Epistemic Counterparts 2: Acquaintance, files, and suitable roles
This is part 2 of a series on epistemic counterpart semantics. Part 1 is here.
I want to defend what I called the "Quine-Kaplan model" of de re belief ascriptions. According to this model, 'S believes that x is F' is true iff there is a suitable role R such that (1) S believes that whatever plays R is F, and (2) in fact, x plays R.
In this post, I mainly want to explain what I mean by a "suitable role". This will also bring to light some arguments in favour of the Quine-Kaplan model.
Let's start with a bit of history.
Quine originally didn't have a restriction to suitable roles. As Kaplan pointed out, this leads to false predictions.
In one of Kaplan's examples, Ralph believes that there is a shortest spy, but he has no idea who that might be. Let's say Jones is the shortest spy. Without the "suitability" restriction, we could infer that Ralph believes that Jones is a spy. After all, (1) Ralph believes that whoever is the shortest spy is a spy, and (2) in fact, Jones is the shortest spy. But intuitively, Ralph does not believe that Jones is a spy.
The lesson is that we need to restrict the relevant roles. 'Ralph believes that Jones is a spy' is false because there is no suitable role such that (1) Ralph believes that whoever plays that role is a spy, and (2) Jones plays that role.
That's not quite how Kaplan put the lesson. Kaplan, and many others, took the example to show that more is required for "singular thought" than a capacity to pick out the relevant object. What else could be required? Some argued that singular thought requires causal acquaintance with the relevant object. (This meshes well with causal theories of linguistic reference.) Others argued that singular thought is a matter of having special mental "files".
But proposals like these don't fit the linguistic data. For example, causal acquaintance is neither necessary nor sufficient for the acceptability of de re belief reports. The same is true for the possession of "mental files".
Van Fraassen (1979) described the following scenario. Susan, a student, reports to her mother a conversation she had with the college registrar. The mother exclaims: 'He must think I am rich!'. Here we have an acceptable de re report even though the registrar is not causally acquainted with the relevant object, Susan's mother. It is also doubtful that the registrar has a "mental file" for Susan's mother. So acquaintance or the existence of a file is not necessary.
Conversely, it is obvious that mere acquaintance, or the mere existence of a file, does not help with the shortest spy problem, and thus isn't sufficient to make a de re report acceptable. If Jones (the shortest spy) once talked to Ralph at a cafe, but Ralph has no idea that the person he talked to was a spy, it still isn't true that Ralph believes that Jones is a spy, although now he is causally acquainted with Jones (and he might have a mental "file" for Jones).
To render acquaintance sufficient, one might suggest that the believer must somehow associate the object of their acquaintance (or the object for which they have a file) with the object of the relevant belief: Ralph would have to know that the person he talked to at the cafe (the object of acquaintance) is the shortest spy (the object of his belief that the shortest spy is a spy). But that still won't do.
I mentioned a counterexample in the previous post. At one point, Mary Harshbarger saw a dark figure approach through the woods and thought it was a bear. In fact it was her husband Mark. Here Mary is causally acquainted with Mark, and believes that the object she is thus-acquainted with is a bear. So 'Mary believed that Mark was a bear' should be acceptable. And it is acceptable – but only in specific contexts. When we don't consider the problematic episode in the forest, but focus on how the relationship between Mary and Mark changed through their marriage (say), it is not acceptable to claim that Mary believed that Mark was a bear.
In general, the acceptability of de re belief reports (and epistemic modals – on which more soon) is highly sensitive to conversational context. To me, this makes it implausible that de re reports capture a psychologically important type of mental state, a special type of singular thought.
In the "Quine-Kaplan model" as outlined above, we can easily accommodate the context-sensitivity of belief reports. We simply have to assume that conversational context affects the range of suitable roles.
So what's a suitable role, relative to a particular context? The main conditions, I think, are the following.
Simplicity. Consider the role of being the US president if it is sunny and otherwise the oldest living donkey. Donald Trump currently plays that role. But suppose Fred falsely believes that it isn't sunny. He also believes what follows from this belief: that whoever plays the above role is a donkey. So (1) Fred believes that whatever plays the role is a donkey, and (2) in fact, Donald Trump plays the role. But we can't conclude that Fred believes that Trump is a donkey.
Why not? Because R is too gerrymandered. Gerrymandered roles are not "suitable".
(This looks like an absolute condition, but one might argue that what counts as simple or gerrymandered depends on context. Imagine creatures who have a simple predicate for the Trump/Donkey role, and who intuit that it expresses a simple and natural property. Perhaps these creatures would be more prepared to say that Fred believes that Trump is a donkey.)
Strength. The more you know about a person, the more we are inclined to ascribe to you a de re belief about that person.
To begin, suppose Alice does not know that the CEO of Amazon is rich, that he is called 'Jeff Bezos', that he owns the Washington Post, or anything like that. She does believe that all rich people are crooks. Does she believe that Jeff Bezos is a crook (understood de re)? In some contexts, the answer is yes – providing another counterexample to the idea that acquaintance or files are required for de re beliefs. But in many other contexts, the answer is no. Now imagine Alice learns that the CEO of Amazon is rich, and infers that he is a crook. Later she learns that the CEO of Amazon is called 'Jeff'. Then she learns that he owns the Washington Post. Then she learns that his surname is 'Bezos'. And so on. The more she learns about Bezos, the more acceptable it becomes to say that she believes him (de re) to be a crook.
To a first approximation, we might define the strength of a role R for a subject (e.g. Alice) as the number of simple roles that are believed (by the subject) to coincide with R, where two roles coincide if they are satisfied by the same things.
For Alice at the end of her enquiry, the roles CEO of Amazon, owner of the Washington Post, famous person called 'Jeff Bezos' all coincide.
All else equal, strong roles are more "suitable".
(Notice that if two roles R1 and R2 coincide for a subject, then on the Quine-Kaplan model, it makes no difference for the truth-value of a belief report (with that subject) whether we use R1 or R2 as the operative role. This might shed light on why strong roles are preferred.)
Acquaintance (?). Some roles reflect the subject's causal acquaintance with the relevant object. For example, person I'm currently looking at, or planet on which I'm standing. Such roles appear to make de re attributions especially natural. If (1) you believe that the person you're looking at is an idiot, and (2) the person you're looking at is Fred, it is natural to say that you believe of Fred that he's an idiot.
So perhaps acquaintance roles (or roles that coincide with acquaintance roles) have a comparatively high degree of "suitability".
I'm not sure about this one though. Acquaintance roles are usually strong, so perhaps the data that seems to favour acquaintance roles can be explained by strength.
Naming (?). Some philosophers claim that it makes a difference whether the subject has a name for the relevant object. This can be motivated as follows. Suppose I introduced the name 'Julius' for the inventor of the zip. Believing that the zip was invented in America, I might say, 'Julius was American'. This sentence expresses a singular proposition; sincerely uttering a sentence implies believing the expressed proposition; it follows that I have a singular belief about the inventor of the zip.
I don't buy this argument. And I don't think there is any clear data suggesting that having a name makes a difference. But there is data that could be interpreted that way, and the account I want to defend can accommodate this interpretation, so I put it on the list.
The idea would be that coinciding with a "naming role" – a role like person called 'Julius' – increases suitability.
Salience. Roles can be made more or less salient by the linguistic and non-linguistic context. Salience contributes to suitability.
This is what happens in the case of Mary and her husband. When we focus on the episode in the forest, the dark figure approaching through the woods role is salient, and so we're inclined to accept that Mary believed that Mark was a bear. But when we focus on Mary's marriage with Mark, other (and stronger) roles are more salient, and we're no longer inclined to say that Mary believed that Mark is a bear.
Charity. Unsurprisingly, general pragmatic considerations affect the domain of suitable roles. In particular, we tend to interpret sentences so that they are neither trivially true nor obviously false. Roles that would result in such an interpretation are normally unsuitable.
To illustrate, return to van Fraassen's example of Susan and her mother. 'He must think I'm rich' is acceptable. But 'he must think I'm you mother' is not. How come? The role that makes the first sentence acceptable is something like Susan's mother. That role would make the second belief report trivial.
I have presented a list of criteria that contribute to the suitability of a role. A full account would have to specify how these criteria are ranked. There's some interesting work on this in chapter 4 of Maria Aloni's dissertation, although her list of criteria differs somewhat from mine. I won't descend into these details.
I still haven't explained how the truth-conditions postulated by the Quine-Kaplan model are compositionally derived. I first wanted to explain what they are. Along they way, I wanted to present some more evidence in favour of the model. The data about the acceptability of de re reports is complex and messy. As far as I can tell, the "Quine-Kaplan model" with the proposed conditions on suitable roles fits the data nicely, and more so than popular versions of the "direct" interpretation.