So at the APA I was planning to attend this session on epistemic modals featuring Kai von Fintel, Thony Gillies and John MacFarlane. Unfortunately events intervened and Kai couldn’t make it. For better or worse, the organisers decided that it would be best if someone else, namely me, presented Kai’s paper. This kind of thing isn’t unusual at APAs, people often read out other people’s papers, but in this case I just had the slides to go off. Fortunately they were very good slides, as you can see here, and combined with some guidance from Thony and Andy Egan, I was able to stumble through fairly easily. I even learned something interesting from the paper.

Imagine the following situation. I’m standing in the hotel lobby, on the phone to Andy who is back in his room. I see some people coming in from outside with wet umbrellas and water dripping off their clothes. It seems I can say either (1) or (2) to Andy. (I assume he’s not looking out his window, so weather news is really news to him.)

(1) It must be raining.
(2) It is raining.

Now change the situation a little. I don’t say anything to Andy on seeing the wet people, but wait until I step outside, then report. Now (2) is OK, but (1) seems dubious. So in this context we have

(1) #It must be raining.
(2) It is raining.

It seems that (1) is not a proper assertion when I have direct visual knowledge of the rain. Philosophers (at least philosophers of my acquaintance) usually take the difference between (1) and (2) to be that (1) is only true if someone salient knows that it is raining. But it seems that (1) requires more, namely that the knowledge is somewhat indirect.

That’s what Kai suggested in the paper. I had various long discussions with Andy Egan, Kenny Easwaran and others about what might count as direct and indirect knowledge. The upshot seemed to be when it comes to perception, most anything except direct visual inspection counts as indirect, which is somewhat interesting.

But the most interesting thing I heard was a point that Ishani made. It seems that for the purposes of this exercise at least, testimonial knowledge counts as direct knowledge. (This contradicts the chart that Kai quotes in his slides.) To see this, imagine that Ishani tells me that the Packers won, and I thereby come to know that the Packers won, and Andy doesn’t know this. It seems I can use (4) but not (3).

(3) #The Packers must have won.
(4) The Packers won.

Note that I can say something like (5).

(5) Ishani said that Packers won, and she’s always right about these things, so the Packers must have won.

It is just standalone (3) that seems odd. But this isn’t really a distinction with (1)/(2), because it doesn’t sound too much worse to say (6).

(6) I have a visual impression of heavy rain, and my visual impressions are always reliable, so it must be raining.

Well, (6) isn’t great, but neither is (5) if Ishani’s reliability about football is something that is too obvious to mention (as it is).

Anyway, I now think I’ve learned two things in under a week. First, English has evidential terms. Second, the ‘swamp epistemology’ of English groups (at least for these purposes) visual observation and testimony together, and separate from inference, and (arguably) non-visual perception. That’s quite a lot to learn from looking at a few small pairs like that.

17 Replies to “Evidential”

  1. It’s worth noting that expressions of epistemic possibility are evidential, too. You can bring this out by putting them under negation.

    Suppose I’m talking to you on the phone, and I know by looking in his office that John isn’t in his office. Then my saying (7) would be odd but my saying (8) wouldn’t be odd.

    (7) # John can’t be in his office.
    (8) John isn’t in his office.

    If I have instead inferred that John isn’t in his office, on the basis of seeing that the lights are off, then it would be fine for me to say (7).

    This suggests that even theories intended to apply only to expressions of epistemic possibility have to accommodate evidentiality facts.

  2. Hi Brian:

    Interesting observation. Seems to me that ‘must’ functions as an entitlement “explicitating” device (to talk Brandomese a bit).

    That’s maybe why the non stand alone ‘must’ is okay when the stand alone one isn’t. If I have “direct” evidence and the entitlment generated by that evidence is unchallenged, then it’s not necessary to explicitate the the claim of entitlemtent. But if the entitlement is challenged, I am thereby invited to make my entitlement explicit.

    If Andy says to you, “Why should I take your word.” he’s invited you to make explicit your entitlement to the belief that it is raining. And the chain of reasoning you launch into about the reliabilty of your perception, etc. culminates in an explicitation of your entitlement in such a way that you can claim to have answered the challenge.

  3. I’m wondering about the relevance of direct/indirect evidence here. Return to the rain example or John in his office. Suppose you’re standing outside with someone insisting that “it isn’t raining, it’s closer to sleeting”. Seems to make sense to say it must be raining given the direct evidence of rainfall. Here the disagreement seems to be over what the direct evidence is (for rain or sleet).
    In the example of John in his office, you might be on the phone with someone insisting that he just spoke with John and John is in his office. Seems to make perfect sense to say that “John can’t be in his office, otherwise I’d be seeing him right now”. But here again you have direct evidence that he is not there.

  4. This is neat! In case you needed it, this, together with the (perfectly felicitous) closing line of Frost’s poem ‘The Census-Taker’, provides more evidence that our knowledge of our own desires is sometimes indirect:

    “It must be I want life to go on living.”


    Do you need the negation to bring out the same feature in statements of epistemic possibility? If I see that John is in his office, then (9) strikes me as inappropriate thing to say to you on the phone (it’s at least misleading, if not outright false):

    (9) John might be in his office.

    But if I’ve come to believe (9) by seeing that the lights are on in John’s office, (9) seems fine. Do you see an important difference between the infelicity of (9) and your (7)?

  5. Mike,

    I don’t think I share your judgment about the rain example. Do you have other cases like it in mind?

    In your “John can’t be in his office, otherwise I’d be seeing him right now” example, the speaker does seem to gesture at an inference the conclusion of which is that John isn’t in his office. At any rate, you’re right that more needs to be said about what the “direct/indirect evidence” distinction amounts to if it’s to explain the evidentiality features of epistemic modals.


    I like the Frost example!

    I would explain the infelicity of (9) in broadly Gricean terms: if you see that John is in his office, then you shouldn’t have said the relatively noncommittal, relatively unhelpful (9). You should have just said “John is in his office.” If you infer that he’s in his office on the basis of seeing that the lights are on, then perhaps the most committal, most helpful thing that it would be appropriate for you to say is that he might be in his office.

    I suppose I have theoretical reasons to think that we need wide scope negation to bring out the evidentiality of epistemic possibility modals. So I would be interested in seeing cases where negation isn’t needed.

  6. Eric—Isn’t (7), which has a wide-scope negation, equivalent to “John must not be in his office,” where the negation is narrow scope? It seems to me asserting an unnegated possibility, might-p, will always be infelicitous whenever your evidence allows you to assert p (for the Gricean reasons you mention). This is different from the contrast between (3) and (4), where (3) is unassertable even though it seems no weaker than (4). So it seems to me that when might-p is unassertable, it’s not for reasons of evidentiality.

    In fact, if I have direct perceptual evidence for p’s possibility that doesn’t establish the truth of p, might-p is assertable (I think). Suppose the doorbell rings, and I look out and see one of the Barber twins. “Is that Tiki Barber?” you ask. “It might be Tiki Barber,” I say.

  7. BTW, I stumbled across this a little while ago, thinking about the differences between “There is a solution” and “There must be a solution.” The combination with the existential creates some interesting cases, I think. Suppose I can see light spilling into the hallway, but I can’t see which office it’s coming from. I can say,

    (10) Someone must be in their office.

    But if I can see that the lights are coming from John’s office, it seems to me felicitous to say

    (11) Someone is in their office.

    (Let’s assume a context in which it doesn’t matter that (11) is less informative than “John is in his office”; say I’m talking on the phone to someone who wants to know whether they’ll be able to get into the office complex or whether it’ll be locked, so she cares if someone is in the office but not who it is.)

    Perhaps this means that there is some sort of relativity or sensitivity to what counts as indirect access; knowing by inference that some particular person is in the office is more direct access to “Someone is in the office” than knowing by inference that someone or other is in the office.

    (In comments to my post, my brother compares this to constructive vs. non-constructive proofs, and I offer a very tentative suggestion linking an account of epistemic possibility to evidentiality.)

  8. Eric,

    You probably would not balk at my saying instead, “it can’t be sleeting, otherwise I’d see it”. This is the analogue of the office example. I think you see this as something inferred from a different direct observation. But both cases seem to include direct observations. I see directly that it’s not sleeting and I see directly that it’s raining. So it’s hard to see the role here of the direct/indirect distinction.

  9. Hi, Brian. First, a comment on the exact nature of your proposed indirectness condition, and particularly on testimony being for these purposes direct. I don’t think it’s the case that whenever you know by testimony, you’re barred from expressing epistemic necessity. I think you’re right that we can in many cases treat knowledge by testimony as being, in the requisite sense, direct. But we can also, at least in some cases, treat testimony as evidence by which we come to know indirectly:

    He must have done it. 38 separate eyewitnesses say so. They can’t all be lying.

    doesn’t seem to bruise the ear.

    Second, on the status of your proposed indirectness requirement. You seem to be construing it as a truth-condition here:

    Philosophers (at least philosophers of my acquaintance) usually take the difference between (1) and (2) to be that (1) is only true if someone salient knows that it is raining. But it seems that (1) requires more, namely that the knowledge is somewhat indirect.

    Well, maybe you’re not committing yourself to the your proposed added condition being a truth condition. Philosophers propose such-and-such truth condititons, you say, but these cases show that indirectness is also required. That would seem to suggest that you think it’s also required for truth. But maybe you’re leaving it open whether it’s required for truth, or just for appropriate assertability.

    In any case, I think there are good reasons to think in your cases, the expressions of epistemic necessity are just inappropriate, but are not false. One good reason to be at least open to the possibility of their being true (though inappropriate) is that, in your cases where []P seems wrong, it also seems somehow wrong to assert >~P, yet it’s unlikely that []P and >~P are both false. So it seems likely that in the case of one of them, it’s not wrong b/c it’s false. Anyway, I think it’s the expression of epistemic necessity that’s really true (though it is inappropriate to assert). I have a Gricean story, but perhaps for here, I should just leave it at that.

  10. I’m sorry. I’m going to ask a question that most of you will probably think is very naive. I find your blog very interesting, Brian, but I have this kind of question a lot, and now I’m asking it.

    I’ll buy the argument that more warrant attaches to some types of evidence than others, and if you tell me that IT MUST BE RAINING and IT IS RAINING indicate different levels or types of evidential support, that’s fine. Tell me what your theory of the difference is, and then we can discuss in particular cases how our separate intuitions apply those to particular cases.

    But we’re interested in warrant, right, and not language. So it seems confusing to say that “It must be raining” means IT MUST BE RAINING, and begin our discussion with our intuitions about the value of evidence as judged through the medium of language. If “It must be raining” sometimes means IT MUST BE RAINING and other times IT IS RAINING, then can’t we put our caps lock on and just get on with our discussion of evidence?

    Put differently, does the fact that English has a particular set of modal words tell us anything about evidence? Maybe if there are modal universals. Maybe. Otherwise no, I’d venture.

    So I see a lot of talk about intuitions about nuances of language, and that’s fine if it’s getting at underlying issues. But a lot of the distinctions that a language allows for are contingent results of the evolution of the language, and don’t tell us anything general about experience, except maybe through a long and circuitous process that’s liable to take us pretty far from our main (philosophical) concerns. Thanks

  11. Now that Brian’s gone this thread is probably dead, but I’ll chime in anyway.

    First of all, Carl, I think that’s an excellent question. I’d quote J.L. Austin saying (approx.) that while ordinary language use is not the last word, it is the first word; nuances of language may provide an interesting place to look for deep thoughts. Or they may not. They can also be interesting in themselves. Other people may not share this view.

    Keith, I like this and agree it sounds OK:

    (12) He must have done it. 38 separate eyewitnesses say so. They can’t all be lying.

    I’ll plug this into my tentative suggestion, that “might” and “must” are peculiarly well suited to conversations in which the participants are trying to figure out some matter, rather than simply sharing information. (This proposal fits nicely with relativism about the modals, in a way I haven’t explained.)

    That would account for why ‘must’ functions as an entitlement explicating device, as Ken said, since in these conversations you need to explain your reasoning. And maybe it accounts for why ‘must p’ suggests that we have indirect/inferential evidence for p: If we had direct evidence, we wouldn’t need to use a locution designed to further inquiry along; we could simply say ‘p’ and end inquiry.

    In some contexts (12) will sound odd. If you ask me “Did he do it?” and I expect you to believe whatever I say, (12) will sound distinctly like protesting too much. But if you’re arguing with me about whether he did it, (12) will sound good. By uttering (12) I offer my evidence that he did it. OTOH it doesn’t seem to me as final as “He did it. Forty-seven people saw him” (this is a mushy intuition).

    So far so good for me. But it also seems that (1) isn’t even good in the context of inquiry. Like Eric, I don’t share Mike’s intuitions that “It must be raining” is OK when you’re both outside and someone is arguing that it’s sleeting. And I think “He must have done it, I saw him” also sounds odd, even in the context of inquiry.

    Perhaps this is because with “I saw him” you’re offering a warrant yourself, and you’re offering it as conclusive (if you didn’t think it was conclusive you’d say “I’m pretty sure I saw him”). Your interlocutor can’t question your access to this warrant. As Ken pointed out, the interlocutor can raise a question—“But you’re not good at identifying people” or “But other witnesses place him far away from the scene.” Then you can explicate an entitlement that goes beyond bare “I saw him” — “It must have been him, I saw his distinctive wing markings.”

    I’m not totally happy about this, though.

    Incidentally, to make a < type &lt;

  12. Like Eric, I don’t share Mike’s intuitions that “It must be raining” is OK when you’re both outside and someone is arguing that it’s sleeting

    Hmm. That’s two (at least) that think I’ve got this wrong. Why do I think it’s right? Maybe the difference is here. Consider a case where your friend has you nearly convinced that his observations are right—it is sleeting. On review of the conditions—again, it seems to me—you could reasonable say “no, wait, it must be raining, just look at the window or ground”.

  13. Mike, it does make a lot more sense to me when the friend has you nearly convinced. Then, as you point out, you can cite something other than your direct perception. In fact, I can even imagine a situation where both of you know that your friend can often convince you of things just through his massive personal magnetism, so you catch a hold of yourself and say to him, “No, it must be raining, I can see it/feel it.” That seems like what Ken envisioned, where you’re basically adverting to the reliability of your perception. (And, as I should have noted in my last post, like (6).)

    So this is a case where you can use “must” in the course of a live argument. Whereas when you’re on the phone to the guy outside there isn’t a live argument — Andy is just asking you for more information. And if you aren’t nearly convinced that it’s sleeting, I’d say that you shouldn’t acknowledge that there’s anything to argue. You’re rejecting the legitimacy of calls for argument.

    In the indirect evidence cases (when you’re in the lobby, or in (10)), there isn’t a live argument either, but maybe the use of ‘must’ indicates that there could be one.

  14. The main problem with natural language sentences is that one needs to explain why the following axiom does not hold:
    (A) []p ↔ p
    In dealing with an artificial language, one pre-selects the kind of logic he wants to use, and so defines the meaning of box [] and ◊ diamond, in terms of necessity versus possibility, or obligation versus permission, etc. But natural languages have chamaleonic modals, as Kai pointed out. So, even if you say that you are dealing with the epistemic, deontic or other interpretation of the English modals, there are still flavours and sub-flavours to be considered.
    So the issue is to explain why, in context where (i.) is true:

    i. It is raining.

    I cannot correctly infer that:

    ii. It must be raining.

    From the Philosophical and Logical side of the issue, we need to say which kind of modal logic is at place and which system, i.e., which axioms are in cause. If we can do that, we can shed more light on the Linguistic side of the issue. And then Logic and Linguistics will help each other. If not, it suffices to use Kai’s analysis, given the empirical evidences at hand.

  15. I have to agree with Carl Seaquist’s post above. The distinction seems simply this:

    1) We generally say “it is raining” if we observe evidence that immediately and necessarily correlates with rain, in our minds. Seeing rain out the window would qualify.

    2) We generally say “it must be raining” if we observe evidence that, after a process of deduction, implies a sufficiently high degree of probability that it is raining. Seeing people walk in the lobby shaking water off umbrellas would qualify.

    On the whole, there’s a great deal we can analyze about language, but a significant quantity of that analyzable content probably won’t yield anything interesting, for several reasons.
    1) Natural language is inherently imprecise; it sacrifices accuracy of information for quantity of information. (Computer geeks, think high bandwidth unreliable memory.) Thus, painstaking precise inquiries into the nuances of language will often end up creating precision where none previously existed, usually out of the intuitions of the philosopher, and thus to no perceivable gain.
    2) Even when such precision seems to be available for discussion, there’s very little guarantee that it’ll tell us anything about ourselves, as Carl points out. You’ll know an intricacy of our language, perhaps; but how does knowing, through and through, the distinction between “it must be raining” and “it is raining” accomplish anything? It doesn’t seem to provide any interesting epistemic insights – at least, the explanations that I’ve heard don’t. At most it seems to provide an example of the obvious fact that human estimations of probability depend highly on the number of (human) logical derivations necessary to arrive at a conclusion from the given input.

    I think it would help all such discussions to carefully and precisely make explicit the consequences of competing viewpoints. If my viewpoint implies nothing interesting but Francisco’s viewpoint implies something interesting, let the debate begin! But if, either way the result isn’t terribly interesting, let’s move on to something more inherently contentful.

  16. Keith de Rose,

    Hi. Reading your post, where you say I have a Gricean story, but perhaps for here, I should just leave it at that. I was wondering:

    Have you any proposal to approach the issue on the grounds of felicity and modality together?

    It is really odd to think that we could have a system that rejects (A1) below:

    (A1) If p is the case then p may be the case.

    Which is to say, it is not intuitive to think that something is true and at the same time it is not possible. Then, it strikes us that if one already states (i.):

    (i.) It is snowing.

    he cannot say:

    (ii.) So it may be snowing.

    The same for the necessity operator: if p has to be the case, so p is the case. If we exclude the deontic reading, we cannot imagine how something has to be true and yet is not true at the same time.

    But if you think in terms of felicity and implicature, then you can approach the issue in a way that, at least for linguists, is satisfactory.

Comments are closed.