representing’ out of which propositional acts and attitudes can be defined, Ridge reverses the order of explanation. First, he provides an account of what it is to believe, desire, assert, fear, and so on, the normative claim that p without any appeal to the proposition that p. Second, he abstracts away from each case to that cognitive event type that is tokened in all and only those acts and attitudes. The idea is that each such act or attitude is (or stands in a relation to) a way of entertaining the normative proposition that p.

Given the highly plausible assumption that events and states are distinct metaphysical categories (Kenny Reference Kenny1963; Vendler Reference Vendler1967; Mourelatos Reference Mourelatos1978; Steward Reference Steward1997), this immediately poses the problem that one cannot derive an event type from various different states. The problem arises because ecumenical expressivism provides only a theory of normative mental states and not of events. Soames's theory avoids any such problem by treating representational acts as explanatorily fundamental. In this way, mental states (e.g., beliefs) can have propositional content by standing in appropriate relations to mental events (e.g., judgments) that token the relevant cognitive acts. The problem for ecumenical expressivism is that normative thought is partly explained in terms of an agent's normative perspective, which is fundamentally a state with no obvious event-counterpart. To be fair, Ridge (Reference Ridge2014: 128n) acknowledges the problem and suggests some solutions. But however the problem is resolved, the objection raised in the next section still holds. So I will assume for the sake of argument that there is some available solution.

Modulo the qualifications of the previous paragraph, ecumenical expressivism claims that being in the state of mind comprising (1_a) and (1_b) is one way of entertaining the proposition that giving to charity is good. Clearly, however, this is not the only way to entertain the proposition. The proposition that giving to charity is good can also be entertained in other propositional attitudes, such as desire, doubt, hope, and so on. However, it is easy to see how the account can be generalized. For example, consider the doubt that giving to charity is good as the complex state of mind comprising:

1. (2_a) A normative perspective

2. (2_b) The doubt that giving to charity is highly ranked by any admissible standard of practical reasoning

Here, the that-clause in (2_b) denotes a representational content indexed to the agent's normative perspective in the same way as (1_b) (Reference Ridge2014: 148). This move can then by applied mutatis mutandis to other attitude types. Further, Ridge notes that it is possible to entertain a normative proposition whereby one ‘simulates’ a normative perspective, where this is run ‘off-line’, such as when one ‘merely entertains’ that p (128).

The proposition that giving to charity is good is then identified as the minimal cognitive event type tokened in relational states <(1_a), (1_b)>, <(2_a), (2_b)>, and so on (again, modulo the above qualifications). In contrast to the representational case, it is somewhat harder to grasp intuitively what this act type consists in. However, it should be clear that it is the act type all instances of which involve the following features: (i) a component with a stable descriptive content, (ii) an actual or simulated normative perspective, and (iii) the structural relation between (i) and (ii) encoded by the concept of being ranked highly by any admissible ultimate standard of practical reasoning. It is in virtue of this concept that the component with stable representational content is necessarily tied to the agent's normative perspective in every possible instance of entertaining the proposition that giving to charity is good.

1.2 The Problem

The real problem for Ridge's theory is that it presupposes rather than explains identity conditions for normative propositions. Whatever a theory of propositions is for, presumably it should explain their identity conditions, i.e., the conditions under which propositions are individuated and distinguished from one another. (Hence the Stanford Encyclopedia of Philosophy page on propositions states: ‘Any good theory of propositions ought to have something to say about when propositions are identical and when they are distinct’ [McGrath and Frank Reference McGrath, Frank and Zalta2018].) For example, on the possible worlds conception, propositional identity is explained in terms of set identity: p and q are identical just in case p and q have the same worlds as elements. Or on the Russellian conception, propositional identity is explained in terms constituents and ordering: roughly, p and q are identical just in case p and q have the same constituents in the same ordering. The ecumenical expressivist conception, on the other hand, fails to provide principled identity conditions for normative propositions. Hence, it is explanatorily inadequate.

To see why Ridge's theory presupposes rather than explains identity conditions for normative propositions, consider the following comparison. We saw above that for Soames propositions are not just representational cognitive acts, but purely representational cognitive acts. This is because there are far more representational cognitive acts than there are propositions: ‘[T]he acts of predicating humanity of Plato (i) on Thursday, (ii) in Peru, (iii) while dancing, (iv) in giving a lecture, or (v) when speaking in a whisper all represent Plato as being human, even though we would not be happy thinking of them as propositions’ (Soames Reference Soames2015: 70, emphasis in the original). If any representational cognitive act were a proposition, then predicating humanity of Plato on Thursday would be a distinct proposition from predicating humanity of Plato. But it is implausible that there is any such proposition. However, if cognitive propositions are purely representational, then the act type of predicating humanity of Plato on Thursday is not a proposition. This is because the condition of the act being on Thursday does not contribute in any way to representing Plato as human.

The ‘purely’ constraint is not an ad hoc fix to a problem of deviant or gerrymandered cases. Rather, it is a consequence of Soames's commitment to explaining the nature of propositions in terms of concrete representational activity. We will explore the significance of this in the next section. For now, however, we can ask the following question of Ridge's theory. For any normative claim p and for the cognitive act types

1. (A₁) entertaining p

2. (A₂) entertaining p on Thursday,

in virtue of what is (A₁) a proposition but (A₂) not a proposition? As far as I can see, Ridge's theory offers no answer to this question and effectively stipulates that only the former kind of act can play the object of attitude role. And this is just to say that the account presupposes rather than explains identity conditions for normative propositions. There are many more normative cognitive act types than normative propositions, and Ridge's theory provides no principled way of delineating the latter class from the former.

One possible response would be to highlight that because the theory of propositions is explanatorily downstream from the theory of attitudes, there is no real worry here. The response goes that since we begin with an account of what it is to judge, doubt, and so on that p and only then go on to abstract the proposition that p from this account, there is no possibility of deriving acts like (A₂) as propositions. However, there is no reason why we cannot provide an account of what it is to judge, doubt, and so on that p on Thursday. Indeed, the account is the same other than the condition that the attitudes occur on a Thursday. It is true that we probably have no interest in giving an account of this kind of attitude. But that is a pragmatic question about our interests. Just as with Ridge's original account, we have a theory of a set of concrete attitudes from which we can abstract the act type (A₂) that is tokened in all and only those attitudes.

Another response would be to find some property that (A₁) possesses and (A₂) lacks in virtue of which (A₁) but not (A₂) is a proposition. Given the hybrid nature of normative thought, perhaps we might say that the act of entertaining a normative proposition must possess the conjunctive property of having some purely representational component, some purely practical component, and no other component, where ‘purely practical’ identifies the relevant action-guiding property in virtue of which normative judgments settle the thing to do. To make good on this suggestion, one would need to cash out the notion of ‘purely practical’ without begging the question about what sort of acts can be purely practical. The notion must be related to guiding action. However, it cannot be related in just any way. For example, suppose that while entertaining the proposition that I ought to ϕ, I simultaneously experience an unconnected occurrent desire to ϕ (or manifestation thereof). Insofar as this desire might determine my actions or intentions to act, on what grounds can we exclude this desire from being ‘practical’ in the relevant sense? Further, suppose I entertain the proposition that I ought to ϕ by imagining that I ought to ϕ. Such an act bears no direct link to action or intention, yet must still be a realization of an act type that has a ‘purely practical’ component. So ‘purely practical’ must somehow include these cases as well.

These considerations are not decisive. For example, perhaps we might primarily appeal to the purely representational component and then derivatively identify the practical components necessarily implicated by the representational component. However, even assuming a suitable notion of ‘purely practical’ can be provided, it is not clear that this response does anything other than restate the original worry. Why is it that only act types with this conjunctive property are normative propositions? The appeal to the property of being purely practical is not simply meant to provide an account of what is distinctive about the nonrepresentational component of normative propositions. Rather, the property was meant to explain (in part) that in virtue of which cognitive act types that possess the conjunctive property are propositions. However, it is not at all clear what such an explanation would look like.

In the next section, I argue that the very form of this kind of explanation is problematic. This is because the approach posits an implausible explanatory bifurcation regarding why representational and normative cognitive acts can play the role of propositions. As we will see, this motivates a desideratum on any expressivist theory of propositions.

3.1 A conceptual role account

Here is the plan. First, I propose that predicational acts be understood in terms of categorization. Second, I argue that thinking of categorization in relation to concepts rather than properties provides a suitably neutral characterization of predication. Third, I argue that a nonrepresentational account of the identity conditions for acts of categorization might be achieved by exploiting resources from conceptual role accounts of content.

What is predication understood as a cognitive act? Plausibly, it is an act of categorization. Indeed, this is how it is understood by extant cognitive act theories. For example: ‘Acts of predication are acts of sorting things into groups. When you predicate a property of an object, you sort that object with other objects in virtue of their similarity with respect to the property’ (Hanks Reference Hanks2015: 64). In order words, propositions are ways of categorizing. Thus, in predicating (say) yellow of my scarf, I am grouping my scarf with other objects, such as rubber ducks, daffodils, and Rothko's No 14 No 10. According to Hanks, this involves sorting these objects in virtue of their similarity to the property of yellowness.

If predication essentially involves categorization according to properties, then this will be of little use to expressivists. However, it is not obvious that the notion of categorization presupposes such an understanding. (Should nominalists deny that there are any acts of categorization?) Moreover, there is a clear sense in which acts of normative predication involve sorting things into groups (the good, the right, etc.) that expressivists can and should accept. Whatever else is going on, or whatever exactly this consists in, no one should deny that when one applies a normative concept to something, one is categorizing that thing according to that concept. (Note that categorizing something according to a concept in this sense need not involve endorsing the categorization or judging it to be true.)

A neutral characterization of predication can be given if we semantically ascend from talking about properties to talking about concepts. On this construal, predication fundamentally involves the application and employment of concepts. That is, we sort things into groups according to our concepts, which provide us with certain rules or principles of categorization. This seems just as true with normative concepts as with any other. To say this is to say very little about the nature of categorization, normative or otherwise. However, the description identifies the relevant phenomenon without presupposing that categorization must be explained in terms of properties or relations. In what follows, I am primarily concerned with exploring the theoretical possibilities that open up if we accept the concept-first approach rather than with arguing for that approach. However, one consideration in its favor is that a concept-first approach might be needed to differentiate between predicational acts involving necessarily coextensive concepts of differing cognitive significance.

In order to develop these observations into a theory of cognitive propositions, we need an account of the identity conditions for acts of categorization. In order to be compatible with expressivism, this account must not individuate acts of categorization in terms of the properties or relations denoted by our concepts. (There are two ways of cashing out the rejection of explaining concepts in terms of the properties they denote. If one takes all properties to be ontologically robust, the rejection amounts to denying that normative concepts denote properties at all. Alternatively, if one allows for ontologically lightweight ‘pleonastic’ properties, then we can say that while normative concepts do denote properties, these properties play no role in explaining the nature of those concepts. Nothing hangs on which approach we take, but I think the latter approach is more in keeping with expressivism's project of accommodating the realist-sounding features of the discourse, cf. Dreier's [Reference Dreier2004] ‘explanation explanation’.) Moreover, the account must be fully general to respect the unity requirement. In the remainder of the paper, I explore the idea that conceptual role accounts of content might provide expressivists with some helpful resources for explaining predication. While I will not defend the conceptual role approach here, it is worth noting that a number of expressivists are sympathetic to such an approach (e.g., Blackburn Reference Blackburn, Devitt and Hanley2006; Båve Reference Båve2013; Köhler Reference Köhler2017; Sinclair Reference Sinclair2017; for discussion see Kalderon Reference Kalderon2007: ch.2; Chrisman Reference Chrisman, McPherson and Plunkett2017).

On one prominent approach to conceptual role accounts of content, concepts are abstract objects individuated by their possession conditions (see Peacocke Reference Peacocke1992; Wedgwood Reference Wedgwood2007; for other approaches to conceptual role theory see Harman Reference Harman1973; Field Reference Field1977; Block Reference Block1986; for an overview see Whiting Reference Whiting2006). The possession conditions for a concept specify the transitions to and from mental states involving that concept that an agent is disposed or rationally committed to making. The idea is clearest when applied to logical concepts. For example, Peacocke (Reference Peacocke1992: 6) proposes that the concept of conjunction is that concept C to possess which an agent must find the following transitions primitively compelling (i.e., not derived from or answerable to anything else):

$$A\lpar C \rpar \quad p,\; q\to p\; C\; q \quad p \; C \; q\; \to \; p \quad p \; C \; q\; \to \; q $$

At first glance, this might look promising because A(C) individuates the concept of conjunction not in terms of what it represents, but in terms of its relational role in cognition. As we will shortly see, there is a complication that arises as to what role, if any, reference plays in a complete account of concepts. I put this complication aside for now. What is important is that the above formula for identity conditions for concepts makes no appeal to their referential properties.

How do concepts fit into the cognitive act view? Concepts are components of thoughts. If concepts are abstract entities, then they are components of propositions. If propositions are cognitive acts, it follows that concepts are components of cognitive acts. A natural way to understand this thought is that propositions are complex acts comprising sub-acts, and these sub-acts are what concepts are (the idea that concepts are event types is developed in detail by Davis [Reference Davis2003]; Hanks [Reference Hanks2015] and Soames [Reference Soames2015] also endorse a structured view of cognitive propositions as complex acts). Indeed, this way seems to demystify how concepts as abstract objects could be involved in concrete thought. In the simplest case, we can think of predication as a complex act type comprising: (i) identification of the predication target, and (ii) application of a rule of categorization to the predication target. What makes it the case that an agent applies some concept C rather than C ′ and thus applies one rule of categorization rather than another is the transitions she is disposed to make or ought to make to and from judgments involving the concept. (Hanks [Reference Hanks2015: 23] and Soames [Reference Soames2015: 23] suggest a tripartite structure for predication that involves a distinct sub-act of identifying or ‘expressing’ a property in addition to the sub-act of predicating that property of the target; given that the present account aims to give nonrepresentational identity conditions for propositions, it has no need to invoke a distinct act over and above the application of the rule of categorization.) The identity conditions of cognitive propositions can then be given in terms of the individuating conceptual roles of their component concepts. Where p and q are cognitive propositions: p and q are identical just in case the constituent concepts in each act of predication license the same transitions to and from mental states with those contents.

This model can explain why cognitive act types like entertaining p on a Thursday are not propositions. This is because the sub-act of ϕ-ing on a Thursday has no constitutive conceptual role—the act type of ϕ-ing on a Thursday is not plausibly individuated in terms of any transitions agents are disposed to make between mental states. So the act type is not a concept and therefore not a propositional constituent. Thus, entertaining p on a Thursday is not wholly individuated by its constitutive conceptual roles and is therefore not a proposition. Moreover, it is worth reemphasizing that concepts are here individuated as abstracta with essential conceptual roles. If they were individuated as mental representations, conceptual roles would arguably not determine the propositional content of a mental state. This might be because such roles differ across times and persons or because of their holistic individuation. The present account faces neither problem.

3.2 A complication

The above framework outlines a way of conceptualizing cognitive propositions that is neutral with respect to whether propositional thought is eo ipso representational. It might therefore seem that expressivists have a framework for theorizing about normative propositions that meets the unity requirement. However, now for the complication. While it is true that conceptual role theorists like Peacocke and Wedgwood individuate concepts by their possession conditions, they make a further claim about such conditions, namely, that to possess a concept is also to know what it is for something to be its semantic value or referent (Peacocke Reference Peacocke1992: 23; Wedgwood Reference Wedgwood2007: 81). In other words, the essential features of a concept together with the world must determine what the concept denotes or refers to. This looks like a problem for expressivists. The conceptual role theory of possession conditions was meant to provide a nonrepresentational explanation of conceptual content. But we now seem to have a view in which a concept's possession conditions determine a robustly representational content, which is what expressivists deny is true of normative concepts.

To illustrate the worry, consider Wedgwood's conceptual role theory for normative concepts. He argues that the concept of the practical ought is the unique concept to possess which an agent must meet (Reference Wedgwood2007: 97):

On the face of it, this looks exactly like the sort of conceptual role that expressivists would endorse. According to A(O), our concept of ‘ought’ is an essentially practical concept not governed by any substantive input conditions for when to accept ought-propositions. However, Wedgwood takes A(O) to determine a robust normative property as the referent of ‘ought’. Specifically, he claims that it determines the weakest property of a proposition in virtue of which it is correct for an agent to make that proposition part of her ideal plan about what to do (Reference Wedgwood2007: 100). While A(O) itself might be specified independently of ‘ought’-predications representing some property, it nonetheless determines a property as the referent of ‘ought’.

Assume for argument's sake that A(O) correctly individuates ‘ought’. If Wedgwood's determination theory for A(O) is correct, then A(O) is incompatible with expressivism. But is the determination theory correct? A number of commentators have suggested that one might accept Wedgwood's theory of possession conditions while rejecting his determination theory (Schroeter and Schroeter Reference Schroeter and Schroeter2003; Sinclair Reference Sinclair2017; Chrisman Reference Chrisman, McPherson and Plunkett2017). First, note that it is not obvious that A(O) presupposes a realist determination theory. Even Peacocke (Reference Peacocke1992: 19) allows for the possibility of antirealist determination theories. Ruling out alternatives from the start seems to beg the question. Further, it is unclear whether A(O) itself supports Wedgwood's determination theory. As Schroeter and Schroeter (Reference Schroeter and Schroeter2003: 201) note, if ‘ought’ denotes a robust property, we would expect a subject who possessed that concept to have some sensitivity to that property. However, A(O) provides no constraints about when a subject can form ought-beliefs (see also Sinclair Reference Sinclair2017: 112f.).

The real question, however, is whether expressivists can provide some alternative determination theory for normative concepts that is compatible with expressivism. To examine this prospect, it is important to see what role the determination theory is meant to play in a theory of concepts. Most centrally, the determination theory for a concept explains the correctness of the transitions specified in the possession conditions for that concept. Returning to conjunction, Peacocke (Reference Peacocke1992: 18) argues that A(C) determines as its referent the classical truth function of conjunction, realistically construed. It does so because this is what makes the transitions specified in A(C) truth-preserving. Because plans and intentions are not truth-apt, Wedgwood (Reference Wedgwood2007: 99ff.) uses a broader notion of correctness to characterize the transition specified by A(O). Details aside, what matters here is that for Wedgwood it is the property referred to by ‘ought’ that explains the correctness of the transitions specified by its conceptual role.

This means that expressivists who accept A(O) but reject Wedgwood's determination theory owe us some explanation of the correctness of the governing transitions. What are their options? One possible answer comes from Sinclair (Reference Sinclair2017: 110), who suggests that we can provide a deflationary determination theory. According to this view, we can trivially determine the truth-conditional content of thoughts in terms of satisfying the deflationary T-schema: the thought that x is F is true iff x is F. Given that expressivists typically accept deflationism of some form, this might seem attractive. Another suggestion comes from Schroeter and Schroeter (Reference Schroeter and Schroeter2003: 201f.), who suggest that the semantic value of an atomic ought-sentence might be something like the semantic value of the performative sentence ‘I hereby intend to do x’. Reformulating this suggestion at the level of concepts and mental content, we might think of this in terms of deciding to plan to do x. Something like this might look attractive to those sympathetic to Gibbard's (Reference Gibbard2003) plan expressivism.

Adequately assessing these approaches would be beyond the scope of this paper. Let me just note some problems that each account faces. First, consider the deflationary account. One worry is that it cannot explain the correctness of inferential transitions because deflated semantic notions cannot do any explanatory work. Putting the general worry aside, it is unclear how a deflationary determination theory for the concept ‘ought’ in particular could explain the correctness governing the transition from believing that ought(p) to intending to bring about p. By contrast, if we understand believing ought(p) as akin to deciding to intend to bring about p, this would plausibly explain the correctness of the transition from believing ought(p) to intending to bring about p. However, it becomes unclear how such a mental state can enter into logical relations with other mental states (though see Gibbard Reference Gibbard2012: appendix 2) or what it means for such a mental state to have a semantic value. One might also worry whether such an approach would respect the unity requirement.

These objections are neither exhaustive nor decisive. They are simply meant to illustrate the commitments incurred by expressivists wanting to adopt the conceptual role framework outlined above. Whether some version of conceptual role expressivism is ultimately viable is a debate for elsewhere. The point here is that if conceptual roles for normative concepts can be given an adequate nonrepresentational explanation, then expressivists have a suitable framework in which to explain cognitive propositions. (Likewise, if the cognitive act view is a plausible view of propositions, then expressivists have a suitable framework for thinking about normative propositions.) Given that a number of expressivists do in fact endorse some sort of conceptual role expressivism, this provides an attractive framework in which to situate an expressivist theory of cognitive propositions.

Moreover, just as conceptual role expressivism can provide support for expressivist cognitive propositions, there is reason to think that the converse holds true as well. Contemporary expressivists prefer to think of conceptual role accounts of content as metasemantic theories that explain that in virtue of which certain mental states have the content that they have (Chrisman Reference Chrisman2016, Reference Chrisman, McPherson and Plunkett2017; Köhler Reference Köhler2017). This leaves a residual question about the nature of content itself that is determined by the metasemantic theory. It is precisely here that a conceptual role metasemantics can be supplemented with a theory of cognitive propositions to explain what the contents of mental states are. Indeed, without some such account, we have not ruled out the possibility that one's nonrepresentational metasemantics determine a robustly representational content, as with Wedgwood's proposal. (Köhler opts for a novel deflationary account of content, whereas Chrisman, as far as I can tell, leaves the question unaddressed.) Thus, the two approaches to thinking about content are complementary. This point would also apply to ecumenical expressivism, in which Ridge endorses the view that expressivism is best seen as a metasemantic thesis. Indeed, Ridge (Reference Ridge2014: 222) also suggests that ecumenical expressivism would fit well with conceptual role accounts of content. The general framework should thus be compatible with various ways of implementing expressivism.