Epistemic counterparts 6: De re/de dicto equivalence
Compare the following three sentences.
(1) I thought my husband was a bear.
(2) Mary thinks her husband is a bear.
(3) I think my husband is a bear.
(1) and (2) are ambiguous between a "de re" reading and a "de dicto" reading. But (3) only seems to have the "de dicto" reading. How come?
According to the semantics I have described in earlier parts of this series, an utterance of (3) is true on its de re reading iff (roughly) there is a suitable role R such that (i) in all the speaker's belief worlds, whatever plays R is a bear, and (ii) in the actual world, the speaker's husband plays R.
These conditions are satisfied, for example, in the scenario in which Mary Beth Harshbarger mistook her husband for a bear. In that context, an utterance of (3) would have been true on its de re reading, and false on its de dicto reading. Why can't we hear the true de re reading?
The answer, I think, is pragmatic. The de re reading of (3) is unassertable unless it coincides with the de dicto reading.
A minimal condition for assertability is that the speaker believes that what she says is true. So the de re reading of (3) is assertable only if at each of the speaker's belief worlds w, there is a suitable role R such that (i) in all belief worlds, whatever plays R is a bear, and (ii) in w, the speaker's husband plays R. Under plausible assumptions, this implies that the de re reading of (3) is assertable only if at each belief world, the speaker's husband is a bear.
Here is why. Suppose the de re reading of (3) is assertable. Let w be any belief world. There is then some role R such that (ii) the speaker's husband plays R at w, and (i) a bear plays R at all belief worlds. Assuming that w is doxastically accessible from itself, it follows that a bear plays R at w. Assuming that R is a "definite" role that can't be multiply realised, it follows that the speaker's husband is a bear at w.
So the de re reading of (3) is assertable only if the de dicto reading of (3) is true: if at each belief world, the speaker's husband is a bear. Since we assume that speakers generally believe what they assert, we therefore can't hear a separate de re reading for (3).
This nicely explains the difference between (1) or (2) and (3).
Now for a complication. An analogous puzzle arises with expressions like 'suspect', 'fear', 'worry', 'hope', or 'might' that quantify existentially rather than universally over the accessible worlds.
For example, while (4) and (5) have two readings (one de dicto, on de re), (6) appears to have only one.
(4) I hoped that my husband was a bear.
(5) Mary hopes her husband is a bear.
(6) I hope my husband is a bear.
Similarly with (7) and (8) vs (9).
(7) My husband might have been a bear.
(8) For all Mary can tell, her husband might be a bear.
(9) My husband might be a bear.
The semantics of 'hope' and 'might' is a little complicated, but most of the complications are (I think) not directly relevant to the present puzzle.
As a rough first stab, let's say that 'S hopes that φ' is true (de dicto) iff (i) φ is true in some of S's belief worlds, and (ii) S assigns positive utility to φ. Accordingly, 'S hopes that x is F' is true (de re) iff there is a suitable role R such that (i) in some of S's belief worlds, whatever plays R is F, (ii) x plays R in the actual world, and (iii) S assigns positive utility to the set of worlds at which whatever plays R is F.
A first stab for (unembedded) 'might' treats it as the dual to 'I believe': 'it might be that φ' is true (de dicto) iff φ is true at some of the speaker's belief worlds. Accordingly, 'it might be that x is F' is true (de re) iff there is a suitable role R such that (i) in some of the speaker's belief worlds, whatever plays R is F, and (ii) x plays R in the actual world.
Now it would be nice if my pragmatic explanation for why we can't hear an ambiguity in (3) carries over to (6) and (9). Unfortunately, it does not.
The problem is a little easier to explain with a different example. Consider (10).
(10) The winning ticket might be a losing ticket.
According to the above semantics, an utterance of (10) is true on its de re reading iff there is a suitable role R such that (i) in some of the speaker's belief worlds, whatever plays R is a losing ticket, and (ii) in the actual world, the winning ticket plays R.
These conditions are satisfied, for example, in a situation in which I have bought a ticket and don't yet know that it is the winning ticket – assuming that the ticket I've bought is a suitable role.
I would like to say that even though the de re reading of (10) is true here, it is not assertable, because I can't know that it is true. Intuitively, the de re reading is true in part because I bought the winning ticket, and in part because there are belief worlds at which I bought a losing ticket. But I can hardly know these two facts. If I knew that I bought the winning ticket, there would no longer be belief worlds at which I bought a losing ticket.
But let's have a closer look.
Let's assume, as before, that a sentence is assertable only if it is believed to be true. Then the de re reading of (10) is assertable only if for each of the speaker's belief worlds w, there is a suitable role R such that (i) in some belief world, whatever plays R is a losing ticket, and (ii) in w, the winning ticket plays R.
It does not follow from these conditions that in some belief world, the winning ticket is a losing ticket. The problem is that the role can vary from world to world.
For a toy example, suppose there are only two tickets: one red, one green. We don't know which of them is the winning ticket. In world 1, the red ticket is the winner; in world 2, it's the green ticket. Presumably the red ticket and the green ticket are suitable roles. If so, then in world 1, there is a suitable role R (namely, the red ticket) such that (i) in some belief world, whatever plays R is the losing ticket, and (ii) in world 1, the winning ticket plays R. Likewise for world 2 with the green ticket as R. So the conditions for the assertability of (10) are satisfied.
The same kind of problem arises for (6) and (9). It looks like we need a different explanation for why the de re reading appears to be unavailable in these cases.
I have come across four proposals in the literature. (I have only recently started seriously thinking about epistemic modals and attitude reports, so I'm still ignorant of much of the recent literature, but I'm slowly catching up.)
Von Fintel and Iatridou 2003 suggest an "Epistemic Containment Principle" which prohibits quantifying into epistemic modals. Assuming that the de re reading of (10) involves quantifying into the modal, this would explain why (10) only has a de dicto reading.
I have four objections. First, the Epistemic Containment Principle is false:
(11) Every day might be our last.
This is a reasonable thing to say, but its only reasonable reading is de re. Second, the ECP doesn't explain the difference between (7) and (8) on the one hand and (9) on the other. Third, the explanation does not carry over to (6). Fourth, the puzzle still arises if we enforce the de re reading by putting the relevant description in a scope island, as in 'my husband is a man who might be a bear'.
Ninan 2018 suggests (in effect) that uttering (10) makes the role the winning ticket salient, so that it becomes the uniquely suitable role.
I have three objections. First, this doesn't explain the difference between (7) or (8) and (9), or between (4) or (5) and (6). Why would the context of (9) make my husband uniquely salient, but not the context of (7)? Second, if the effect is just a matter of salience, one would expect it to go away if a different role is made salient. But it is very hard to hear (6) or (9) or (10) as expressing anything other than their de dicto reading, no matter how much we try to make other roles salient. Third, in the ticket example, the winning ticket is actually a fairly poor candidate for a suitable role. The ticket I've bought is much better – at least by the criteria from part 2.
Yalcin 2015 suggests that the puzzle raised by (10) is solved by a Veltman-style dynamic semantics for 'might'. Roughly, the idea is that if we say that some x is F, we can't go on to say that x might be non-F without "crashing the context": the first utterance removes worlds where the relevant thing is not-F from the live possibilities, so the "test" expressed by the second utterance is guaranteed to fail.
I have two objections. First, this explanation does not seem to carry over to (6). Second, Yalcin assumes a simplistic "singular" analysis of 'it might be that x is F' that does not involve any roles or modes of presentation. I don't think such an analysis is adequate in light of other puzzles about de re epistemic modals (e.g. those discussed in Ninan 2018, on which more later).
Aloni 2001 presents a much more complicated dynamic semantics for 'might' that does take into account modes of presentation. Roughly, on her account, every singular term is associated with a suitable role. When we say that x is F and x might be non-F, the variable 'x' is associated with a fixed role, settling how its value is tracked across the worlds in the relevant information state. With that assumption, my simple diagnosis of (3) carries over to (6), (9), and (10). We don't really need the complicated dynamic semantics.
Recall that the problem for my explanation in the case of (10) arose because the "suitable role" for the winning ticket could vary from belief world to belief world. If the role is contextually fixed by the utterance context, plausibly no such variation is possible.
The assumption that trans-world identity conditions are contextually associated with singular terms is standard in individual concept semantics, and popular in counterpart-theoretic semantics. But I don't think we can make this assumption when the relevant possibilities are epistemic. As I mentioned in the previous post of this series, the relevant role can vary from world to world, as in (11):
(11) There is a man whom Mia would have believed to be an idiot if she had seen him at the pub but not if she had seen him at the beach.
We also want to allow different people to think of an object under different modes of presentation:
(12) Ralph and Fred both believe that Ortcutt is a spy.
Now, to deal with (6), (9), and (10) we don't really need the relevant role to be fixed across counterfactual worlds and different epistemic subjects. It would suffice if it were fixed across the speaker's epistemic alternatives.
That is, it would be enough if we could assume that when I say 'the winning ticket might be a loosing ticket', then the way the winning ticket is tracked across the live possibilities does not depend on which of these possibilities is actual.
Unfortunately, I'm not sure how to make this work.
My fall-back explanation involves two additions to the criteria for suitable roles.
First, if an object is introduced in a certain way, then that way is a good candidate for a suitable role. To illustrate, suppose Fred knows that Sydney has a harbour, and falsely believes that it is Australia's capital. Then compare (13) and (14).
(13) Canberra, which is Australia's capital, is a city that Fred believes to have a harbour.
(14) Australia's capital is a city that Fred believes to have a harbour.
To my ears, (14) sounds better than (13), even though both make the Australia's capital role salient. This suggests that the way an object is introduced has a special influence on the domain of suitable roles that goes beyond mere raising to salience.
Second, if the relevant subject currently expresses a thought in which she picks out an object under some role, then that role tends to be suitable. More generally, if a role plays an active role in the subject's mind it arguably becomes more suitable.
In cases like (6) and (9) and (10), these conditions combine to strengthen the suitability of the role by which the relevant object is picked out, making it the dominant choice for a suitable rule.
(It's worth noting all three cases arguably have a marginal acceptable reading that involves other roles.)