Thursday, November 04, 2010

Motivating Attitudes De Dicto and De Se

According to the Property Theory of content, the contents of our beliefs, desires, etc. aren't propositions, but properties. The contents of de se, or irreducibly first-personal, beliefs are properties we self-ascribe; if I believe that I myself am the messy shopper, I self-ascribe the property of being the messy shopper. If I merely believe de dicto (or even de re) that Jason is the messy shopper, then what I believe is the property of being such that Jason is the messy shopper. If I'm suffering from selective amnesia and have forgotten who I am, I may believe the latter without believing the former.

I like the property theory. I'd like to believe it. I've even defended it against objections. But I'm worried that it's motivations are going to over-generate. The property theory can be thought of as taking thoughts best expressed with one indexical expression -- "I" -- and inserting a slot into the content of this thought for the bit associated with that expression. (E.g., "I am the messy shopper" expresses the content ____ is the messy shopper.) I'm worried the argument that gets us to add in this slot is going to drive us to add in other slots as well.

Consider what I take to be the strongest case for the property theory: Lewis's ("Attitudes De Dicto and De Se", 1979) case of the two gods. Zeus lives at the top of the mountain; Poseidon lives at the bottom of the deepest ocean. They both know all the true propositions. But neither knows who he is. Zeus knows that Zeus is at the top of the tallest mountain; but he doesn't know that he is at the top of the tallest mountain. Since he knows all the true propositions (Lewis argues), and since if he did know that he was at the top of the mountain he'd have a (new) true belief, whatever content Zeus fails to be belief-related to must not be a proposition. But properties: those could do the job. Zeus could believe all the propositions but not believe the property being on top of the mountain.

Properties aren't the only way to handle Lewis's two-gods case. We could instead have belief as a triadic relation between, roughly, a believer, a proposition believed, and a way of presenting that proposition to oneself. Then Zeus might believe the proposition that Zeus is on top of the mountain under one mode of presentation, but not under another, first-personal, mode. Why prefer the property theory to this one? Neil Feit (Beliefs About the Self, 2008) argues (inter alia) that the property theory is just more streamlined, more elegant, than the triadic theory. We'll come back to this in a mo.

Here's the case that's worrying me. We have one god, who is sitting in front of two ghostly spheres -- call them Bo and Luke. They're intrinsic duplicates and, gosh, wouldn't you know it, they're occupying the exact same region right now. But one of them is going to move here in a minute.

Beings like us will have a hard time ostending one of the co-located spheres. But that's no problem for a god! So this god ostends one of them and says, "I wonder if that one is going to be the one that moves in a minute."

It looks like we can repeat the Lewis-style worries here. Our curious god -- call her Daisy -- might well know that Bo is going to move in a minute, but not know whether she is ostending Bo or Luke. Indeed, it looks like she might know all propositions, but still not know whether that sphere is going to move in a minute. So -- by parity of reasoning -- if Lewis's gods case drives us to add a slot in for irreducible "I"-thoughts, shouldn't the Bo and Luke case drive us to add in a slot for irreducibly demonstrative thoughts? But I take it this would be a disaster (once we see the trick, it's a good bet this will get out of hand pretty soon), so we should resist drawing the property-theory lesson from Lewis's two gods case.

I expect the property theorist to respond: "If we're already property theorists, we can find a property that Daisy doesn't believe: the property of ostending Bo. Once she comes to know that property, since she also knows that Bo will move in a minute, she will be in a cognitive state that she is not in now --- and it's one that can serve the role of 'knowing that that sphere will move in a minute'".

But the simplest version of this won't work. Suppose Daisy wonders, "Will it be that sphere or that one which moves in a minute?", respectively ostending Bo and Luke in the process. Even if she knows that she has ostended Bo during her wondering, this won't improve her cognitive state (because she has also ostended Luke). So the property theorist will have to resort to a more linguistically fine-grained property for Daisy to believe, one along the lines of "the property of having first ostended Bo and then ostended Luke", or something like that.

I don't have any argument the property theorist can't make this move work. I rather suspect he can. What I want to point out now is that the property theorist is now relying on properties that seem to be close to the triadic theorist's modes of presenting a proposition. That is: there will need to be some sort of quasi-syntactical specification of the thought that Daisy is having, so that Daisy can learn how parts of this thought are related to the world (e.g., that this part is related to Bo, and that one is related to Luke). This isn't the same thing as the triadic theorist's view by a long shot; but it makes use of many of the same sorts of resources.

But once we're going down this line as property theorists to deal with Daisy's ignorance, what happens to the objection to the triadic theorist's treatment of Zeus's ignorance? The triadic theorist, in essence, says that Zeus doesn't know his mental tokens of "I" pick out Zeus; the property theorist (on the envisaged response) says that Daisy doesn't know that her (particular) mental tokens of "that" pick out Bo and Luke, respectively. If we're going down this line anyway, why not be triadic theorists from the get-go? Maybe the triadic theory is ugly, but if the property theory has to partake of this same ugliness, then there's no argument from ugliness in favor of properties over modes of presentation. And the property theory in fact looks worse, because the triadic theorist can treat what seem like similar phenomena -- indexical ignorance -- in a similar fashion, whereas the property theorist treats some cases of indexical ignorance very differently than others.


Shen-yi Liao said...

I am certain that I am misunderstanding the case, but I'd like to see where I am getting it wrong.

Daisy's ignorance here seems to be about which world she is in. That is, contrary to the stipulation, she doesn't seem to know all propositions because she doesn't know everything about the world. In particular, she does not know whether this is a world where she ostends Bo before Luke or a world where she ostends Luke before Bo.

(I am assuming here that Daisy, by the god stipulation, is not ignorant about Bo and Luke. She might know them by acquaintance through their names or through their essences.)

Jason said...

I don't think you're misunderstanding the case. I haven't fleshed it out enough to tell us enough about whether she can consistently have the sort of knowledge you point to.

If Daisy can distinguish a unique token of "that" which refers to Bo and is embedded in a sentence about spheres moving in a minute, she will have some (metalinguistic) knowledge that can go proxy for her knowledge that that sphere will move in a minute.

But (i): she might not be able to do this. First, her tokens might not be temporally extended; perhaps she can token thoughts instantaneously. Then there won't be any "first" token to speak of. And even if her tokens are temporally extended, she might token the sentence twice, simultaneously (as a god, she can perform simultaneous tokenings) where the only difference is that in the second tokening the ostentions are reversed. Then she'll know that one of her first tokenings ostended Bo and one ostended Luke, but that won't let her tie those tokens down to movers and non-movers.

Moreover (ii): suppose she's in a vanilla case, with just one temporally extended tokening. Then the property theorist can identify her demonstrative knowledge with some metalinguistic knowledge. What I was thinking was that this would be a serious cost of the view -- a cost that triadic theorists don't face -- because it would mean demonstrative knowledge would require metalinguistic capacities. My two year old daughter can think demonstrative thoughts to herself, or so say I, but has no metalinguistic concepts at all.

So on this front we might re-do the case with Daisy*, who isn't omniscient, but only omniscient of all non-metalinguistic facts. She lacks the demonstrative knowledge. So: either demonstrative knowledge isn't propositional, or it's metalinguistic -- and I take it either horn is a bad horn for the property theorist to be on.

Shen-yi Liao said...

That helps, I think. But still, I want to press on the metalinguistic facts horn.

It's not yet clear to me which facts count as metalinguistic facts and which count as non-metalinguistic facts. Are facts about how people use names (including Daisy* herself) metalinguistic facts? This seems to be too broad, and denies propositional (or, umm, property-al) knowledge that involve names.

So I was thinking that all Daisy, and Daisy*, needs to know are what the names "Bo" and "Luke" refer to, via acquaintance or essence, then we can run a story about reducing de re to de se as Lewis does at the end of "Attitudes de dicto and de se".

In fact, I am not sure whether the property theorist would not want to be led to the other horn either. I think the answer that Lewis would want to give is that, yes, demonstrative knowledge isn't propositional -- it's property-al because de re reduces to de se.

Jason said...

Last things first: since the property theorist thinks all contents are properties, they deny propositional knowledge; I should have phrased the Daisy case so that she has all (non-metalinguistic?) property knowledge: for every property P, she believes P iff she instantiates P (modulo some complications we can ignore here). The idea is that there still isn't a property that she believes which can count as knowledge that that sphere will move. So if we follow the property theorist's initial strategy, we turn contents into relations.

Now, onto the metalinguistic horn: I'm not sure what you mean by "denying metalinguistic knowledge". Let's distinguish explicit metalinguistic knowledge --- like knowledge that "red" picks out a color, say --- from implicit knowledge, the sort we manifest by being competent language users. I take it we can be competent users of "that" without knowing what our various tokens of it pick out. (Suppose want to pick a student at random to answer a question; I close my eyes, point in the general direction of some students, and say 'You!'.) And I was thinking if we supposed Daisy* had no explicit metalinguistic knowledge, then your maneuver wouldn't help her -- and the argument would show that, if property theory was right and triadicism wrong, demonstrative knowledge would need to be explicitly metalinguistic.

I'll grant that you can run a reduction-of-the-de-re type story by appealing to metalinguistic knowledge. But I was thinking that was a cost of the view -- a cost not shared by triadic theories. (You believe something under a guise, which looks like a metalinguistic entity in some form or another, but you need not have any explicit beliefs about that metalinguistic entity to believe something under it.) So even if Lewis already likes such a theory, that's to say that Lewis has already incurred a cost the triadic theorists don't have to pay.

Why think it's a cost? Well, here's one reason: familiar Kaplan-style arguments seem to show that demonstrative knowledge isn't metalinguistic knowledge, because the demonstrative knowledge has a different modal profile than the metalinguistic knowledge. Suppose I know that that is going to move in one minute. It seems that I know something that would have been true regardless of whether I ostended anything just then (or even existed). But the metalinguistic knowledge is contingent on my ostending something. So the two seem to be distinct.

Anonymous said...
This comment has been removed by the author.
Anonymous said...
This comment has been removed by the author.
Neil Feit said...

Jason, I think property-theorists can handle your case with the same moves that are already needed for making sense of de re belief. For example if I have a de re belief, of my cat, that she is old, then what I believe (i.e., self-ascribe) is a property like this: being R-related to something old. Here, R is a relation that I in fact bear to my cat, maybe it's the relation x sees y. (It's what Lewis calls a relation of acquaintance.) In your case, what Daisy doesn't believe - but merely wonders - is a property like this: being R-related to a sphere that will soon move. (Notice that neither Bo nor Luke figures into this property.) She does not know, then, whether she is R-related to Bo or to Luke. There are some assumptions here that making reference by ostension involves cognizing a relation that one bears to the ostended object; you might not be comfortable with that. And sure, by using relations of acquaintance the property theorist makes use of the same sorts of resources as the triadic theorist, but such resources are needed by everybody to make any sense of belief reports (especially ones with proper names etc.). I am biased, but with the property theory we still only have one, two-place relation between subjects and their belief contents.
PS - Thanks for the book mention.