Remarks on Mylan Engel’s Epistemic Contextualism and the Problem of Knowing What One Says

Read as comments in reply to Mylan Engel’s presentation at the annual Midsouth Philosophy Conference (24 February 2006).

Mylan Engel’s engaging essay poses a serious challenge for contextualism in epistemology, in the form of a problem: the problem of knowing what one says (before she says it). Now, I’m not a contextualist or even an epistemologist by trade, so it’s reasonably open to question whether I’m in much of a position to answer the challenge. What I’d like to do here is offer a few remarks and ask a few questions, from a few different angles, to get a clearer view of what the precise contours of the challenge are. I suspect that contextualists might have at least some chance of answering it; but more importantly, whether they can or not, I’d be interested to hear some more on the specific standards that contextualism is being measured against in order for it to be found wanting. (¶ 1)

Contextualists like to motivate their view by using it to handle our conflicting intuitions about skeptical hypotheses. They consider it a particular virtue of their view that it handles not only global skepticism based on obscure, Satanic error-possibilities, but also local and mundane cases, like the conflicting intuitions that we have between the view that Poor Phil knows he’ll never be a multimillionaire (I mean, come on) and the view that he doesn’t know that he won’t win the lottery (if he knows he’ll lose, why buy the ticket?). Mylan criticizes two prominent versions of this solution — Stewart Cohen’s and Keith DeRose’s — by posing a skeptical challenge of his own: if we view knowledge-claims as context-sensitive, and accept Cohen’s or DeRose’s characterization of what the context in question amounts to, then the content of our own knowledge-ascriptions partly depends on facts we won’t know until after we make the ascription. Our ignorance of salient aspects of the context means that we must fail to know what we are asserting, or even whether we are asserting anything at all, until after we’ve already asserted it. Thus it remains an utter mystery why the speakers think their utterances are true, when they don’t even know what the truth conditions (if any) of those utterances are. (¶ 2)

There are four main things that I’d like to know about his line of reasoning. (¶ 3)

  1. First, Mylan’s challenge to these two versions of contextualism does nearly all of its work with one particular feature of them: the after-the-fact semantic veto power that each version grants to listeners over speakers’ attempts to shift epistemic standards. If listeners have the choice to either play along with the shift or refuse to, then they have the power to make epistemic claims false (on Cohen’s version), or neither true nor false (on DeRose’s version) by refusing to accept the new truth-conditions for knowledge-ascription that the context-shifting speaker proposes, thus forcing the speaker, despite her best efforts, to say something under the old standards that she’d intended to say under the new standards, or else to say something in the gap, which fails to state anything at all. (¶ 4)

    What I’d be interested to know, though, is whether semantic veto power is essential to (ascriber) contextualism, or a peculiar feature of Cohen’s and DeRose’s versions of it. Mylan offers four defining commitments for contextualism: (1) no context-independent standard of knowledge, (2) the abandonment of any non-trivial general theory of knowledge in favor of metalinguistic observations on truth-conditions in specific conversational contexts, (3) attention to the ascriber’s context, not the would-be knower’s; and (4) the possibility for seemingly contradictory knowledge claims to both be uttered truly at the same time (as long as the speakers utter them in contexts with different standards in play). But, as far as I can tell, after-the-fact semantic veto power isn’t necessarily entailed by any of these claims, either singly or jointly. It seems to be a fifth independent and separable feature of Cohen’s and DeRose’s view: (5) that conversational context at a given time is partly determined by future contingent features of the conversation relative to that time. (¶ 5)

    If that’s so, and there isn’t any deeper connection that I’m missing, isn’t it open for contextualists to hold on to their position by keeping (1)-(4) but dropping semantic veto power? That would mean giving up on Cohen’s and DeRose’s accounts (such as they are) of conversational context, and in order to avoid the criticism mounted here, it would mean that the conversational context of a knowledge-ascription has to be something determined entirely by past and present features of the conversation, without depending on future contingents about how the listener responds. But is that cost unacceptable for contextualists? Should it be? (¶ 6)

  2. Secondly, suppose they aren’t willing to give up semantic veto power, for whatever reason. One of Cohen’s frequent replies to critics (such as Hawthorne and Jason Stanley) has been to point out how particular objections to epistemic contextualism don’t actually make any use of the fact that the view is an epistemological one; that is, they would apply just as well (or just as poorly) against any claim that the conditions for ascribing any term are context-sensitive. Cohen’s favorite examples here are tall and flat. Maybe he’s wrong to suggest that tall and flat are uncontroversially context-sensitive; but, wrong or right, it’s important to be clear on whether the debate is one within epistemology, or a broader debate about language that just happens to be using epistemological cases as illustrative examples. I have a similar question about Mylan’s rather different brief against Cohen and DeRose in this essay: given the central role that semantic veto power plays in Mylan’s worries, do his objections apply just as well to any theory that leaves the truth-conditions (if any) of any statement open to the risk of being affected by future contingents? (¶ 7)

  3. Part of the reason I ask is that is that if the worry here is general — and the case of knowledge-ascriptions just a particular application to an increasingly common view taken from the literature — then how does the worry apply to other cases in the literature where philosophers have argued that truth-conditions of other kinds of statements depend in part on future contingents? Kripke argued that any reasonable theory of truth-ascriptions will have to admit and deal with risky statements that might turn out to be gappy in spite of your best efforts — if I say, The first thing Mylan’s going to say in his reply to me is false, I’m liable to the risk that he might begin by saying What you said about my reply is true, making my statement part of a looped liar paradox and therefore neither true nor false, on pain of contradiction. That gives you, in effect, semantic veto power over whether my truth-ascribing predictions have any coherent truth conditions at all. Still, we do make risky predictions like this, all the time. Does a general worry about semantic veto power leave it an utter mystery, why the speakers think their utterances are true here? Or is there some way that we make do? If there is, couldn’t there be an equivalent way that we make do in the case of knowledge ascriptions? Or is there a specific problem with risky knowledge-ascriptions that keeps us from treating them in whatever way we treat risky truth-ascriptions?

    (¶ 8)
  4. Finally, it’s worth noting that the problem Mylan poses for epistemological contextualists to answer is framed as an epistemological problem, involving the difficulty of knowing what the truth-conditions (if any) for knowledge ascriptions are before you make them. For Mylan to make that claim, he has to help himself to the locally skeptical premise: (¶ 9)

    (SR) I don’t know ahead of time whether my listener will accept the epistemic standard I propose. (¶ 10)

    But that, of course, is a knowledge ascription like any other knowledge ascription. So if you do accept epistemic contextualism, then the most natural reply to the problem is to ask under what standards we fail to know how listeners will respond in a given conversational context. Of course listeners might refuse to accept mundane knowledge claims by invoking outlandish error possibilities about painted pumas or vatted brains, and somebody considering insurance plans might snort that she knows she won’t develop a debilitating illness in the next ten years, despite your best advice to shift her standards upwards for this decision. People might say all kinds of crazy things. These are genuine error-possibilities; but we’re not completely out at sea, are we? There are ways of forming fallible rational expectations about how your listeners will react — the sort of Verstehen that comes from personal acquaintence and sympathetic engagement, or if nothing else some kind of inductive generalization about how different people react in different circumstances. And if you’ve formed those expectations, then it seems like you have some evidence for or against the claim that your listener will accept the standard you try to fix. So the question is how much evidence you need to say that you know how they’ll react, and for a contextualist that’s a context-sensitive question. There are at least three broad categories of standards you could invoke: (¶ 11)

    (SRi) I don’t infallibly know ahead of time whether my listener will accept the epistemic standard I propose. (¶ 12)

    (SRs) I don’t meet strict standards for knowing ahead of time whether my listener will accept the epistemic standard I propose. (¶ 13)

    (SRm) I don’t meet *mundane standards* for knowing ahead of time whether my listener will accept the epistemic standard I propose. (¶ 14)

    So which standard is Mylan holding knowedge-ascribers to? There’s a lot of room for differences of degree and kind within SRs and SRm, but it seems like in most conversational contexts we can, at least, live up to some acceptable but low standard, even though we can’t usually live up to strict standards and won’t ever live up to infallibilistic standards. I mean come on, Mylan, we know that most dockworkers aren’t going to take Hume seriously most of the time, and that people who are already looking for insurance are going to want to take unlikely catastrophes into account. (¶ 15)

    But if this move is available to Cohen and DeRose, then the problem of knowing what one says vanishes unless and until we conversationally elevate the standards to SRs or SRi. But to get by with knowledge ascriptions, do we need infallible or strict-standards knowledge of which, among a limited set of closely related truth-conditions, our interlocutor is likely to play along with? We do have some evidence and the penalty for getting it wrong seems pretty low. Can’t that establish a conversational context in which rough-and-ready evidence about the epistemic standards likely to be in play, would be good enough to justify saying that we know what our knowledge-ascription means? (¶ 16)

    One possible counterpoint would be a suggestion that something special about the truth-conditions of our statements might raise the bar for knowledge, just in virtue of the subject-matter. Maybe some subject-matters shift epistemic contexts all on their own. You probably need scientific standards of evidence to make even minimally responsible knowledge-claims about quarks or gluons, that you wouldn’t need for making knowledge-claims about books or rocks in mundane contexts; reason being that there aren’t mundane contexts where you talk about quarks or gluons, and just mentioning them shifts the conversational context. Does something special about the truth-conditions for your own statements demand strict or even infallibilistic standards for knowing them, just in virtue of talking about them? There are some views out there that might suggest this — for example, internalism about meaning and transparency of the mind, which together entail that you need to have infallible knowledge of the content of your own assertions insofar as you can count as having that content at all. (I’m not implying that this is Mylan’s view; take this only as an illustration of one possible view that would suggest this line of reasoning.) But any move along the lines of the requirements of the subject-matter would seem to make the debate here more about the nature of propositional attitudes and our relation to them than it is about epistemology. (¶ 17)

    I mention this because Mylan frames the issue primarily as an epistemological problem — the problem of knowing what one says. But the counterintuitive results he points out for Cohen and DeRose each have two components — the epistemological complaint that we often don’t know what (if anything) we assert until after we assert it, but also a complaint more about the location of the truth-makers, that it is ultimately the listeners, not the speaker, who determine whether the speaker actually says anything [or what the speaker says]. Is the objection to epistemic contextualism ultimately epistemological or metaphysical? If we’re coming to this with a certain set of metaphysical views about where we should find the truth-makers of claims about the content of our own statements, or with a set of views about what that location means for the sort of standards to which we need to hold knowledge-claims about the content of our own statements, then that would help explain the standards being invoked against Cohen and DeRose. But just as far as it did explain them, it would also bring us back to the second and third questions — whether the debate here is ultimately about some more general set of questions about risky sentences and the requirements on determining content in general, and whether folks already committed to an external semantic veto power might be willing to reject the commitments that might be used against them here. (¶ 18)

    If the objection turns out to be primarily epistemological rather than metaphysical, then I wonder why the contextualist can’t simply make the contextualist answer that we know what we say according to some standards and not according to others, and that this gives us something good enough to get by on, while also explaining away the pull of Mylan’s skeptical intuitions. If the objection helps itself to some further set of metaphysical and epistemological commitments outside of the Five Pillars of Contextualism that we’ve already mentioned, then I wonder how far the utter mystery is internal to the contextualist view, and how far it involves a set of further commitments that they may already be inclined to reject. (¶ 19)