3.1. The role of that-clauses in attributions of meaning

The first step is to identify what role that-clauses play in meaning-attributions. To do so, we should consider paradigmatic uses of that-clauses in meaning-attributions. One is the use of that-clauses in the context of translation. Translation is a situation in which one faces a sentence in a foreign language and tries to assign meaning to it: one tries to determine whether two sentences in two different languages mean the same thing. In such contexts that-clauses are used as follows:

• (1) ‘Heinrich ist ein Imker’ (in German) means that Heinrich is a bee-keeper.Footnote ¹⁴

What is going on in (1) is that we mention a German sentence and then give its meaning by introducing a that-clause. Specifically, we consider a sentence in a language to be assigned meaning to – the object language –, and then assign meaning to that sentence by pairing it with a sentence in the language in which meaning is to be assigned – the meta-language. But, we do not pair those sentences in just any way. Rather, we modify the meta-language sentence using a that-clause and then assign it to the object language sentence. And, it seems exactly this modification that allows the meta-language sentence to play its meaning-giving role in (1). Neither merely mentioning nor using the sentence would straightforwardly do the job. Consequently, in this kind of situation that-clauses play purely intra-linguistic roles, as a tool for semantic ascent: they allow assignment of meta-language sentences to object language sentences in a meaning-giving way.

What goes for translation goes for interpretation generally. When interpreting what certain sentences mean in the mouth of other speakers of English, we can assign meaning in the same way. For example, when we interpret another speaker of English, call her ‘Helene, ’ we can employ that-clauses as follows:

• (2) ‘Heinrich is a bee-keeper’ (in Helene's idiolect) means that Heinrich is a bee-keeper.

Again, in (2) we consider sentences in an object language and then assign meaning to them by pairing them with a sentence modified by a that-clause in a meta-language. And, once more it is this modification that allows the meta-language sentence to play its meaning-giving role. So, in the case of interpretation that-clauses play a purely intra-linguistic role as well.

Hence, we’ve identified the role that-clauses play in meaning-attributions: they play an intra-linguistic role in translation and interpretation. We now need an account that explains how that-clauses (and meaning-attributions generally) play this role, and one on which that-clauses can play this role in some non-representational fashion.

A first important thing to note is a constraint on such accounts. Accounting for the role of that-clauses in meaning-attributions is part of meta-semantics. Specifically, it is part of an attempt to cash out the notion of ‘literal meaning.’ This notion, however, is already associated with certain properties of sentences. A suitably theory-neutral description of these properties can be given in terms of the role meaning-attributions play. Specifically, they are that sub-set M of the properties of a sentence S ₁, for which it holds that some sentence S ₂ is a good translation or interpretation of S ₁ if and only if S ₂ has M. I will call these properties the ‘meaning-explananda.’ The constraint on satisfactory accounts of meaning-attributions is that they need to say something about what accounts for the meaning-explananda.

Of course, deflationary views of that-clauses reject the common and straightforward way of doing so: introducing an entity (e.g. a proposition) into our ontology that is the meaning of the relevant kinds of sentences (and, hence, represented by the that-clause) and whose various distinctive features account for the meaning-explananda and, hence, the role of that-clauses in translation and interpretation. This would be representationalism about that-clauses, and so is an explanation that is closed to deflationary views of meaning-attributions. Still, even on deflationary accounts there should be some properties of sentences that account for the meaning-explananda, properties fully describable in terms not mentioning meanings. I will call these the ‘basic properties.’ An example for a basic property might, for example, be the property of being used in a certain way by competent speakers.

It is important to note that commitment to a view on which the meaning-explananda are fully accounted for in terms of basic properties is in principle perfectly compatible with the following commitments: First, that explanations of the meaning-explananda that mention meaning (e.g. semantics) are perfectly legitimate. Second, that such explanations are the best explanations (currently or ever) available to us (given our cognitive and other epistemic limitations, time constraints, etc.), so that it is completely legitimate to carry out semantics employing the semantic notions, and to assume such an enterprise to be the best project available to us to account for the compositionality of natural languages. Third, that we could never engage in interpretation or translation solely in terms mentioning only basic properties instead of using terms mentioning meaning. Fourth, that ordinary speakers could never fully spell out what those properties are. All that this view amounts to is that what accounts for the meaning-explananda fundamentally are facts fully describable in terms not mentioning meaning. What would a view of this kind say about meaning-attributions, though? I think a plausible approach – which can be fleshed out along deflationary lines – is Wilfrid Sellars (Reference Sellars1954) account of meaning-attributions.

It is plausible to read Sellars (e.g. Reference Sellars1954, Reference Sellars1969) as adopting a view on which the meaning-explananda can be fully accounted for in terms not mentioning meanings, namely, in terms of rules about and patterns of linguistic behaviour. However, Sellars did not doubt the legitimacy of meaning-attributions. Instead, he thought that their legitimacy could be accounted for without expanding the metaphysical commitments of his explanatory account.

On his account, meaning-attributions are illustrating sortals: ‘Means that p’ is a sortal phrase, which illustrates particular basic properties, and in this way, allows classifying sentences in accordance with those properties. What does this mean? According to this account, meaning-attributions are a tool for semantic ascent: by use of a meaning-attribution we introduce a meta-linguistic phrase (‘means that p’) that allows us to classify sentences in the object language as having certain basic properties. These are those (or relevantly similar) basic properties as possessed by the sentence in the meta-language we have transformed into a that-clause. However, meaning-attributions play this role not by explicitly telling us what the relevant basic properties are. Rather they do this by illustrating these basic properties using the meta-language sentence that has these basic properties as an example. It is the role of that-clauses in this context to pick out the basic properties of sentences. To transform a declarative sentence S into a that-clause is to transform it into a meta-linguistic predicate that picks out the basic properties that S has in our language, where S serves as an illustration for the basic properties relevant in that context. The word ‘means’ in a meaning-attribution, according to Sellars, is merely a special copula telling us that the object language sentence has the basic properties picked out by the that-clause.

Let me make this clear with an example. Take,

• (1) ‘Heinrich ist ein Imker’ (in German) means that Heinrich is a bee-keeper.

On Sellars’ account, (1) attributes certain basic properties to the German sentence ‘Heinrich ist ein Imker.’ (1) does this by using the English sentence ‘Heinrich is a bee-keeper’ as an example to illustrate the basic properties in question. Indicating that this is what the sentence is used for is the job of transforming the sentence into a that-clause. More specifically, by transforming ‘Heinrich is a bee-keeper, ’ we are introducing a predicate that serves to pick out the relevant basic properties, by using the sentence as an illustrative example for something with those properties. ‘Means, ’ then, merely attributes those properties to the German sentence.

Sellars’ account satisfies the first desideratum, as it accounts for the role of that-clauses in translation and interpretation. Whether it can also offer a deflationary account of that-clauses, though, depends on whether it can also satisfy the second desideratum, i.e. whether it is compatible with that-clauses playing their role in some non-representational fashion. So, let's investigate that issue.

3.2. Non-representationalism about that-clauses

I should start by noting that Sellars’ account is compatible with a reductionist approach, on which facts about literal meanings are nothing but facts about basic properties. Such an approach would not fit the second desideratum, as it would postulate that there are literal meanings represented by that-clauses, which it identifies with basic properties. Endorsing such a reductionist view would also not help meta-semantic expressivism's quest toward compatibility with truth-conditional semantics, because a combination of those views would imply Psychologized Semantics. After all, on an expressivist account, the most likely candidate for the basic property of a sentence is the property of expressing a certain mental state, and so if literal meanings and basic properties are the same, we are back to Psychologized Semantics.

While the Sellarsian account is compatible with reductionism, though, it also does not entail such a reduction, i.e. it does not entail that facts about literal meanings are nothing but facts about basic properties and so does, hence, not entail that the relation in which that-clauses stand to basic properties is the one that would be employed by representationalists.Footnote ¹⁵ So, the account is compatible with a non-reductionist treatment, which is not, at least on first sight, incompatible with the second desideratum. Consequently, I will here only consider the non-reductionist version of Sellars’ account, which I will call the ‘Sellarsian account.’

Before I show that the Sellarsian account satisfies the second desideratum, let me note, though, that there are actually good reasons independently of my project to not understand the account reductionistically. This is so, because of the particular relation in which that-clauses stand to basic properties, which does not seem to me well understood in terms of representation. On Sellars’ account, meaning-attributions assign basic properties to sentence S ₁, by using another sentence S ₂ as an illustrating example. So, the way in which talking about S ₂ picks out properties should be akin to the way in which taking as an example of generally picks out properties. Taking as an example, though, has peculiar features.

To take some x as an example of some F is to be in a state where certain aspects of x are salient to oneself as those features something has in virtue of being an instance of F. Assume one takes David Lewis’ papers as an example of papers with a high chance of getting published in high quality philosophy journals. In this case, certain aspects of David Lewis’ papers will be salient to one as those features that papers have in virtue of having a high chance of getting published in high quality philosophy journals.

Certain aspects of an object being ‘salient’ in this way just means that one is disposed to give a special role to those aspects in determining whether other objects are members of the same general kind (the kind the first object is being taken as an example of). This special role is to treat those aspects as a reference-point in determining whether other objects fall within that general kind. This means that one's judgements about objects being identical (or relevantly similar) to x in these aspects will have some weight in settling one's judgements with regards to whether some object falls within the relevant kind. Of course, what weight one gives to these judgements can vary across cases: sometimes, one might already have extensive and explicit knowledge of the properties that instances of a general kind have and so needs to consider examples as reference-points only to confirm one's judgements about particular cases. In other cases, however, the only way one can explicitly think about a certain general kind is via such reference-points.

What taking as an example of does, is to feed on our ability to pick up on features and aspects of objects in ways that are certainly far more extensive than our ability to make explicit, both in language and in thought, to provide resources for new and often quite economical ways of thinking and talking about those features. For example, taking as an example actually makes it easier for us to deliberate about and determine whether some object x belongs to some general kind F. As long as we can rely on certain reference-points with regards to that general kind, we can determine whether x is an F by determining whether x is identical or sufficiently similar to the reference-points in relevant respects. Second, it makes it possible (via its expression or the invitation to others to come to be in that state) to communicate to others that some x belongs to a general kind F, even when we cannot spell out the features shared by instances of F or the features of x in virtue of which it is an instance of F. The only thing required is that one can presuppose a common reference-point, or give others a common reference-point by pointing out an object in which, for example, the relevant features are easier to pick out than in x. Proceeding from there one can then rely on one's audience's abilities to pick out similarities to lead them to the conclusion that x is indeed an F.

However, if taking as an example of functions in the ways presented here, it will not be appropriate to think that it represents the properties that it picks out. The relationship that holds between instances of taking as an example of and the properties picked out by its instances is importantly different from the relationship between ordinary descriptive beliefs and the properties they pick out. If this is correct, this provides a very good reason to read Sellars’ account in a non-reductive way. After all, on this account uses of that-clauses are just a particular instance of taking as an example of. With these remarks out of the way, let me now turn to the question whether the Sellarsian account satisfies the second desideratum.

It turns out that it does: the account mentions nowhere entities in a way that would play into a representationalist's hand. The only entities mentioned on the Sellarsian account are basic properties. However, the relation in which that-clauses stand to these properties forecloses a reduction of that-clauses in terms of those properties. So, that-clauses do not represent basic properties. But, the Sellarsian account makes no mention of any other entities either! Consequently, while we can give an account of that-clauses in terms of their function, it will not be the function a representationalist about that-clauses would appeal to. According to the Sellarsian account, the best explanation of that-clauses makes no reference to that-clauses functioning to represent entities.

Note also, that on the Sellarsian account, there will be no informative analysis of that-clauses. We can only characterize that-clauses in a relatively platitudinous way. The same goes for the verb ‘means’ in the context of meaning-attributions. While ‘means’ functions in particular ways in a meaning-attribution, ‘means’ will not be analysable in terms of this function. So, according to the Sellarsian account, there will not be much of an informative answer to the question what it is that is assigned to ‘Heinrich ist ein Imker’ in

• (1) ‘Heinrich ist ein Imker, ’ means that Heinrich is a beekeeper.

except for the answer that that Heinrich is a beekeeper is assigned to ‘Heinrich ist ein Imker’ as its meaning.

This means, of course, that the Sellarsian account of that-clauses is a deflationary view about that-clauses. What does the view say about propositional contents, though? Note first, that the Sellarsian view is compatible with that-clauses legitimately functioning syntactically as referential terms. On this account, that-clauses pick out basic properties by using some sentence as an example for something with those (or relevantly similar) properties. What will be most salient in the context of using a that-clause, consequently, are those features of sentences relevant for its having particular basic properties. Thus, that-clauses will behave in a way that is structurally isomorphic to the basic properties of the sentence following the ‘that.’ This means, though, that that-clauses will behave, linguistically, like referential singular terms that refer to something that supervenes on, but does not reduce to, basic properties. Consequently, the Sellarsian account offers an explanation of why that-clauses function syntactically this way, even though the explanation of why that-clauses are in our vocabulary makes no reference to such entities.

If that-clauses behave in this way, though, we can introduce other ways of talking about what that-clauses refer to.Footnote ¹⁶ In particular, we can use the label ‘propositional contents’ to refer to what is attributed by that-clauses. So, for example, rather than saying

we might now, at least in some contexts, say

It is important to note, however, that the metaphysical commitments of claims made by sentences such as (P) will not go beyond those undertaken by claims invoking sentences such as (M). For example, the way in which sentences like (P) figure in giving the meaning of sentences must be derivative from the way sentences such as (M) do so. In fact, it does not seem too far-fetched to say that in contexts in which it is appropriate to use (P), it simply provides alternative ways of providing information already provided by (M). Of course, this is compatible with sentences such as (P) being more precise or expressively more powerful than simple meaning-attributions. This is especially the case when they employ formal machinery such as sets of possible worlds to make the notion of a proposition more precise. However, even if such sentences do the work of ordinary meaning-attributions much better, they do not convey information that is of a different type to the information already conveyed by meaning-attributions.

What should we say about the status of these ‘propositions’? I think the following stance is most reasonable: talk of such contents is extremely useful and harmless in many cases, but we should be cautious not to read any metaphysical implications into it. This view fits nicely with viewing semantics and the invocation of propositional contents in that context as a theoretical modelling exercise set up specifically to capture very specific features of sentences in natural languages. Models can, in many cases, plausibly be thought of as useful fictions and the approach to propositional contents suggested here fits this approach nicely.

With the Sellarsian account of that-clauses, we now have a deflationary account of that-clauses that allows talk of propositional contents.Footnote ¹⁷ Furthermore, this account has no particular commitments that would rule out that that-clauses or propositional contents could be attributed in a meaning-giving way to normative sentences even if expressivism was true. Hence, it seems to fit our third requirement as well. Whether it does, however, still needs to be investigated more explicitly. After all, the Sellarsian account has particular commitments about meaning, and it is unclear how these relate to expressivism's commitments about meaning-constituting properties.

3.3. The fit with meta-semantic expressivism

On the Sellarsian account, meaning-attributions are a tool to pick out basic properties and assign them to sentences without needing to be able to specify them. If we accept this account of meaning-attributions, though, what is the relation between sentences’ meaning, their basic properties, and their meaning-constituting properties?

On the Sellarsian account, a sentence's meaning does not reduce to its basic properties. Nor does it reduce to anything fully describable in terms not mentioning meaning. However, the sentence's meaning supervenes on its basic properties. This makes these properties the most plausible candidate for those properties in virtue of which sentences have their meaning. Consequently, the relationship between sentences’ meaning, their basic properties, and their meaning-constituting properties is that sentences’ basic properties just are their meaning-constituting properties. In what follows, I will, consequently, use ‘basic properties’ and ‘meaning-constituting properties’ interchangeably.

This identity between the basic and the meaning-constituting properties is good news, as it makes clear how the Sellarsian account combines with meta-semantic expressivism. On the combination of those two, the meaning-explananda of declarative sentences would be fully accounted for in terms of the mental states that uses of those sentences express. That-clauses in meaning-attributions would then have the function of picking out the mental states expressed by the use of those sentences followed by the that-. And, meaning-attributions would serve to use that-clauses to assign the property of expressing a mental state to sentences. However, the meaning of a declarative sentence, on this account, would consist in a propositional content, where this is to be understood along deflationary lines presented here. Furthermore, the sentence will have this propositional content in virtue of expressing a mental state, so that the property of expressing a mental state will be the meaning-constituting property of that sentence. Of course, this raises a couple of further issues, which I will address for a clearer view of the picture on the table.

According to expressivism, the meaning-constituting properties involve mental states. On the traditional understanding of expressivism, these will themselves have propositional contents, but not necessarily the same propositional contents as the sentences expressing them. For example, if expressivists held – the toy-view – that

• (3) Abortion is morally wrong

expresses a desire that no one commit an abortion, (3) would have the content that abortion is morally wrong, but this content would supervene on a desire with the content that no one commit an abortion. Even though this kind of structure might appear strange, let me note that it is not necessarily so. Michael Ridge (Reference Ridge2014, 125–131), for example, has argued that normative sentences express beliefs that share the content of the sentence, but that these beliefs are constituted by a non-normative belief and desire pair that have different contents from the normative belief and the sentence.

On Ridge's account, however, the same notion of content figures in all of these cases, which is made possible by his particular account of propositions. One might ask, though, whether expressivists who use the Sellarsian approach instead are not committed to using two different notions of content: one for linguistic items and one for mental states. After all, on the above picture, the mental content does significant theoretical work that seems foreclosed by the deflationary approach to content suggested here. This indicates that expressivists cannot use the Sellarian account to account for both linguistic and mental content.

It is unclear how devastating this would actually be. After all, the best accounts of mental and linguistic content might not converge. It seems to me, however, that there is actually a feasible alternative to biting the bullet on this. This is to reject the assumption that for expressivists’ theoretical purposes their commitments in the philosophy of mind require invoking mental content. This is a common way to understand expressivism – after all, expressivism has traditionally been understood as the view that normative judgements are conative attitudes –, but it is not forced on expressivists. Indeed, it seems that what should do the work of the meaning-constituting properties should be the same as that which explains in virtue of what normative judgements have their contents. After all, most contemporary expressivists agree that normative judgements are beliefs and so they will need an account that explains in virtue of what these beliefs have their contents. While they can invoke other propositional attitudes here, I see no reason why they need to. Most fundamentally, what accounts for the content of all mental states should not be something that would itself be described in terms of contents anyway, so why not invoke it to account for the contents of normative beliefs directly, if this is possible?

In fact, there is a perfectly viable approach in the philosophy of mind for accounting for mental content that would allow cashing out expressivism's second thesis, without using a notion of content that is separate from the Sellarsian account. Before I present the view, I should highlight that the approach departs in some significant ways from traditional conceptions of expressivism. In particular, it undergoes no commitment to the view that normative judgements are partially constituted by familiar conative attitudes such as desires, emotions, or intentions. Rather, the view suggests a different way of cashing out expressivism's commitments about normative judgements, though one fully in line with the motivations for expressivism.

The approach I have in mind uses a certain school within functionalism about the mind. According to functionalism about the mind, mental states are dispositional states fundamentally characterized by their causal role within a mental economy. Functionalism itself is compatible with a notion of content figuring irreducibly in the functional characterization of mental states. However, there is a school within functionalism according to which, on the most fundamental level, propositional attitudes can be fully characterized without making reference to propositional content. This is conceptual role semantics.Footnote ¹⁸ According to conceptual role semantics, mental states can be fully characterized by their role within a mental economy. On this view, the features of mental states we would normally account for by invoking mental content can be fully integrated into that state's functional role (which then explains in virtue of what the state has its content).

Using conceptual role semantics, expressivists can plausibly cash out their second thesis purely in terms of functional roles.Footnote ¹⁹ For these purposes, we can first notice that the functional role of a state can be helpfully distinguished into three parts. First, ‘mind-entry’ conditions, which specify the state's role in the procession of sensory stimuli. Second, ‘mind-to-mind’ conditions, which specify the state's role in the processes of deliberation, reasoning, and inference. Third, ‘mind-exit’ conditions, which specify the state's role in the production of bodily behaviour.

Using this distinction, expressivists can suggest that normative judgements are characterized by very minimal mind-entry conditions, mind-to-mind conditions just sufficiently robust to account for the role normative judgements play in reasoning and deliberation, and very robust mind-exit conditions. She might then contrast this with the role that characterizes paradigmatically representational mental states, which would involve significantly robust mind-entry conditions, mind-to-mind conditions sufficiently robust to account for the role these judgements play in reasoning and deliberation, but very limited mind-exit conditions.

If expressivists endorse this view, their commitments in the philosophy of mind and, hence, their view of the meaning-constituting properties, would not require them to invoke a separate notion of content from that provided by the Sellarsian account. On this view, it is the functional role of the mental state (or at least a part of it) expressed by a sentence that figures in its meaning-constituting properties. Combined with the Sellarsian view, we get the result that it is this functional role that is illustrated and attributed by that-clauses and on which the content of that sentence supervenes. No additional notion of content is required. Of course, this does not tell us how the account can be generalized to mental content and so does not tell us whether expressivists can use the same account for both semantic and mental content.Footnote ²⁰ The suggestions here suffice to show, however, that such an approach cannot be ruled out solely based on expressivism's meta-ethical commitments. The worry that expressivists need two separate notions of content, if they endorse the Sellarsian account does not, consequently, hold water.