Semantic Stipulation and Knowledge De Re
Semantic Stipulation and Knowledge De Re
Chris Tillman
University of Manitoba
Joshua Spencer
Syracuse University
We need only be able to understand the descriptive predicates in order to form a representation of a possibly unperceived object. I think this seemed almost magical to Russell, and it is. Russell’s knowledge by description focused on definite descriptions. But names carry the same power. We need only be able to ‘understand’ the name to have a representation of the individual named, and to understand names seems to require even less than to understand descriptive predicates. (Kaplan 2005, 997)
1. Introduction
Suppose we rush into the Department of Homeland Security and enthusiastically inform its occupants that we know the precise, present location of Osama bin Laden. We confidently tell them that, if Osama is presently located, then, presently, Osama is located at L! We know this because ‘L’ is the name that we have introduced for Osama’s precise present location, if such a location exists. Furthermore, we claim, they can come to know this information on the basis of our testimony. We present the baffled agents with copies of Naming and Necessity and offer to divulge any other highly sensitive information they might require. We then make off to collect our reward for revealing Osama’s whereabouts.
Something has gone wrong. It seems we cannot gain knowledge of Osama’s location as a result of semantic stipulation.1 Worse still, if we can manage this feat, then the only obstacles to our near-omniscience are that our brains are too small and that we will not live long enough.2 Barring these considerations, we could similarly come to know of an arbitrary object o that we are capable of naming that it is F, where F is any feature we can apprehend. Quasi-omniscience on the cheap! And there is nothing special about us. You can do the same thing if we can.
On the other hand, it is very hard to see what, if anything, stops us from having knowledge via semantic stipulation. Urbain Leverrier apparently named Neptune in just this way.3 So it seems that, at least in some cases, one can introduce names by stipulation which results in new belief and knowledge. What, if anything, differentiates the “good” cases from the “bad” ones?
2. The Puzzle
In Naming and Necessity, Kripke (1980, 56, 79-80) argued that names could be introduced by reference-fixing description. ‘L’ is a purported example of such a name. The role of the description is merely to fix the reference; it does not “give the meaning” of ‘L’. Kripke’s discussion strongly suggests that one can know de re the “object-involving” singular propositions encoded by sentences of the form N is F, where N is a name introduced by the reference-fixing description F, simply by reflecting on the fact that the referent of N must satisfy F, given the method of introducing N.4 Call such sentences ‘K-sentences’ (‘Kripke-sentences’). Call the introduction of a name by reference-fixing description ‘K-stipulation’. Via K-stipulation, it would seem we could gain knowledge of all sorts of propositions that, intuitively, we do not know. So we must reject one of the assumptions that lead to this problematic knowledge, or otherwise explain away the problem.
Let us state the problem, and the assumptions that lead to it, more precisely. We specify what it is to be a K-name as follows:
If N is a name, then: N is a K-name iff (i) N is introduced via a reference-fixing description, F, (ii) if N refers, N refers rigidly, (iii) N is entirely lacking in descriptive content, and (iv) the sentence N is F semantically encodes a singular proposition <o, P>, where o is the individual picked out by N and P is the semantic content of F.5
(Roughly: A K-name is a Millian name introduced by reference-fixing description.)
Now we can state the assumptions that lead to our problem:
Naming by Description [ND]:
There are agents who grasp reference-fixing descriptions. If an agent grasps such a reference-fixing description for an individual, then it is possible for an agent to introduce a K-name for that individual.
(Roughly: K-names can be introduced via K-stipulation.)
Naming/Attitude Link [NA]:
Necessarily, any agent who introduces a K-name, N, via reference-fixing description F, and who considers the sentence N is F is thereby able to assert and bear other (cognitive) propositional attitudes to the singular proposition <o, P>, where o is the individual picked out by N and P is the semantic content of F. Specifically, the agent can believe that proposition by accepting the sentence N is F.
(Roughly: K-stipulation puts one in position to assert and believe contents of K-sentences.)
Attitude/Knowledge Link [AK]:
Necessarily, any agent who introduces a K-name N via reference-fixing description F, and who considers the sentence N is F, thereby coming to believe the singular proposition <o, P>, where o is the individual picked out by N and P is the semantic content of F is, absent any relevant defeaters, justified (perhaps a priori) in her belief. If the agent believes that proposition, then, barring Gettier-type cases and epistemologically relevant monkey business, the agent knows that proposition.
(Roughly: If an agent can believe the content of a K-sentence, then she can also know it.)
It follows from these three principles that it is possible for an agent to come to know singular propositions simply by introducing appropriate names via K-stipulation and considering appropriate sentences. In addition to Osama’s whereabouts, it seems that via K-stipulation we can know who the next president of the United States will be (if there is one), which ticket will win the lottery (provided it has a unique winner), the color of your underwear (provided you have underwear with a unique color), etc.6 But such knowledge seems rather more difficult to come by; intuitively, it is not the sort of thing that can be had on the basis of what seems to be a semantic trick.
3. What the Puzzle is a Puzzle About
Before we consider replies that we think should be taken seriously, some remarks are in order. Something superficially similar to the problem we wish to consider is sometimes called “the problem of the contingent a priori.” But the problem we wish to consider is not the so-called problem of the contingent a priori. That is, in part, because we believe there is no problem of the contingent a priori. Such propositions surely exist. For any proposition P, P iff actually P is contingent and a priori. Perhaps appropriate uses of ‘I am here now’, ‘I exist’, or the F is actually F, where the F is non-rigid, express contingent a priori propositions as well.
To reinforce the point, note that the problem we are concerned with has nothing to do with contingency. Suppose that P is a necessary feature of some object, o. Via K-stipulation, and granting our assumptions, an agent can come to know that o is P, though that proposition is necessary. For example, if we introduce the name ‘Primo’ for the largest prime number under a googolplex, we can come to know that Primo is the largest prime number under a googolplex even though that proposition is necessary.
More to the point, it does nothing to blunt the intuitive force of our original example if we K-stipulate that ‘L*’ refers to the actual precise present location of Osama bin Laden, if Mr. Bin Laden is precisely presently located. Homeland Security should be no less interested in learning that necessary proposition than the contingent proposition that Osama is located at L.7 The real problem with K-sentences, as we see it, is, very roughly, that they allow for objectionable epistemic shortcuts. In responding to the problem, our task is to see whether there is a plausible way to block the shortcuts.
4. Replies
Corresponding to our three assumptions [ND], [NA], and [AK], there are three main replies. We will consider each in turn.8
4.1. The “No Name” Reply
The “No Name” reply denies [ND]. According to this reply, attempts to introduce a name by way of a reference-fixing description fail to introduce a K-name.9
The most popular version of the “No Name” reply holds that names introduced via K-stipulation function semantically like descriptions. They are “descriptive names”—name-like in syntax only.10, 11 As others have observed, this reply is vulnerable to Kripke’s semantic objection to descriptivism.12 Consider again Leverrier, who K-stipulated that ‘Neptune’ refers to the unique planet causing perturbations in the orbit of Uranus, if any. Leverrier might have told his astronomically ignorant father that he is searching for a planet, Neptune. Leverrier’s father might in turn brag to his friends that his son is searching for a planet, Neptune. He could very well succeed in referring to Neptune even if he is ignorant of ‘Neptune’’s reference-fixing description and even if he only associates the indefinite description ‘a planet’ with the name. To reinforce the point, we might even suppose Leverrier’s father lacks the conceptual resources to grasp complex descriptions like ‘the unique planet causing perturbations in the orbit of Uranus’.13
Finally, when explaining how a definite description might fix the reference of a name without giving its meaning, Kripke (1980, 55-57) observes that we may introduce a name for an object via description and go on to wonder various things about it. For example, let ‘Tobie’ refer to the first child born in July 2051 in Manitoba, if any. When we consider whether that individual will be a famous philosopher, we are not merely wondering whether the first child born in July 2051 in Manitoba, if any, will be a famous philosopher. We can coherently claim that Tobie might not have been the first child born in July 2051 in Manitoba. We might also believe of Tobie that it might be a famous philosopher.14
Suppose, however, that ‘Tobie’ was in fact a descriptive name. Note that the occurrence of ‘it’ in the preceding attitude description is a device of direct reference; the ascription is true iff we believe (roughly) that x might be a famous philosopher, under an assignment of the referent of ‘Tobie’ to x. If this is correct, then de re ascriptions allow us to introduce devices of direct reference that co-refer with the allegedly descriptive names.15 It is easy to reintroduce our puzzle using these devices of direct reference rather than the allegedly descriptive names. We conclude that descriptivism about names introduced via K-stipulation is useless for explaining the apparent problems that arise from what seems to be de re knowledge via K-stipulation.16
4.2 The “No Attitude” Reply
The “No Attitude” reply denies [NA]. According to its proponents, it is possible for an agent to introduce a K-name via K-stipulation, but it is not possible for an agent who introduces such a name to thereby bear cognitive attitudes toward any singular proposition solely via the relevant K-sentence(s). Proponents of this reply tend to hold that K-stipulators who lack appropriate acquaintance with the objects involved in propositions encoded by K-sentences cannot believe, or bear other cognitive attitudes, toward those propositions.
There are various accounts of what acquaintance amounts to, beginning (perhaps) with Bertrand Russell’s (1910/11) theory of direct acquaintance. According to Russell, one can only be directly acquainted with universals, sense data, and perhaps one’s self. On this account, one can only have de re thoughts about these sorts of things. David Kaplan (1968) suggested instead that one must be en rapport with an object. This involves possessing a name for object o that (i) denotes o, (ii) is a “vivid” name of o, and (iii) is, in some technical sense, a name of o. (A name that is a “vivid designator” is, in some sense, supposed to be the analogue in belief contexts of a rigid designator in modal contexts.) Kaplan’s account required that one must be or have been in perceptual or otherwise causal contact with an object in order to have de re thoughts involving it. Keith Donnellan (1977) and Nathan Salmon (2004, 2010) hold that Kaplan’s third condition alone is necessary and sufficient: one must have a name of that object, in a special sense of ‘name of’, where the relevant object enters into the correct “genetic” account of how one came to learn the term.17 Alternatively, numerous philosophers have proposed that one must have appropriate knowledge-wh of an object in order to have the relevant de re thoughts.18
Extant accounts of acquaintance are not currently in anything like an acceptable shape. Practically no one accepts Russell’s account, for obvious reasons. Accounts that appeal to vivid names or being a name of in a technical sense are deeply obscure.19 And any account on which one can introduce K-names but cannot bear cognitive attitudes toward semantic contents of K-sentences leave troubling unanswered questions about the content of mental states that result from accepting a K-sentence, or, in some cases, the semantic content of the K-sentences themselves. The trouble is this: it seems the only ways to answer these questions involve supposing the relevant mental state contents and K-sentence contents must be either purely descriptive or empty.20
Troubling content questions aside, any account that merely requires perceptual or otherwise causal contact with an object in order to have the relevant de re thoughts cannot satisfactorily solve our problem. As Thomas Ryckman (1993, 329) has pointed out, it was possible for Leverrier to have hypothesized that a planet was causing perturbations in Mars’s orbit, and it was possible for him to K-stipulate that ‘Spock’ is the name of the unique such planet, if any. Spock, it turns out, is Earth, with which Leverrier is perceptually and otherwise causally acquainted. And as Salmon (1987, 202) suggests, one could even introduce via a perceptual demonstrative a directly referential name for the particular length of an object that one is observing (Let that length be named ‘Leonard’!). The resulting knowledge, e.g. that the precise length of that stick is Leonard, or that Spock is the planet causing perturbations in the orbit of Mars, strikes us as problematic in just the same way the alleged knowledge that Osama is located at L.
Furthermore, we think the “causal contact” condition is clearly not necessary. ‘Tomorrow is another day’ seems to express a truth that we assert, believe, and know, though in the relevant respects ‘tomorrow’ seems to function semantically like a proper name (with respect to a context) and we have no perceptual or otherwise causal contact with tomorrow.21 Other apparent examples of de re thought of merely future objects abound.22
Finally, consider the popular suggestion that one must have knowledge-wh of an object in order to know propositions concerning it. We think this account does not yield the desired result. David Braun (2006) argues (correctly, by our lights) that knowing an answer to ‘Who is A?’ is sufficient to know who A is. So on Braun’s account, merely knowing that L is on Earth is sufficient to know where L is. This, in turn, is sufficient for us to meet the proposed condition for having de re thoughts about L. (‘On Earth’ would be an answer to ‘Where is L?’ though questioners would not likely find that answer satisfying.) So if Braun is correct, then proposing that one must have knowledge-wh in order to have de re thought imposes practically no constraint on de re thought at all.
We might, of course, be mistaken in our assessment of Braun’s view. Perhaps the more common view that knowledge-wh is interest-relative or context-sensitive is correct.23 Still, we think an eyewitness may have the relevant de re thoughts when distinguishing the culprit from the decoys in a police line-up even if typical speakers would dissent from ‘The eyewitness knows who the culprit is’. The eyewitness might later strike an unlikely friendship with the culprit and truly report ‘Even before I learned who you were I already believed you would get caught.’24 So even if knowledge-wh is interest-relative or context-sensitive, there still seem to be situations in which one might have a de re attitude without the supposedly requisite knowledge-wh.
We conclude the proposed condition is not necessary. And though we think it is plausible that it is sufficient, we doubt proponents of the condition would agree that the condition is merely sufficient. Suppose while discussing this paper with a colleague, a bystander overhears you use the name ‘Tobie’. The bystander interrupts, asking ‘Who is Tobie?’ We think that in this context it is perfectly acceptable, and correct, to answer ‘Tobie is the individual that Joshua Spencer and Chris Tillman claim will be the first person born in July 2051 in Manitoba.’ We doubt proponents of the condition would agree that the bystander is now in a position to have de re thoughts about Tobie.
We conclude that the proposed necessary conditions on de re attitudes are all underdeveloped or inadequate to their aims. Perhaps the proponents of the “No Attitude” reply will simply claim that there are certain necessary conditions on de re attitudes that we have not yet discovered but will explain why we can’t bear attitudes toward propositions via K-stipulation. Contrary to this optimism, we believe that the “No Attitude” reply has implausible consequences, none of which will be made less implausible by any putatively necessary condition on possessing de re attitudes. The problem, put simply, is this: K-stipulators are in a position to understand K-sentences. Understanding a sentence is sufficient to be in a position to assert its content. Asserting the content of a sentence is sufficient for being in a position to bear cognitive attitudes (like belief) to that same content. Any version of the “No Attitude” reply will have to deny one of these highly plausible claims.
It is not plausible that K-stipulators do not understand the relevant K-sentences. We know, for instance, that ‘Obama is presently located at L’ means that Osama is presently located at L.25 We know that ‘L is Joshua Spencer’s back pocket’ is false and that ‘Osama bin Laden is presently located at L’ does not mean that dogs bark or that Clemens smokes. We know that the sentence is a sentence of a fragment of English (or at least our idiolects) and that we are competent speakers of that fragment of English (or at least our idiolect).26 Any more demanding constraints on understanding lead to incorrect results.27
An objector may charge that if ‘L’ refers to cave 8000 in Afghanistan, we do not know, in virtue of knowing the relevant K-sentence, that Osama bin Laden is presently located at cave 8000 in Afghanistan. But if the objector is granting that the relevant K-sentence encodes a singular proposition, the objector has merely presented another instance of Frege’s (1892) puzzle of cognitive significance. But K-names are “Millian” names—devices of direct reference. So Millians should accept that this objection shows K-stipulators do not understand K-sentences only if Millians also admit that Frege’s puzzle refutes Millianism.
According to another common objection to the claim that K-stipulators understand K-sentences, expressions mean what they do because there is some “common core” of shared beliefs among the relevant language-users. This in turn suggests the following constraint on what it takes to understand an expression: If e is an expression and S is a subject, then S understands e only if S possesses the common core of beliefs that are shared about e (or its semantic content) in the relevant linguistic community.
We agree with Williamson (2008) that the “common core of shared beliefs” view of meaning is false: Van McGee (1985) means modus ponens by his use of ‘modus ponens’. Graham Priest (2006) means true by his uses of ‘true’, and so on. This is so even though they deny what are perhaps the most central beliefs about these expressions and their semantic contents. We know of no plausible necessary conditions on understanding an expression apart from a roughly disquotational account paired with an inchoate intention to speak the relevant public language. Eccentrics about (e.g.) ‘truth’ are eccentric because they have unusual beliefs about ‘truth’ and truth, not because they have uncontroversial beliefs about something in their own private language. We conclude that proponents of the “No Attitude” reply should accept that K-stipulators understand K-sentences.
As we see it, the remaining live options for proponents of the “No Attitude” reply involve denying either that understanding a sentence is sufficient to be in a position to assert its content or that asserting the content of a sentence is sufficient for being in a position to bear cognitive attitudes (like belief) to that same content. We think that having to deny either one of these claims is painful enough to motivate looking elsewhere for a solution to our problem.
Nathan Salmon (2004, 2010) disagrees. On his view, K-stipulation puts us in a position to assert, but not believe, the relevant singular propositions. If Salmon is correct, we can tell Homeland Security agents where Osama is, but we do not believe what we say, and neither do they.
We demur. Salmon knows what he is talking about. That is, he doesn’t know what he’s talking about when he says things that entail that he doesn’t know what he’s talking about. Rather, he knows what he’s talking about when he talks about things by using a K-name. Given the intimate relation between assertion and “inward assent”, it is implausible to suppose that one may assert P but not inwardly assent to P. Intuitively, inward assent is the “mental analogue” of assertion, and inward assent is sufficient for belief.28 Suppose you assert P. Salmon grants that, even if the vehicle of assertion is a K-sentence, P may be a singular proposition. Suppose you understand every term in the utterance involved in your assertion. Now suppose you announce that you concur with what was said, nod approval to it, and seemingly bear this, that, and the other favourable attitude toward it. On Salmon’s account, if the vehicle of assertion is a K-sentence, then you fail to believe P, even after going through all of this rigamarole.
The usual function of assertion is to report beliefs. (Typically, assertion minus belief is lying.) But, on Salmon’s account, a K-stipulator who sincerely assertively utters the relevant K-sentence fails thereby to report any belief she has. Salmon’s view severs natural connections between belief and assertion, language and thought.29 On that ground alone, we are motivated to see if there is a better account.
4.3. The “No Knowledge” Reply
The “No Knowledge” Reply accepts that K-names can be introduced by K-stipulation [ND] and that K-stipulation puts one in position to assert and believe contents of K-sentences [NA], but rejects that one can thereby come to know those propositions [AK].
We think several philosophers who consider the epistemic status of the propositions encoded by K-sentences for K-stipulators become unduly fixated on the putative a priori nature of that knowledge. A number of philosophers hold that, in cases of K-stipulation in which the stipulator is in perceptual or otherwise causal contact with the relevant object, knowledge can be gained through K-stipulation. But according to these philosophers, it is invariably a posteriori. But by our lights, it is not that the propositions in question are known a priori that is objectionable. It’s that they are known at all.
To see that the problem is not really a problem of a prioricity, suppose Goldbach’s Conjecture is false. Then, presumably, there is a least number that is even, non-prime, and is not the sum of two primes. If such a number exists, call it ‘Numero’. If ‘Numero’ has a referent, then ‘Numero is the least even non-prime that is not the sum of two primes’ expresses a truth that we can putatively come to know via K-stipulation. Moreover, assuming that there is a proof that Numero is the least even non-prime, that truth is knowable a priori (if anything is). But if this putative knowledge via K-stipulation is problematic, then it is problematic in just the same way as other putative bits of knowledge via K-stipulation. Though the proposition, if any, is knowable a priori, if anything is, K-stipulation seems to allow that we could know it without doing the requisite work that is, intuitively, required for the relevant knowledge.
In order for a reply that focuses on the epistemic relationship between K-stipulators and the propositions encoded by K-sentences to truly be a solution to the puzzle, it must reject that the K-stipulators know the propositions in question simply by inferences involving information about the K-stipulation itself. Any reply that fails to do so is inadequate. And as far as we know, no one has endorsed such a “No-Knowledge” reply.
4.4. The “Accept the Conclusion” Reply: Latitudinarianism
To our knowledge, Robin Jeshion (2000) is virtually alone in accepting the conclusion that one can gain de re knowledge via K-stipulation.30 According to Jeshion, Millians who accept that the sole semantic content of a proper name is its referent should distinguish between contents on the one hand, and vehicles of content on the other.31 Furthermore, Millians should employ this distinction to address well-known problems, like Frege’s, that Millians face. Such Millians should take belief to be a mediated relation; S believes P in virtue of accepting some (perhaps mental) sentence or other, or via some vehicle of content or other.32 We may express the relationship between the agent, the proposition, and the vehicle via Salmon’s (1986, 111) BEL relation:
[BEL] If S is an agent and P is a proposition, then [S believes P iff there is a vehicle v such that S is familiar with P via v and BEL(S,P,v)]
S may believe P via one vehicle and not others. S may even rationally believe P via one vehicle and its negation via another.
Some comments are in order. First, it is important to stress that BEL is distinct from belief. The latter is a binary relation that an agent bears to a proposition, and the former is a ternary relation between an agent, a proposition, and a vehicle of content. The view on offer is not that attributions of belief or knowledge involve covert reference to vehicles, or even that they conversationally implicate anything about vehicles. By analogy, if you stand in some two-place punching relation to Mike Tyson, you also stand in some three-place relation involving you, Tyson, and your triquetral bone. If you say ‘I will punch you’, however, you are not making covert reference to your triquetral bone, nor are you (typically) conversationally implicating anything about said bone.
[BEL] quantifies over vehicles of content. ‘Vehicles of content’ is a term of art. So we owe some account of what vehicles of content are. We might suppose, for simplicity, that natural language sentences are themselves vehicles of content. (Or, more plausibly, their mental analogues.) We believe the proposition that Osama is bad via one vehicle by accepting the sentence ‘Osama is bad’. We can express the same proposition via another vehicle by saying ‘You are bad’ to Osama. Osama can believe this same proposition via a different vehicle by accepting the sentence ‘I am bad’. But neither of us believes the proposition that Osama is bad by accepting the sentence ‘I am bad’, since neither of us is Osama. Nor do we believe the proposition by accepting the Vietnamese counterpart of the English sentence ‘Obama is bad’, since we understand very little Vietnamese.33
Equipped with vehicles of content, Jeshion (2000, 309-11) proposes that the proper explanation of how K-stipulation puts one in a position to have “shortcut” knowledge is that epistemological properties of propositions, like a prioricity, are primarily properties not of propositions alone, but of pairs of propositions and vehicles. A proposition may be known a priori by being part of a <proposition, vehicle> pair that is a priori.
We have a number of worries about Jeshion’s proposal.34 But the most important one is that Jeshion’s proposal is at best incomplete. Jeshion, like a number of others, focuses extensively on the putative a priori status of one’s knowledge of the content of K-sentences. She attempts to explain away the belief that such propositions are knowable a priori by noting that many sentences used to express such propositions would not be vehicles for knowing those propositions a priori. However, as we have tried to stress, the putative a prioricity of our knowledge of propositions encoded by K-sentences is not what is primarily troubling about these cases. Rather, it is the fact that these propositions are known at all, or, at the very least, that they are known simply by K-stipulation, that is troubling. If Jeshion is to have a compete account, she must explain why these cases do not seem like cases of knowledge when in fact they are.
5. Two New Proposals
We have tried to make clear why we are dissatisfied with the current proposed solutions to the puzzle of naming by K-stipulation. We have two new proposals to offer. The first version accept [ND], [NA], and [AK] and hence accepts that we can know the contents of K-sentences simply by reflecting on our K-stipulations. On this version of our proposal we must explain why people are reluctant to attribute knowledge in such cases. The second proposal involves rejecting [AK] on the grounds that there is a further condition on knowledge that has not been met by those who come to believe the contents of K-sentences simply by reflecting on their K-stipulations. We will discuss the first proposal first and then augment the proposal for those who prefer to reject [AK] over accepting the conclusion that K-stipulators do know the contents of K-sentences by reflecting on their K-stipulations.
5.1. Information Isolation
Let us say, very roughly, that someone’s belief is informationally isolated for a particular practical purpose iff that person cannot use that information toward accomplishing that purpose. So, for example, although we might be able to use the information that Osama is at L to help us find L (simply by looking for Osama), we cannot usefully employ that information to help us find Osama. So, or belief that Osama is at L is informationally isolated for the purpose of finding Osama.35
We believe this idea of information isolation is fairly intuitive. In fact, we believe there are cases of information isolation that arise independently of K-stipulation. Suppose Greg and Andrew go on vacation to the island of Meropis. Andrew has just eaten Venenifer Pomum, a very poisonous fruit. Greg correctly believes, as a result of reading his travel guide, that Curatio Peganus contains an antidote to the poison of Venenifer Pomum. Moreover, Greg and Andrew are surrounded by the plant that contains the antidote. However, since Greg attained his belief merely by reading the sentence ‘Curatio Peganus contains the antidote to the poison of Venenifer Pomum’ in his travel guide, he is unable to act on this information and save his friend. The belief that Greg gained by reading the travel guide is informationally isolated for the purpose of saving Andrew’s life.36
That same bit of information, however, may not be informationally isolated for other purposes. If Greg were on a quiz show and the host asked him ‘Which plant contains an antidote to the poison of Venenifer Pomum?’ Greg would readily answer ‘Curatio Peganus’. His belief is not informationally isolated for the purpose of answering this particular quiz show question.
That is the rough, yet intuitive idea. But we can consider what being infomationally isolated amounts to in a bit more detail. Here is our proposal:
[II] S’s belief that P is informationally isolated under vehicle v for practical purpose A iff (i) S cannot, while considering P solely via vehicle v, use P toward accomplishing A, (ii) P is relevant to doing A, and (iii) it is not in virtue of independent constraints on S’s doing A that S cannot A.37
Note that, unlike our rough characterization, our account of information isolation is relativized to vehicles of content. We believe a useful account of information isolation must be so relativized given a commitment to Millianism.38 To see why, suppose that Berossus is a Babylonian astronomer who has a desire that he expresses with the Babylonian equivalent of ‘I want to observe Phosphorus’.39 Berossus accepts ‘Hesperus is visible in the evening’ and thereby believes that Hesperus is visible in the evening. So if Millianism is correct, he believes that Phosphorus is visible in the evening.40 However, when he considers acting on the desire he expresses with the sentence ‘I want to observe Phosphorus’ and considers the sentence ‘Hesperus is visible in the evening’, he is not moved to look up in the sky shortly after sunset. His belief of Venus that it is visible in the evening is infomationally isolated under the vehicle ‘Hesperus is visible in the evening’ for the purpose of observing Venus shortly after sunset.41
As we see it, Millians must explain why Berossus does not act in the expected way, given his beliefs. A first-pass explanation by a Millian who endorses vehicles of content is that Berossus’s vehicles “mismatch” in a way that prevents him from satisfying his desire by looking for Phosphorus in the evening sky. We also think that problems stemming from K-stipulation are particularly peculiar instances of what the Millian should see as a more general phenomenon. We think it is a virtue of a Millian view that endorses vehicles of content that it can provide an explanation of why we are reluctant to attribute knowledge in certain cases of K-stipulation that simply falls out of the more general account of failure to act on beliefs.
One might object that a better account of Berossus’s failure to act is that there is some information that Berossus is lacking. That is, one might think that his failure to act has nothing to do with the vehicles under which Berossus believes that Venus is visible in the evening. Rather, Berossus failed to act because he fails to believe some further proposition that links his belief up with his desire. Perhaps he fails to believe that Hesperus is Phosphorus.
We think this reply is mistaken. Berossus does believe that Hesperus is Phosphorus, since he accepts the sentence ‘Hesperus is Hesperus’ and that sentence encodes the proposition that Hesperus is Phosphorus. Moreover, for any belief that Berossus putatively lacks which might be pertinent to his failure to act, we might suppose that he does in fact have that belief under some vehicle or other.42 Hence it seems to us that Berossus’s failure to act is not explained by his lacking some pertinent information. Rather, his failure to act is explained by facts that advert to the vehicles via which he believes what he does.
Finally, we believe that the notion of information isolation just introduced is merely a limit case of a relation that comes in degrees. It seems plausible that sometimes a person can use one of her belief while considering it solely under a particular vehicle for a purpose, but it is a bit difficult for her to do so. And sometimes a person can so use a belief but it is very difficult to do so. For example, remember that Greg believes that Curatio Peganus contains an antidote to the poison of Venenifer Pomum, but he believes this solely on the basis of reading that sentence in a trustworthy travel guide. We might suppose, though, that once Andrew eats the poisonous fruit, Greg tries to look up more information on the antidote. Unfortunately, the only further information he can find is a blurry picture of a plant with purple flowers labelled ‘Curatio Peganus’. With this further information, Greg may be able to pick out the plant that contains the antidote. But it will be very difficult for him to do so. In this instance, it seems that Greg’s belief that Curatio Peganus contains the antidote to Venenifer Pomum is very isolated, under the vehicle by which Greg believes it, for the purpose of saving Andrew’s life. But it is not completely informationally isolated for that purpose.
It may be rather procrustean to do so, but let us suppose we can measure the degree of difficulty of various tasks so that something that is a bit difficult has a lesser degree of difficulty than something that is pretty difficult, which in turn has a lesser degree of difficulty than something that is very difficult. Then we might introduce a gradable notion of information isolation as follows:
[IIG] S’s belief that P is informationally isolated under vehicle v to degree D for practical purpose A iff (a) D = 100% and S’s belief that P is informationally isolated under vehicle v for practical purpose A or (b(i)) with D degree of difficulty S can, while considering P solely via v, use P toward accomplishing A, and (b(ii)) P is relevant to doing A, and (b(iii)) it is not in virtue of independent constraints on S’s doing A that S cannot A.
With these notions of information isolation at hand, we can now introduce our first solution to the puzzle.
5.2. Assertion and Information Isolation
Assertions give rise to expectations. It is typical to expect of someone who claims to know the present location of Osama to be in a position to track him down. We are clearly in no such position. It is typical to expect of someone who claims to know which ticket will win the lottery to be in a position to pick that ticket. We are clearly in no such position. And often our interest in whether one knows a proposition is rooted in our interest in whether they can act in certain ways on that information. Our first solution is to accept that K-stipulators can have de re knowledge via K-stipulation. We claim to have various bits of knowledge by reflecting on our semantic stipulations, though we are in no position to act in the appropriate way. Our account of information isolation explains why we cannot so act. And we think it is conversationally misleading for us to convey that we know certain things when we are in no position to act in the way that others expect one who has such knowledge to act.
Using the notion of information isolation, we can make this idea more precise. Let A be a practical purpose that conversational participants expect us to be in a position to carry out given that we know a particular proposition P. Then, if we are considering P solely under vehicle v, and we convey that we know P even though our belief that P is informationally isolated under v for purpose A, then we are being conversationally misleading.
Thus, for example, when we run into the Department of Homeland Security and say that we know where Osama bin Laden is, we are being conversationally misleading. This is so for three reasons. First, our listeners expect us to be in a position to track Osama down (or help them to do so) given that we know where he is. Second, the only vehicle by which we consider the proposition that Osama is presently located at L is the K-sentence we gained through K-stipulation. And third, our belief that Osama is presently located at L is informationally isolated under that vehicle for the purposes of tracking Osama down. We are inclined to believe that we do know where Osama is presently located via K-stipulation. But we think others are inclined to deny we have knowledge because it is (typically) conversationally misleading to convey that we have such knowledge.
Some might be inclined to say we violate a norm of assertion when we assert things that we know only by way of K-stipulation. We are happy to accommodate such a position by formulating our view as a norm of assertion:
[AII] S is conversationally permitted to assert P in a conversation only if for any practical purpose A which participants in the conversation expect one who knows P to be in a position to fulfill, for some vehicle v under which S considers P, S’s belief P is not informationally isolated under v for A.
For example, suppose we introduce ‘Tickie’ as a name for the winning New York lottery ticket for this week and come to believe that Tickie is the winning New York lottery ticket, simply by reflecting on our K-stipulation. If we go around asserting that Tickie is the winning ticket, we have violated a norm of assertion. This is because conversational participants would expect us to be in a position to buy the winning lottery ticket given that we know it is the winner. And yet, for every vehicle under which we consider that Tickie is the winning lottery ticket, our belief is informationally isolated under that vehicle for the practical purpose of purchasing the winning ticket. Again, we are inclined to believe that we do know that Tickie is the winning ticket via K-stipulation. However, we think others are inclined to deny we have such knowledge because we violate a norm of assertion when we assert the contents of K-sentences.
It is important to keep in mind that this phenomenon occurs independently of K-stipulations and that since information isolation comes in degrees, so too does permissibility of assertion. Consider Greg and Andrew again. Greg’s belief that Curatio Peganus contains an antidote to the poison of Venenifer Pomum is informationally isolated to a high degree. Given that fact, Greg should be very careful about how he conveys this particular belief so as not to misleadingly convey that he can cure Andrew. It might be best for him to say that Curatio Peganus contains an antidote to the poison of Venenifer Pomum but that he cannot identify a sample. After Greg has looked up the fuzzy picture of Curatio Peganus in his guide book, it might be more appropriate for him to assert that he can identify a sample as long as he makes it clear that it will be very difficult for him to do so.
5.3 Knowledge and Information Isolation
Some people might be reluctant to admit that K-stipulators can know the content of K-sentences just by reflecting on their K-stipulation. Such people will find our pragmatic explanation of why we are reluctant to attribute knowledge in such cases inadequate at best. Those people want to hold on to the claim that K-stipulators cannot know the content of K-sentences when those K-stipulators reflect on their K-stipulation. For those who want to hold on to such a view we have a second proposal. The second proposal rejects [AK]. One might claim that a certain necessary condition on knowledge is not met when someone comes to believe a proposition via K-stipulation. We can use the ideas presented above to introduce such a necessary condition on knowledge.
[NCK] S knows that P only if for every practical purpose, A, that S expects one who knows that P to be in a position to accomplish and for some vehicle, v, under which S considers P, S’s belief that P is not informationally isolated under v for A.
Given this principle, a person may believe the content of a K-sentence and may even be justified in believing that content simply by reflecting on his K-stipulation. However, that person does not know the content of that K-sentence because for every salient practical purpose, A, and for every vehicle, v, under which he considers the content of K-sentences, that person’s belief in the content of K-sentences is informationally isolated under v for A.43 Hence, on this view, [AK] is false.
Notice that in [NCK] we have focused on those practical purposes that the putative knower expects one to be in a position to accomplish given that one has the knowledge in question. For example, if we expect one to be in a position to track down Osama given that one knows his precise present location, then tracking down Osama is a practical purpose that is salient to whether we know that Osama is presently precisely located at L. Hence, on this view, unless there is some vehicle under which we consider the proposition that Osama is precisely presently located at L such that our belief in that proposition is not information isolated under that vehicle for the purpose of tracking down Osama, then we do not know that Osama is precisely presently located at L.
On behalf of those who wish to reject [AK] by endorsing [NCK] and claiming that we do not know the contents of K-sentences, we offer the following argument. It seems that [NCK] follows from the norm of assertion [A] given fairly plausible principles. Suppose that belief is merely like inward assertion and, hence, governed by the same norms as assertion. Then the norm captured by [A] also applies to belief. But, when one inwardly asserts, or believes a proposition, the only conversational participant is the believer herself. Hence, if the believer expects one who knows the proposition in question to be in a position to do an action, then that action is a practical purpose salient to the norm. Putting this all together, it seems that we get the following norm of belief:
[NB] S is doxastically permitted to believe that P only if for every practical purpose, A, which S expects one who knows that P to be in a position to accomplish, and for every vehicle, v, under which S considers P, S’s belief that P is not informationally isolated under v for A.
However, given that doxastically permitted belief is a necessary condition for knowledge, [NCK] immediately follows. Hence, it seems that if the norm of assertion [A] is true, then so is principle [NCK].44
Moreover, recently, John Hawthorne and Jason Stanley have argued that a proposition must be usable as a premise in practical reasoning in order for it to be known.45 We take the principle [NCK] to be one way of capturing this idea.46 If we are right, then the various motivations for Hawthorne and Stanley’s position may be motivations in favor of [NCK] as well.
5. Conclusion
The puzzle we discuss in this paper has received a significant amount of attention. However, we do not believe that any of the extant putative solutions to the puzzle ultimately work. Some of those solutions are subject to Kripke-style objections to descriptivism. Some solutions are committed to an objectionable disconnect between understanding, language, and thought. And, finally, some solutions mistakenly assume that the problem is a problem of a prioricity when in fact it is a problem of objectionable epistemic shortcuts. Our proposals have none of these problems. Moreover, our proposals explain why we are reluctant to say that people know the contents of K-sentences by reflecting on their acts of K-stipulation. On one version of our proposal, it would be misleading to say that K-stipulators know the content of K-sentences. On the other version of our proposal, K-stipulators simply don’t know the contents of K-sentences. Both versions of our proposal crucially rely on the notion of information isolation. We believe that this notion is the key to solving the puzzle discussed in this paper. Instead of placing implausible constraints on genuine naming or severing natural connections between understanding, language, and thought, we hold that the right reaction to the puzzle is that information obtained via K-stipulation is useless. If it is knowledge, it is junk knowledge. If it is not knowledge, that is because it is junk.47, 48
Notes
1. Strictly speaking, the putative knowledge would not be a priori. The proposal is that we would know by reflecting on our semantic stipulation that if Osama is precisely presently located, then he is precisely presently located at L. Our knowledge that he is precisely presently located is presumably “empirical”. If so, our knowledge that Osama is precisely presently located at L is not a priori. Additionally, Osama may fail to be precisely located anywhere, for various reasons. Really what we have in mind is a claim concerning his more-or-less precise, present location. There is no mystery involved in supposing that we know he is presently located in the Milky Way or on Earth, for example.
2. Maybe there are cardinality problems as well. Let’s ignore that; things are bad enough already!
3. Camille Flammarion alleges that Leverrier, confident in his calculations and the word of others, never bothered to actually look at Neptune through a telescope.
4. Scott Soames (2003, 411-412) argues that Kripke (unpublished) denies that we can gain knowledge through K-stipulation. Soames’s Kripke seems to think this is because the alleged knowledge flunks the sort of test for de re knowledge employed explicitly in Donnellan (1977) and implicitly in Sosa (1970).
5. For the sake of readability and brevity, we will be lax with use and mention when it is unlikely we will be misunderstood. For the same reasons we sometimes write as if certain schematic sentences have truth values, encode propositions, etc. We also advertised an account of what it is to be a K-name, but then provided a mere biconditional. We did this because we are unaware of any notational device that would allow us to unambiguously state the claim we wished to make. (We doubt ‘=df’ would suffice, for instance.)
6. We are assuming that knowing that o is P is sufficient for knowing which o is P. It is also no accident that the propositions in question are conditional: as mentioned above, the puzzle does not purport to show that one can know a priori (e.g.) where Osama is located. Rather, one can know (perhaps a priori) that if Osama is located, then Osama is located at L. Given sufficient evidence for the antecedent, one can justifiably believe the consequent, if the puzzling argument is sound. But the justification for the antecedent will almost surely be “empirical”.
7. Here we are echoing points made in Jeshion (2000, 298-302) and (2001, 113).
8. This paper was accepted for publication before the publication of Jeshion (2010). As a result, we were unable to explicitly consider a number of the papers in it that directly bear on problems stemming from K-stipulation. The one exception is Salmon (2010), manuscripts of which have been circulating for some time.
9. Proponents of this reply include Chalmers (2002), Evans (1979), Kaplan (1968), Reimer (2004), Russell (1910/11), and Stalnaker (1978). Some proponents of this reply are tricky to categorize. Russell, for instance, thought that the semantic content of one’s (verbal or mental) use of a name of an object with which one is not acquainted is purely descriptive. But Russell’s road to this conclusion strongly resembles certain motivations cited by proponents of the “No Attitude” reply who deny that one can have de re cognitive attitudes that involve objects with which one is not acquainted. Or consider the sort of positions advocated by Nathan Salmon (2004, 2010) and Scott Soames (2003, 2005). We find Salmon curiously silent on the content of certain apparently meaningful mental states. We find Soames curiously silent on the semantic content of apparel meaningful K-sentences. We assume that whatever view they would adopt would also put them in the “No Name” camp. We think problems of categorization are not very serious, however. Being a part of more categories means being subject to more problems.
Jeshion (2004) seems to endorse a version of the “No Name” reply for at least some cases of K-stipulation. She argues that, in order to introduce a name, certain preparatory conditions must be met. The prospective namer has to have the authority to name the object in question, for example. We disagree. We hereby name her unique child, if any, ‘Beatrice’. It is unclear to us why we cannot wonder of that person whether that person will be a senator as an adult. It also seems clear to us that if uses of ‘Beatrice’ grounded in the dubbing “caught on”, then it would be uncontroversial that ‘Beatrice’ is a name for Jeshion’s child (if any). So we see no good reason to suppose that ‘Beatrice’ is not now a name for that child. (But see Reimer (2004) for an opposing view.) Finally, we note that, in the experimental spirit, one of us named a colleague’s children in just this way. The names caught on and now seem to function uncontroversially in a name-like way.
10. Proponents of the “No Name” reply may alternatively hold that K-sentences fail to encode propositions, or that they encode semantically defective “gappy” propositions, which we might represent like this:
<Osama, BEING-PRESENTLY-LOCATED-AT, ___>
(See Braun (2005) and Caplan (2002) for an explication and defense of this view of the propositional content of sentences that contain non-referring proper names. Braun and Caplan do not endorse the view that all K-sentences encode gappy propositions.) We know of no one who has defended this option in print, and we suspect this is for good reason. For starters, K-sentences pass the Frege-Geach test (Frege (1918) and Geach (1958, 54)), which is strong evidence that they do not systematically fail to encode propositions. The following inferences, for example, seem perfectly valid:
Osama bin Laden is presently located at L
\Something is presently located at L
Osama bin Laden is presently located at L
\Osama bin Laden is presently located somewhere
Osama bin Laden is presently located at L
I am not presently located at L
\I am not presently located where Osama bin Laden is
It is also apparent that K-sentences can also figure into valid arguments in which a K-sentence is embedded in the antecedent of a conditional. Finally, “atomic” gappy propositions—those expressed by sentences with a simple subject-predicate structure—are all either false or lack a truth value. But one of the very few claims with near-universal assent in this debate is that K-stipulation can lead to knowledge that the relevant K-sentences, if meaningful, are true.
11. We take Chalmers (2002) and Stalnaker (1978) to hold sophisticated variants of this view. We will not separately consider their “two-dimensional” accounts here.
12. See especially Jeshion (2004, 597-598). For Kripke’s semantic arguments against descriptivism, see Kripke (1980, 80-87).
13. It is not implausible that Leverrier might have named Neptune via a description that involved his complex calculations. If he did, very few users of the name were or are in a position to grasp the relevant reference-fixing description.
14. Some may object that Tobie does not yet exist, and that we cannot refer to what does not yet exist. We think parallel points would apply were we to have used the reference-fixing description ‘the first child, if any, born in July 2010 in Manitoba’. (Thanks here to an anonymous referee.)
15. It is worth noting that descriptivism about names introduced via K-stipulation does not provide a fully general solution to the problem, even supposing its other difficulties were met. David Kaplan (1978) introduced a device that he stipulated to be directly referential: ‘dthat’. We may avoid the issue of the semantics of K-names altogether by just using ‘dthat’ in place of them. ‘Dthat’ functions by combining with definite descriptions so that the resulting compound expression semantically contributes only the denotation of the description, if any. So ‘Osama bin Laden is presently located at dthat(the present location of Osama bin Laden)’ semantically encodes a singular proposition concerning that very location. We admit that proponents of the “No Name” reply are unlikely to be more comfortable with ‘dthat’ than they are with names introduced by K-stipulation. We note, however, that one enemy of K-names, Scott Soames, makes free use of ‘dthat’ (Soames 2005, passim). Soames is not technically a descriptivist about K-names, however.
16. It is also worth pointing out that these latter points apply with equal force to “wide-scopists”. Suppose ‘Neptune might have existed and not been the cause of perturbations in the orbit of Uranus’ were synonymous with ‘The planet causing perturbations in the orbit of Uranus is such that it might have existed and not been the cause of perturbations in the orbit of Uranus’. The truth of the latter still requires the existence (and truth) of the relevant singular proposition.
17. Donnellan (1970) adopts the condition for both assertion and belief, while Salmon (2004, 2010) endorses latitudinarianism about assertion but Donnellan’s condition on belief.
18. I.e., one must know who or what object o is. Perhaps the first proponent of this view was Hintikka (1962). See also Sosa (1970), Boër and Lycan (1975, 1986), Quine (1977), and especially Salmon (1987). Scott Soames (2003, 2005) holds that the F cannot semantically fix the referent of ‘L’ for subject S unless S believes, independently of K-stipulation, of the intended referent of ‘L’ that it is the F. Other acquaintance theorists include Bach (1987/1994), Burge (1977), Evans (1982), Lewis (1983), Recanati (1993), and Reimer (2004).
19. Kaplan has since rejected his (1968) account. See Kaplan (1978, 2005). According to Salmon (2010), some descriptions count as names of what they denote, and others do not. These special descriptions are names in the sense that they license the entertaining of singular thoughts about their denotations. Rather than providing an “acquaintance” condition, this proposal tells us what things we are acquainted with. Given the proposal’s highly schematic nature it is difficult to evaluate it in more detail here.
20. For problems with the first proposal, see our criticisms of the “No Name” reply above. For problems with the second proposal, see note 10.
21. Sosa (1970, 889) makes a similar point.
22. See Sosa (1970, 889), Kaplan (1973, 516-17 fn. 19), and Kaplan (1978, 252).
23. Proponents of the popular view include Boër and Lycan (1975, 1986), Quine (1977), Sosa (1970), and Salmon (1987), among many others.
24. These examples are inspired by Salmon (2004, 246-49). Kaplan (2005) remarks in a similar vein:
We certainly comprehend the sentence ‘Timbuktu was a trading center for myrrh’ (it wasn’t) without knowing much of anything about Timbuktu or myrrh, and without ever having seen either. We don’t have to know who Aeschylus is to comprehend his name well enough to talk about him, as is shown by the fact that if we incorrectly identify him as a student of Plato, it doesn’t affect our ability to use the name to refer to him. (998)
25. Donnellan (1977) and Salmon (1987) both deny this.
26. Here we are trying to skirt very thorny issues concerning the semantics of an expression versus the semantics of an expression occurrence or use.
27. See also Kaplan (2005) in connection with this point.
28. Our own view is more along the lines of Salmon (1986):
One intuitively appealing picture that is entrenched in philosophical tradition depicts belief as a type of inward assent, or a disposition toward inward assent, to a piece of information. To believe P is to concur covertly with, to endorse mentally, to nod approval to, the information that P. In fact, the adoption of some such favourable attitude toward a piece of information is both necessary and sufficient for belief. That is just what belief is. (80)
29. Here we are following Jeshion (2001, 129).
30. Though they do not explicitly discuss K-stipulation, the latitudinarian views expressed in Sosa (1970) and Quine (1977) are, in important respects, similar in spirit. After rejecting Kaplan’s (1968) proposal involving vivid designators, Quine worries that we are left without any resource to mark out the seemingly vital contrast between merely believing there are spies and suspecting a specific person. His reaction: “At first this seems intolerable, but it grows on one.” (1977, 10)
31. Other terminology that is often used for what we mean by ‘vehicles’ includes ‘modes of presentation’, ‘modes of presentation of a proposition’, ‘propositional guises’, and ‘ways of believing a proposition’.
32. In addition to Jeshion, defenders of this sort of Millianism include Braun (2001a, 2001b, 2002), Caplan (2007), Salmon (1986), Soames (1987), and Tillman (2005, 2010). Perhaps its most devoted non-Millian detractor is Schiffer (2003, 2006).
33. The proposal that vehicles of content are natural language sentences is not plausible given Kripke’s (1979) Paderewski case and others similar to it. A more sophisticated proposal may take vehicles of content to be sentences in the language of thought or perhaps some other sort of mental entity that is equipped to play the relevant role. See Tillman (2010) for further discussion.
34. Though on Jeshion’s account we are told that a prioricity is primarily a property of propositions, we are left with no account of the epistemological status of propositions simpliciter. Two accounts suggest themselves, and neither is plausible. According to the first, a proposition is a priori simpliciter for an agent iff there exists some pair of that proposition and a vehicle that constitutes a belief state of the agent that is a priori. Jeshion should reject this account, however, since one of the main motivations for her proposal was that Millians should avoid “forfeiting the function that the notion of a prioricity has traditionally played in characterizing a possible way in which the objects of justification may be justified” (2000, 309). This account fails on this score. According to it, practically nothing winnows the domain of a priori propositions. Practically any singular proposition is knowable a priori on this account, since there is some vehicle that, when paired with it, is a priori. After all, new K-sentences are easy enough to construct.
According to the second account, if a <proposition, vehicle> pair is a priori, then so is its first member, and if a <proposition, vehicle> pair is a posterior, then so is its first member. While the first account renders Jeshion’s proposal superfluous (given her motivation for the account), the second renders it incoherent. According to the second account, any proposition knowable via both an a priori and an a posteriori <proposition, vehicle> pair that has the corresponding epistemological property will be both a priori and a posteriori. But on the traditional assumption, being knowable a priori and being knowable a posteriori are exhaustive and exclusive over the domain of knowable propositions. So no proposition has that distinction.
A proponent of Jeshion’s view might reject the propriety of talking about the a prioricity of a proposition simpliciter at all. (Jeshion herself does not endorse this view.) The previous worries go away if a prioricity is only applicable to <proposition, vehicle> pairs. But those who are attracted to the claim that (e.g.) some propositions are contingent a priori while others are necessary a priori should resist this result.
35. Our belief that Osama is at L is not informationally isolated for the purposes of finding L. However, that is not a purpose that is typically salient in a conversation.
36. One might object that there is an easy diagnosis of Greg’s difficulty that need not advert to vehicles of content, as [II] does. After all, Greg believes the general proposition that all samples of Curatio Peganus contain an antidote to Venenifer Pomum, and all he needs to put his belief into action is to acquire a de re belief of a plant in his surroundings that it is a sample of Curatio Peganus. (Thanks here to an anonymous referee). We disagree. Greg’s guidebook may indicate that there is a particular sample of Curatio Peganus named ‘Saviour’ very nearby. We think he could thereby have a de re belief of a plant in his surrounding that it is a sample of Curatio Peganus. Merely acquiring the relevant belief is not enough on our view; his new belief that Saviour is a nearby sample needs to fail to be informationally isolated for the purpose of saving Andrew’s life.
37. Some remarks are also in order about (ii) and (iii). (ii) is a relevance constraint on information isolation. In order for it to be the case that someone’s belief is informationally isolated under a particular vehicle for a particular purpose, it must be that the information is relevant to that purpose for that person. For example, the claim that Osama is at L is relevant to finding Osama. It is also relevant to finding L. But it is not relevant to buying a chocolate bar.
Now consider condition (iii). Suppose that scientists come into contact with intelligent alien life forms. The scientists are able to exchange information with the aliens. As a result of the information they receive, the scientists justifiably believe that those aliens are on a planet orbiting Alpha Centauri. Now consider the following claim:
[Claim] Those aliens with whom scientists have been communicating are on a planet orbiting Alpha Centauri.
We admit that [Claim] might be relevant to going to visit those aliens. However, the scientists are constrained by physical limitations from going to Alpha Centauri. Moreover, those physical limitations are independent of the information at hand. Hence, even though the scientists cannot, while considering the content of [Claim] under any vehicle, use [Claim] toward fulfilling the purpose of going to visit the aliens, they cannot do so because of physical limitations. But, intuitively, their belief is not thereby informationally isolated. Scientists who believe the aliens are on that planet by introducing the name ‘L’ for the present planetary location of the aliens, however, have a belief that is informationally isolated in virtue of the vehicle by which they believe what they do. (Thanks here to an anonymous referee.)
38. For further arguments to this effect, see Braun (2001a, 2001b, 2002), Caplan (2007), Salmon (1986), and Tillman (2005, 2010).
39. For simplicity we will continue to suppose that vehicles of content are ordinary language sentences and that one believes a proposition under a vehicle by accepting a sentence that encodes that proposition. But see note 33.
40. Strictly speaking, Millianism alone does not secure the inference. We suppress here standard neo-Russellian assumptions about propositions. We are also assuming without argument that Kit Fine’s (2007) account of propositions, according to which Venus is the sole semantic content of ‘Hesperus’ and ‘Phosphorus’, though the proposition that Hesperus is Hesperus is distinct from the proposition that Hesperus is Phosphorus, is incorrect.
41. The case can be sharpened by supposing that Berossus encountered a foreign astronomer who told him that there is a planet called ‘Undomiel’ which is visible in the evening. As a result of his encounter, Berossus comes to accept the sentence ‘Undomiel is visible in the evening’ and thereby believe the content of that sentence. It turns out that Undomiel is Venus. So Berossus believes that Venus is visible in the evening by way of two vehicles of content; he believes it by way of the ‘Hesperus’ vehicle and he believes it by way of the ‘Undomiel’ vehicle. Moreover, Berossus may have a desire that he expresses with the sentence ‘I want to observe Undomiel’ and when he considers the sentence ‘Undomiel is visible in the evening’, he is moved to look up in the sky shortly after sunset. Hence, although Berossus’s belief that Venus is visible in the evening is informationally isolated under the ‘Hesperus’ vehicle of thought for the purposes of observing Venus shortly after sunset, that same belief is not informationally isolated under the ‘Undomiel’ vehicle for the purpose of observing Hesperus/Phosphorus shortly after sunset.
42. One might think that the problem is that Berossus lacks the information, and hence fails to believe, that ‘Hesperus’ and ‘Phosphorus’ co-refer. However, Berossus might have encountered a logician who gives names to words. The logician might have introduced the name ‘Bob’ for ‘Hesperus’ and ‘Carol’ for ‘Phosphorus’. The logician then tells Berossus that Bob and Carol co-refer. If Berossus takes this logician to be an authority, and accepts that Bob and Carol co-refer, then he has thereby accepted that ‘Hesperus’ and ‘Phosphorus’ co-refer. So even if the metalinguistic claim about ‘Hesperus’ and ‘Phosphorus’ is pertinent to Berossus’s actions (though we think it is not), we might suppose that he in fact believes this metalinguistic claim, too. He simply believes it under the wrong kind of vehicle, given his purposes.
43. Here we are also assuming the only vehicles by which he considers the content of K-sentences are vehicles introduced when he introduced K-names that appear in those K-sentences by way of semantic stipulation.
44. If something is wrong with this argument, it is probably that inward assertion does not convey the same things that outward assertion conveys. Normally, when one outwardly asserts that P, one conveys that one knows that P to the other conversational participants. Hence, one’s assertion gives rise to expectations in the listeners that one will be in a position to accomplish those things that they expect someone who knows that P to be in a position to accomplish. However, inward assertion may not convey to oneself that one knows a proposition. Hence, inward assertion may not give rise to the same expectations of oneself that outward assertion gives rise to. We are inclined to believe that this is the best response to the argument outlined in the paragraph above.
45. See Hawthorne (2004), Stanley (2005), and Hawthorne and Stanley (2008). Their intent is more specific than the claim that it must be possible for P to be a premise in some piece of practical reasoning or other in order for it to be known. Regardless of the details of their proposals, we are here construing the condition in such a way that it is not satisfied in the present case.
46. It is worth noting that it is compatible with Hawthorne and Stanley’s being right that what is expressed by our K-sentences are justified, true beliefs. And it’s surprising, but we think correct, that one can get that via K-stipulation.
47. Strictly speaking, we should say that the vehicle via which the relevant proposition is known is junk, in the relevant sense. (Thanks here to an anonymous referee.)
48. Earlier versions of this paper were presented at the 2006 World Philosophy Day conference at the University of Manitoba, to the Department of Philosophy at the University of Winnipeg, the Department of Linguistics at the University of Manitoba, the Central Meeting of the American Philosophical Association, and the Inland Northwest Philosophy Conference. Thanks to audience members at those talks. Thanks to David Braun, Ben Caplan, Stephen Crowley, Andre Gallois, Alyssa Ney, Sarah Moss, Jonathan Sutton, and Brandt Van der Gaast for discussion. Thanks especially to Bradley Armour-Garb, Dan Korman, David Shier, and two anonymous referees for very helpful written comments. Finally, thanks to Dimitrios and Elisabeth Dentsoras for putting up with the first author naming their daughters, Kratos and Phobos.
References
Bach, K. 1987/1994. Thought and Reference. Oxford: Clarendon Press.
Beaney, M., ed. 2001. The Frege Reader. Oxford: Blackwell.
Bezuidenhout, A., and Reimer, M., ed. 2004. Descriptions and Beyond. Oxford: Oxford University Press.
Boër, S., and Lycan, W. 1975. Knowing Who. Philosophical Studies 28: 299-344.
Boër, S., and Lycan, W. 1986. Knowing Who. Cambridge, MA: MIT Press.
Braun, D. 2001a. Russellianism and Explanation. Philosophical Perspectives 15: 253-89.
Braun, D. 2001b. Russellianism and Prediction. Philosophical Studies 105: 59-105.
Braun, D. 2002. Cognitive Significance, Attitude Ascriptions, and Ways of Believing. Philosophical Studies 108: 65-81.
Braun, D. 2005. Empty Names, Fictional Names, Mythical Names. Noûs 39: 596-631.
Braun, D. 2006. Now You Know Who Hong Oak Yun Is. Philosophical Issues 16: 24-42.
Burge, T. 1977. Belief De Re. The Journal of Philosophy 74: 338-362.
Caplan, B. 2002. Empty Names. Unpublished dissertation. UCLA.
Caplan, B. 2007. Millian Descriptivism. Philosophical Studies 113: 181-98.
Chalmers, D. 2002. On Sense and Intension. Philosophical Perspectives 16: 135-82.
Donnellan, K. 1977. The Contingent A Priori and Rigid Designators. Midwest Studies in Philosophy 2(1): 12-27.
Evans, G. 1979. Reference and Contingency. In Evans 1985.
Evans, G. 1982. The Varieties of Reference. Oxford: Clarendon Press.
Evans, G., ed. 1985. Collected Papers. Oxford: Oxford University Press.
Fine, K. 2007. Semantic Relationism. Oxford: Wiley-Blackwell.
Frege, G. 1892. On Sinn and Bedeutung. In Beaney 2001.
Frege, G. 1918. Negation. In Beaney 2001.
Geach, P. 1958. Imperative and Deontic Logic. Analysis 18(3): 49-56.
Hawthorne, J. 2004. Knowledge and Lotteries. Oxford: Oxford University Press.
Hawthorne, J., and Stanley, J. 2008. Knowledge and Action. The Journal of Philosophy 105(10): 571-90.
Hintikka, J. 1962. Knowledge and Belief. Ithaca, NY: Cornell University Press.
Hintikka, J., Moravscik, J., and Suppes, P., eds. 1973. Approaches to Natural Language. Dortrecht: D. Reidel.
Jeshion, R. 2000. Ways of Taking a Meter. Philosophical Studies 99(3): 298-318.
Jeshion, R. 2001. Donnellan on Neptune. Philosophy and Phenomenological Research 63(1): 111-35.
Jeshion, R. 2004. Descriptively Descriptive Names. In Bezuidenout and Reimer 2004.
Jeshion, R., ed. 2010. New Essays on Singular Thought. Oxford: Oxford University Press.
Kaplan, D. 1968. Quantifying In. Synthese 19(1-2): 178-214.
Kaplan, D. 1973. Bob and Carol and Ted and Alice. In Hintikka, Moravscik, and Suppes 1973.
Kaplan, D. 1978. Dthat. Syntax and Semantics 9: 221-53. New York: Academic Press.
Kaplan, D. 2005. Reading ‘On Denoting’ on its Centenary. Mind 114(456): 933-1003.
Kripke, S. 1979. A Puzzle about Belief. In Margalit 1979.
Kripke, S. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press.
Lewis, David. 1983. Individuation by Acquaintance and by Stipulation. Philosophical Review 92: 3-32.
Margalit, A., ed. 1979. Meaning and Use. Dortrecht: D. Reidel.
McGee, V. 1985. A Counterexample to Modus Ponens. The Journal of Philosophy 82(9): 462-71.
Priest, G. 2006. In Contradiction: A Study of the Transconsistent. Oxford: Oxford University Press.
Quine, W.V.O. 1977. Intensions Revisited. Midwest Studies in Philosophy2(1): 5-11.
Recanati, F. 1993. Direct Reference: from Language to Thought. Oxford: Blackwell.
Reimer, M. 2004. Descriptively Introduced Names. In Bezeudenheit and Reimer 2004.
Russell, B. 1910/11. Knowledge by Acquaintance, Knowledge by Description. Proceedings of the Aristotelian Society 11: 108-28.
Ryckman, T. 1993. Contingency, A Prioricity, and Acquaintance. Philosophy and Phenomenological Research 53(2): 323-43.
Salmon, N. 1986. Frege’s Puzzle. Cambridge, MA: MIT Press/Brandon Books.
Salmon, N. 1987/88. How to Measure the Standard Metre Bar. Proceedings of the Aristotelian Society 88: 193-217.
Salmon, N. 1998. Nonexistence. Noûs 32(3): 277-319. Reprinted in Salmon 2005. References are to Salmon 2005.
Salmon, N. 2004. The Good, the Bad, and the Ugly. In Bezeudenheit and Reimer 2004.
Salmon, N. 2005. Metaphysics, Mathematics, and Meaning: Philosophical Papers Vol. 1. Oxford: Oxford University Press.
Salmon, N. 2010. “Three Perspectives on Quantifying In.” In Jeshion 2010.
Schiffer, S. 2003. The Things We Mean. Oxford: Oxford University Press.
Schiffer, S. 2006. A Problem for a Direct Reference Theory of Belief Reports. Noûs 40(2): 361-368.
Soames, S. 1987. Direct Reference, Propositional Attitudes, and Semantic Content. Philosophical Topics 15: 47-87.
Soames, S. 2003. Philosophical Analysis in the Twentieth Century Vol. 2: The Age of Meaning. Princeton, NJ: Princeton University Press.
Soames, S. 2005. Reference and Description: The Case against Two-Dimensional Semantics. Princeton, NJ: Princeton University Press.
Sosa, E. 1970. Propositional Attitudes De Dicto and De Re. The Journal of Philosophy 67(21): 883-96.
Stalnaker, R. 1978. Assertion. Syntax and Semantics 9: 315-32. New York: Academic Press.
Stanley, J. 2005. Knowledge and Practical Interests. Oxford: Oxford University Press.
Tillman, C. 2005. A Millian Propositional Guise for One Puzzling English Gal. Analysis 65(3): 251-58.
Tillman, C. 2010. Reconciling Justificatory Internalism and Content Externalism. Synthese. Online First: 23 October, 2010. DOI: 10.1007/s11229-010-9827-y.
Williamson, T. 2008. The Philosophy of Philosophy. Oxford: Wiley-Blackwell.