The Varieties of Russellianism

Abstract

Russellianism is the view that the meaning of a proper name is the individual designated by the name. Together with other plausible assumptions, Russellianism entails the following: Sentences containing proper names express Russellian propositions, which involve the individual designated by the name as a direct constituent, and which can be represented as sets of individuals and properties. Moreover, as they occur in ordinary belief reports, ‘that’-clauses designate Russellian propositions. Such belief reports are true if and only if the subject of the belief report bears the belief relation to the proposition designated by the ‘that’-clause. In defending this doctrine, some Russellians appeal to propositional guises, which, roughly speaking, are ways of grasping propositions. However, some Russellians don’t appeal to such entities. In this paper, I explain the varieties of Russellianism and then argue for Modest Russellianism: Believing a Russellian proposition is essentially mediated by guises, so that an agent can’t believe a Russellian proposition without standing in some appropriate relation to both the proposition and a guise. Nonetheless, guises don’t feature in the semantics of ordinary belief reports, so that an adequate account of the meaning of such belief reports needn’t invoke guises.

1 Introduction

According to Russellianism, the semantic content of a proper name is the individual designated by the name.¹ Together with other plausible assumptions, Russellianism entails the Russellian Theory of Belief.

RTB: Sentences containing proper names express Russellian propositions, which involve the individual designated by the name as a direct constituent, and which can be represented as sets of individuals and properties. Moreover, as they occur in ordinary belief reports, ‘that’-clauses designate Russellian propositions. Such belief reports are true if and only if the subject of the belief report bears the belief relation to the proposition designated by the ‘that’-clause.

According to this doctrine, the proposition expressed by ‘Jack is handsome’ can be represented as < Jack, handsome > . Moreover, the ‘that’-clause occurring in ‘Jill believes that Jack is handsome’ designates < Jack, handsome > . Finally, this belief report is true if and only if Jill believes < Jack, handsome > .

In defending this doctrine, some Russellians appeal to propositional guises, which, roughly speaking, are ways of grasping propositions (alternatively, it might be said that guises are ways of apprehending propositions, ways of entertaining propositions, or ways in which propositions are represented to agents). These entities are also sometimes known as conceptions, notions, or modes of presentation. However, some Russellians don’t appeal to such entities. Thus, there are the following two varieties of Russellianism.

RTB1 Believing a Russellian proposition is essentially mediated by guises, so that an agent can’t believe a Russellian proposition without standing in some appropriate relation to both the proposition and a guise. Moreover, guises feature in the semantics of ordinary belief reports, so that an adequate account of the meaning of such belief reports needs to invoke guises.

RTB2 Believing a Russellian proposition is an unmediated relation between an agent and that proposition. Moreover, guises don’t feature in the semantics of ordinary belief reports, so that an adequate account of the meaning of such belief reports needn’t invoke guises.

According to RTB1, Jill can’t believe < Jack, handsome > without standing in some appropriate relation to < Jack, handsome > and a guise. It might be held that Jill needs to assent to the proposition through some guise. Moreover, guises feature in the semantics of ‘Jill believes that Jack is handsome’. It might be held that ‘Jill believes that Jack is handsome’ quantifies over guises, and so it’s semantically equivalent to a certain existential statement. According to RTB2, however, Jill’s believing < Jack, handsome > is a relation that obtains directly between her and the proposition, unmediated by guises. Moreover, guises have no role to play in the semantics of ‘Jill believes that Jack is handsome’. The former doctrine is famously defended by Nathan Salmon (1986), while the latter doctrine is perhaps most thoroughly defended by Jonathan Berg (2012).

However, there is a third option available to Russellians, between the two extremes of RTB1 and RTB2.

RTB3 Believing a Russellian proposition is essentially mediated by guises, so that an agent can’t believe a Russellian proposition without standing in some appropriate relation to both the proposition and a guise. Nonetheless, guises don’t feature in the semantics of ordinary belief reports, so that an adequate account of the meaning of such belief reports needn’t invoke guises.

We might say that, according to RTB3, guises feature in the metaphysics of belief even though they don’t feature in the semantics of belief reports. Strictly speaking, there is also a fourth option available to Russellians, according to which guises feature in the semantics of belief reports even though they don’t feature in the metaphysics of belief. However, I know of no considerations in favor of this fourth option, whereas I know of several considerations in favor of the third option. It’s the purpose of this paper to lay out these considerations. These considerations may not be decisive, but they’re strong enough that Russellians should consider RTB3 as seriously as they consider other options.

In Section 2, I rehearse a familiar argument for the claim that guises feature in the metaphysics of belief. This argument should also clarify what I mean when I make this claim. Then I consider Berg’s response to this argument, finding his response to be unsatisfactory. In Section 3, I consider arguments for the claim that guises feature in the semantics of belief reports. The main argument that I consider comes from Salmon and concerns the difficult topic of suspended judgment. Ultimately, I find these arguments to be inconclusive. In Section 4, I suggest some reasons for holding that guises don’t feature in the semantics of belief reports. In Section 5, I offer final remarks.

Obviously, Salmon and Berg are my main targets even though there are many other proponents of RTB1 and RTB2. This is because Salmon and Berg are excellent representatives of their respective views, and, due to space limitations, I can’t address everyone who’s contributed to the vast literature on RTB. That said, I’ll have things to say here and there about other theorists. Also, I want to make clear that this paper assumes the truth of RTB. It does not attempt to provide an explanation of anti-Russellian intuitions about the substitutivity of names. These intuitions form the basis of Frege’s puzzle, which poses the most famous challenge to RTB. I’m neutral about whether a fully adequate explanation of anti-Russellian intuitions would invoke pragmatic phenomena or purely psychological phenomena, or even whether there is a fully adequate explanation.² That said, Frege’s puzzle is relevant to our discussion, as we’ll see in the following section.

2 Guises and the Metaphysics of Belief 

Here I defend the view that believing a Russellian proposition is essentially mediated by guises. Some might think that this view is obviously correct. More generally, some might think that it’s obviously correct that there is “no mentation without mediation,” to use Kaplan’s memorable slogan (2003). Alas, an actual argument is in order, since some Russellians have rejected this slogan.

One of the most popular arguments is based on the explanation of behavior. Suppose that Lois is in a “Frege’s puzzle situation” with respect to a certain man, whom she knows as both ‘Superman’ and ‘Clark’, unaware that these names are co-designative. She exhibits two different patterns of behavior when she encounters this man. Suppose that he kisses Lois while wearing his famous blue suit and red cape. Lois embraces him. Suppose that he kisses Lois while wearing his dull brown suit and thick glasses. Lois slaps him. From a Russellian perspective, it seems that the best way of accounting for this difference in Lois’s behavior is along the following lines. The same Russellian proposition is expressed by both ‘Superman is kissing me’ and ‘Clark is kissing me’, which we can represent as < Superman, kissing > .³ However, Lois grasps < Superman, kissing > through different guises, which explains why Lois embraces Superman in the first scenario and slaps him in the second scenario. We might say, moreover, that these guises enter into the individuation of Lois’s beliefs. The belief she would express by saying ‘Superman is kissing me’ is different from the belief she would express by saying ‘Clark is kissing me’. Although these beliefs have the same content (the Russellian proposition), they differ because of the way in which that content is represented. This is the sense in which guises feature in the metaphysics of belief.

Berg (2012, pp. 111–115) is unpersuaded. He suggests an alternative account of Lois’s behavior. Note that Berg doesn’t reject the very idea of guises. In fact, he sketches a theory of guise-like entities, which he calls conceptions.⁴ According to Berg, conceptions are sets of predicates (or properties) that one believes to be jointly instantiated. Let’s assume that Lois associates ‘Superman’ with ‘wears a red cape’, ‘protects the city’, and so on. Let’s assume further that she associates ‘Clark’ with ‘wears thick glasses’, ‘works as a reporter’, and so on. These two sets of predicates correspond to her two conceptions of Superman. So, how does Berg account for the difference in Lois’s behavior toward him? Berg insists this difference doesn’t need to be explained in terms of Lois’s belief in < Superman, kissing > , or any other belief whose content involves Superman as a direct constituent. These beliefs are identical in each of the scenarios described above, but Lois’s overall doxastic states are different. Focusing on the second scenario, Lois also has a belief she would express by saying ‘There is someone who wears thick glasses, works as a reporter … and he is kissing me’. This belief, among others, is absent in the first scenario, replaced with the corresponding belief involving her other conception of Superman. Note that Berg’s explanation does invoke conceptions, but not in a way that commits him to the view that conceptions feature essentially in the metaphysics of belief. He can continue to hold that Lois’s belief in < Superman, kissing > isn’t mediated by conceptions (or guises or what have you). Although conceptions exist, agents don’t believe propositions through conceptions.

This account is unsatisfactory for several reasons. In the first place, it doesn’t respect how agents themselves would explain their behavior. If Lois were asked why she slapped the man, she would presumably say something along the lines of ‘Clark was kissing me’. She presumably wouldn’t say ‘There was someone who wears thick glasses, works as a reporter … and he was kissing me’. Our original explanation of Lois’s behavior, which individuated Lois’s beliefs in terms of guises, does respect how Lois herself would explain her behavior. The belief she would express by saying ‘Clark was kissing me’ is what produced her behavior, at least in part.⁵

Of course, this evidence against Berg’s account isn’t dispositive, since he can insist that an adequate explanation of Lois’s behavior needn’t reflect the way that she herself would explain her behavior. Moreover, it must be admitted that if Lois were to forget the name of the man who kissed her, then she might resort to general descriptions of the sort suggested by Berg. But there is another problem with Berg’s account. Even if it works in the Lois case, it can’t be easily extended to other cases. Lois’s conceptions of Superman are unique conceptions, in that she believes each set of predicates to be instantiated by exactly one individual (2012, p. 113, n. 12). But suppose that an agent is in a Frege’s puzzle situation with respect to an individual even though the agent doesn’t have a unique conception of that individual. This is perfectly possible. As Kripke (1980, pp. 80–81) pointed out, most ordinary agents don’t associate names with uniquely identifying properties. An agent might associate both ‘Cicero’ and ‘Tully’ with ‘Roman orator’, not knowing that these names are co-designative, and not presuming that there is exactly one Roman orator. Suppose that this agent is asked whether Cicero is Cicero. The agent nods in agreement. Suppose that the agent is asked whether Cicero is Tully. The agent shrugs his shoulders. The agent exhibits two different patterns of behavior with respect to the same individual, even though the agent doesn’t have a unique conception of this individual.

It’s unclear how Berg would handle this case. There is only one Russellian proposition, but the agent nods when presented with it in the first scenario and shrugs when presented with it in the second scenario. As far as I can see, Berg’s only option is going metalinguistic. Indeed, Berg often appeals to metalinguistic properties in defending his account. Perhaps the agent nods because he has the belief that ‘Cicero is Cicero’ is true, whereas the agent shrugs because he lacks the belief that ‘Cicero is Tully’ is true (alternatively, we might say that the agent nods because he has the belief that there is someone designated by ‘Cicero’, whereas the agent shrugs because he lacks the belief that there is someone designated by both ‘Cicero’ and ‘Tully’). Unfortunately for Berg, this maneuver won’t work in other cases, such as cases where the agent has no name for the individual in question. Suppose that I encounter the same woman on two separate occasions, not knowing that I’ve encountered only one woman rather than two different women. Suppose, further, that I failed to acquire the woman’s name on these occasions. It seems possible for me to assent to the Russellian proposition that she is identical to herself in some circumstances, even though I might suspend judgment when presented with the same Russellian proposition in other circumstances. However, if my memory is foggy enough, then I might not have any description of the woman other than ‘women I met yesterday’.

It seems most promising to say that, although there is only one Russellian proposition, I grasp this proposition through two guises, and believe this proposition through only one of those guises. As many Russellians have observed, we need to distinguish between the content of belief and the way in which that content is represented. At this point it would be natural to request a more detailed account of guises. We know something about the role they’re supposed to play in RTB1 and RTB3, but we don’t know much about their fundamental nature. What exactly are guises? Our discussion has already provided some clues, but a more thorough discussion awaits us in Section 4. I don’t attempt to establish a fully developed theory of guises, but I consider several possible theories, some more plausible than others.

I’ve argued that Frege’s puzzle situations provide good reasons to invoke guises in the metaphysics of belief. This doesn’t mean that I take myself to have solved Frege’s puzzle. So far I’ve said little about the nature of guises (but, again, see Section 4 for further discussion). Moreover, I’ve suggested no explanation of the apparent difference between ‘Cicero is Cicero’ and ‘Cicero is Tully’, or between ‘Lois believes that Superman is kissing her’ and ‘Lois believes that Clark is kissing her’. I’m neutral here about the correct way of handling these apparent substitution failures. I should mention, however, that even if Russellians can handle these apparent substitution failures without invoking guises, that doesn’t undermine the considerations adduced above. Thus, I disagree with Thau (2002, pp. 104–107), another proponent of RTB2, who suggests that if we can explain apparent substitution failures by appealing to pragmatic phenomena, then we can explain Lois’s different patterns of behavior in the same way. Very briefly, Thau’s idea is that ‘Lois believes that Superman is kissing her’ and ‘Lois believes that Clark is kissing her’ pragmatically convey different propositions, and so speakers mistakenly conclude that these belief reports have different explanatory powers (‘Lois believes that Superman kisses her’ explains why she embraces Superman, while ‘Lois believes that Clark is kissing her’ explains why she slaps him). However, as Berg himself notes (2012, p. 112, n. 9), this account fails to address the issue at hand. Even if Thau’s pragmatic explanation of apparent substitution failure is correct, there is still the question of why Lois would embrace Superman in certain scenarios and slap him in certain other scenarios, when there is no difference in the Russellian contents of her beliefs.⁶ On this point, at least, I’m in agreement with Berg. Where we disagree is whether Lois’s beliefs about Superman can be individuated entirely in terms of their Russellian contents, or whether they need to be individuated partly in terms of guises.

3 Guises and the Semantics of Belief Reports

Now I’ll consider the claim that guises feature in the semantics of ordinary belief reports. Some might argue that if guises feature in the fundamental nature of belief, then they obviously feature in the semantics of ordinary belief reports. This, however, is not necessarily the case. As Kaplan (2003) pointed out, sperm cells feature in the fundamental nature of fatherhood, but they don’t seem to feature in the semantics of ordinary fatherhood reports. For example, it doesn’t seem that ‘Andrew fathered Lillian’ quantifies over sperm cells, or anything else for that matter.

It might be complained that this analogy is inapt. Empirical research led to the discovery of sperm cells, not the sort of philosophical considerations that I offered in support of guises. Now, I see that there is a difference here, but I don’t see that this difference makes a difference. The point stands that certain relations might involve certain entities even though those entities don’t enter into the semantics of sentences reporting those relations. Anyway, the same basic point can be made without invoking empirically discovered entities. Through purely philosophical reflection we might conclude that walking essentially involves ways of walking. But, as Eaker (2009) has pointed out, it would be premature to conclude that ways of walking enter into the semantics of ‘Lillian walked to Andrew’.⁷

Some might argue that guises must be invoked in the semantics of belief reports in order to explain the consistency of ‘Lois believes that Superman flies’ and ‘Lois doesn’t believe that Clark flies’. This, however, would seem to misunderstand the commitments of RTB. Assuming that the latter belief report is understood as ‘It is not the case that Lois believes that Clark flies’, then it is not consistent with ‘Lois believes that Superman flies’, for it expresses the same thing as ‘It is not the case that Lois believes that Superman flies’. There is therefore no consistency to be explained. On the other hand, if ‘Lois doesn’t believe that Clark flies’ is understood as ‘Lois believes that Clark doesn’t fly’ or equivalently ‘Lois disbelieves that Clark flies’, then it is consistent with ‘Lois believes that Superman flies’. It is also consistent with ‘Lois believes that Clark flies’, which expresses the same thing as ‘Lois believes that Superman flies’. But the consistency of these belief reports need not be explained by introducing guises into their semantics. They are consistent because it is possible that Lois believes a certain Russellian proposition and also its negation. Indeed, it is possible that Lois rationally believes a certain Russellian proposition and also its negation. For we can suppose that she grasps the proposition and its negation under appropriately different guises. But this psychological fact about her need not be reflected in the semantics of belief reports about her, just as the fact that fatherhood involves sperm cells need not be reflected in the semantics of ‘Andrew fathered Lillian’.⁸

Salmon (1986, pp. 92–113) suggests a more compelling argument for the claim that guises feature in the semantics of belief reports. On his view, belief reports quantify over guises. His argument is based on suspended judgment.⁹ Plausibly, suspending judgment with respect to a proposition requires neither believing it nor disbelieving it. Suppose that Lois believes that Superman flies, but that she suspends judgment about whether Clark flies (to speak as Lois herself would speak). On pain of contradiction, Russellians can’t say that Lois believes the Russellian proposition < Superman, flies > while also saying that she neither believes nor disbelieves < Superman, flies > . In this sort of case, how can Russellians accommodate the truth of ‘Lois suspends judgment about whether Clark flies’ without committing themselves to a contradiction? According to Salmon, we should dispense with the assumption that suspending judgment requires neither believing nor disbelieving. Instead, we should invoke a ternary BEL relation between agents, propositions, and guises. An agent stands in the BEL relation to a proposition and a guise if and only if the agent assents to the proposition through the guise. Belief, disbelief, and suspended judgment are all analyzed in terms of BEL. Belief reports are analyzed as existential generalizations over guises. In general, ‘A believes p’ is analyzed as ‘∃x[A grasps p through x & BEL(A, p, x)]’, while ‘A disbelieves p’ is analyzed as ‘∃x[A grasps ~ p through x & BEL(A, ~ p, x)]’. Accordingly, Russellians can represent Lois’s belief by means of:

• (1) ∃x[Lois grasps < Superman, flies > through x & BEL(Lois, < Superman, flies > , x)].

  As for Lois’s state of suspended judgment, BEL allows Salmon to introduce the notion of withheld belief. An agent withholds belief from a proposition if and only if there is a guise through which the agent grasps the proposition and the agent does not assent to the proposition through the guise (Salmon, 1986, p. 111). Formally, ‘A withholds belief from p’ can be understood as ‘∃x[A grasps p through x & ~ BEL(A, p, x)]’, where the negation operator occurs before ‘BEL’ and takes narrow scope with respect to the quantifier. On Salmon’s view, suspending judgment requires withholding belief. Note, however, that Russellians cannot straightforwardly analyze ‘Lois suspends judgment about whether Clark flies’ as:

• (2) ∃x[Lois grasps < Superman, flies > through x & ~ BEL(Lois, < Superman, flies > , x)] & ∃x[Lois grasps ~  < Superman, flies > through x & ~ BEL(Lois, ~  < Superman, flies > , x)].¹⁰

  This says that Lois withholds belief from both < Superman, flies > and the negation of < Superman, flies > , which is consistent with (1). The problem with (2) is that it is true even in scenarios where Lois does not suspend judgment. Consider a scenario where Lois believes that Superman flies and disbelieves that Clark flies (to speak as Lois herself would speak), but where Lois does not suspend judgment at all. In this scenario, there are only two guises through which Lois grasps < Superman, flies > , one of which is associated with ‘Superman’ and one of which is associated with ‘Clark’. If we were to ask Lois whether Superman flies, she would sincerely answer in the affirmative. If we were to ask Lois whether Clark flies, she would sincerely answer in the negative. There is no guise through which she suspends judgment on the question. Still, it is true that Lois withholds belief from both < Superman, flies > and the negation of < Superman, flies > . For there is a guise, associated with ‘Clark flies’, such that Lois grasps < Superman, flies > through that guise and does not assent to < Superman, flies > through that guise. Indeed, Lois dissents from < Superman, flies > when it is presented to her through that guise. Moreover, there is a guise, associated with ‘Superman does not fly’, such that Lois grasps the negation of < Superman, flies > through that guise and does not assent to the negation of < Superman, flies > through that guise. Indeed, Lois dissents from the negation of < Superman, flies > when it is present to her through that guise. Thus, (2) is true in this scenario, even though Lois does not suspend judgment.¹¹

  According to Salmon, if we want to accurately represent suspended judgment, then we should invoke a function over guises called ‘Neg’. For any given way of grasping a proposition, the Neg function delivers “the corresponding way” of grasping the negation of that proposition (1995, p. 8). In scenarios where Lois suspends judgment, this fact is represented by means of:

• (3) ∃x[Lois grasps < Superman, flies > through x & ~ BEL(Lois, < Superman, flies > , x) & Lois grasps ~  < Superman, flies > through Neg(x) & ~ BEL(Lois, ~  < Superman, flies > , Neg(x)],

  which is consistent with (1). In general, ‘A suspends judgment about whether p’ is analyzed as ‘∃x[A grasps p through x & ~ BEL(A, p, x) & A grasps ~ p through Neg(x) & ~ BEL(A, ~ p, Neg(x)]’, which is consistent with ‘∃x[A grasps p through x & BEL(A, p, x)]’ as well as ‘∃x[A grasps ~ p through x & BEL(A, ~ p, x)]’.¹²

I don’t claim to have any decisive refutation of this analysis, but it’s unsatisfying for several reasons. As I’ve argued elsewhere Atkins (2017), this analysis encounters difficulties in specifying “the corresponding way” of grasping the negation of a proposition. For any guise x that is a way of grasping < Superman, flies > , it is unclear which guise is supposed to be Neg(x). For there are many ways of grasping the negation of < Superman, flies > . The objection, then, is that Neg is undefined in this case. One tempting reply is that Neg(x) is the same as x together with some additional way of grasping the negation function. However, the objection is not avoided entirely. For there are arguably many ways of grasping the negation function just as there are many ways of grasping a proposition. Salmon has suggested that, although there are many ways of grasping the negation function, one of these ways is privileged in that it identifies the negation function. Thus, Neg(x) is the same as x together with this identifying way of grasping the negation function. In general, we can say that a way of grasping something is identifying if and only if it reveals who or what is being grasped.¹³ This is an interesting suggestion, but problems persist. In the first place, knowing who and knowing what are arguably context-sensitive notions (Boër & Lycan, 1986; Quine, 1979). For example, there is arguably no context-independent fact about whether I know who the Zodiac Killer is. Relative to some contexts it is appropriate to say that I know who the Zodiac Killer is (a casual conversation about true crime podcasts) and relative to other contexts it is inappropriate to say that I know who the Zodiac Killer is (a police interrogation). If knowing who and knowing what are context-sensitive in this way, then Neg will be undefined unless some relevant context is specified. Setting aside the issue of context-sensitivity, we are given no reason for thinking that there is only one identifying way of grasping something. In the absence of further argument, why not say that there are many identifying ways of grasping something, each identifying it in a different way (revealing its identity in a different way)? If so, then it is possible that there are many identifying ways of grasping the negation function and the original objection resurfaces.

It is also worth noting that ‘A believes p’ and ‘A suspends judgment about whether p’ simply do not seem to be complex quantificational formulas. It would be desirable for Russellians to have a less byzantine account, especially one that respects the surface-level form of these sentences.¹⁴ There are several options available to Russellians. One might follow Crawford (2004a, b) and represent Lois’s state of suspended judgment in terms of her higher-order beliefs (see Tillman (2005) and Atkins (2017) for criticisms, but also Masny (2020) for a defense of a version of Crawford’s view). My own view, which Friedman defends for independent reasons (2013a, b) and which Berg anticipates (2012, p. 122), is that suspended judgment should be treated as a sui generis attitude, not properly explained in terms of belief. Russellians can then analyze ‘A believes p’ as ‘B(A, p)’, and ‘A suspends judgment about whether p’ as ‘SJ(A, p)’, where ‘B’ and ‘SJ’ are both taken to be semantically primitive, not analyzed in terms of any further predicates.¹⁵ On this analysis, there is no inconsistency between ‘Lois believes that Superman flies’ and ‘Lois suspends judgment about whether Clark flies’. Moreover, this analysis reflects more faithfully the surface-level form of ‘Lois believes that Superman flies’ and ‘Lois suspends judgment about whether Clark flies’. It might be worried that this analysis entails that ‘Lois believes that Clark flies’ and ‘Lois suspends judgment about whether Clark flies’ are consistent. However, this is no more problematic than the standard Russellian claim that ‘Lois believes that Clark flies’ and ‘Lois disbelieves that Clark flies’ are consistent. As I argued above, the consistency can be explained at the psychological level in terms of the different ways that Lois grasps the Russellian proposition (see Atkins (2017) for elaboration). That said, this analysis of suspended judgment has costs. We must reject the plausible assumption that suspending judgment is a matter of neither believing nor disbelieving. But Salmon’s account of suspended judgment also rejects this assumption. In any event, the availability of alternative accounts of suspended judgment undermines Salmon’s argument that belief reports quantify over guises.

4 Guises and the Semantics of Belief Reports (Continued)

I’ve considered arguments for the claim that guises feature in the semantics of ordinary belief reports, finding them to be inconclusive. In this section I suggest some reasons for holding that guises don’t feature in the semantics of ordinary belief reports. Following the lead of Schiffer (1992) and Ostertag (2009), let’s consider belief reports that have quantified noun phrases in subject position. Suppose the following is true:

• (4) Everyone who lives in New York City believes that Donald Trump is crooked.

  If specific guises feature in the semantics of belief reports, then (4) would entail that there is some specific way in which every New Yorker grasps the proposition that Donald Trump is crooked, which is presumably false and which does not seem to be entailed by (4). To avoid this problem, we may, of course, follow Salmon in analyzing (4) as follows:

• (5) ∀y[y lives in New York City ⊃ ∃x[y grasps < Trump, crooked > through x & BEL(y, < Trump, crooked > , x)]].¹⁶

But, even though (5) captures the intuitive truth conditions of (4), there remains the problem, mentioned in Section 3, that such an analysis doesn’t reflect the surface-level form of ordinary belief reports, since ordinary belief reports don’t seem to involve existential generalizations of the sort that we find in (5).¹⁷ It’s simpler to analyze ‘Every F believes p’ as ‘∀y[Fy ⊃ B(y, p)]’, rather than as ‘∀y[Fy ⊃ ∃x[y grasps p through x & BEL(y, p, x)]]’. And we Russellians are free to adopt the simpler analysis once we accept the idea that guises are semantically irrelevant, even if they might be psychologically relevant.

I want to suggest one more argument against the claim that guises feature in the semantics of ordinary belief reports, inspired by Braun’s (1998, pp. 567–568) argument against Salmon’s (1986, pp. 114–118) solution to Frege’s puzzle. Very briefly, Salmon’s solution to Frege’s puzzle is that, although belief reports are generalizations over guises, speakers routinely use belief reports to communicate propositions involving specific guises. This pragmatic phenomenon results in the anti-Russellian intuitions that motivate Frege’s puzzle. According to Braun, however, it’s unreasonable to insist that ordinary speakers routinely entertain and assert such propositions, since such propositions would be too sophisticated. Here I want to extend Braun’s point. When devising a propositional semantics for a part of ordinary language, theorists should try not to invoke propositions that involve “exotic entities”—that is, theoretical entities that ordinary speakers can’t be expected to already grasp. It’s unreasonable to insist that most ordinary speakers routinely entertain and assert propositions involving exotic entities, since such propositions would be too sophisticated. Following Braun, I submit that most ordinary speakers don’t possess even a rudimentary understanding of guises, characterized loosely as ways of grasping propositions. Guises should be invoked in the semantics of ordinary belief reports only if necessary. But, as I argued in Section3, guises aren’t required in the semantics of ordinary belief reports, even if, as I argued in Section2, they’re required in the metaphysics of belief. We should conclude, at least tentatively, that guises don’t feature in the semantics of ordinary belief reports, even as entities over which ordinary belief reports supposedly quantify.

To be clear, the point here is not that theorists always go wrong when invoking exotic entities. I suppose that linguistic characters, understood as functions from contexts to propositional contents, are exotic entities and theorists are certainly justified when they invoke characters. But linguistic characters seem necessary to explain the function of indexicals, and, importantly, theorists don’t insist that characters feature in the propositions that ordinary speakers routinely entertain and assert. Theorists don’t insist that ordinary speakers routinely designate characters or quantify over characters. In contrast, RTB1 insist that guises feature in the propositions that ordinary speakers routinely entertain and assert—ordinary speakers are thought to somehow designate or quantify over guises.

It might be objected that properties are often said to feature in the propositions that ordinary speakers routinely entertain and assert, and such abstract entities are sometimes thought to be peculiar. Indeed, Quine (1960) found them objectionable enough that he denied their very existence. However, the theoretical status of properties seems to be rather different from that of guises. Most ordinary speakers possess a rudimentary understanding of properties, even if they’ve never ruminated on the potentially puzzling nature of properties and even if they’ve never bothered to develop a metaphysical theory of properties. In contrast, ordinary speakers don’t possess a rudimentary understanding of guises, at least if guises are characterized as ways of grasping propositions.

At this point, however, we need to consider the following possibility. Most ordinary speakers don’t possess a rudimentary understanding of guises, characterized as ways of grasping propositions, but perhaps guises can be identified with entities that ordinary speakers can be expected to already grasp. For example, it’s often thought that guises are sentences in natural language, and these aren’t such exotic entities. It’s not terribly implausible that ordinary speakers routinely entertain and assert propositions about sentences.

To evaluate this possibility, we need to revisit a question that we encountered in Section 2: What are guises? Any viable theory needs to meet the Fundamental Principle of Guises.

FPG: If a rational agent believes p and ~p, then there is a guise through which the agent believes p, a guise through which the agent believes ~p, and the first guise is different from the second guise. In other words, if an agent believes p and disbelieves p, then the agent does so through two different guises. Moreover, if an agent believes p and suspends judgment about whether p, then the agent does so through two different guises.¹⁸

In what follows I consider various theories, including the theory that guises are sentences in natural language. I argue that some of these theories fail to satisfy FPG, and so aren’t viable. The remaining theories seem to be viable, but they identify guises with exotic entities. Several of the following points have been made by others, but here I wish to bring them together in an argument against the general claim that guises feature in the semantics of ordinary belief reports.

Let’s begin with the common assumption, mentioned above, that guises are sentences in natural language.¹⁹ One immediate problem is that it seems possible for certain animals and young children to believe propositions, even though they don’t associate propositions with sentences. Even fully grown adults, if not raised in the appropriate environment, will fail to acquire a language, and yet it seems possible for such adults to believe propositions. Another problem is the notorious Paderewski case. There is only one proposition expressed by ‘Paderewski is talented’, which we can represent as < Paderewski, talent > . Peter comes to both believe and disbelieve < Paderewski, talent > , being under the mistaken impression that ‘Paderewski’ is the name of two different individuals. Peter grasps < Paderewski, talent > through different guises, believing it through one guise and disbelieving it through the other guise, but Peter associates < Paderewski, talent > with only one sentence in natural language. Thus, the theory that guises are sentences in natural language is inconsistent with FPG.²⁰

Although linguistic characters are exotic entities, several prominent theorists have explored the possibility that characters can be deployed against Frege’s puzzle.²¹ So, it should be mentioned that the FPG also rules out the theory that guises are characters. Peter grasps < Paderewski, talent > through different guises, but doesn’t associate < Paderewski, talent > with different characters. Instead he associates it with a single sentence, ‘Paderewski is talented’, which doesn’t even contain indexicals. Of course, the same point can be made with the Lois case. She grasps < Superman, flies > through different guises, and although she associates < Superman, flies > with different sentences, these sentences don’t contain indexicals, and so it’s not possible to identify the relevant guises with characters. Even if we restrict our attention to cases where indexicals are involved, this theory of guises seems to falter. Consider two utterances of ‘He is about to be attacked’, where a single individual is designated, but where this isn’t obvious. Since it may seem that different individuals are under discussion, the first utterance and second utterance may correspond to different guises. But we can’t identify these guises with characters, since there is only one indexical involved, and so only one character involved.²²

It might be suggested that guises are pieces of descriptive information. It’s not implausible that ordinary speakers routinely entertain and assert propositions involving descriptive information. Indeed, this seems to happen whenever descriptions are involved in linguistic communication. There are two ways of fleshing out this theory of guises. We can say these pieces of descriptive information pick out unique individuals (the French general who was exiled to Elba and defeated at the Battle of Waterloo in 1815 is short) or we can say they’re general enough to pick out several different individuals (the French general is short). The problem with the former option is that an agent can believe propositions about an individual without having any description that uniquely identifies that individual. This was the point of the Cicero case that we discussed in Section 2. This leaves the latter option, but this option also runs afoul of FPG. Consider again the agent who associates both ‘Cicero’ and ‘Tully’ with ‘Roman orator’. The agent nods when he is asked whether Cicero is Cicero and the agent shrugs when he is asked whether Cicero is Tully. According to the present theory, there must be two different pieces of descriptive information, one through which the agent believes the relevant Russellian proposition and one through which the agent suspends judgment. But this doesn’t seem to be the case. The agent associates ‘Cicero is Cicero’ and ‘Cicero is Tully’ with the same descriptive information (the Roman orator is the Roman orator). We might try to salvage the theory by going metalinguistic (saying that the agent believes the proposition through the Roman orator whose name is ‘Cicero’ is the Roman orator whose name is ‘Cicero’ and suspends judgment through the Roman orator whose name is ‘Cicero’ is the Roman orator whose name is ‘Tully’). But this maneuver encounters a familiar problem. It fails to extend to cases where the agent has failed to acquire a name for the relevant individual, such as the case that I described in Section 2, where I encounter the same woman on two separate occasions, don’t acquire a name for the woman, and don’t know that I’ve encountered only one woman rather than two different women.

Fortunately, there are more promising theories. It’s possible to identify guises with mental entities. The most natural version of this theory would identify guises with mental representations physically realized in the brain. The idea here is that guises are sentences in Mentalese, the language of thought, rather than sentences in natural language. These Mentalese sentences would be individuated in terms of the functional role that they occupy in an agent. A closely related account would identify guises with mental files, which are sometimes characterized as “clusters of information.” Importantly, however, mental files are not individuated simply in terms of the information they store. The same mental file can persist even as new information is introduced and old information is discarded. Alternatively, guises might be identified with modes of acquaintance, understood as evidential chains that connect an agent with the object of the agent’s thought.²³ Now, I’m neutral about which of these views is correct, or whether any of them is correct.²⁴ My point in bringing them up is that they stand a better chance of satisfying FPG. Yet they invoke entities that ordinary speakers can’t be expected to routinely designate or quantify over. Even sophisticated speakers are unfamiliar with evidential chains, mental files, and sentences in the language of thought. These are highly theoretical entities, after all. It seems unreasonable to insist that ordinary speakers are somehow talking about such entities when engaged in the mundane practice of reporting beliefs.

5 Conclusion

It seems to me that, among the varieties of Russellianism, the best option is RTB3, but I haven’t tried to establish this conclusion. I’ve merely tried to show that RTB3 should be considered as seriously as RTB1 and RTB2. That said, if the main contentions of this paper turn out to be correct, then a natural division of labor emerges. Russellians should clarify the nature of guises. This is a problem in the philosophy of mind. Russellians should also explain the apparent difference between ‘Cicero is Cicero’ and ‘Cicero is Tully’, and between ‘Lois believes that Superman is kissing her’ and ‘Lois believes that Clark is kissing her’. This is a problem in the philosophy of language. Russellians shouldn’t expect a single solution to both problems.