Philosophical Studies

, Volume 163, Issue 2, pp 561–574

On nonindexical contextualism



DOI: 10.1007/s11098-011-9831-1

Cite this article as:
Davis, W.A. Philos Stud (2013) 163: 561. doi:10.1007/s11098-011-9831-1


MacFarlane distinguishes “context sensitivity” from “indexicality,” and argues that “nonindexical contextualism” has significant advantages over the standard indexical form. MacFarlane’s substantive thesis is that the extension of an expression may depend on an epistemic standard variable even though its content does not. Focusing on ‘knows,’ I will argue against the possibility of extension dependence without content dependence when factors such as meaning, time, and world are held constant, and show that MacFarlane’s nonindexical contextualism provides no advantages over indexical contextualism. The discussion will shed light on the definition of indexicals as well as the meaning of ‘knows,’ and highlight important constraints on the way meaning can be represented in semantics.



1 Context sensitivity versus indexicality

As MacFarlane (2007, pp. 247–8; 2009, p. 232) defines his terms, an expression is context sensitive iff its extension at a context depends on features of the context, and indexical iff its content at a context depends on features of the context. Here MacFarlane is focusing on contexts of use, although he is best known for his work on contexts of assessment (e.g., MacFarlane 2005a). MacFarlane takes contexts (of use) to be characterized by a speaker, a place, a time, a world, an epistemic standard, and possibly other factors. He allows that the world characterizing a context may be non-actual.

MacFarlane is aware that ‘indexical’ is commonly used as a synonym of ‘context sensitive.’ Indeed, indexicals are usually defined as “linguistic expressions whose reference shifts from context to context” (Braun 2007, p. 1), with ‘reference’ and ‘extension’ used interchangeably.1 The big problem with this definition is that it is hard to find any term in natural language whose extension does not depend on and vary with the context in some way. First, all contingent expressions have different extensions when used in different possible worlds. Thus the extension of ‘the 43rd president of the United States’ is George W. Bush at any context in the actual world, but Al Gore at any context in a world in which Gore won the 2000 presidential election. Second, non-eternal terms have different extensions at different times. Thus when used in 1920, the extension of ‘airplane’ did not include any jets, whereas its extension in 2010 includes many jets. Third, every natural language expression I’ve ever examined is ambiguous, in the broadest sense of having multiple meanings or interpretations. The extension of an ambiguous term at a context depends on the sense with which it is used in that context. Thus the extension of ‘plane’ in an actual context depends on whether it means “airplane,” “wood plane,” or “geometric plane” on that occasion. Since ‘plane’ could have had, or could come to have, different meanings, its extension in a non-actual context depends on what meaning it has in that world at the time.

The terms ‘context sensitive’ and ‘indexical’ are customarily applied to terms like ‘I,’ ‘here,’ ‘now,’ ‘he,’ ‘local,’ and ‘foreign’ to denote the way in which they differ markedly from expressions like ‘plane’ and ‘the 43rd president of the United States.’ What is special about the terms commonly classified as indexical or context sensitive is that their extension varies with the context of use even when their linguistic meanings and the world and time of evaluation are held constant. The extension of ‘the 43rd president of the United States’ does not vary with the context when those three factors are held constant. As long as we hold fixed its meaning and the world it is evaluated in, ‘the 43rd president of the United States’ will have the same referent no matter which world it is used in.

MacFarlane (2009, p. 233) thinks it is obvious that the content of a sentence can vary from context to context without its extension varying. He cites Tomorrow comes after today, claiming that it has the same extension (True) in every context even though its content varies from context to context. This sentence is true in every context when the ‘tomorrow’ and ‘today’ have their typical deictic use. But with other uses of these terms, the sentence may be false. Like the most typical indexicals, ‘today’ can be used demonstratively or anaphorically as well as deictically. One might use ‘tomorrow’ deictically on December 1 to refer to the next day while using ‘today’ demonstratively, pointing at December 25 on a calendar. As used in this context, ‘Tomorrow comes after today’ is false. Or we might use ‘today’ anaphorically, as in ‘December 25: today comes before tomorrow,’ which would also be false if ‘tomorrow’ is used deictically on December 1. Sentences with indexicals that are true in every context are surprisingly rare. One example is I am myself, in which the syntax and semantics of the reflexive pronoun guarantee that it is coreferential with the other first person pronoun. The fact that ‘I am myself’ would commonly be classified as indexical can be accommodated by classifying a sentence as indexical if either it or one of its components has an extension that varies with the context of use even when its meaning and the circumstances of evaluation are constant.

The question I wish to investigate, however, is whether the converse possibility exists. Can the extension of a term or sentence vary with the context of use without its content varying, when meaning, world, and time are held constant? What I will call MacFarlane’s thesis answers “Yes”.2 MacFarlane further argues that epistemic contextualists would be on firmer ground if they claimed that what varies with the context is the extension of ‘know’ rather than its content. We will follow MacFarlane in calling this thesis nonindexical contextualism. MacFarlane suggests that other expressions may be extension variable without being content variable, but we will focus on ‘know.’

2 Content

Can extension vary independently of content even when meaning, world, and time are held constant? In order to investigate this claim, we need to understand what MacFarlane means by ‘content.’ On one common interpretation, the content of an expression is its linguistic meaning. So interpreted, MacFarlane’s thesis would be old news, proved by the existence of the personal pronoun, whose extension varies when used by different people even though it is used with the same linguistic meaning and evaluated at the same world at the same time. MacFarlane (2009, p. 232, fn. 2) has a highly original thesis because he adopts an even more common definition: “the content of a sentence at a context is a proposition, and… the content of a subsentential expression is the contribution it makes to the content of the sentences containing it.” MacFarlane does not say much about what he takes propositions to be, except for a parenthetical remark that they are “what is said or asserted or believed” (2009: 248). He indicates later that he takes the word ‘know’ to express a relation, which has an intension. Relations and intensions are thus candidates for the contents words contribute to the contents expressed by sentences.

Intensions are commonly defined as functions from worlds to extensions. MacFarlane follows Kaplan (1977, p. 494) in taking intensions to be functions from circumstances of evaluation to extensions. When we ask whether a sentence is true in a given context, the circumstance of evaluation includes the world within which that context occurs. We can also ask whether the sentence, as used in that context, is true in other circumstances. As used by me now, ‘I am here’ is true because it is true in the actual world; but it is not true in a world in which Wayne Davis is at the beach. Kaplan (1977, pp. 512–3) was clear that when ‘I am here’ as I used it is evaluated in another world, we hold fixed the linguistic meaning and content it had in my context, and see whether that content is true at the other world. If factors other than worlds influence the extension of a sentence, then they need to be included in circumstances of evaluation, and intensions need to be defined as functions from circumstances to worlds. Kaplan included times as well as worlds, and allowed that additional parameters may be needed. MacFarlane’s view is that circumstances of evaluation include an epistemic standard parameter to handle epistemic terms. On contextualist views, the value of the epistemic standard parameter is set by the context.

The thesis that extension may vary independent of content even when meaning, world, and time are held constant may now seem to follow directly from three premises: (i) circumstances of evaluation include parameters other than worlds and times; (ii) intensions are functions from circumstances of evaluation to extensions; and (iii) intensions are or represent contents. But in fact, a fourth premise is necessary: (iv) the value of an intension function is sometimes affected by the values of parameters other than the world and time. That is, intension functions are not constant over the other parameters. Thus inclusion of an epistemic standard parameter in circumstances of evaluation will entail that the extension of an instance of ‘S knows p’ may vary independently of its content even when world and time are held constant only if the values of that parameter affect the truth-value of the sentence. In order to isolate the effect of the standards parameter, we will focus on the actual world and the present time.

On contextualism, the epistemic standard parameter is determined by factors such as the interests of the speaker, what is at stake for the speaker, and what is salient to the speaker. The problems confronting contextualism are due to the fact that it makes the truth-value of knowledge claims dependent on features of the speaker’s context that are independent of the truth of ‘p.’ So the problems facing contextualism are evidence that the extension of ‘S knows p at t’ does not vary with such truth-independent features of the speaker’s context. Why does MacFarlane believe that the extension of a knowledge claim could vary independently of its content? How does he think allowing nonindexical forms of contextualism is advantageous? We will turn now to these questions.

3 Knowledge claims

The indexical contextualist believes that the extension of ‘S knows p at t’ varies from context to context because its content does. The thesis can be formulated well-enough for present purposes as follows:
  1. (1)

    When it is used in a context C, ‘S knows p at t’ expresses the proposition that ‘p’ is true and S’s epistemic position with respect to ‘p’ at t meets e(C).3


MacFarlane characterizes e(C) as the epistemic standard “in play” at C. On indexical contextualism, despite having only three explicit argument places, the sentence ‘S knows p at t’ expresses a four-place relation in any context, the fourth relata e being supplied by the context. In the same way, sentences of the form ‘S knows p’ are well-formed with only two argument places because context supplies the time of utterance as the third relata via the present tense.

Indexical contextualism faces a number of problems. Suppose Sam is in one context, and Barry another, and imagine them saying the following.
(2) Sam:

Alice knows her car is in the driveway


Sam said that Alice knows her car is in the driveway

Given what Sam said, Barry’s report seems perfectly correct. But on the version of indexical contextualism under consideration, Barry’s report is false if Barry is in a high standards context while Sam is in a low standards context. In that case, Sam said Alice meets a low epistemic standard, but Barry described Sam as saying that she met a high standard. As MacFarlane puts it, if this form of contextualism is correct, then:

The word “know” expresses a different relation in Barry’s mouth than it did in Sam’s mouth. So Barry’s report must misrepresent the content of Sam’s claim…. (MacFarlane 2009, p. 239).4

While MacFarlane’s claim is true for indexical contextualism as MacFarlane defined e(C), other indexical forms of contextualism could allow e(C) to be either the epistemic standard of the speaker’s context or that of the subject’s context when ‘know’ is in a subordinate clause. That is how many indexicals behave. In Barry’s report in (3), for example, ‘nearby’ may refer to a location near Barry, or to one near Alice (Cappelen and Hawthorne 2009, p. 34).
(3) Sam:

Alice’s car is nearby


Sam said that Alice’s car is nearby

As long e(C) varies from one context to another, (1) defines an indexical contextualism.

That said, (2) provides an objection even when indexical contextualism is generalized in this way. For whereas any clear indexical can block an indirect speech report, ‘know’ cannot. Whereas Barry’s report in (3) has an interpretation on which it is incorrect, his report in (2) has no incorrect interpretation. ‘Know’ cannot express a different relation in Barry’s mouth than in Sam’s.

MacFarlane (2009, p. 240) believes that nonindexical contextualism can avoid this problem. He suggests the following theory:

[T]he word “know” expresses the same relation at every context of use: the relation a subject S stands in to a proposition p and a time t just in case S knows that p at t. This relation, like others, has an intension: a function from circumstances of evaluation (world/epistemic standard pairs) to extensions. We can specify this intension roughly as follows: a subject S stands in the knowing relation to p at t at (w, e) iff p is true at (w, e) and S is in a strong enough epistemic position at w and t with respect to p to satisfy the standard e. (MacFarlane 2009, p. 236).

The problem with what MacFarlane says here is that if the truth-value of the proposition expressed by ‘S knows p at t’ in every context depends on a variable e that is independent of S, p, and t, then there is no unique relation a subject S stands in to a proposition p and a time t just in case S knows p at t. For S’s epistemic position at t with respect to p may meet the standard e(C1) without meeting the standard e(C2).

To make this concrete, suppose the bank will be open on Saturday, and suppose Alice has moderately good evidence that it will be. The time is the present, and the world is the actual world. Suppose Alan is talking about Alice in a context A with low standards, while Bob is talking about Alice in a context B with high standards. Then Alice meets the epistemic standards of A but not of B. Can we say whether Alice stands in the knowing relation to the proposition that her car is in the driveway? Not given the truth-conditions MacFarlane provided. According to those truth-conditions, all we can say is that Alice stands in the knowing relation to that proposition and e(A), but does not stand in the knowing relation to that proposition and e(B). The intension MacFarlane has provided is that of a four-place relation between S, p, t, and e. So he has not described a form of contextualism according to which ‘knows’ expresses the same three-place relation in every context.5

To see the problem in another way, consider the following definition of a relation, in which q(x) is the measure of quantity q possessed by x. To make sure we rely on the definition provided, and not our antecedent understanding, I use the nonsense verb ‘grear.’
  1. (4)

    x grears y at t in q iff q(x) > 1.25q(y) at t.

We can thus say that Alan grears Bill at the present time in height, while Bill grears Alan in net worth. These truth-conditions define a four-place relation, not a three-place relation. Consequently, Alan grears Bill at the present time makes no sense without specifying a quantity. We cannot say that ‘x grears y at t’ expresses the relation of x grearing y at a time t.

Perhaps the nonindexical contextualist could maintain that ‘knows’ always expresses the same four-place relation, the one represented by the intension function MacFarlane defined. This would be a relation between a subject S, a proposition p, a time t, and an epistemic standard e. The problem is that any substitution instance of ‘S knows p at t’ will specify only three of the four relata. So ‘Alice knows her car is in the driveway now’ would not express a proposition; it would be equivalent to the open sentence ‘Alice knows her car is in the driveway now at e,’ where ‘e’ is a free variable. What it expresses may be true at some epistemic standards and false at others, but cannot be true or false itself. Contextualism maintains, though, that instances of ‘S knows p at t’ express propositions, which can be true or false. Any adequate semantics must do so. The contextualist cannot maintain that ‘knows’ expresses a four-place relation between S, p, t, and e without being an indexicalist, holding that the value of e, and therefore the proposition expressed, is determined by features of the context of use.

MacFarlane says that ‘know’ expresses “the relation a subject S stands in to a proposition p and a time t just in case S knows that p at t.” His truth-conditions may seem to tell us when S knows p at t given that they tell us when S knows p at t at any circumstance of evaluation (w, e). But (5)(b) cannot be validly inferred from (5)(a).
  1. (5) (a)

    S knows p at t at (w, e).

  2. (b)

    S knows p at t.

This is particularly clear given that w may be a non-actual world. ‘Alice knows she has three hands at a world in which she has three hands’ does not entail that Alice knows she has three hands. But even when we let w be the actual world, there is no way to get from (5)(a) to (5)(b) on contextualism. There is no uniquely “correct” epistemic standard according to contextualism. If S knows p at t at e1 but not at e2, we have no basis for inferring that S either does or does not know p at t.

MacFarlane’s truth-conditions do suffice to define a function m from pairs (w, e) to truth-values for each n-tuple (S, p, t): mspt(w, e) = T iff p is true at (w, e) and S is in a strong enough epistemic position at w and t with respect to p to satisfy the standard e. MacFarlane suggests that assigning such a function to instances of ‘S knows p at t’ accounts for their semantics. It does not. An adequate semantics must do more than tell us when an instance of ‘S knows p at t’ is true at a circumstance of evaluation, it must also tell us when it is true. If we assign ‘S knows p at t’ a standard intension function kspt(w) from worlds to truth-values, we have done this. For it would tells us that ‘S knows p at t’ is true when kspt(α) = T, where α is the actual world. But because there is no “correct” epistemic standard according to contextualism, there is no special epistemic standard ε such that ‘S knows p at t’ is true iff mspt(α, ε) = T. If we assign a Kaplan-style function nspt(w, t), we can also satisfy this requirement by picking a function constant over t. Then ‘S knows p at t’ is true when mspt(α, t) = T for every t. Times are needed in circumstances of evaluation to assign extensions to terms, which have different extensions at different times (recall ‘airplane’). When a sentence has a temporal reference provided either explicitly by an adverbial phrase ‘at t’ or indexically by tense, it does not have different truth-values at different times; so the time parameter in a circumstance of evaluation is idle. The whole point of contextualism, however, is that the truth-value of ‘S knows p at t’ varies with e(C).

Perhaps the contextualist can provide a formal semantics for monadic truth in terms of truth at a circumstance of evaluation (Cf. MacFarlane 2009, p. 243; Cappelen and Hawthorne 2009, pp. 13–14). MacFarlane describes the semantics of a nonindexical expression by assigning it an intension, a function from circumstances of evaluation to extensions. ‘True’ applies to sentences (or the propositions they express), so the intension for the monadic predicate ‘true’ would be a function from circumstances of evaluation to sets of sentences. On MacFarlane’s view, circumstances of evaluation contain an epistemic standard and a world. So the semantics of ‘true’ would be described by assigning it a function from pairs (w, e) to sets of sentences. Specifically, the intension of ‘true’ will assign to (w, e) the set of sentences that are true at (w, e). To be adequate, the semantics must assign ‘Σ is true’ to the extension of ‘true’ at (w, e) iff it assigns Σ to the extension of ‘true’ at (w, e), at least when Σ expresses a fixed non-paradoxical proposition. This will secure the result that instances of the disquotation schema (6) are in the extension of ‘true’ at every (w, e).
  1. (6)

    ‘S knows p at t’ is true iff S knows p at t.

In general, an adequate semantics for ‘true’ must satisfy the following condition:
  1. (7)

    ‘p’ is true iff p is true at all circumstances of evaluation.

The success of this account depends on the assumption that the semantics of ‘true’ or any other term can be adequately represented by assigning it an intension that varies with an epistemic standard parameter. Why should we accept that assumption? The fact that it yields results like (6) and (7) may appear to provide a good reason, but it does not. For as we have seen, the nonindexical contextualist semantics described above does not tell us when S knows p at t. Consequently it does not tell us when ‘S knows p at t’ is true even given (6). The most we can get from (6) and nonindexical contextualism is that ‘S knows p at t is true’ is true at (w, e) iff ‘S knows p at t’ is true at (w, e).
Consider ‘grear’ again. Let the circumstances of evaluation include the quantity parameter q. Suppose a semantics assigns ‘x grears y at t is true’ to the extension of ‘true’ at q iff it assigns ‘x grears y at t’ to the extension of ‘true’ at q. This will secure the result that instances of disquotation schema (8) are in the extension of ‘true’ at every (w, e):
  1. (8)

    ‘x grears y at t’ is true iff x grears y at t.

(8) does not tell us when ‘x grears y at t’ is true because the right side of (8) makes no sense given definition (4). So the fact that a semantics yields a result like (8) is proof of its inadequacy. (6) is of course meaningful and true in English. So the fact that a formal semantics generates it may seem like a good sign. But given the nonindexical contextualist semantics MacFarlane has provided, it should be no more meaningful than (8).

As a final demonstration of the limitations on putting relata in the circumstances of evaluation, consider a semantics in which the circumstances are quadruples (p, t, w, e) consisting of a proposition, time, world, and epistemic standard. Use MacFarlane’s truth-conditions for the knowing relation to define a function from (p, t, w, e) to truth-values for each subject S: ms(p, t, w, e) = T iff p is true at (w, e) and S is in a strong enough epistemic position at w and t with respect to p to satisfy the standard e. Now assign each instance of ‘S knows’ the intension ms(p, t, w, e). Does this enable us to say that ‘S knows’ expresses the same property in every context, the property of knowing, which is the property S possesses at (p, t, w, e) iff p is true at (w, e) and S is in a strong enough epistemic position at w and t with respect to p to satisfy the standard e? Does this nonindexical contextualism tell us when Alice has that property? Certainly not. If ‘knows’ did express a non-relational property, furthermore, it would be hard to understand how it could take a propositional object.6

4 Conceptual contents

MacFarlane could perhaps avoid this problem by maintaining that propositions are composed of concepts rather than relations. He could hold that the verb ‘know’ expresses a concept that combines with a subject concept, a propositional concept, a temporal concept, and an epistemic standard concept to form the proposition that S knows p at t by epistemic standard e. What works best for MacFarlane’s uniformity claim is the thesis that ‘S knows p at t’ expresses the same complex of concepts on each occasion of use. MacFarlane would have to maintain that this complex is an incomplete proposition in the sense that it (i) does not have a truth-value and is not an object of belief in itself, but (ii) can be a part of propositions that are completed by the addition of concepts representing standard e.7

There are two problems with this conceptual approach. First, as MacFarlane (2009, p. 240) notes following Stanley (2004, p. 138), if ‘S knows p at t’ did express such an incomplete proposition, we would expect to be able to expand sentences of the form ‘S knows p at t’ by adding clauses that express the different standards concepts. But we never say things like (9), which seems if anything to be an oblique way of saying that Alice does not really know.
  1. (9)

    Alice knows her car is in the driveway by low standards, but not by high standards.

On the suggested version of nonindexical contextualism, though, we should be able to hear sentence (9) as reconciling the apparently contradictory claims in (10), just as we can hear ‘Alice is strong by little league standards but not by big league standards’ as reconciling the apparent contradiction between utterances of ‘Alice is strong’ and ‘Alice is not strong.’
(10) Sam:

Alice knows her car is in the driveway


Alice does not know her car is in the driveway

But we cannot. We hear the two claims in (10) as intractably contradictory.8
The second problem, which MacFarlane did not bring up, is that if what ‘S knows p at t’ expresses on all occasions is a propositional part that is not an object of belief and lacks a truth-value, then it is not that what Sam said in (2), which is what Barry said he said. This is not the way we understand (2). What Sam said is either true or false, and can be believed or disbelieved. On nonindexical contextualism, ‘Alice knows her car is in the driveway’ should be much like (11).
  1. (11)

    Alice finished.


‘Alice finished’ is a complete sentence, unlike ‘Alice completed.’ Nevertheless, there is a complex of concepts that ‘Alice finished’ expresses in every context. But that conceptual complex is not a proposition and has no truth-value. ‘Alice finished’ is also used to express a proposition, but different ones on different occasions. ‘Alice knows her car is in the driveway’ does not need to be “completed” by reference to an epistemic standard the way ‘Alice finished’ needs to be completed by reference to an object.

5 Sentence occurrences

MacFarlane says that the key to understanding nonindexical contextualism lies in a rule provided by Kaplan.

  1. (12)

    If c is a context, then an occurrence of φ in c is true iff the content expressed by φ in this context is true when evaluated with respect to the circumstance of the context. (Kaplan 1977, p. 522) (MacFarlane 2009, p. 236; see also MacFarlane 2007, 246.)

Let φ be ‘Alice knows her car is in the driveway (at the present time).’ According to nonindexical contextualism, this sentence expresses the same content in every context. Let that content be K. We imagined that φ occurs in both A and B. MacFarlane wants us to suppose that K is true relative to e(A) and false relative to e(B). Why then should we say that the occurrence of φ in A is true just because the content K it expresses is true relative to e(A) when K is false relative to e(B)? On contextualism, no standard can be singled out as the correct standard. And on nonindexical contextualism, the meaning of φ has no indexical element referring to features of the context the way ‘I am here’ does. Moreover, what φ expresses entails nothing about A, B, or standards of evidence. How can the occurrence of a sentence—a mere string of letters or speech sounds—get a truth-value independent of its meaning and content? If the speakers in A and B meant the very same thing by φ, and used it to express exactly the same content, then how could one occurrence be true while the other is false?
As we noted in §3, ‘φ is true’ cannot be inferred from ‘φ is true at E’ when E contains a non-actual world or an epistemic standard parameter. Kaplan’s rule works fine for Kaplan provided we apply it to actual occurrences. Thus the following inference is valid when the circumstances of evaluation contain nothing but a time and a world:
  1. (13) (a)

    Wayne Davis uttered ‘I am in Kirchberg’ on August 9, 2011.

  2. (b)

    The occurrence of ‘I am in Kirchberg’ in that context was true at the circumstances of the context.

  3. (c)

    ∴ The occurrence of ‘I am in Kirchberg’ in that context was true.

The premises of (13) imply the conclusion because premise (a) entails that the world of the context is the actual world. The premises in (14) imply the conclusion for the same reason if the circumstances of evaluation contain nothing but a world and a time.
  1. (14) (a)

    Sam uttered ‘Alice knows her car is in the driveway’ on August 9, 2011.

  2. (b)

    The occurrence of ‘Alice knows her car is in the driveway’ in that context was true at the circumstances of the context.

  3. (c)

    ∴ The occurrence of ‘Alice knows her car was in the driveway’ in that context was true.

But on nonindexical contextualism, the circumstances of evaluation contain an epistemic standard parameter. From the fact that a knowledge claim is true at a particular standard we cannot infer that it is true. So (14) is invalid on nonindexical contextualism.
To see an independent problem, we need to assume for the sake of argument that the nonindexical contextualist could adopt Kaplan’s rule. It is natural to identify the utterance of an expression with the production of an occurrence,9 so that the utterance of φ in C is true iff the occurrence of φ in C is true, as MacFarlane (2009, pp. 247, fn. 20; 248) maintains.10 It is similarly natural to identify the use of a sentence with making it occur in order to express a proposition or thought, and to count the sentence, as used in C or by S, to be true iff the proposition expressed is. But then nonindexical contextualism with Kaplan’s rule would allow occurrences of (15) and (16) to be true when Sam is in a low standards context and the speaker is in a high standards context (cf. Kompa 2002).
  1. (15)

    Sam’s utterance of ‘Alice knows her car is in the driveway,’ and the occurrence of that sentence in his context, are true; but Alice does not know her car is in the driveway.

  2. (16)

    As used by Sam, ‘Alice knows her car is in the driveway’ is true, but Alice does not know her car is in the driveway.

But in fact, such occurrences seem contradictory.11 Kaplan’s rule (12) would make nonindexical contextualism acquire all of the well-known problems for indexical contextualism that arise from its making the truth of an utterance depend on standards prevailing in the context of utterance. It would entail that Sam’s utterance of ‘Alice knows her car is in the driveway’ does not contradict Tom’s utterance of its negation when their contexts differ only because the standards in Tom’s context are higher than those in Sam’s. And it would allow that occurrences of (17) may be true when the speaker is in a low standards context while Alice is in a high standards context.
  1. (17)

    Alice knows that her car is in the driveway, but she does not have enough evidence to assert that her car is in the driveway.

Sentences like (17) sound as inconsistent as the others.
MacFarlane addresses the problem posed by sentences like (15), and responds as follows.

This certainly sounds weird on first hearing, but I’m not sure we should be bothered by it once we realize that utterance truth is a technical notion. In ordinary speech, people predicate truth of propositions (that is, of what is said or asserted or believed), not of utterances. If utterance truth is a technical notion, we had better make sure our intuitions about it are in line with our theories, not the other way around. Rejecting a theory because it makes predictions about utterance truth that “sound funny” is not sound methodology. (MacFarlane 2009, p. 248).

It is true that many linguists and philosophers of language use ‘utterance truth’ and ‘utterance meaning’ in technical ways, and that ‘sentence occurrence’ is uncommon in everyday speech. But (15) nonetheless seems to be a meaningful English sentence that we would naturally interpret to be false. As for (16), there is nothing at all out of the ordinary about saying that a sentence, as used on a given occasion or by a particular speaker, is true or false. The intuitive absurdity of (16) cannot be chalked up to its employment of any term with a technical sense that deviates from its conventional sense. Sentence truth is not a technical notion. The contradictoriness of (17) is similarly immune to MacFarlane’s response, and shows that the problem with nonindexical contextualism is its contextualism.12

If it were correct, moreover, MacFarlane’s response would deprive nonindexical contextualism of any apparent importance (Cappelen and Hawthorne 2009, p. 23). In particular, the main source of support for contextualism would be completely lost. That support is provided by the linguistic intuition that knowledge claims may be true in low standards contexts and false in high standards contexts (see Cohen 1986, 1988; DeRose 2009). Without that intuition, contextualism would never have been proposed. We saw in previous sections that nonindexical contextualism either provides no account of what speakers say when they use ‘S knows p at t,’ or it rules that what they say is neither true nor false. That is why Kaplan’s rule regarding the truth of a sentence occurrence is the “key” to understanding MacFarlane’s thesis. For Kaplan’s rule provides monadic truth-conditions for sentence occurrences even though, on nonindexical contextualism, the content they express is only true relative to some contexts and false relative to others.13 If Kaplan’s rule were merely an arbitrary stipulation, defining a concept rarely if ever employed by natural language users, then it would provide no ability to explain linguistic intuitions, and would thus be an idle component of a semantic theory.

To recap, the nonindexical contextualist cannot use Kaplan’s rule for sentence occurrences, because truth at a circumstance does not entail truth when the circumstance includes an epistemic standard. If Kaplan’s rule could be used, nonindexical contextualism would inherit many of the objectionable consequences of indexical contextualism. Without Kaplan’s rule, however, nonindexical contextualism provides no account of when knowledge claims are true or false. It is not possible for the truth-value of a sentence to vary with the context of use when it expresses the same proposition and is evaluated at the same world.


See Bar-Hillel (1954, p. 359), Gale (1967, p. 151), Montague (1968, p. 103), Lewis (1970, p. 184ff), Stalnaker (1976, p. 229), Kaplan (1977, pp. 490, 506), Lyons (1977, pp. 106, 637, 646), Levinson (1994, 2003), Leezenberg (1994, p. 1648), Perry (1997, p. 586), Forbes (2003, p. 87), Corazza (2003, p. 593ff), Cappelen and Lepore (2005, p. 89) Gonzáles-Romero (2005, p. 261). Related definitions characterize indexicals as expressions whose reference varies with the token, occurrence, or utterance; see Burks (1949, pp. 685–7), Reichenbach (1947, p. 284), Gale (1964, p. 227), Searle (1983, p. 221), Perry (2003, p. 378); contrast Kaplan (1977, p. 522).


MacFarlane cites Kompa (2002) and Ludlow (2005, p. 27) as possible antecedents.


This formulation is adapted from MacFarlane (2009, p. 236). For it to be plausible, we also need to assume that S’s epistemic position with respect to p meets e(C) only if S believes p, and is not in a Gettier situation.


MacFarlane credits Hawthorne (2004) and Cappelen and Lepore (2005) for this argument.


MacFarlane (2009, p. §2) uses “temporalism” as a motivating example. Inferring from that example, MacFarlane might take temporalism to maintain that ‘knows’ expresses the same two-place relation in every context, the relation S stands in to p iff S knows p. Given MacFarlane’s truth-conditions, and the independence of t as well as e of S and p, there is no such two-place relation.


‘S knows’ can be used elliptically. If Sam says, “Does Alice know her husband is having an affair?” Barry can reply “She knows.” When so used, ‘S knows’ expresses different propositions on different occasions.


MacFarlane could fill this out by using the theory of thoughts and propositions I developed in Davis (2003).


As MacFarlane (2009, p. 240) observes in connection with a slightly different example (one of DeRose’s “abominable conjunctions”), nothing in contextualism prevents the epistemic standard from shifting from one conjunct to another since they occur in different contexts (as demonstrative pronouns show). So it allows an explicit contradiction like ‘Alice knows her car is in the driveway but does not know it is’ to come out true. MacFarlane claims that nonindexical contextualism explains why it always seems contradictory: “the context of use determines a single epistemic standard that is relevant to the evaluation of both occurrences of ‘know.’” This is an independent postulate, however, not something entailed by nonindexical contextualism. The indexical contextualist could postulate the same thing.


As is customary, we are using ‘utter’ in a generic sense that includes writing and signing. If it is restricted to spoken utterance, then to utter an expression is to make it occur by speaking.


Things are less obvious if a sentence occurs more than once in a context, which is why I changed MacFarlane’s ‘an’ to ‘the.’


See Cappelen and Lepore (2005, p. 105ff), MacFarlane (2005, p. 202ff), Hawthorne (2004, pp. 107, fn. 125), Davis (2005, p. 39, 2007, pp. 400–01), Bach (2005, p. 60), Stanley (2005, pp. 54ff, 115, 119ff), Fantl and McGrath (2009, p. 180ff). Contrast DeRose (1992, pp. 925–6, 2009, pp. 207–12), Cohen (2005, pp. 205–206), Blome-Tillmann (2008, pp. 34, 37).


MacFarlane thought that Kompa 2002 lapsed into inconsistency when she claimed that her theory had the “unpleasant” consequence of counting Ascriber C says something true in uttering ‘A knows that p’ but A doesn’t know that p, his (30), as true in certain cases. According to MacFarlane (2009, p. 248), ‘if “says something true” means “expresses a proposition that is true,” which I think is the most natural reading, then nonindexical contextualism does not predict that (30) is true.’ But ‘says something true’ also means “speaks truly,” which on nonindexical contextualism means uttering a sentence that is true in the context of utterance, and requires only that the proposition expressed be true relative to that context.


I critique MacFarlane’s relativism in “Knowledge Claims and the Context of Assessment,” which was presented at the “Contexts, Perspectives, and Relative Truth” conference held at the University of Bonn, June 9–11, 2011.



I thank members of the audience at the Kirchberg Symposium, especially John Greco and Stewart Cohen, for comments that helped improve my argument significantly.

Copyright information

© Springer Science+Business Media B.V. 2011