As briefly as I can muster here. I take the evident facts to be these. First, sometimes we as speakers apply to far-away subjects who are no party to our conversation (or to our past selves, who were in very different contexts than we are in as we speak) standards that are appropriate to the subjects’ conversational contexts. This often happens, for instance, when we are discussing practical decisions those subjects face(d): “She should do that only if she knows…” Here, if the subject is/was in a high-stakes context, we’ll tend to apply high standards to her.

But, second, and equally evident, in some contexts we’ll apply standards to far-away subjects (or our own past selves) standards that are appropriate to our own present context. We’ll do this, for instance, in cases where we’re discussing absent subjects as potential sources of information to us. Thus, suppose we’re in a high-stakes bank context, wondering whether to rush over to the bank on Friday aft., having admitted that we “don’t know” it’s open on Saturday. Suppose that in this high-stakes context we won’t count ourselves as knowing that on the basis of having been at the bank two weeks ago on Saturday and having found it open with posted hours that include Sat. morning hours, but only if we’ve made some special check, like calling a bank employee who’s told us that the hours haven’t changed, that he himself is scheduled to work Sat. morning, etc. Now, if we’re asked whether someone not present at the conversation might “know” whether it’s open on Sat., where the thought here is that if she does “know” we can call her up, find out, & then “know” ourselves, we’ll apply high standards to her: If one of us knows that this absent subject doesn’t meet the extra-high standards, he’ll say that this absent subject, like us, “doesn’t know.” [Here, for reasons I won’t take the time to explain, it’s important that we won’t just refrain from saying she “knows”, but will go so far as to say she “doesn’t know”.] And we’ll do this even if we happen to know or suspect that this absent subject is herself now in a low-standards context.

Non-contextualist RA, and similar non-contextualist views, cannot handle the second evident fact above. On these views, the subject’s context sets the standards that must be met if anyone in any context is to truthfully call the subject a knower. So they predict we won’t apply our own high standards to the absent subject, but will always apply to the absent subject standards appropriate to her own context — and, I suppose, if we have no idea what kind of context she’s in, the prediction is that we’ll be agnostic as to whether she knows. (If you think that maybe our thinking of her, though it happens far away from her, may still be part of her “practical context,” and think this might get the non-contextualist off the hook, this escape quickly comes to grief over easily constuctible cases — like ones where the absent subject somehow knows that we’re talking about her in a high-stakes setting, but her own context calls on her to apply low standards to herself — where speakers happily ignore what they know is going on in far away contexts and apply to themselves standards appropriate to their own local context.) But the evident fact here is that in such a case (where we’re thinking of the subject as potential informant), we will deny that she “knows” if she doesn’t meet the high standards of our own context.

Contextualism, of course, has no problem with the second fact above. Plus, IT HAS NO PROBLEM WITH THE FIRST FACT, either. There’s nothing in contextualism to prevent speakers’ contexts from selecting standards appropriate to the subject’s practical context. Sometimes we do so, sometimes (and about as often, so far as I can tell) we apply our own standards to them. That’s because sometimes our conversational purposes (and I’ve mentioned a purpose or two for each side above)call on us to apply the one type of standard; sometimes the other.

We really need a theory that can handle both of these evident facts.

The second kind of case is interesting. It really is important for the argument that knowledge claims about the person in the low-standards situation are denied rather than just not asserted, but maybe that’s what the data says. In any case, there really need to be lots and lots more cases, as few as possible involving present-tense unembedded first-person knowledge ascriptions (which IMHO really don’t discriminate between lots of rival theories) before it’s clear to me which side has the better of the story here. Ideally we want a simple theory which explains (a) the original bank cases, (b) the versions of them given by Keith here, © the Woody-Hannah-Suzanna conversation in the post, (d) the manager-marketing director conversation in the update and (e) to (z) more examples that people are yet to generate before we can confidently say which direction the data point in.

And I should add, I suspect everyone here is eventually going to have to say that the folk theory needs massaging at some point to be made plausible or even consistent. If the RA theorist has to say that people sometimes make negated knowledge claims when really they should be staying silent, that might not be too costly an error claim to make, because we know that in this area negation is not well behaved. (Compare the frequent use of “I don’t believe” to mean “I believe not”, considered a horrible gaffe by syntactic prescriptivists, and an interesting transformational fact by descriptivists.)

Contextualists, myself included, certainly have stressed cases involving first-person cases far too much. That’s been a mistake. Part of the reason for it is that the live rival to contextualism has been what we might now call “Classical Invariantism,” which agrees with contextualism that it’s the speaker’s context that sets the standards for knowledge attributions (first-, second-, or third-person)(though it can, and should, also agree with contextualism that the speaker’s context will often select standards approriate to the subject’s context), but insists, against contextualism, that these varying standards govern only whether it’s appropriate (and not true) to attribute “knowledge” to the subject.

It’s in very recent times that a new form of invariantism — what might be called “Subject-sensitive invariantism,” though it often goes by the misleading name of “subject contextualism” (I think I’m among those who have used that misleading name) — has gotten a lot of attention. On such views, the subject’s context sets the standards that govern whether any speaker can truthfully attribute knowledge to that subject — thus, you can’t get the results contextualists insist on: that one speaker in one context can truthfully ascribe “knowledge” to a subject while another speaker in another context can truthfully deny “knowledge” to that same subject at the same time.

Now, I think it’s not just an historical accident that Classical Invariantism was seen as the real threat. I think it is the more threatening form of invariantism, and have long thought so on the basis of cases for reasons like those in my above comment. Subject sensitive invariantism seems a strange concoction to me in the way it goes so far to make first-person knowledge claims come out true, yet has to go in for a big-time error theory about third-person attributions. It seems to miss the fact that the very conversational factors that make us raise the standards on ourselves also often make us raise the standards on subjects not present at our conversation.

But SSI certainly is getting a lot of attention nowadays, and for that reason I suppose can’t anymore just be brushed aside.

Here are some reasons why non-contextualist versions of RA are getting attention now. First, there is a lot of work now on trying to figure out when an expression is context-sensitive. It seems that “know” fails virtually every test people have come up with. On the other hand, Keith and other contextualists have rightly brought our attention to certain cases (e.g. bank cases) that suggest that there can be true knowledge ascriptions that seemingly contradict one another. So this naturally leads people like myself and Hawthorne to explore views that can capture bank case intuitions semantically, as contextualists rightly urge, while not committing ourselves to the context-sensitivity of ‘know’. Furthermore, in developing a non-contextualist version of RA, I for one have realized that there are a lot of resources there that haven’t been properly explored.
My own interest in non-contextualist versions of RA also comes from a general interest in theories that are more restrictive but account for much of the same core data. The contextualist framework is much more flexible — it can explain cases that the non-contextualist theory can’t. But one might worry that it’s too permissive, and can get us ‘out’ of contradictions that are, in fact, contradictions.

I’d agree that SSI views are are well-motivated (*very* well-motivated) if I agreed with that bit about “knows” failing all these tests for context-sensitivity. This would of course have to be fought out on a test-by-test basis, but my own judgment has been that all the tests “knows” fails that I’ve seen (and I’ve seen a lot!) are also failed by other expressions that are obviously context-sensitive, and so are not good tests at all.

I’m not sure there will be one single test that manages to set all context-sensitive expressions on one side and all non-context-sensitive expressions on the other side (this is my worry with Cappelen and Lepore’s latest paper, where they seek to give just such a criterion). So I agree there is not going to be a single criterion test against the context-sensitivity of “know” (or rather, I’m skeptical that there will be). But if “know” is context-sensitive, then there should be some class of (relatively) uncontroversially context-sensitive expressions with which it patterns. It’s here that I have worries — I worry that there is no such class.
So, Keith, it looks like we’re agreed on the methodology — any consideration that militates against the context-sensitivity of “know” must also not militate against the context-sensitivity of (e.g.) “flat” or “tall” or “many”. But I think I do have arguments that if “know” is context-sensitive, then it is context-sensitive in a hitherto undiscovered way. Some of these arguments I gave in the paper in the contextualism conference at UMass; others will be in the papers I’ll put on my website when I get out of the Daintree rainforest in Australia (where I currently am) and back to Michigan later in the week.
I’ve been communicating with Stewart Cohen about some of these arguments — one thing he has been saying in these useful e-mails is that “know” is context-sensitive in the same sense as “hit” is context-sensitive. I think we’re at a stalemate — I don’t see any sense in which “hit” is context-sensitive, so I think I win, and he seems to think that “hit” is context-sensitive on the basis of considerations I don’t really understand, so he wins.
I think what Cohen is detecting with “hit” is that it is vague (either that or that “hit” could be used to mean something slightly different), not that it is context-sensitive. Since I tend to be an epistemicist about vagueness, I’m not sure this position helps evade skepticism.
But it looks like we’re all agreed on methodology. The question is then whether “know” is context-sensitive like (relatively) uncontroversial context-sensitive expressions. The second question is related. If “know” is context-sensitive rather than interest-relative (the word I use for non-contextualist RA, following Delia Graff’s related theory of vagueness), then its kinematics should be governed by salience. But I think salience isn’t a good model here. Keith, you’ve been saying that the contextualist doesn’t have to appeal to salience to explain how “know” shifts its content across contexts. But if salience governs all other context-sensitive expressions, this suggests again (if salience is the wrong model for “know”) that “know” isn’t context-sensitive (if you recall, this was the point Jeff King raised in discussion of your paper Exploding Scoreboard paper last October).
The appeal of the interest-relative account (Hawthorne’s SSI — though I think that’s an unhelpful name) is that it explains why salience is the wrong model. So, the worries with the context-sensitive model for “know” (namely, that it suggests an inappropriate model for how the content of “know” gets fixed relative to a context) again provide support for the interest-relative view.

OK, but it’s one thing to say that “knows” fails every test people have come up with, which will likely make folks think each of these tests is a good test for context-sensitivity, and damned if “knows” doesn’t fail each & every time we put it to the test!, and quite another to say, “We have all these different ways of classifying the how terms behave, and, though “knows” behaves like some (other) context-sensitive terms in some respects, and some others in other respects, we’re not finding any particular uncontroversially context-sensitive terms that match how “knows” behaves with respect to all of these these different aspects.”

My skepticism about salience explanations is quite general: Not just regarding accounts of “knows”, but throughout a lot of recent philosophy of language, I find myself rebelling against appeals to “salience.” Why is particular content for “tall” expressed on a particular occasion? I can understand someone appealing here to the fact that, given certain conversational purposes, it would be helpful or useful to use “tall” with that particular content on the occasion (though I will generally think that it’s various conversational maneuvers that make “tall” have the content it has, and the fact that that content would be useful just explains, on those occasions where speakers are wise enough to employ a very useful content, why they chose to execute maneuvers that put into play that content, rather than others). And I can understand — and will be more sympathetic to — accounts that appeal in various ways to the conversational moves the speakers have made: “It has that content because the speakers said this, and there is a rule of conversational kinematics by which, when that move is made in such a situation, the content for a relative adjective moves in such-and-such a way…” But what I don’t get is what’s added by putting in salience as an extra step: “His executing such a conversational maneuver, by such-and-such a rule or generalization, made that particular content for ‘tall’ SALIENT, and because it was salient, that was the content for his use”, “Because, for these reasons, that content was the most useful one to use, that made that content SALIENT in the context, and because it was salient, that’s the content that was expressed.” I generally feel that adding in the salience part doesn’t advance my understanding at all; the explanations are just as good (and therefore better, if you know what I mean) without the salience part. Well, I’m not trying to win over converts here to my general skepticism about salience explanations. I’m just speculating (I don’t recall the exchange securely enough to do more than that) that the reason I would express skepticism about the role of salience in response to Jeff’s question was probably because of this general skepticism of mine toward salience explanations. And so I don’t think I’m thereby making “knows” some kind of isolated freak. If I’m convinced that my general skepticism is mis-placed, then I’ll be as open to allowing salience a role in “knows” as in “tall” and the like.

Two Quick commments on Brian’s