Saturday, July 25, 2009

DeRose on 'knowledge' norm of assertion

In his 2002 paper "Assertion, Knowledge, and Context," Keith DeRose gave an argument for contextualism about 'knows' that took basically this form: knowledge is the norm of assertion; assertability varies according to context; therefore, knowledge varies according to context.

This was a pretty confused argument -- though of course this is much clearer in retrospect, with the advantage of years of engagement with SSI. The problem is that contextualism is a thesis about the word 'knows', not about knowledge, while 'knowledge is the norm of assertion' seems like it must be a thesis about knowledge, not about English. In fact, something like a knowledge norm for assertion, combined with the observation that what you're allowed to assert depends on your situation, provides a pretty good argument for SSI; I take it to be exactly parallel to the main argument for SSI that Stanley and Fantl and McGrath give.

In chapter 3 of his new book The Case for Contextualism, DeRose essentially reproduces the content of that 2002 paper, but he does add about two new pages of material designed to correct this aspect of the original. Now, in contrast to earlier, he recognizes the need to clarify the statement of the knowledge norm of assertion, if it is to be understood in contextualist terms. He gives us:
The Relativized Knowledge Account of Assertion (KAA-R): A speaker, S, is well-enough positioned with respect to p to be able to properly assert that p if and only if S knows that p according to the standards for knowledge that are in place as S makes her assertion. (99)

Just to flag a broader source of uneasiness than is my focus in this post, I'm not really clear just what 'standards' are in this context, and I'm surprised by the invocation of 'knows that p according to such-and-such standards', which I guess is meant to be invariantist. But I'm taking the right-hand side of KAA-R to be equivalent to: 'S knows p' expresses a truth relative to the context in which S is participating. The central feature of KAA-R is that it's the subject's context that is relevant for whether the KAA-R condition is met; we're not being contextualists about 'well-enough positioned' or anything like that. DeRose contrasts KAA-R with KAA-R2:
KAA-R2: A subject, S1, is well-enough positioned with respect to p to make true the claim of a speaker, S2, that S1 is ' warranted in asserting that p' if and only if S1 knows that p according to the standards for knowledge that are in place in S2's context as S2 makes her claim. (99)

KAA-R2 is presented rather confusingly, I think. As far as I can tell, it's equivalent to this much more easily-digestable thesis: In all contexts, "S is well-enough positioned with respect to p to assert that p iff S knows that p" is true. But anyway, that's a side note. DeRose doesn't say why he prefers KAA-R over rivals like KAA-R2 (unless he does so later than the point I've read up to), but only mentions KAA-R2 to illustrate that KAA-R is his preferred gloss on the knowledge norm of assertion.

I have some concerns about KAA-R as a statement of the knowledge norm for a contextualist. Here are two; one I'm more confident about, and one I'm more tentative about.

(1) Given contextualism about 'knows', KAA-R is consistent with the truth of many sentences of the form "S doesn't know that p but it's entirely appropriate for her to assert that p." KAA-R predicts the truth of such sentences any time the attributor is in a skeptical context and S is in a nonskeptical one. But if you accept the knowledge norm of assertion, you should not think that sentences like that are ever true. Therefore, KAA-R does not capture the knowledge norm of assertion. A similar problem befalls sentences like "S knows that p, but she is not in a strong enough epistemic position to assert that p." KAA-R2 enjoys a clear advantage here.

(2) KAA-R cites the "standards for knowledge that are in place as S makes her assertion". I worry whether there will be a unique such set of standards -- it seems like there might not be, especially since S's assertion is just the assertion that p; S isn't attributing or denying knowledge. If part of what sets the relevant standards is a kind of scoreboard/accommodating process, then the act of attributing or denying knowledge plays a key role in fixing the relevant standards. Maybe DeRose just thinks this can never happen; I guess we need to know more about what it is in virtue of which the operative standards in a given context are what they are. KAA-R is in trouble, I think, if there are possible situations where one could say "S knows p" or one could say "S doesn't know p", and either would express a truth, because the context would shift to accommodate it. I'm inclined to think there are  cases like this. If that's right, then there might not be any operative 'standards for knowledge' in S's context.

5 comments:

  1. Hi Jonathan,

    I don't think it (i.e. DeRose's 2002) was a confused argument. It's not logically valid. But it is suggestive. And in fairness to DeRose, he has said that he had long assumed SSI was a non-starter because of its problem handling third-person attributions. (See "The Problem with Subject-Sensitive Invariantism," PPR, March 2004, esp. p. 348.) And if SSI is false, then it does at least become difficult to see how one might resist contextualism while hanging on to KAA.

    Ironically, the third-person attributions DeRose has in mind pertain to someone in a skeptical context talking about someone in a non-skeptical context. That's just the sort of case you go on to say causes trouble for KAA-R.

    How's about this?

    KAA-R3: An utterance of 'S may assert Q' expresses a truth just in case an utterance of 'S knows that Q' expresses a truth.

    With that in hand, the context-sensitivity of 'S may assert Q' gives you context-sensitivity for 'S knows that Q'.

    ReplyDelete
  2. John, I'm confused by your comment.

    First point of confusion: is your KAA-R3 meant to be different from DeRose's KAA-R2 or of my attempted gloss of it? I can think of two ways to interpret your KAA-R3; one makes it very unappealing, given contextualism, and the other, I think, makes it the same as KAA-R3. The reading that makes it the same as the previous version, I think, is the one where we're restricting it to cases where the two utterances are made in the same context. The reading that makes it very unappealing does not include that implicit restriction; it just says that some utterance of the former is true just in case some utterance of the latter is. I take it this view is very unattractive. (It also doesn't license the inference from the context-sensitivity of the former to that of the latter.)

    Second point of confusion: DeRose doesn't ever say that 'S may assert Q' is context-sensitive; nor do any of his examples point in that direction. And given his endorsement of KAA-R over KAA-R2, it looks a lot like he's assuming language like that to be context-invariant.

    I'm also confused about the way we're talking about SSI, but I think I'll save that confusion for a full post at a later date.

    ReplyDelete
  3. Jonathan,

    I don't understand KAA-R2, so I can't say whether my proposal is equivalent to it. My proposal is very similar to your gloss on KAA-R2 (which, I take it, was more than just a "side-note" to be set aside?)

    I offered KAA-R3 as a way around the third-person attributions involving an attributor/attributee context split (one in a HIGH context, the other in a LOW). I believe this is the sense in which you find it unnatractive (given contextualism). Why?

    "(It also doesn’t license the inference from the context-sensitivity of the former to that of the latter.)" I'm unable to see why it would fail. Suppose that, in all contexts, 'S is tall enough to play quarterback' is true just in case 'S is tall enough to play point guard' is true. And suppose that the truth of 'S is tall enough to play PG' varies with context. It follows that the truth of 'S is tall enough to play QB' varies with context. Any reason why it would work differently for 'S may assert P' and 'S knows P'?

    You're right, DeRose never says that 'S may assert Q' is context sensitive. Did my comment suggest that he did?

    ReplyDelete
  4. John, you ask:

    You’re right, DeRose never says that ‘S may assert Q’ is context sensitive. Did my comment suggest that he did?

    Well, I guess I'm not totally sure what you were doing with your comment. I thought you were making the 'How about this?' suggestion as 'How about this as an interpretation of what DeRose was up to?', thus supporting your earlier claim that the original argument was not confused. Since DeRose doesn't seem to think that 'may assert' is context-sensitive, I take it to be pretty clear that, as a matter of DeRose exegesis, that's wrong. But maybe you were making an independent suggestion on your own behalf; if that's the case, then I think you're probably saying basically the same thing I'm saying in the post.

    I don't think, contrary to your last comment, that you mean to be embracing the terrible disambiguation of KAA-R3 I mentioned in my last comment. I think I wasn't clear about it. Here's what I meant. You said:

    KAA-R3: An utterance of ‘S may assert Q’ expresses a truth just in case an utterance of ‘S knows that Q’ expresses a truth.

    This can be disambiguated in two ways; the question is whether we're meaning to hold the context fixed. If you hold it fixed, you get something equivalent to KAA-R2:

    KAA-R3-fixed: In any given context C, an utterance of ‘S may assert Q’ expresses a truth in C just in case an utterance of ‘S knows that Q’ expresses a truth in C.

    If you leave contexts unconstrained, you get this very weird principle:

    KAA-R3-loose: An utterance of ‘S may assert Q’ expresses a truth in some context just in case an utterance of ‘S knows that Q’ expresses a truth in some context.

    KAA-R3-loose doesn't say anything about context-sensitivity, even conditional on the context-sensitivity of 'may assert'. All it says that if 'may assert' is true somewhere (whether or not it's context-sensitive), then 'knows' is true somewhere (whether or not it's context-sensitive). KAA-R3-fixed, by contrast, does ensure that if one is context-sensitive, so is the other.

    The main point of my post was that DeRose's embracing of KAA-R over KAA-R2 isn't particularly motivated, and that it raises some problems that KAA-R2 handles better. If, as I think, you intend your KAA-R3 to be KAA-R3-fixed, it sounds like you're saying the same thing I am with respect to how to capture the knowledge norm.

    But there's only an argument for context-sensitivity of 'knows' here if we have good reason to think that 'S may assert Q' is context-sensitive. Indeed, it's a bit disingenuous to talk as if the issue is whether we have a good argument for contextualism; if we have good reason to think that 'S may assert Q' is not context-sensitive, then we have a powerful argument against contextualism -- it's one that both Williamson and Hawthorne have made. So the contextualist had better (a) reject the knowledge norm, (b) reject KAA-R2 as the proper formulation of the knowledge norm, or (c) make plausible the context-sensitivity of 'S may assert Q'. DeRose has tried to choose option (b); I've argued that this is not a good choice. It sounds now like you are agreeing with me.

    ReplyDelete
  5. Yes, I do believe we mostly agree, Jonathan.

    Mostly as an afterthought, I think that if warranted assertability is factive, it makes option (c) look more attractive.

    ReplyDelete