3.1. Metaphysical Problems with Platonic Propositions

Regarding the contents of utterances and thoughts as abstract objects raises difficult questions. A crucial one is: how can they be the contents of concrete utterances and thoughts they cannot interact with, given the causal closure of the physical world? As Devitt argues, “because [Platonic] propositions and the like can play no causal role in mind and language, we have the best of reasons for thinking that they are not part of mental and linguistic reality” (1996, p. 210).

Devitt stresses that the contents of utterances and thoughts play causal roles in the spatio-temporal world. People behave the way they do because of the contents of their beliefs, desires and other mental states. People’s mental states and behavior are in turn influenced by the contents of other people’s utterances and thoughts. Suppose that Mark believes that it is raining and shares his belief with Nancy by uttering “It is raining”. The content of Mark’s belief plays a causal role in the explanations of why he picks up an umbrella and of why he utters that sentence, while the content of his utterance plays a causal role in the explanations of how Nancy comes to believe that it is raining-gaining information about the weather if Mark is reliable-and of why she also picks up an umbrella (^{Devitt 2013, p. 138}).

Crucially, our folk-psychological explanations of such facts rely on ascribing contents to people’s thoughts and utterances. As Devitt emphasizes, two of the main purposes for which we use ordinary content ascriptions are to explain and predict people’s behavior and to account for how their utterances and mental states serve as guides to reality (^{1996, pp. 57-60}). According to Devitt, these ascriptions succeed to the extent that they track contents which constitute causally efficacious features of the world. So, there are compelling reasons to regard contents as worldly natural phenomena:

A commitment to [...] Platonic propositions would be appropriate [...] only if these propositions do real explanatory work. And it is hard to see how entities that are outside space-time could do any such work. Thoughts and the utterances that express them are parts of the natural physical world. How, then, could they be related to entities outside space-time? Even if they could, how could their being so related do any work? Our task is to ascribe properties to thoughts and utterances that can explain their causal role in the spatio-temporal world. Relations to objects outside the causal order surely could not do this. (^{2013, pp. 140-141})

The contents of utterances and thoughts play causal roles in the spatio-temporal world that Platonic propositions cannot play. So, the contents of utterances and thoughts cannot be Platonic propositions.

Since ascriptions are indispensable to our folk-psychological explanations of behavior and cognition, and since those explanations are often successful, there is strong evidence-as Devitt argues- that thoughts and utterances really do possess the kinds of contents we ascribe to them (^{1996, p. 71}). But if those contents are real and causally efficacious, they cannot be Platonic propositions-and so the correct account of true content ascriptions cannot treat them as about such objects. (Recall that the existence of propositions requires not just SA, but also true ascriptions that actually refer to or quantify over them.) This implies that ascriptions must be able to be either literally true and about worldly semantic properties, or at least non-literally convey truths about such properties. The challenge is to provide an account of ascriptions that does justice to their explanatory utility and to the naturalistic constraint that contents must be part of the causal fabric of the world.

3.2. A Predicative Analysis of Particular Ascriptions

Motivated by his metaphysical considerations, Devitt suggests an alternative analysis according to which TCs do not refer to propositions. He proposes that particular ascriptions have the same form as “A loves some B”, instead of “A loves B”, so that TCs function as indefinite singular terms rather than referential ones (^{1996, p. 56}; ^{2013, p. 149}). According to DA, “S v’s that p” asserts a relation between S and some object-a token thought or a token utterance, depending on the attitude verb “v”-which has the representational property or content specified by the predicative TC (^{1996, p. 56}; ^{2013, p. 149}).² For example, “Nancy believes that it will rain” would be equivalent to “Nancy has a belief which has the content that it will rain”. A natural interpretation of Devitt’s analogy with “A loves some B” is that the TC functions as the monadic predicate “x hasthe-content-that-p”-though he does not explicitly formulate it in these terms.

But Devitt also proposes that the TC specifies a semantic property by means of synonymy or sameness of content with the sentence it contains:

The expression ‘that p’, used to ascribe a putative meaning, seems to ascribe that property in virtue of the fact that its content sentence, ‘p’, has the property or one very like it. So there always seems to be an “intimate link” between the two properties. However [...] we can abstract from this link, treating [...] ‘that p’ like any arbitrary expression for a property. (^{1996, p. 82})

This suggests that a particular attitude ascription is true if and only if some thought or utterance of S has the same content as “p”.³ Devitt argues that the “intimate link” is usually one of identity between the content of the ascribed token thought or utterance and that of the content sentence token (^{1996, pp. 116, 121, 123})-where the spatio-temporally located tokens involved are meaningful tokens, not meaningless tokens stripped of their semantic properties: “By talking of meaningful tokens, we can avoid talk of ‘propositions’” (^{1996, p. 13}). This proposal treats the TC as the dyadic predicate “x has-thesame-content-as y”, where the second place is ocuppied by a singular term for the content sentence token. For example, “Nancy believes that it will rain” would be true just in case Nancy has a belief that has the same content as the token of “it will rain” embedded in the ascription.⁴

Devitt proposes two versions of DA: a monadic one (DA _M ), which gives a coarse-grained account of the predicative role of TCs, and a dyadic one (DA _D ), which analyzes how TCs specify representational properties. Although Devitt doesn’t explicitly mark the distinction, his appeal to synonymy points toward DA _D , while his analogy with “A loves some B” and his claim that we can abstract away from the intimate link support DA _M .

Part of the appeal of SA stems from particular ascriptions seeming to have the form Rab. But, as Devitt notes, “an attitude ascription no more appears to have the logical form of “Iago kissed Desdemona” than it appears to have that of “Iago kissed a woman” and, if DA is correct, ascriptions do not “commit us to [Platonic] propositions” but to “mental states [and utterances] with representational properties” (^{Devitt 2013, p. 149}). Devitt argues that DA is as linguistically plausible as SA, and has the added metaphysical advantage of treating the contents ascribed as properties of concrete utterances and thoughts, rather than as abstract objects. Given that true ascriptions specify contents that play causal roles in the spatio-temporal world, Devitt’s analysis offers a plausible way to understand the ascribed contents as worldly phenomena that satisfy this naturalistic constraint.⁵

The contents ascribed uncontroversially include truth conditions. For Devitt, however, these are not inherited from abstract propositions but are worldly properties had independently by token utterances and thoughts, including the content sentence tokens embedded in ascriptions. Devitt further argues that the properties ascribed include modes of reference, on the grounds that meaningful tokens with the same truth conditions can differ in their explanatory roles. A person may act on the belief expressed by “Mark Twain is at the Town Hall” but not on the belief expressed by “Samuel Clemens is at the Town Hall”, even though “Mark Twain” and “Samuel Clemens” co-refer. ^{Devitt (2020}, ²⁰¹², ²⁰¹⁵⁾ takes such differences as evidence that modes of reference expressed in content ascriptions help account for action, and are therefore among the semantic properties ascribed.⁶

3.3. General Ascriptions without Propositions

General ascriptions like (3) and (4) below seem to quantify over propositions. But, as ^{Devitt (1996, pp. 212-213)} points out, paraphrases that serve the same purposes and clearly do not quantify over propositions are “readily available”. For example, (3) and (4) can be paraphrased as (3) and (4), respectively:

(3’) and (4’) quantify over meaningful tokens. Yet they serve the same purposes as (3) and (4). For example, (3’) conveys the same information as (3) about the shared content of Copernicus’ and Galileo’s beliefs. (4’) likewise serves the same explanatory purpose as (4) in accounting for Nancy’s picking up an umbrella after hearing Mary utter “It will rain”-although it better serves this purpose if (4) is read as quantifying over Platonic propositions, which cannot play a causal role in such an explanation. Notice that the paraphrases predicate that pairs of meaningful tokens have the same content. This neatly matches DA _D ’s reliance on synonymy between tokens to explain how particular ascriptions specify contents.

Despite his non-propositionalist analysis of particular ascriptions, Devitt concedes that general ascriptions like (3) and (4) quantify over propositions-and thus regards them as literally false (^{1996, p. 213}), given the metaphysical considerations discussed above. Yet because such general ascriptions appear “explanatorily useful” and “generally successful”, he suggests that they are mere manners of speaking that can be replaced “in the Quinean way” by paraphrases that serve the same purposes “without the objectionable commitment” (^{1996, p. 213})-that is, without quantifying over propositions, which cannot play the causal roles in virtue of which the ascriptions are explanatorily useful. The original sentences and their paraphrases are equivalent “in some sense” but cannot be synonymous, since they differ in their ontological commitments (^{Devitt 2010, p. 25}). Stephen ^{Yablo (1998, p. 248n)} makes a useful suggestion for such cases: “a paraphrase of S expresses in literal terms what S says metaphorically”, which explains how a “paraphrase ‘says the same’ as what it paraphrases” when their literal meanings differ. This is compatible with the Devittian analysis of general ascriptions (DA _G ), which we can thus characterize as follows: paraphrases like (3) and (4) express literally what (3) and (4) convey figuratively. DA _G shows how general ascriptions can convey truths about worldly semantic properties, thereby preserving their causal-explanatory role in accounting for behavior and cognition.

4.1. Semantic Substitution Failures

If “that p” refers to a proposition, it should be replaceable by “the proposition that p” without a change in meaning. But such substitution often fails. (5a) cannot be paraphrased as (5b):

(5b) reports an odd attitude that clearly differs from the one reported by (5a). Similar failures occur with “expect”, “mention”, and “remember”. Such failures seem to provide evidence against SA’s claim that TCs are singular terms referring to propositions (^{Prior 1971}; ^{Moltmann 2020}; ²⁰¹³; ²⁰⁰³; ^{Hofweber 2016}; ^{Matthews 2022}).

A variant of the problem arises in general ascriptions. If the quantifiers in the relevant contexts range over propositions, they should be replaceable with quantifiers explicitly restricted to propositions, again without a change in meaning. But consider:

In the context of an inference from (5a) to (6a), (6a) cannot be paraphrased as (6b), which does not have the same truth conditions and is not entailed by (5a). These failures seem to provide evidence against SA’s treatment of quantifiers as ranging over propositions (^{Prior 1971, p. 16}; ^{D’Ambrosio 2023}, ²⁰²¹).⁷

In defense of SA, Jeffrey ^{King (2007}, ch. 5) proposes that verbs like “fear” are polysemous: a TC triggers a “content relation” reading, while a noun phrase (NP) like “the proposition that p” triggers an “object relation”-even though they co-refer. But this undermines the inference from (5a) to (6a), since the NP “something” would trigger a different reading than the TC (^{Hofweber 2016, p. 218}).⁸ A further problem is that no ambiguity or zeugma arises in sentences like “Nancy fears the surgery and that it will be expensive”, even though they combine both an NP and a TC as complements (^{King 2007, p. 219}; ^{Nebel 2019}, p. 77). Moreover, King’s proposal seemingly fails to solve the quantificational variant, since the complements in (6a) and (6b) are both NPs (^{D’Ambrosio 2021, section 3}).

Jacob ^{Nebel (2019)} proposes that SA is right about TCs referring to propositions, but that “the proposition that p” refers instead to a function from situations to propositions-thus explaining substitution failure. As I see it, “the proposition that p” plainly purports to refer to a proposition, just like “the rumor that p” purports to refer to a rumor. As ^{Nebel (2019, p. 90)} himself notes, however, his reasoning entails that “the rumor that p” cannot refer to a rumor either! The view also implies that sentences like “The rumor that p surprised/hurt S” can never be true, since functions are causally inert (^{Güngör 2022}). Furthermore, it does not seem to solve the quantificational variant, since (5a) would license (6a) as quantifying over propositions, and thus (6b) (^{D’Ambrosio 2023, section 4}).⁹

4.2. Marks of Concreteness

Further doubts about SA arise from a range of linguistic phenomena-including the kinds of predicates that quantifiers and attitude nouns naturally take, the entities these nouns appear to refer to, and the syntactic role of TCs in noun constructions.

Consider “John thought something daring” and “Nancy said something nice”. In both cases, the quantifier appears to range over concrete contentful entities-such as a token thought or a particular remark-that are described as daring or nice, which are not properties plausibly predicated of propositions (^{Moltmann 2013, pp. 142- 143}). Other examples involve predicates like “unusual”, “surprising”, “persistent”, or “unwelcome”. This suggests that general ascriptions may quantify over meaningful tokens-such as thoughts, fears, remarks, or claims-rather than propositions.

Attitude nouns such as “belief”, “claim”, and “hope” appear to refer directly to such entities. They combine naturally with similar predicates: “John’s belief is unusual”, “Mary’s claim was mean”, “Sarah’s hope is persistent”. Again, these predicates are not plausibly interpreted as applying to abstract propositions, but to concrete mental states or speech acts with content (^{Moltmann 2013, p. 136}).

Moreover, in expressions like “S’s belief/claim/fear/hope that p”, the TC is optional-which suggests that it may not function as an argument (^{van Elswyk 2023, p. 64}). A possible reply is to treat the TC as a singular term in apposition-like the name in “my aunt Susy”-so that, even if it is not an argument, it can still be taken to refer to a proposition. But this would require attitude nouns also to refer to propositions, since terms in apposition co-refer-which conflicts with the kinds of predicates these nouns naturally take.

Finally, it is noteworthy that verbal and nominal ascriptions alternate in ways that preserve meaning. “Nancy believes that it will rain” has noun-based counterparts like “Nancy has the belief that it will rain” and “Nancy’s belief is that it will rain”. Similarly, “Nancy claimed that it will rain” alternates with “Nancy made the claim that it will rain” and “Nancy’s claim was that it will rain”. In the nominal ascriptions, the TC does not seem to function as an argument and is unlikely to be a singular term for a proposition-if the above considerations are correct. But the verbal and nominal variants appear closely matched in meaning (^{Moltmann 2017, section 3.3.2}; ^{Matthews 2022, section 3}), suggesting that, perhaps, the verbal variants do not involve referential TCs in argument positions either.

6.1. Propositionalist Replies to the Explanatory Argument

Advocates of Platonic propositions often contend that their opponents are motivated by a general and unjustified aversion to abstract objects as either mysterious, unknowable, or lacking adequate identity conditions (^{Bealer 1998, pp. 1-2}; ^{Schiffer 2008, pp. 283-284}; ^{Horwich 1998, p. 90}; ^{2010, p. 33}). But even if we grant that positing abstract objects can be justified in some cases-such as the case of sets or numbers-such objects do not play causal roles in the spatio-temporal world. Indeed, most contemporary Platonists agree that abstract objects are causally inert. Yet, advocates of Platonic propositions often do not even address the issue of the causal roles of contents when defending the abstractness of propositions. In his survey of the main objections to abstract propositions, ^{Schiffer (2008, pp. 283-284)} considers only one concern related to causality: the epistemological worry that we could not know about abstracta if we cannot causally interact with them. He rejects the causal theory of knowledge that motivates this concern and concludes it is unproblematic that propositions are causally inert. The same point is made by ^{Horwich (1998, p. 90)}. This does not address the concern about the causal roles of contents in the explanation of behavior. The strongest case against Platonic propositions is often simply ignored.

In fairness, ^{Schiffer (2003)} does discuss the issue, as I point out below.

One possible reply is to deny that contents, understood as abstract propositions, are causally inert. But this conflicts with the widely accepted principle of the causal closure of the physical world, which holds that every physical effect has a fully sufficient physical cause. This principle is supported by physical science, especially by laws like the conservation of energy, which appear to exclude non-physical causes. Since abstract objects are not physical, allowing them to play causal roles would require giving up this principle-a move most philosophers, including Platonists, reject.¹²

A second reply is to reject the key premise that contents play genuine explanatory roles in behavior and cognition. But this move undermines the main reason for positing contents at all-namely, the explanatory success of belief-desire psychology. If contents are not doing explanatory work, then it is unclear why we should be realists rather than eliminativists about them-whether they are conceived as abstract propositions or otherwise. Indeed, the main argument against eliminativism relies on the success of “propositional” attitude explanations (^{Fodor 1987}; ^{Devitt 1996, p. 254}).

A third and more sophisticated reply grants that contents play an essential explanatory role, but denies that they must therefore be causally efficacious. On this view, contents are like numbers in physics: abstract entities that help structure our theories and support explanations, even if they do not cause anything. ^{Schiffer (2003, ch. 8)}, for example, defends a version of this view. Notice that this position is compatible with contentful inner states playing causal roles. What it denies is not the causal efficacy of such states, but the causal role of their contents-taken to be abstract. This distinction is important to keep in mind. Even if we accept that some abstract entities can figure in successful explanations without having causal powers, however, it remains unclear why we should regard them as real rather than as useful fictions. Fictional or metaphorical constructs can aid explanation-for instance, we often talk about the “average taxpayer”, or treat bodies as if all their mass were located at a single point-their center of gravity. In such cases, the explanatory utility of the construct does not by itself imply that the entity is real. As ^{Yablo (1998, p. 259n)} suggests, one of the marks of metaphorical or fictional talk is that it is “oftentimes paraphrasable away with no felt loss of subject matter”.

This point is central to debates over abstract entities. Hartry ^{Field (1980)}, for instance, attempted to paraphrase away references to numbers from physical theory, arguing that mathematical posits might be useful but dispensable. Although many doubt that his project can ultimately succeed, the case of propositions is comparatively weaker than the case for numbers. As shown in section 3.2 and section 3.3, both particular and general content ascriptions that are often taken to refer to or quantify over propositions can be paraphrased into variants that are clearly not about propositions, without any apparent loss of subject matter or explanatory power. These paraphrases preserve the explanatory roles of contents while avoiding commitment to abstract entities. While some have argued that numbers and propositions may be useful fictions even if they cannot be paraphrased away (^{Balaguer 1998a}; ^1998b), the case of propositions does not require such a concession. If ordinary ascriptions do happen to involve reference to or quantification over propositions, their paraphrasability- and the fact that propositions make no causal contribution-at the very least suggests the plausibility of a fictionalist reading, on which propositions are useful but avoidable posits, not genuine explanatory entities.

These considerations do not amount to a knock-down argument against propositionalism, but they show that Platonic realism is not the only reasonable view-and that there are good reasons to explore alternative frameworks. Moreover, as I explain in section 6.2, Platonic theories face a further metaphysical difficulty: the very idea that concrete thoughts and utterances stand in relations to abstract propositions gives rise to a regress that such theories are ill-equipped to resolve.

6.2. Shared Contents and Abstract Objects

The idea that different utterances and thoughts can share a single content is often taken to support the view that contents are abstract propositions-entities related to each token as its shared meaning. This line of reasoning exemplifies a classical “One over Many” pattern: when several particulars appear to share a property, we are invited to posit a relation to a single abstract object to account for what they have in common. As I have argued elsewhere (^{Withrington 2023}), this reasoning takes for granted a controversial Platonic or transcendent realist framework and inherits its familiar problem-a regress of relations (^{Armstrong 1978}).

The problem is this. If a token utterance or thought has a given content in virtue of bearing a relation to an abstract proposition, then that relation must itself be part of the explanation. But in virtue of what does the token bear that relation to that proposition? Many tokens are said to stand in the same relation to the same abstract proposition, so we again confront a One over Many: how can these be multiple instances of the same relation? The Platonist must appeal to a second, higher-order relation to explain the first-and so on ad infinitum. The explanatory strategy of grounding shared contents in relations to abstract objects appears to generate a vicious regress- undermining metaphysical explanation, rather than securing it.

Alternative approaches aim to avoid this outcome. Immanent realism holds that shared properties are instantiated universals wholly present in each instance; ^{Armstrong (1978)} develops a non-relational version of this view to block the regress. Trope nominalism avoids transcendent entities by construing properties as particularized tropes, and some authors (e.g., ^{Maurin 2002}) have explored whether resemblance among tropes can ground apparent commonalities without triggering a similar regress. Whether such strategies ultimately succeed remains an open question. If they do not, this may suggest that the very demand for a unifying entity is misguided. Perhaps there is no need to posit anything beyond the fact that each instance has the property it has-e.g., each thought and utterance that p has the property of representing that p. This points to a better explanatory strategy: focusing on what it is for a token to have a property. In some cases, this requires explanation: when the property is not fundamental, it may be reduced to more basic properties (as being water is explained by being composed of H₂O molecules). But once we reach fundamental properties-those not further reducible-there is no metaphysical mystery about what it is for something to have them. If two tokens share such a property, that simply means each has it independently; there is nothing more to explain. As ^{Quine (1961)} and ^{Devitt (2010)} argue, the “One over Many” problem dissolves once we reject the assumption that shared predications must be explained by positing common metaphysical entities: abstract, immanent, or of any kind.

Devitt himself has not applied this strategy to the case of contents, but doing so aligns with his Quinean nominalism about properties. More broadly, any non-Platonic approach to shared contents treats them as worldly features of concrete thoughts and utterances. On such views, content properties are not relations to abstract objects but concrete features of individual tokens-offering the best prospects for avoiding both the regress of relations and the causal argument.

Of course, representational properties are not fundamental. What we seek is not merely to reject abstracta, but to understand what kinds of worldly features constitute contents. A naturalized semantics aims to provide such an explanation-one that identifies the underlying non-semantic properties in virtue of which a token representation has the content it does. Different proposals appeal to the informational or causal links of representations (^{Fodor 1990}), their functional or inferential roles (^{Horwich 2005}), or their etiological functions (^{Millikan 2017}; ^{Neander 2017}). This is not the place to engage with the issue, but it is worth noting that several promising proposals have been developed in recent years.¹³

7.1. Making Explicit the Logical Forms

To assess the adequacy of Devitt’s analyses, we must first make explicit the logical forms they imply. In the case of particular content ascriptions, Devitt offers two proposals.

The monadic analysis draws directly on the analogy between particular ascriptions and “A loves some B”. The logical form of the latter is the existentially quantified ∃x(Lax∧Bx), where B functions as a monadic predicate. Similarly, DA _M treats the TC as a one-place predicate expressing a content property. For instance, the logical forms of “S believes that p” and “S said that p” are ∃x(Bxs∧Px) and ∃x(Uxs∧Px)-where ∃x ranges over thought or utterance tokens, Bxy and Uxy symbolize “x is-a-token-belief-of y” and “x isa-token-assertive-utterance-of y”, respectively, and Px symbolizes “x has-the-content-that-p”.

On the dyadic analysis, the content is specified by synonymy with the content sentence token. According to DA _D , the logical forms capturing the truth-conditions of “S believes that p” and “S said that p” are ∃x(Bxs∧SCxp and ∃x(Uxs∧SCxp-where SCxy symbolizes “x has-the-same-content-as y” and its second place is occupied by a singular term for the content sentence token (not for a proposition or a sentence type), symbolized by enclosing it in guillemet quotation marks.

For general ascriptions, Devitt proposes paraphrases that avoid quantification over propositions by quantifying instead over meaningful tokens and predicating sameness of content among them. According to DA _G , “There is something S and R believe” is paraphrased as “Some belief of S has the same content as some belief of R”, with the logical form ∃x∃yBxs∧Byr∧SCxy, while “S believes everything R says” becomes “If R assertively utters any sentence, S has a belief with the same content”, with the form ∀xUxr→∃yBys∧SCxy-where all variables range over meaningful tokens.

Devitt does not articulate these forms explicitly, but they follow naturally from his paraphrases and ontological commitments. They aim to show how content ascriptions can be analyzed in terms of the semantic properties of concrete tokens, without invoking abstract propositions. Whether these forms succeed in avoiding propositional commitments-or merely disguise them-is the next question.

7.2. Predicative Propositionalism, Synonymy, and the Concealment Problem

A monadic predicational analysis like DA _M may appear to avoid commitment to propositions. However, this appearance can be decep tive.¹⁴ Even if the TC functions as a predicate, the property it predicates may still involve a relation to a proposition. This raises what I call the Concealment Problem: the possibility that reference to propositions is merely hidden behind the surface form of a predication.

Consider an analysis according to which “S believes that p” asserts that there is some x such that x is a belief of S and x has as its content the proposition that p. This analysis has the form ∃xBxs∧Cxp- where Cxy symbolizes “x has y as its content” and p refers to a proposition. This is a predicational version of propositionalism (PP) that abandons SA and differs from DA _D only in the nature of its second dyadic conjunct. While DA _D appeals to synonymy with the content sentence token, PP predicates a relation to a proposition. Notice that we could abstract away from the predicated relation by treating “x has-as-its-content-the-proposition-that-p” as a monadic predicate Px. The resulting logical form of the ascription would be ∃xBxs∧Px-the same form ascribed by DA _M !

A serious concern is that by treating TCs as monadic predicates, a non-propositional analysis may only mask reference to propositions. Because of this Concealment Problem, DA _M does not suffice to avoid commitment to propositions. The availability of DA _D to provide a non-propositional account of how the TC specifies a semantic property is crucial. While PP suggests that the semantic property is a relation to a proposition, DA _D proposes that it is specified by a relation of synonymy with the content sentence token. The risk extends to any predicational analysis that purports to avoid propositions but is coarse-grained or monadic.

Tobias ^{Rosefeldt (2008, p. 320)} discusses PP, and Hüseyin Güngör recommends it as an alternative to SA: “that-clauses are not singular terms that refer to propositions-they are predicates denoting relations between contentful entities and propositions” (2022, section 5).¹⁵ Yet the difficulty this view poses for non-propositionalist analyses is rarely addressed explicitly.

Devitt suggests that we can “abstract from” the “intimate link” between the content ascribed and that of the content sentence token. But this is mistaken: such abstraction leads to a monadic predicate that no longer guarantees avoidance of propositional commitments.

His genuinely non-propositional analysis of particular ascriptions is thus DA _D , not DA _M .

Finally, it may be objected that DA _D and DA _G do not truly avoid propositions, since they rely on synonymy and sameness of content presupposes a relation to a proposition. This would amount to another form of the Concealment Problem. As argued in section 6.2, however, once we reject the problematic assumptions underlying this One over Many line of reasoning, synonymy can be treated as a worldly relation among tokens, not one mediated by an abstract object.

7.3. Accounting for the Inferences

DA owes an account of inferences like (1) and (2). I here show that their validity can be accounted for by combining DA _D with DA _G . (I set aside DA _M , since it faces the Concealment Problem.) DA suggests the following paraphrases of (1) and (2):

(1’) and (2’) seem intuitively valid. But to account for their formal validity, two issues must be addressed.

One issue is that the inferences rely on the fact that having the same content is an equivalence relation: it is symmetric, transitive and reflexive. When translating (1’) and (2’) into first-order logic, an extra premise must be added to define SCxy as symmetric ∀x∀ySCxy→SCyx, transitive ∀x∀y∀zSCxy∧SCyz→SCxz and reflexive ∀xSCxx.

Another issue arises because DA _D employs quotations of tokens. Since each quotation in (1) and (2) refers to a different token, there are two quotational singular terms in each of the inferences, instead of two tokens of the same singular term, which seems to make the inferences invalid. I propose this solution: (1’) and (2’) are enthymemes that assume the premise that the quoted sentence

tokens they contain are synonymous.¹⁶

Consequently, we can represent (1) and (2) as having the following valid forms:

The first premises define SCxy as an equivalence relation. The second premises state that the quoted tokens have the same content. Both inferences are invalid if either of these two premises is omitted. Unlike (1SA) and (2SA), the forms derived from DA avoid propositions, which is a theoretical virtue under naturalistic assumptions. While DA succeeds in formally validating the relevant patterns, the question remains whether it offers a plausible analysis of what content ascriptions literally say in ordinary language-or merely of what they non-literally convey.

8.1. Replies on Behalf of the Standard Analysis

An initially promising reply to semantic substitution failure concedes that TCs do not always refer to propositions, but to different kinds of entities depending on the verb: propositions in the case of “believes”, possibilities or eventualities in the case of “fears” (e.g., “John fears the possibility that it will rain”), and facts in the case of “remembers” (e.g., “Mary remembers the fact that Nancy kissed Peter”). This move limits the scope of SA but preserves it for a core class of propositional attitude verbs. However, even this concessive proposal faces difficulties. First, some central verbs still exhibit semantic substitution failure: “John tells Nancy that it will rain” is not equivalent to “John tells Nancy the proposition that it will rain”, even though both are well-formed. Second, Fregean cases arise with allegedly nonpropositional verbs: “Lois fears that Superman will not come” may be true, while “Lois fears that Clark Kent will not come” is false. This suggests that what is ascribed is not a relation to a fact or a possibility, but a particular way the subject represents the situation- something that contents capture and facts or eventualities do not.

A possible reply to the marks of concreteness problem is to claim that predicates like “daring” or “surprising” apply to acts or states when quantifiers or attitude nouns are involved, while predicates like “true” or “false” apply to propositions. On this view, verbs like “believe” and “say” are polysemous: in “John believes something unusual”, the quantifier ranges over belief states; in “John believes something true”, over propositions. A similar ambiguity would be posited for attitude nouns such as “belief” and “claim”. But this reply is undermined by the naturalness of co-predication (^{Moltmann 2013, pp. 136-137}). In “John’s belief that nothing exists is unusual and false”, both predicates apply without ambiguity or zeugma, even though one is said to apply to a state and the other to a proposition. The same happens with quantifiers: “John said something unusual and true” sounds entirely natural. These cases suggest that the pred icates apply to a single concrete bearer of content, such as a belief or a remark-not to distinct readings of an ambiguous term.

A final reply holds that attitude nouns preserve the argument structure of their verbal counterparts by including a null complement-so “John’s belief” would be analyzed as “John’s belief [0]”. This move is meant to explain the optionality of TCs in noun constructions, while preserving the claim that TCs are arguments of the noun. But even with an overt TC, this proposal runs into trouble. In “John’s belief that nothing exists is unusual”, neither “John” nor the TC seems to denote what is said to be unusual-it is the belief itself. Yet the analysis provides no syntactic position, singular term, or quantified variable for this bearer of the predicate. The problem is even sharper in co-predication cases like “Copernicus’ claim that the Earth moves was surprising but true”. Both predicates naturally apply to a single entity, with no indication of ambiguity or syntactic doubling. The idea that the dyadic argument structure of SA is preserved in these nominalizations cannot account for how these predications work.

While SA may yet find adequate responses to these challenges, it cannot be assumed as the default account. The search for alternative analyses is warranted.¹⁷

8.2. Linguistically Plausible Non-Propositionalist Alternative

DA is not only a plausible alternative to SA with clear metaphysical advantages, as Devitt argues. There are also reasons to believe that it may offer a more linguistically plausible account of particular ascriptions. For example, DA avoids SA’s problem of semantic substitution failure and provides a principled diagnosis: (5a) cannot be paraphrased as (5b) because TCs are predicates specifying representational properties, not singular terms referring to propositions. DA thus supports a uniform and plausible analysis of ascriptions with all attitude verbs. It suggests that the logical form of (5a) is either ∃x(Fxn∧Rx)-where Fxy symbolizes “x is a fear of y” and Rx symbolizes “x has the content that it will rain”-or ∃x(Fxn∧SCxit will rain).

Moreover, DA avoids SA’s difficulties with TCs complementing attitude nouns. The TCs in “Nancy has the fear that it will rain” and “Nancy’s fear is that it will rain” seem not to behave like arguments or singular terms referring to propositions. DA treats such cases as structurally akin to (5a), where the ascription involves a fear and the TC again functions as a predicate rather than an argument. On this view, verb and noun alternatives are not only equivalent in meaning, but have closely related logical forms-just as they appear to.

Recent linguistic work supports this line. Keir ^{Moulton (2015)} proposes a “uniform” account on which TCs work as predicates both when combined with “non-argument taking nouns” like “belief” and with verbs like “believes”, where they apply to a quantified variable introduced by the verb. On his analysis, the truth-conditions of “S believes that p” are that there is an x such that x is believed by S and x has the content that p (^{van Elswyk 2020, p. 2870}). Since this x corresponds to the same kind of object as the noun “belief”, it is natural to interpret it as a belief of S.¹⁸ Similarly, Boban ^{Arsenijevic (2009)} proposes that attitude verb and noun alternatives share the same underlying syntax-e.g., “S believes that p” and “S has the belief that p” both have the structure S [ _VP have [ _NP belief [ _CP that p]]], since “believes” is equivalent to “have-belief” (^{2009, p. 43})-and suggests that TCs specifying contents do not work as arguments but as relative clauses modifying nouns (^{2009, pp. 44-46}). Relative clauses play predicative roles and are not singular terms.¹⁹

In light of the linguistic evidence against SA’s treatment of quantifiers (section 4 and section 8.1), Devitt’s concession that quantifiers literally range over propositions may be open to challenge. The quantifiers seem to truthfully compose with predicates that apply to contentful entities like beliefs or remarks, but not to propositions (^{Moltmann 2013}). In “Copernicus and Galileo thought something daring but true”, what is said to be daring and true are presumably a thought of Copernicus and a thought of Galileo-given their shared content-not a proposition. The quantificational version of semantic substitution failure (^{Prior 1971}; ^{D’Ambrosio 2023}) further supports this view. “Nancy fears everything Mary fears” cannot be paraphrased as “Nancy fears every proposition Mary fears”, which is not equivalent in any sense. But it does seem equivalent to a Devittian paraphrase like “If Mary has any fear, Nancy has a fear with the same content”. So I tentatively suggest that DA _G may also capture the literal truth-conditions of general ascriptions of shared contents.²⁰

8.3. The Linguistic Plausibility of Predicative Propositionalism

It is important to acknowledge that PP also avoids the core linguistic problems of SA. PP avoids the problem of semantic substitution failure. Indeed, ^{Güngör (2022, section 5)} argues for his version of PP based on this advantage over SA. PP agrees with DA that (5a) cannot be paraphrased as (5b) because TCs are predicates instead of singular terms. According to PP, the logical form of (5a) is ∃x(Fxn∧Cxit will rain). PP also avoids SA’s problem with the

seemingly non-argumentative roles of TCs complementing attitude nouns. According to PP, “Nancy has the fear that it will rain” and “Nancy’s fear is that it will rain” share the same form as (5a), where the TC works as a predicate instead of an argument. Like DA, PP treats attitude verb and noun alternatives as synonymous.

PP may even avoid SA’s problem with the application of predicates like “unusual”. In “John’s belief that nothing exists is unusual”, what is said to be unusual according to PP is a belief whose content is the proposition that nothing exists, not the proposition itself, which cannot be unusual. But PP faces problems with causal predicates. Consider “Bill was hurt by John’s claim that he is incompetent”. Surely John’s claim is said to have hurt Bill in virtue of its content. But a proposition presumably cannot causally contribute to the claim having this effect.²¹ This is not just a linguistic problem with causal predicates, but also an instance of the metaphysical problems stressed by Devitt. The proponent of abstract propositions can appeal to PP to account for the linguistic data, but must then regard the causal claims we make about contents as strictly speaking false. To the extent that such claims seem true, DA has an important advantage.

Both DA and PP avoid the core linguistic problems of SA. But insofar as propositions are abstract objects, PP is undermined by the metaphysical consideration that the contents we ascribe play causal roles in the spatio-temporal world-especially in the explanation of intentional behavior. Suppose (5a) is true. If the content ascribed to Nancy’s fear were an abstract object, it could not play a causal role in the explanation of her behavior governed by this fear-e.g., of why she is constantly looking at the sky. The advantage of DA is that it treats contents as properties that can play causal roles in the world.

If, however, PP provides the correct analysis of the literal meaning of ascriptions, DA would still offer a plausible account of the truths these ascriptions convey in a non-literal way. It is worth noting that PP enjoys no special linguistic advantage over DA: at best, the two are tied, but DA has the further benefit of being able to treat ascriptions of content-based causation as literally true when they seem so-e.g., “Nancy is looking at the sky because she fears that it will rain”. The original rationale for favoring SA-namely, that it best fits the surface grammar of ascriptions-no longer applies in favor of this new propositionalist analysis.