Keywords

Epistemic contextualism concerning knowledge says that the truth conditions of knowledge attributions (including denials of knowledge) vary with the context of the attributor (cf. Cohen 1987; Lewis 1996; DeRose 1999). There have been recently quite a number of objections to contextualism (cf., e.g., Rysiew 2009). One objection, however, has not been discussed much at all even though it might be the most serious one so far: the so-called “Factivity Objection” according to which contextualism is inconsistent at its core. This objection has been developed mainly by Brendel (2003, 2005) and Wright (2005); see also, from a different perspective, Lihoreau and Rebuschi (2009) (cf. also short passages in Luper (2003, pp. 196–7), Veber (2004, pp. 268–269), Brueckner (2004), Engel (2005, pp. 58, 63), Kompa (2005, pp. 18–19, 25–26), Kallestrup (2005), Steup (2005, Sect. 1–2, 6), and Montminy (2008)). In my Baumann (2008) I defended the idea that there is a problem (cf. Brueckner and Buford 2009; Baumann 2010; Brueckner and Buford 2010 for an exchange on this) but also proposed a solution, namely a relationalist version of contextualism. In this paper I will first present the problem and then discuss some proposed solutions (some of them denying that there is a problem in the first place) before I move on to my own proposal of a solution.

4.1 The Problem

Consider two knowledge attributors, Ordi and Spec. Ordi finds herself in an ordinary and not so demanding context while Spec finds herself in a much more demanding (but not skeptical) context S. According to contextualists, the epistemic standards for the correct attribution of knowledge vary with the context of the attributor. A given knowledge sentence can be true as uttered in one context but false as uttered in another context; it can be true (false) in one context while its negation can also be true (false) in another context. So, let us assume that the following is true:

(1):

Ordi’s utterance of “Spec knows that Spec has hands” in O is true,

and

(2):

Spec’s utterance of “Spec knows that Spec has hands” in S is not true.

Suppose now that Spec is a contextualist about knowledge. Since Spec does not find herself in a dramatic skeptical context, she can still correctly claim knowledge of some propositions. For instance, she can correctly claim to know (1). Why should this not be possible for Spec (see below)? Thus, we get

(3):

Spec’s utterance of “Spec knows that (1)” in S is true.

Furthermore, a plausible general disquotation principle tells us that

“p”is true → :

p

while the principle of factivity of knowledge tells us that

(S knows that p) → :

p.

Both combined and adjusted for contextualism (I won’t go into the details here) we get

(DF):

“S knows that p” (as uttered in some context) is true → p.

And (DF) applied to (1) gives us

(4):

Ordi’s utterance of “Spec knows that Spec has hands” in O is true → Spec has hands.

There is certainly no problem for Spec to correctly claim knowledge of (4). Hence, we get

(5):

Spec’s utterance of “Spec knows that (4)” in S is true.

Finally, we should assume a closure principle which is adapted to the needs of the contextualist:

(Closure):

For all contexts C, speakers S and propositions p, q: [“S knows that p” (as uttered in C) is true and “S knows that (p → q)” (as uttered in C) is true] → “S knows that q” (as uttered in C) is true.

A more adequate closure principle would look a bit more complicated (see Baumann 2011) but for our purposes here we can stick with this simple version.

The important point here is that from (3) and (5) plus (Closure) it follows that

(6):

Spec’s utterance of “Spec knows that Spec has hands” in S is true.

And (6) contradicts (2). This is the threat of inconsistency contextualism faces.

What can the contextualist do about this? Something must be given up. (1)–(3) express what a contextualist seems committed to in our example. Principles of disquotation, factivity and closure should not be given up (certainly not just in order to save some epistemological theory). (5) is unproblematic. But (3), (5) and (Closure) entail (6) which contradicts (2). Should we therefore give up the weakest link, namely contextualism?

4.2 Ways Out?

Interestingly, many people deny that there is a problem for contextualism in the first place and argue that (3) is false. One way to do this is to point out that if Spec finds herself in a skeptical context, then she could not correctly claim any knowledge in that context. However, a demanding context like S need not be a skeptical one (see above). It also won’t help much to argue that (3) must be false because it leads, 3 together with the unproblematic (5) and (Closure) to a contradiction for contextualism. This would be a case of begging the question against the view that there is a contradiction here. Similarly implausible is the idea that since (2) is true, (6) must be false; since (6) follows from (3), (5) and (Closure) and since (5) and Closure are non-negotiable, (3) must be false. Again, this kind of move begs the question against the inconsistency objection.

More interesting is another way to argue against (3) (for all this see also Brueckner and Buford 2009, 2010). Roughly, one could say that in order for A to know that B knows that p, A needs to know that p herself:

  • (A knows that B knows that p) → 

  • A knows that p.

Applied to our case concerning contextualism we get:

  1. (3) →

     (6).

And since (6) is false, according to contextualism, (3) must be false, too, so the argument goes.

If this is not just the question begging point above again – (6) is false because (2) is true –, then there must be something else behind this argument against (3). The most promising move I can think of is based on a principle of epistemic priority according to which (in its more straightforward non-contextualist version)

(EP):

(A knows that B knows that p) → A knows (independently from and prior to the knowledge that B knows that p) that p.

A contextualist version of such a principle of epistemic priority would look a bit more complicated; I won’t go into this here because the basic point does not depend on such variations.

The problem is simply that (EP) is not true. Consider this example. I know, four having read a reliable newspaper, that Andrew Wiles found out, proved and came to know that Fermat’s conjecture is true. I myself thus came to know that Fermat’s conjecture is true. However, I was able to come to know that Wiles knows that Fermat was right without independently – and prior to reading the papers – knowing that Fermat was right (for the contextualist, all this will be a bit more complicated to express; see below for some aspects). (EP) is false and the argument against (3) relying on it does not go through.

But can it really be correct to say in a demanding context that “S knows that B knows that p” – when it is not correct to say in that demanding context that “S knows that p”? Yes, that S meets the demanding standards for “knowledge” of B’s epistemic situation concerning p does not entail that S also meets the demanding standards for “knowledge” concerning p. Even if (i) B knows that p entails (ii) p, it does not follow that meeting the epistemic standards relevant in a certain context for (i) entails meeting the epistemic standards relevant in that context for (ii) (for more on this, cf. Baumann 2008).

But doesn’t all this neglect an important temporal aspect (cf. Brueckner and Buford 2010)? Sure, at an earlier time t1 (2) might be true. But then, at t2, Spec learns that (1) is the case and engages in some reasoning, arriving at a later time t3 at a relevant conclusion concerning herself having hands such that (6) is true. At t3 (but not at t1) (6) will be true and (2) will be false – while at t1 (2) is true and (6) is false. Hence, the contradiction evaporates. However, our problem does not depend on such an equivocation. (2) as well as (3) can easily be true at the same time t1: (2) is true because Spec does not meet the relevant standards while (3) could be true on the basis of testimony (see above). (5) can also easily be true at the same time. Given (Closure) which is also true (not just at t1), (6) just follows and must thus be true at t1, too. Nothing changes if one replaces the non-dynamic (Closure) by some dynamic principle of closure with some temporal dimension (some transmission principle, for instance cf. Baumann 2011).

However, there is a kernel of truth in this kind of objection which will be brought out in the solution proposed below. Before I go into that, I should stress that denying (3) comes with serious costs. Under certain quite common conditions one subject cannot be correctly said to “know” another subject’s epistemic situation:

(Restrict):

For all subjects A and B, for all propositions p, and for all pairs of contexts O and S such that “A knows that p” is true in O but not true (false) in S: “S knows that ‘A knows that p’ is true in O” is not true in S.

This restriction (cf. Baumann 2008, p. 583) is very severe. Not only does it limit the statability of contextualism in non-trivial ways. It also seems like a truism that subjects can in principle and under non-extreme circumstances come to know about other subjects’ epistemic situation without there being farreaching and systematic restrictions to this kind of knowledge. Giving up such a principle requires very good independent reasons. It is not enough to want to save some pet epistemic theory.

Adhering to something like (Restrict) or denying (3) also seems to commit the epistemologist to accepting utterances of abominable conjunctions (cf. DeRose 1995, pp. 27–29) by an epistemic subject in a demanding context, like Spec:

I don’t know whether Ordi knows I have hands but if my standards were less strict I might well know that!

4.3 A Relationist Solution

Here is a different thought which helps us dissolve the above paradox of contextualism. For independent reasons it is plausible to analyze the knowledge-relation not as a binary relation between a subject and a proposition (“S knows that p”) but rather as a ternary relation between a subject, a proposition and an epistemic standard (“S knows – with respect to standard S – that p”; cf. Schaffer 2005 or Steup 2005, Sects. 2, 6). Let us, for the sake of simplicity, use the terms “knowledge-relative-to-low-standards” or “knowledge-low” on the one hand and “knowledge-relative-to-high-standards” or “knowledge-high” on the other hand (cf. Sosa 2004, pp. 43–44; Bach 2005, pp. 58–59; Cohen 2005, pp. 201–204). We can thus reformulate (1) and (2) above in the following way:

(1*):

Spec knows-low that Spec has hands

and

(2*):

Spec does not know-high that Spec has hands.

(1*) and (2*) do not themselves have context-sensitive truth conditions but that’s fine and compatible with contextualism taken as a view about the truth conditions of our ordinary sentences and utterances.

(3) can be replaced by

(3*):

Spec knows-high that (1*).

Or, spelled out in more detail: Spec knows-high that Spec knows-low that Spec has hands. This makes more specific sense of the more general remarks above that there can be different standards for knowing that “S knows that p” is true and for knowing that “p” is true.

(DF) as well as (4) and (5) also need slight modifications:

(DF*):

(S knows-relative-to-some-standard that p) → p,

(4*):

(Spec knows-low that Spec has hands) → Spec has hands,

(5*):

Spec knows-high that (4*).

Now, the crucial question here concerns the adequate closure principle. Given that a subject might meet high epistemic standards concerning “p” as well as concerning “p → q” but only low epistemic standards concerning “q”, we have to modify (Closure) along the following lines (I won’t give the non-relativized version here and will restrict myself to the case of two kinds of knowledge only):

(Closure*):

For all subjects S, knowledge relations knows-low and know-high, and for all propositions p and q: If (i) S knows-high that p and if (ii) S knows-high that (p → q), then (iii) S knows-low that q.

Here is the application to our case of knowledge that someone knows:

(Closure**):

For all subjects O and S, knowledge relations knows-low and know-high, and for all propositions p: If (i) S knows-high that O knows-low that p and if (ii) S knows-high that [(O knows-low that p) → p], then (iii) S knows-low that p.

(Closure**) seems plausible as soon as one accepts contextualism; hence, the contextualist should be allowed to help himself to such a modification of (Closure).

(Closure**), (3*) and (5*) entail that

(6**):

Spec knows-low that Spec has hands.

However, (Closure**), (3*) and (5*) do not entail

(6):

Spec’s utterance of “Spec knows that Spec has hands” in S is true

or, in other words,

(6*):

Spec knows-high that Spec has hands.

So, since we can only infer (6**) but not (6*) we do not get a contradition. (6**) is, in contrast to (6*), perfectly compatible with

(2*):

Spec does not know-high that Spec has hands.

A relational version of contextualism thus avoids our contradiction; it shows how the argument for the contradiction between (6) and (2) equivocates on subtle context differences. This is a major advantage of relationalist contextualism over other versions of contextualism (e.g., purely indexical versions according to which “know” functions more or less like an essentially indexical expression). Relationalism can thus account for cross-context attributions – something that poses a serious difficulty for other versions of contextualism. It also explains in what sense subjects “lose” their knowledge when the attributor moves into a more demanding context and in what sense they “keep” it: Even if the subject may not know-high that p they might still know-low that p. Focusing exclusively on knowledge-high, the attributor might forget that there is still knowledge-low left. But contexts are not completely closed: Even if the context determines high standards one can still acknowledge, within that context, that the subject’s epistemic position still satisfies less demanding standards.

Someone might doubt the generality of our solution and argue that we’re still facing a problem. Assume again that Spec, the contextualist, finds herself in a demanding context. She accepts (see above)

(1):

Ordi’s utterance of “Spec knows that Spec has hands” in O is true

from which she can infer, using

(DF):

“S knows that p” (as uttered in some context) is true → p

that (given her knowledge that she is Spec)

(7):

I have hands.

At the same time, given (2) Spec also accepts that (in her words)

(8):

I don’t know that I have hands.

This also commits Spec, in so far as she is rational, to the conjunction of (7) and (8), namely

(9):

I have hands but I don’t know that.

This, however, is clearly Moore-paradoxical and not acceptable (though not inconsistent) (cf. for this kind of objection, Williamson (2001, pp. 26–27) and Kallestrup (2005, pp. 249–50), as well as Hans Kamp in conversation).

However, given what I said above, the solution of this problem is straightforward. Spec, the contextualist in a demanding context, is not committed to (8) but only to

(8*):

I don’t know-high that I have hands.

No utterance of

(10):

I have hands but I don’t know-high that

is Moore-paradoxical or problematic in any way.

Similar things can be said about the problems arising with the change of context over time, especially when the attributor moves from a less demanding to a more demanding context. Doesn’t it sound incoherent to say something of the form “Yesterday, in the pub, I knew that I have hands but now, in conversation with a skeptic, I don’t know it any more!” (cf. DeRose 2009, Chap. 6)? No, and if one were more precise and less misleading one would rather say something along the lines of “Yesterday, in the pub, I knew by low standards that I have hands but now, talking to the skeptic, I realize that I don’t know-high that”.

4.4 Conclusion

In its most common form contextualism faces a serious objection: that it is simply inconsistent. I have tried to show that this objection needs to be taken seriously and that there is a way out for the contextualist if he opts for a particular version of contextualism: relationalist contextualism. This view can account for cross-context attributions of knowledge – something which creates problems for other versions of contextualism.