1. Epistemic Contextualism
Epistemic contextualism can be formulated as follows:
The Contextualist Thesis: Where attributer A in context C1 utters ‘S knows that p,’ and attributer B in context C2 utters that ‘S does not know that p,’ both attributions can be true. The simultaneous truth of both attributions is explained by reference to a difference in the epistemic standards governing context C1 and context C2, such that the semantic content or meaning of A’s claim is different from the semantic content of B’s claim.Footnote 1 In fact, attributer B is not denying the semantic content of A’s claim. Thus, it is possible for both claims to be true simultaneously without involving a contradiction.
Note that in this formulation A and B ascribe both knowledge of p and lack of knowledge of p, respectively, at the same time, to the same agent (namely S). S’s epistemic position is also the same in both contexts C1 and C2. So the change in epistemic standards stems from the change in the attributer’s context. Thus, the epistemic contextualist holds that the standards required for knowledge vary relative to the contexts of knowledge attributers. Differences in contexts of attribution bring about differences in the standards required for knowledge. Differences in the standards required for knowledge, in turn, yield differences in the semantic contents of knowledge attributing sentences. Finally, differences in the semantic contents of knowledge attributing sentences make it possible for apparently contradictory knowledge attributing sentences to have the same truth-value without involving a contradiction. The following story illustrates the epistemic contextualist thesis:
Mary and John are at the L.A. airport contemplating taking a certain flight to New York. They want to know whether the flight has a layover in Chicago. They overhear someone ask a passenger Smith if he knows whether the flight stops in Chicago. Smith looks at the flight itinerary he got from the travel agent and responds, ‘Yes, I know—it does stop in Chicago.’ It turns out that Mary and John have a very important business contact they have to make at the Chicago airport. Mary says, ‘How reliable is that itinerary? It could contain a misprint. They could have changed the schedule at the last minute.’ Mary and John agree that Smith doesn’t really know that the plane will stop in Chicago. They decide to check with the airline agent.Footnote 2
In this example, we are asked to distinguish between two claims:
(1) Smith’s claim: I know that the plane stops in Chicago.
(2) Mary’s claim: Smith does not know that the plane stops in Chicago.Footnote 3
According to epistemic contextualists, both claims are intuitively true. Contextualists cite differences in contextually provided epistemic standards to explain how they can both be true without involving contradiction. In the context in which Smith makes his claim, the contextualist tells us, it is of no particular importance to Smith whether or not the plane stops in Chicago. Smith’s interest in the question is entirely incidental. Smith therefore satisfies the epistemic standard governing that context because the circumstances under which flight schedules are printed rarely result in erroneous schedules and, given Smith’s casual interest in the question, seeing the Chicago stop in print is all the evidence he needs to justify his knowledge claim. In the context in which Mary is making her claim, however, it is important whether the plane stops in Chicago. If it were not to stop in Chicago, she would miss an important business contact. Mary’s interest is perhaps vital to her career, or at least to the success of an important assignment. So Smith cannot be ascribed knowledge on the basis of the epistemic standard governing this latter context, in which the importance of the information to the knowledge attributer (in this case, Mary) is such that she rightly worries whether the circumstances in which flight schedules are printed allow for errors. In Stewart Cohen’s terms, what distinguishes John and Mary’s context of attribution from Smith’s is that certainly possibilities—such as that of a misprint in the flight schedule—become salient once the importance of making the meeting in Chicago comes into play. In this latter context, it is not appropriate to eliminate possibilities such as misprints (which Mary mentions), sudden cancellations or route changes due to mechanical problems, and so on.
Thus, a difference in epistemic standards results in a difference in the semantic content of knowledge claims; that is, the contribution of the term ‘know’ to the semantic content of Mary’s claim is different from the contribution of ‘know’ to the semantic content of Smith’s.
Now the question arises as to what justifies the thesis that knowledge attributions are context-sensitive. The epistemic contextualist’s proposal is that the semantic content of knowledge attributions can vary from context to context because the epistemic standards governing distinct contexts can vary. But the real question is how the epistemic standard governing a given context helps to determine the semantic content of the relevant knowledge attributing sentence uttered in that context. The answer that contextualists offer to this question determines the semantic basis of their view.
2. The Semantic Basis of Epistemic Contextualism
Cohen draws a distinction between two types of semantic bases that are apparently available to underpin the contextualist account:
How from the point of view of formal semantics should we think of this context-sensitivity of knowledge ascriptions? We could think of it as a kind of indexicality … But we could instead view the knowledge predicate as expressing the same relation in every context …. So we could think of knowledge as a three-place relation between a person, a proposition, and a standard.Footnote 4
Using Kent Bach’s terminology,Footnote 5 we can formulate the above distinction as follows:
Semantic Basis 1: The epistemic standard in a given context is an essential part of the semantic content of the verb ‘to know.’ For epistemic standard D, the semantic content of a sentence such as ‘S knows that p’ is the proposition S knows Dthat p. On this construal, the knowledge relation is a two-place relation whose relata are a person S and a proposition p. “So the knowledge predicate will express different relations (corresponding to different standards) in different contexts.”Footnote 6 On such a reading, ‘know’ denotes a contextually based epistemic standard much as the pronoun ‘I’ does.
Semantic Basis 2: On this construal, the semantic content of ‘S knows that p’ will be S knows relative to D that p,Footnote 7 in which case, knowledge is a three-place relation whose relata are a person S, a proposition p, and an epistemic standard D. ‘Know’ will express the same relation in every context, but since D changes from context to context, the semantic content of ‘S knows that p’ varies by context.Footnote 8
In the contextualist literature, two ways of interpreting this Semantic Basis have emerged:
SB 2a: The epistemic standard is an unarticulated constituent. That is, there is no syntactic element denoting the epistemic standard. So, the proposition expressed by a particular knowledge attribution is incomplete. But that standard, which is specified in virtue of the context, is part of the semantic content of a given token of a knowledge attribution.
SB 2b: The epistemic standard is a hidden variable that is not articulated in the surface grammar of a knowledge attributing sentence, but is part of its underlying logical form. That is, such a sentence does express a complete proposition in virtue of a value provided for this variable.
It is interesting to note that, Cohen disclaims interest in working out the semantics of knowledge attributions in detail. He believes that doing so is not important to defending epistemic contextualism.Footnote 9 I argue that epistemic contextualism cannot be properly understood without a full account of the semantics of knowledge attributions. Epistemic contextualists are obligated to provide an adequate account of these semantics according to which a contextually derived epistemic standard completes the semantic content of a knowledge attribution. The first construal (Semantic Basis 1) commits us to the view that many arguments using the word ‘know’ that ordinary language users would judge valid, are in fact invalid due to unperceived drift in the meaning of the word ‘know.’ This construal has been challenged by philosophers of language and epistemologists.Footnote 10 They hold that there is compelling linguistic evidence that the term ‘know’ cannot be taken as an indexical term. Without explicating these arguments, I will, for the purposes of this paper, accept them and leave it to others to pursue them further. The second construal (represented by the disjunction of SB2a and SB2b), however, has attracted much less attention.Footnote 11 In what follows, then, I propose to examine and criticize SB2a and SB2b. My aim in this paper is to show that neither of them provides an acceptable semantic basis for epistemic contextualism.Footnote 12 To this aim, in Section 3, I will argue that Semantic Basis 2 (SB2) faces a general problem, a problem that both SB2a and SB2b share. In Sections 4 and 5, I will discuss worries to which each version of SB2 needs to respond.
3. Meaning, Assertion, and Epistemic Contextualism
According to the traditional picture of the relation between meaning and assertion (what is said), if speaker A asserts a sentence s relative to a context of assertion C, he expresses in C the proposition that is semantically expressed by s in C. More crucially, the proposition expressed by the assertion of s is the semantic content of s. The traditional picture of the relationship between meaning and assertion can be formulated as follows:
A sincere, reflective, competent speaker who assertively utters s (speaking literally, non-ironically, non-metaphorically, and without conversational implicatures cancelling the normal force of the remark) in a context C says (or asserts), perhaps among other things, what s ‘says’ in C (also known as the semantic content of s in C).Footnote 13
According to this traditional picture, contextualists claim, when speaker A utters the sentence ‘S knows that p,’ what is said by A is the semantic content of the uttered sentence. In other words, when A utters the sentence ‘S knows that p,’ what is asserted is assumed to be the semantic content of ‘S knows that p.’ Contextualists maintain, of course, that the semantic content of the uttered sentence is determined in part by the context of assertion, which is the context in which speaker A utters it assertively. Therefore, given the traditional picture of the relation between meaning and assertion, the underlying structure of assertion of knowledge attributions according to epistemic contextualism can be set out as follows:
Linguistic Convention (LC): If speaker A assertively utters the sentence ‘S knows that p’ in context C, then A asserts the proposition semantically expressed by ‘S knows that p’ in context C.
Suppose that speaker A in context C utters ‘S knows that p,’ and that in context C epistemic standard D governs. Now, given LC and Semantic Basis 2, we can reconstruct the contextualist’s argument as follows:
The Contextualist’s Argument (CA):
(3) (CA1) Speaker A in context C asserts ‘S knows that p’; therefore, he asserts the proposition semantically expressed by ‘S knows that p’ in context C.
(4) (CA2) The proposition semantically expressed by the sentence ‘S knows that p’ in context C is the proposition that, relative to standard D, or according to D, S knows that p.
(5) Therefore, (CA3) A asserts the proposition that S relative to standard D knows that p.
The question now is: how does mentioning epistemic standard D when stating the truth conditions for ‘S knows that p’ contribute to those truth conditions? What role does D play and how does it play it?
Now, consider the airport case. In this example, Smith assertively utters (6), in context C and Mary and John assertively utter (7) in context C*
(6) I know that the plane stops in Chicago.
(7) Smith does not know that the plane stops in Chicago.
Suppose that in context C epistemic standard D governs, whereas in context C*, epistemic standard D* governs. According to the contextualist, both (6) and (7) can be true without contradiction because the epistemic standards required for knowledge attribution differ in C and C*.
I argue that the intuitive basis for CA2 is weak and therefore unpersuasive, calling the soundness of the argument into question. As was noted, contextualists believe that speaker A, by asserting the sentence ‘S knows that p’ in context C, says what the sentence itself says in C. Contextualists believe that what the sentence ‘S knows that p’ expresses in context C is that S knows relative to D that p.Footnote 14 Consider the airport example: as was noted, the proposition expressed by Smith’s assertion is the proposition semantically expressed by ‘I know that the plane stops in Chicago,’ while the proposition expressed by Mary’s assertion is the proposition semantically expressed by ‘Smith does not know that the plane stops in Chicago.’
There is an intuitive test called a ‘disquotational indirect report’ that is designed to reveal what competent speakers of a language grasp by a sentence in that language. Intuitively, it seems that what speakers assert can be understood by considering disquotational indirect reports, that is, reports such as ‘A said that p’ or ‘in uttering s, A said that p.’15 Applying the test to reveal what competent speakers of a language understand by a sentence s in that language entails considering indirect reports about s.Footnote 16 Here a disquotational indirect report reveals that Semantic Basis 2 does not have the weight of intuition on its side. Now, consider the following indirect reports about Mary’s assertion:
• Mary said that Smith does not know that the plane stops in Chicago whereas Smith said that he knows that the plane stops in Chicago.
• Mary said that Smith’s reasons are insufficient for us to say that he knows that the plane will stop in Chicago, but I think that Smith’s reasons are sufficient for us to say that he knows that the plane will stop in Chicago (in a context in which an epistemologist is narrating the airport story for her friend).
• Mary said that that man does not know whether the plane stops in Chicago (in a context in which a passenger, pointing to Smith, is narrating the story for his son).
• Mary said that I do not know that the plane stops in Chicago, but I do know that the plane stops in Chicago (in a context in which Smith is narrating the story for a nearby passenger).
Each of these indirect reports, regarding our respective contexts, is, intuitively, potentially true. Intuitively, that is, each of the reports could be a correct description of the context in which Mary utters the sentence, ‘Smith does not know that the plane stops in Chicago.’ Each of the disquotations could have been true, depending of which factors had become salient to the knowledge attribution as the context shifted. Thus, in view of the preceding observations, the assertive content of the sentence uttered by Mary could be indicated by any of the above reports, based on the nature of disquotational indirect reports.Footnote 17 Regarding the above-mentioned indirect reports, each of the propositions expressed after the that clause could express the content of Mary’s assertion. In other words, according to the disquotational indirect report, Mary might have asserted either of the following propositions:
(8) Smith does not know that the plane stops in Chicago.
(9) Smith, relative to our situation, does not know that the plane stops in Chicago.
Contextualists believe that the best explanation of what is happening in the airport example is that the semantic content of Smith’s assertion is different from the semantic content of Mary’s assertion. In fact, contextualists claim that Mary is not denying the semantic content of Smith’s assertion. So, contextualists believe that, for instance, the content of Mary’s assertion is the proposition that ‘Smith, relative to the standard D, does not know that the plane stops in Chicago.’ But proposition (8), which is a plausible disquotation of what Mary asserts, could express the content of that assertion as well, and it provides no indication of what epistemic standard might apply in that context. Now, the question is: on what grounds do contextualists believe that what Mary has asserted is specifically the proposition that ‘Smith, relative to the standard D (or relative to our situation), does not know that the plane stops in Chicago’ and not proposition (8), which does not mention standard D? Contextualists must still explain why Mary has asserted the proposition that ‘Smith, relative to the standard D, does not know that the plane stops in Chicago.’ If contextualists want to argue for this claim, they should go for the following:
The semantic content of Mary’s assertion is the proposition that ‘Smith, relative to standard D, does not know that the plane stops in Chicago,’ but Mary, by asserting the sentence ‘Smith does not know that the plane stops in Chicago,’ could have asserted other propositions in other contexts.
If so, why do contextualists think that Mary’s and Smith’s assertions are both intuitively true and conclude from this that the semantic contents of these assertions are different? If the assertorial content differs from the semantic content,Footnote 18 how can they close the gap between the intuitive truth of an assertion and the truth of its semantic content? Also, if contextualists endorse the claim above, the question would then become: why should we accept that the semantic content of the sentence uttered by Mary is, for example, the proposition that ‘Smith, relative to standard D, does not know that the plane stops in Chicago’ and not something else? Why can we not say, with the invariantists,Footnote 19 that the semantic content of the sentence uttered by Mary is ‘Smith does not know that the plane stops in Chicago’ simpliciter, that is, without allowing the epistemic standard to enter into the semantic content of knowledge attributions?
In response, the contextualist can say that only if the semantic contents of the sentences asserted by Mary and Smith are different can we explain our intuitions about cases like the airport example. That is right, but invariantists can also say that the semantic content of the sentence uttered by Mary is ‘Smith does not know that the plane stops in Chicago,’ and then explain our intuition by invoking the distinction between the semantics and pragmatics of the sentences uttered by Mary and Smith, viz., in terms of the implicatures of these sentences.Footnote 20 Also, if we accept that the assertorial content differs from the semantic content, we can say that the assertorial content of the sentence uttered by Mary is ‘Smith, relative to our situation, does not know that the plane stops in Chicago.’ In this case, the epistemic standard would come into play due to the pragmatics of the situation. Therefore, while the semantic contents of the knowledge attributions do not vary across contexts (including those suggested as hypothetical contexts at the airport), their assertorial content does vary as those contexts vary. Why can epistemic contextualists not accept this account of the distinction between the assertorial content and the semantic content?
Suppose, however, that we grant the truth of the contextualist response. Here we discover where the problem actually lies. The contextualist believes that, in the airport example, both Mary’s and Smith’s assertions can be intuitively true and, in explaining these intuitions, he appeals to the semantic content of these assertions. In order to privilege an indirect report according to which (9), which contains the desired reference to context, indicates the semantic content of Mary’s utterance over the other possible indirect reports, the contextualist must argue that the reference to context is built into the underlying structure of assertion in the case of knowledge claims; but this seems to amount to a rejection of (LC), since the simple symmetry in (LC) between utterance and expression is now broken. The contextualist seems committed, that is, to the following variant:
LC': If speaker A assertively utters the sentence ‘S knows that p’ in context C, then A asserts the proposition semantically expressed by ‘S knows that p’ only if the proposition semantically expressed by ‘S knows that p’ is determined by epistemic standard D in context C.
If contextualists are committed to LC', it begs the question at issue. After separating the assertorial content from the semantic content, we have no other path to epistemic contextualism than to simply assume that two apparently contradictory assertions can both be true because differing epistemic standards are included in their respective truth conditions. The burden of proof is therefore on contextualists to demonstrate that if two apparently contradictory assertions satisfy LC they can both be true because of differences in epistemic standards.
Now, suppose that the second premise (CA2) of the argument (CA) is correct and that the sentence ‘S knows that p’ is true iff S knows, relative to D, that p. In the next two sections, I will argue that neither account of the semantic basis of epistemic contextualism—that an epistemic standard operates as an unarticulated constituent in such attributions or that it operates as a hidden variable—provides a semantic theory that supports epistemic contextualism.
4. Epistemic Standards as Unarticulated Constituents
Most epistemologists hold that the sentence ‘S knows that p’ is true iff S bears a knowledge relation to p, but how shall we determine whether S bears such a relation to p? Epistemologists try to answer this question by analyzing knowledge attributions. In their analysis, the verb ‘to know’ is taken to denote a two-place relation whose relata are a person S and a proposition p that is invariant across contexts. According to truth-conditional semantics, the meaning of a sentence is its truth conditions.Footnote 21 Accordingly, the truth conditions of knowledge attributions are:
‘S knows that p’ is true iff S knows that p (or S is in the knowledge relation to p).
It seems that epistemic contextualists agree with the majority of epistemologists in endorsing a truth-conditional conception of meaning.Footnote 22 They claim that the contextual provision of the epistemic standard plays a role in determining the semantic content of knowledge sentences used as knowledge attributions. They hold, following Semantic Basis 2, that the knowledge predicate expresses the same relation in every context. There are two ways to interpret this Semantic Basis:
SB 2a: The epistemic standard is an unarticulated constituent.
SB 2b: The epistemic standard is a hidden variable.
I examine SB2a in this section and SB2b in the next section. According to SB2a, no epistemic standard appears at the syntactic level of knowledge attributions, but such a standard is needed to complete their semantic content. That is, to complete the content of such sentences we need a further constituent (namely, the epistemic standard), which is determined by context (i.e., the context of the attributer). An unarticulated constituent is any part of the semantic content of a sentence that is syntactically unarticulated. For example, consider the sentence, ‘it is raining’: the semantic content of this sentence is not yet complete, and so it lacks a truth-value. The relevant location must be contextually specified before the sentence can be said to be true or false. For instance, if the sentence, ‘it is raining,’ is uttered in Paris, then its semantic content is the proposition, ‘it is raining in Paris,’ even if the person uttering the sentence does not mention Paris. If we apply this model to knowledge attributions, we can say that the epistemic standard is an unarticulated constituent iff it is ‘an element which appears at the semantic level but does not appear anywhere at the syntactic level.’23 In this case, ‘know’ has two argument places (S and p), but designates a three-place relation between a person S, a proposition p, and an epistemic standard D. On this account, the knowledge predicate expresses the same relation in every context, but there is a constituent, D (namely, the epistemic standard), that changes from context to context. Thus, we can state the truth conditions of knowledge attributions as follows:
S knows that p is true iff the denotation of ‘know’ takes D to be true, where D is the epistemic standard in the context of an assertion of ‘S knows that p’.
Now the question is: why should we assume that there is a constituent in the analysis of knowledge attributions that contributes to their semantic content but is not articulated at the syntactic level? Epistemic contextualists have not explicitly answered this question.Footnote 24 Nevertheless, considering the examples they provide to support their view (e.g., the airport example), and considering how they use the contextualist thesis to refute skepticism, a schematic description of the contextualist claim such as the one below can be constructed on their behalf.
(10) Knowledge ascribing sentences, in themselves, lack complete semantic content.
(11) For knowledge ascribing sentences to have complete semantic content, a further constituent is needed.
(12) An epistemic standard is this further constituent.
(13) Such an epistemic standard is determined by the context of the attributer.
The first proposition, (10), is the key one in this set. The epistemic contextualist’s reason for believing this proposition seems to be the intuition provoked by scenarios such as the airport case. Let us assume that it is true and that the epistemic standard is an unarticulated constituent. Now the question is: what is the criterion for detecting an unarticulated constituent?
Contextualist philosophers of language such as François Recanati hold that unarticulated constituents that are semantically relevant but syntactically unarticulated must be specified in virtue of features of the context, not in virtue of linguistic (that is, syntactic) features of the sentence, properly analyzed. Thus, Recanati says: “In context, it may be that the unarticulated constituent is ‘required’; but then it is required in virtue of features of the context, not in virtue of the linguistic properties of the expression-type.”Footnote 25
On Recanati’s view, if a sentence expresses a semantically incomplete proposition in every context, and ‘Mary dances’ is such a sentence, then it needs a constituent to become complete. Note, however, that for such a sentence this constituent does not depend on context, in virtue of its being needed for every sentence that expresses that proposition. Context does not enter into the analysis. Therefore, for such sentences this constituent must be sought in linguistic structure or syntax (but not necessarily in the surface grammar of the natural language sentence); in fact, we can say that such a constituent must somehow be articulated at some level in the syntax of the sentence. Whereas, as mentioned earlier, an unarticulated constituent in what Recanati calls the ‘strong sense’—the sense in which he is most interested because in his view it demonstrates the truth of semantic contextualism—is (by definition) syntactically unarticulated and is specified only in virtue of features of context. Recanati offers the following:
Criterion for a Syntactically Articulated Constituent:
Suppose a given sentence s needs a constituent for its content to be complete. Now if we cannot conceive of a context where s (without that constituent) has a complete content, then that constituent is not an unarticulated constituent; rather, it is articulated (at some level) in the syntax of the sentence.Footnote 26
To reinforce this idea, we can add to Recanati’s examples that of the transitive verb ‘to kiss.’ To understand the meaning of ‘S kisses’ a further constituent is needed. This sentence lacks a complete content unless some indication of whom or what S kisses is added to it. But this sentence needs this constituent in every context in which this sentence is uttered. Therefore, according to the criterion offered by Recanati, this constituent is not an unarticulated constituent. Rather, this constituent is imbedded in the structure of the verb ‘to kiss.’ Roughly speaking, based on truth-conditional semantics, ‘S kisses’ is true iff S kisses something.
Now, given Recanati’s insight about unarticulated constituents, I can offer the following:
Criterion for an Unarticulated Constituent:
Suppose a given sentence s needs a constituent for its content to be complete. Now if e is a constituent that is not syntactically articulated and we can conceive of a context where s (without constituent e) has a complete content, then that constituent is an unarticulated constituent.Footnote 27
Now the problem is whether an epistemic standard conceived of as an unarticulated constituent satisfies the above criterion. For an epistemic standard to meet the criterion, it must be shown that the content of knowledge attributions is complete in at least one context.
Suppose that epistemic contextualists claim that the content of knowledge attributions is complete in some cases without reference to a context. This claim would be odd. For it amounts to saying that context is semantically redundant in some cases, and this does not fit with the general contextualist view, according to which the context determines the epistemic standard in every case. So, the epistemic standard cannot satisfy the Criterion for an Unarticulated Constituent, since it is needed in every context. Therefore, epistemic contextualists cannot accept a version of the truth-conditional semantics of knowledge attributions, according to which the epistemic standard is an unarticulated constituent. Of course, epistemic contextualists can appeal to the distinction between metaphysical and communicational unarticulatedness.Footnote 28 They can say that the epistemic standard is unarticulated only in the metaphysical sense, not in the strong communicational sense. According to this view, there is a structural position in the logical form of knowledge attributions that is occupied by epistemic standards. In this case, although the epistemic standard is not articulated in the surface grammar of the sentence, it is part of the underlying logical form. If the epistemic contextualist follows this line of reasoning, the claim amounts to regarding the epistemic standard as a hidden variable that is part of the underlying logical form of knowledge attributions. In the next section, I examine this account.
5. Epistemic Standard as a Hidden Variable
I now consider the other likely account of the semantic basis of epistemic contextualism, according to which the epistemic standard that completes the semantic content of a knowledge attribution is not an unarticulated constituent of such an attribution but is instead a hidden variable. In this case, it seems therefore that epistemic contextualists must hold that the verb ‘to know’ is linguistically a three-place predicate, that is, it expresses a three-place relation whose relata are a person S, a proposition p, and an epistemic standard D (KSpD). On this account, contextualists must hold that an epistemic standard is a hidden variable iff (1) the epistemic standard is a component of the logical form of ‘S knows that p’ and has semantic value and (2) it is an element of the context of the attributer to the truth-conditions of ‘S knows that p.’ Contextualists might try to provide some evidence supporting this claim by driving an analogy between the predicate ‘know’ and such comparative adjectives as ‘tall’ and ‘flat.’ As mentioned, Cohen does not attempt to provide a full account of the semantics of knowledge attributions; nevertheless, he claims that there is a very close relationship between the verb ‘to know’ and gradable adjectives or predicates.Footnote 29 He writes, “I am not proposing a semantic theory for predicates of this kind. I am just proposing that we view the knowledge predicate as a predicate of this kind”.Footnote 30
Consider the sentence, ‘John is tall.’ This sentence seems to lack complete content unless a comparison class is specified. Suppose that John is 1.8 metres tall. Now if I utter this sentence in reference to a group of philosophers whose average height is 1.75 metres, the content of the uttered sentence will be true. Whereas, if I utter the same sentence in reference to a basketball team whose average height is 1.95 metres, the content of the uttered sentence will be false. According to this model, ‘tall’ is a context-insensitive binary predicate (x is tall for a y) that has an associated hidden variable y whose value in a context is a comparison property or a class. The value of the hidden variable can be supplied by an articulated predicate, as in the sentence ‘John is tall for a basketball player,’ or it can be supplied by the context in which the sentence is uttered. Thus, the comparison class is built into the linguistic or logical structure of such comparative adjectives as ‘tall’ and ‘flat.’
If an epistemic standard is a hidden variable, however, it is difficult to see how an attributer’s beliefs and intentions determine its value. Even if the value of the variable depends on the context, that is, even if the epistemic standard that determines whether S is in the knowledge relation regarding p varies with the context, that variable is part of the concept of what it is to know a proposition. Whether S is in the knowledge relation to p seems unrelated to an attributer’s standard. Here, then, it seems possible to argue that the epistemic standard is a function of the correct analysis of knowledge. If a context is involved, it is the context of S’s epistemic position, not the context of attribution. How, then, should we state the truth conditions of the assertion ‘S knows that p’ if its logical form is S knows that p relative to D or S knows that p for a D? Consider the following two formulations of its truth conditions:
(14) ‘S knows that p’ is true iff S knows that p according to the epistemic standard determined by the context in which an attributer asserts that ‘S knows that p’.
(15) ‘S knows that p’ is true iff S knows that p according to D, where D is the epistemic standard in the context of S.
Now the question is: why do contextualists believe that the truth conditions for ‘S knows that p’ are given in (14) and not in (15)? In (14), the role played by the context of the attributer is prominent, while in (15) the epistemic standard applies to S’s relation to p with no reference to an attribution made by someone else. Clearly, an epistemic contextualist such as Cohen is committed to the view that the standards required for knowledge vary with the contexts of knowledge attributers.Footnote 31
In this case, however, it seems that the epistemic standard mentioned in (14) would operate as or be similar to an unarticulated constituent. Let’s return to Cohen’s airport example. In that case, Mary’s epistemic standard (according to which the remote possibility of a misprint on a flight schedule becomes salient) is part of the meaning of her assertion that Smith does not know that the plane stops in Chicago. This means that, if I hear what Mary says but fail to grasp that she has that epistemic standard in mind, then I fail to grasp what she meant to say. But that outcome is clearly unwelcome to epistemic contextualists, because it implies that the epistemic standard is merely an unarticulated constituent, not a hidden variable that is articulated in the logical form of the attribution. And that means, in turn, that contextually provided epistemic standards, while they might affect how hearers understand attributions of knowledge, are not logically required by the semantics of knowledge attributions. In other words, if Mary’s negative attribution to Smith is true, it is the knowledge relation itself that changes in the context of attribution, not the epistemic standard that, according to epistemic contextualism, applies to the same knowledge relation in both contexts. Yet epistemic contextualism seems to depend on there being a single knowledge relation that is determined by reference to contextually provided epistemic standards. Here the shift in context (from subject to attributor) introduces a new relation altogether.
In addition, the claim that the verb ‘to know’ is a three-place predicate is not supported by linguistic evidence. To make this point clear, let me compare the verb ‘to know’ in the sentence ‘S knows that p’ and the verb ‘to rain’ in the sentence, ‘it is raining.’ When someone utters, ‘it is raining,’ he means that it is raining in a particular location. The hearer also understands the sentence as meaning that it is raining in a particular location. When indirectly reporting such sentences, the place must be mentioned in order to convey the full meaning. For example, suppose that John is speaking with his friend Mary on the phone and he tells her, ‘it is raining.’ Suppose that John is in London when uttering the sentence and that Mary is in Paris when hearing him uttering it. Competent English speakers, including both John and Mary, understand this sentence as meaning that ‘it is raining in London.’ For, when reporting what John said, Mary says, ‘John said that it is raining in London.’ Now let’s apply the disquotational indirect report test to such sentences as ‘S knows that p’: competent English speakers, when reporting someone’s, say Smith’s, utterance of the above-mentioned sentence, do not refer to the epistemic standard. They do not say, ‘Smith said that he knows relative to standard D that p.’ That is, in ordinary circumstances, competent speakers do not sense that something is missing from such a disquotational knowledge claim, whereas they do sense that something is missing when someone says only, ‘it is raining.’ They wonder where it is raining, but they typically do not wonder what the epistemic standard is in the case of ‘know.’ This provides prima facie linguistic evidence that runs against the claim that the verb ‘to know’ is a three-place relation.
Epistemic contextualists might respond that, in most cases, listeners do not consciously think of the comparison class or missing information that completes the semantic content of what others say, yet they have no problem understanding such sentences. We often report that somebody said, for example, that John is not tall without realizing that it matters whether the speaker had in mind a comparison class of adult males or of basketball players. However, a moment’s reflection reminds us that in some cases we should not disquote without making the comparison class explicit. Contextualists might contend that sentences of the form ‘S knows that p’ are similar insofar as the epistemic standards that pertain to their truth conditions are understood but not consciously entertained when such sentences are uttered assertively. Why should we not treat the case of disquoting knowledge claims in the same way? As we have seen, Cohen himself insists that ‘to know’ and words that operate as ‘tall’ does are predicates of the ‘this’—meaning the same—‘kind.’ If we do not reject the context-sensitivity of a certain group of expressions although they at first fail the disquotational indirect report test, then there is no reason to do so in the case of ‘know.’ This suggests perhaps that the contextualist can still try to use the case of gradable adjectives/predicates to show that the preceding objections fail, in the following way: if the objections are sound, they show that the semantic content of gradable adjectives does not include a contextually provided constituent; the semantic content of a gradable adjective does include a contextually provided constituent; so, the objections are unsound.
It seems, however, that there is a very good match between things that are metaphysically gradable—that is, things of such a nature that they can be graded—and expressions that can link to modifiers (adverbs such as ‘extremely’ and ‘very’) and comparisons that grade them. Take, for example, the property of being flat. By its nature, this property comes in grades—one surface can be flatter than another surface—and we can say ‘x is very flat,’ or ‘x is extremely flat.’ Similarly, consider the property of being straight. By its nature, this property comes in grades—one thing can be straighter than another thing. But it is also the case that, in language, we can express these facts. We can say, ‘this is straighter than that.’32
Now consider these sentences: ‘S very knows that snow is white,’ ‘this proposition is knowner than that proposition,’ ‘S knows that snow is white but not much,’ and so on. These sentences seem linguistically odd. Competent English language users do not usually utter these sentences. It seems that we cannot modify the knowledge predicate with such comparative and superlative forms. We do, of course, compare the relative merits of knowledge claims in remarks such as, ‘he knows more than she does,’ ‘he knows best,’ but what we are stating in the first case is that someone knows a greater number of propositions than someone else and in the second case that someone is in a better position to know than is someone else. We can also say that one proposition is better-known than another, but what we mean in such cases is that more people know the first than know the second. There are also phrases like ‘a student knows a theorem better than another student,’ or ‘one theorem is better known than another.’ In such cases, too, it seems that we just mean that the student has a better understanding of the theorem, or knows more truths about the theorem.
It seems odd, then, to suggest that a predicate is metaphysically gradable but that we lack the resources to express it as such. If the usage of such a predicate makes it metaphysically gradable, our language should reflect that to match up with the facts.Footnote 33
On the other hand, when competent English speakers utter sentences such as, ‘S knows that p,’ they do not seem to be consciously ignoring epistemic standards. Rather, they are apparently unaware of this kind of context-sensitivity and are disinclined to make epistemic standards explicit, even on second thought.Footnote 34 This then appears to mark at least one sense in which the analogy between knowledge attributions and attributions of gradable adjectives/predicates such as ‘tall’ and ‘flat’ breaks down and may then excuse epistemic contextualists from defending the view against this line of argument. I am willing to concede that, even if contextualists appeal to the context-sensitivity of gradable adjectives as a model illustrating the role of hidden variables, it does not automatically follow that contextualists regard ‘know’ also as a gradable adjective. Perhaps the context-sensitivity of ‘know’ is somehow more subtle than that of gradable adjectives, which might explain the linguistic dissimilarities that I have demonstrated. That does not, however, relieve contextualists of the burden of explaining precisely what constitutes the context-sensitivity of ‘know.’ It seems as though contextualists must simply embrace the context-sensitivity of ‘know’ sui generis, but such a ‘proposal is bound to be somewhat ad hoc.’35
Although it emerged from epistemological considerations and purports to have solved some perennial epistemological problems, epistemic contextualism is actually a semantic thesis. It tries to solve epistemological problems (such as the problem of skepticism) by means of a thesis about the meaning of knowledge attributing sentences. On this view, the semantic content (truth conditions) of knowledge attributing sentences changes with the context of the knowledge attributer. In the context of skepticism, where the epistemic standards are high, the content of the sentence, ‘S does not know that p,’ can be true; whereas in everyday contexts, where the epistemic standards are low, the semantic content of the sentence, ‘S knows that p,’ can be true. The simultaneous truth of these two knowledge attributions is possible, according to epistemic contextualists, because the difference in the epistemic standards governing these two kinds of contexts changes the semantic contents of the sentences. This is epistemic contextualism’s general strategy for solving the problem of skepticism. However, for this strategy to succeed, epistemic contextualists must provide a full account of the semantic basis of the thesis.
In this paper, I examined two possible semantic bases of epistemic contextualism, namely the claim that the epistemic standard is an unarticulated constituent of the semantic content of knowledge ascribing sentences and the claim that the epistemic standard is a hidden variable. I have argued that, in the first case, an unarticulated constituent is inconsistent with epistemic contextualism and that, in the second case, the role of hidden variables in determining the truth conditions of knowledge attributions remains unclear. I conclude that epistemic contextualism is an incomplete doctrine.
Acknowledgements:
I would like to express my deep gratitude to my dissertation advisor, Hamid Vahid, for his invaluable support and guidance. I would also like to sincerely thank Hashem Morvarid for his great help in editing the language of the paper. I am also thankful to Sajed Tayebi, Amir Saemi, and anonymous reviewers of this journal for their insightful comments on earlier drafts of the paper.
