For Pacific APA session on Chalmers’ Two-Dimensional Semantics. Chalmers' reply is here.

3/30/02

 

 

Mental Content and Epistemic Two-Dimensional Semantics

 

Stephen Schiffer

New York University

 

David’s epistemic understanding of two-dimensional semantics has these two features.  First, although he considers at least two construals of epistemically possible worlds, on one of them they are centered metaphysically possible worlds.  Second, David intends epistemic two-dimensional semantics to yield a theory of propositional-attitude content, as well as having application to the semantics of natural language expressions.  These two features come together in David’s “The Components of Content,” where he deploys the apparatus of epistemic two-dimensional semantics to provide an account of propositional-attitude content, and where, for reasons of simplicity and familiarity, epistemically possible worlds are taken to be centered metaphysically possible worlds.  My talk is about this account of content.

            A brief sketch of a restatement of David’s theory that’s faithful to his own preferred way of stating his theory might go something like this:

Thoughts are propositional-attitude state tokens, and they are composed of concepts, where a concept is the analog in thought of a term in language.  Thoughts and concepts have content, and the content of a thought is determined by its structure and the contents of its component concepts.  The content of a concept decomposes into its epistemic content, given by its epistemic intension, a function from epistemically possible worlds onto extensions, and its subjunctive content, given by its subjunctive intension, a function from metaphysically possible worlds onto extensions, and likewise for the content of a thought, where the extensions of thoughts are truth-values.  Thus, the epistemic content of a thought is a kind of truth-conditional content.  Another defining feature of epistemic contents is that they are narrow contents, contents determined by internal states of thinkers’ cognitive systems (e.g., by a combination of a subject’s functional organization and phenomenology).  Still another defining feature of epistemic contents is that they reflect rational relations between thoughts, thereby making them what are nowadays called “modes of presentation” of the objects and properties our thoughts are about.  Correlative to these defining features are certain others.  Epistemic intensions are such that given sufficient information abut the actual world, we are in a position to know whether our thoughts are true, and a thought’s epistemic content is rationally prior to any knowledge of the thinker’s environment, thus enabling it to capture the way a thought’s truth-value depends on the character of the environment, and so is independent of that environment itself.  And if the epistemic intension of one thought entails that of another, a thinker should in principle be able to infer the second from the first by (idealized) a priori reasoning.  Other features and consequences of this decomposition of mental content into epistemic and subjunctive content are …. 

Now, given that there are epistemic contents as defined in terms of the foregoing features, and given that they are functions from epistemically possible worlds onto extensions, what, we need next to ask, are these “epistemically possible worlds”?  Well, one possible answer is that they are centered metaphysically possible worlds.  Subjunctive intensions present less of a problem, because they are plausibly modeled in the familiar way as functions from metaphysically possible worlds onto extensions.  But there are such-and-such prima facie problems with taking epistemic worlds to be centered worlds, and so other models need to be explored….

            But I’m not very happy with this style of presentation.  Thoughts do have subjunctive contents, since that’s merely to say that we can often ascribe to thoughts the truth-values they would have were certain states of affairs to obtain, and for certain purposes it may be benign, although perhaps not very instructive, to represent the subjunctive aspect of a thought’s content as a function from possible worlds into truth-values.  But one can reasonably doubt whether there are such things as epistemic contents, and I don’t think David does much to motivate their existence; doesn’t, that is to say, do much to show that there are things that have the properties definitive of “epistemic contents.”  He more or less simply postulates epistemic contents, shows what work they could do if they did exist, and wonders how they should be construed.  This might be an OK way to proceed were it not for the existence of antecedently good reasons to doubt that there can be such things as epistemic contents.  For this and other reasons, it seems to me that a more convincing and perspicuous way of proceeding would be as follows.  First, show that state tokens, characterized in a way that makes clear that they have only contingently the contents they have, stand in certain relations to certain two-dimensional formal structures, characterized without the use of tendentious labels that carry unearned implications.  Second, proceed from there to argue that those structures, standing in those relations to thought states, exhaust the contents of thoughts, determine the truth-values those thought states both actually have and have in counterfactual circumstances of evaluation, and have all the other properties pertaining to rationality and “narrowness” that David would like them to have. 

            So here’s how I’m going to proceed.  First, using my own theoretically neutral labels, I’m going to offer a bald statement of a theory, which I’ll call Theory C, that I take to be entailed by David’s theory of mental content when epistemic worlds are taken to be centered metaphysically possible worlds.  I focus on centered worlds because, for reasons I can’t go into, I think they’re best suited to David’s overall semantic and metaphysical project.  Because my allotted time is very limited, I won’t try to justify the attribution of Theory C to David but will rely on his correcting me if at any point I get him wrong.  Having stated Theory C, I will then state what I take to be certain problems for it.

                       

Theory C

 

Let’s assume that we think in an inner system of mental representation, a lingua mentis I’ll call Mentalese (this assumption is intended to represent the cash-value of David’s claim that thought tokens are sentence-like things whose basic components are things he calls “concepts,” where those things are bearers of content and not themselves contents).  As a useful heuristic, we can pretend that Mentalese is neural English, the neural expressions standing to spoken English expressions in roughly the way the spoken expressions stand to their orthographic counterparts.  I’ll refer to Mentalese expressions by way of referring to their orthographic counterparts, and I’ll assume that for any kind of semantic value—an extension, an intension, or whatever—the semantic value of a complex expression is determined, roughly speaking, by its structure and the semantic values of its component expressions.  Theory C asserts the following.

 

1.             For every Mentalese expression token e (capable of having an extension, a qualification henceforth to be implicit) there is a function f , which I’ll call e’s 1-intension, such that, for any x and any centered world w, if f(w) = x, then x is the extension of e in w, when w is considered as actual (I assume this audience knows what that means).  In the event, I’ll call x the 1-extension of e in w.  Some examples by way of rough illustration.  The 1-intension of ‘Julius’ is that function which maps a centered world onto the inventor of the zipper in that world, if there is a unique such person, and is undefined otherwise.  Thus, as Bob Stalnaker has recently revealed, the 1-extension of ‘Julius’ in the actual world is a certain Whitcomb L. Judson, but David is the 1-extension of ‘Julius’ in a centered world in which he alone invented the zipper.  The 1-intension of ‘water’ is that function which maps a centered world onto whatever in that world is the watery stuff, if anything is, and is undefined otherwise.  Thus, the actual 1-extension of ‘water’ is H2O, but XYZ is the 1-extension of ‘water’ in any centered world in which it’s the watery stuff.  The 1-intension of ‘circular’ is that function which maps a centered world onto the class of things that are circular in that world.  And the 1-intension of ‘dog’ is that function which maps a centered world onto the class of things in that world which belong to the zoological species of the such-and-such looking and behaving creatures in that world.  Thus, the actual 1-extension of ‘dog’ is the class of dogs, i.e., the class of things belonging to the species Canis familiaris, but there are centered worlds in which dogs don’t comprise the 1-extension of ‘dog’, as the such-and-such looking and behaving creatures in that world belong to a zoological species other than Canis familiaris.

 

2.             For every Mentalese expression token e, there is a relation R such that, for any f, f is e’s 1-intension iff R(e, f).  Two important things are true of R.  First, it holds contingently between the things it relates.  That a Mentalese expression token has a particular 1-intension has at the least something to do with the conceptual role of the expression type of which the token is a token.  Second, bearing R to a 1-intension is an individualistic property, in that whether a thinker’s expression bears R to a particular f is determined entirely by the internal state of the thinker, thereby securing that 1-intensions are so-called narrow contents.

 

 

3.             Expressions also have another kind of intension, which determines another kind of extension.  For every Mentalese expression token e there is a function g, which I’ll call its 2-intension, such that, for any x and any possible world w, if g(w) = x, then x is the extension of e in w, when w is considered as counterfactual (I assume this audience knows what that means).  In the event, we can call x the 2-extension of e in w.  There are two kinds of 2-intensions and two correspondingly different ways in which an expression’s 2-intension may be determined.

Some expressions, such as ‘circular’, have the same 2-intension in every centered world.  For such expressions, their 2-intensions are directly determined by their 1-intensions, and there’s a sense in which these 2-intensions are mere copies of the 1-intensions that determine them.  This is because both the 1-extension and the 2-extension is determined by the same condition.  For example, the 1-intension of ‘circular’ maps each centered world onto the class of circular things in that world, and the 2-intension of that word maps every world onto the class of circular things in that world. 

Other expressions—including those that are rigid designators—have different 2-intensions in different centered worlds.  For such an expression, its 2-intension in a given centered world is determined by its 1-extension in that world, and the expression can have different 1-extensions in different centered worlds.  If in these cases the expression’s 1-extension in a centered world is x, then its 2-intension in that centered world is that function which maps every world onto x (or perhaps onto x in every world in which x exists, a qualification I’ll henceforth ignore).  In these cases, the expression’s 1-intension determines a certain uniqueness property for each centered world—a property of the form the property of being the only thing that’s such-and-such—and the 1-extension in a given centered world is whatever in that world has the uniqueness property.  Such a uniqueness property may be a composite property containing an ineliminable occurrence of whatever agent or time (or whatever else) belongs to the “center” of a particular centered world.  The expression’s 2-intension in that centered world will map each possible world onto the thing in the centered world that has the uniqueness property determined by the expression’s 1-intension.  Some examples by way of rough illustration.  The 1-intension of ‘Julius’ determines the 1-extension of the name in a centered world to be whatever in that world has the property of being the unique inventor of the zipper.  Since Whitcomb L. Judson actually has that property, he’s the actual 1-extension of ‘Julius’, and therefore the actual 2-intension is that function which maps every possible world onto Judson.  Thus, ‘Julius’ has Judson as its 2-extension in every possible world.  In a centered world in which David is the 1-extension of ‘Julius’, then the name’s 2-intension in that centered world maps every world onto David.  Let’s call the 1- and 2-intensions of sentences propositions.  Then we see that the 1-proposition expressed by ‘If Julius exists, then Julius invented the zipper’ is the necessary proposition that (so to say) if the inventor of the zipper exists, then he or she invented the zipper, whereas the 2-proposition expressed is the contingent proposition that (so to say) if Judson exists, then Judson invented the zipper.  The 1-intension of ‘water’ determines the 1-extension of the word in a centered world to be whatever in that world has the property of being the unique watery stuff in that world.  Since H2O actually has that property, it’s the actual 1-extension of ‘water’, and, accordingly, its actual 2-intension is that function which maps every possible world onto H2O.   But in a centered world in which XYZ is the 1-extension of ‘water’, then in that world its 2-intension maps every world onto XYZ.  In this way we see that the 1-proposition expressed by ‘Water is H2O’ is the contingent proposition that (so to say) the watery stuff in our environment is H2O, whereas that sentence’s 2-proposition is the necessary proposition that (so to say) H2O is H2O. 

 

4.             There are very important epistemological and psychological asymmetries between 1- and 2-propositions, and some of the more striking ones are inventoried here and in the points to follow.  The first point has virtually been made already, but it’s worth repeating.  It’s that whereas 1-propositions are always narrow contents, 2 propositions, when they’re not mere copies of 1-intensions, are typically wide contents.  That all 1-propositions are narrow contents follows from the claim, made in glossing what it is for a function to be an expression’s 1-intension, that having a certain 1-intension always supervenes on what’s in the head of the person for whom the expression has that 1-intension.  That many 2-propositions are wide contents is an obvious consequence of the way in which 2-intensions are determined.

 

5.             If p is a 1-proposition, then it’s in principle possible to know a priori whether p is metaphysically necessary, impossible, or possible.  It follows that all metaphysically necessary 1-propositions are a priori, in the sense that anyone who can think such a proposition is capable in principle of knowing it a priori.  This is not true of metaphysically necessary 2-propositions that aren’t mere copies of the 1-propositions which determine them.  The points at play here give a neat way of explicating the necessary a posteriori/contingent a priori distinction as applied to Mentalese sentences.  Necessary a posteriori sentences have necessary 2-propositions but contingent 1-propositions, while contingent a priori sentences have necessary 1-propositions but contingent 2-propositions.  Thus, ‘Water is H2O’ is necessary a posteriori; it has a contingent 1-proposition that determines a necessary 2-proposition.  One counts as knowing both propositions by virtue of knowing the contingent proposition that H2O is both self-identical and the watery stuff in one’s environment.  And ‘If Julius exists, then Julius invented the zipper’ is contingent a priori by virtue of having a necessary 1-proposition and a contingent 2-proposition.  

 

6.             The point made in (5) has a very important corollary.  It’s that it must always be possible to think a 1-proposition without thinking of it under another proposition that determines it in the way that 1-propositions determine 2-propositions.  If this weren’t so, then it couldn’t be claimed that the modal status of a 1-proposition is always knowable a priori.  David has another way of getting at the same point.  He says that a term is “semantically neutral” if it’s not vulnerable to Twin Earth thought experiments, which is to say that its 2-intension is a mere copy of its 1-intension.  ‘Circular’ is a semantically neutral term.  Then we can restate the present corollary by saying that for every 1-proposition there is an idealized language in which that proposition is the 1-proposition of a sentence that contains only semantically neutral terms or indexicals that refer to things, such as oneself and the current moment, that are needed to specify the centers of relevant centered worlds.

 

 

7.             Rationality relations are determined by 1-propositions, in the following sense. You can’t rationally have in your belief box at the same time two sentences where p is the 1-proposition of one of the sentences and not-p is the 1-proposition of the other.  But you can rationally have in your belief box at the same time two sentences where p is the 2-proposition of one of the sentences and not-p is the 2-proposition of the other.  For example, suppose the 1-proposition of one sentence is the proposition that the F is H and the 1-proposition of another sentence is that the G isn’t H.  Suppose further that, unbeknown to you, the F = the G, thereby securing that the two sentences have the same 2-proposition.  In this case, nothing has yet been ascribed to you that impugns your rationality.

 

8.             The rationality claim just made implies an account of modes of presentation, since it offers an explanation of rational belief and disbelief, as when someone rationally believes both that George Eliot adored groundhogs and that Mary Ann Evans didn’t adore woodchucks, notwithstanding that George Eliot was Mary Ann Evans and the property of being a groundhog is the property of being a woodchuck.  Given the way this works, we may take 1-intensions to be modes of presentation of the 1-extensions they determine, where the notion of a mode of presentation is playing the Fregean role of being that in terms of which one explicates, e.g., how it is that one can rationally believe that George Eliot adored groundhogs and that Mary Ann Evans didn’t adore woodchucks, notwithstanding the identities involved.

 

 

9.             Finally, Theory C deploys its two-dimensional apparatus to explicate epistemic necessity/possibility by way of reducing it to metaphysical necessity/possibility.  To see this, consider the following two sentences in my belief box, both of which are true:

(a)         Necessarily, Ed McBain is Evan Hunter.

(b)        Possibly, Ed McBain isn’t Evan Hunter

But how can it be that both sentences are true?  As we know, it can be if the necessity operator in (a) registers metaphysical necessity while the possibility operator in (b) registers epistemic possibility.  So much is a datum for any theory to explain; the question is how to explain it.  What Theory C claims is that (a) and (b) are true because

Ed McBain is Evan Hunter

has a contingent 1-proposition and necessary 2-proposition.  This would be the case if, say, the 1-proposition expressed were (so to say) that the author of the 87th Precinct novels is the author of Blackboard Jungle.  In general, the idea goes, an epistemically possible proposition is either a metaphysically possible 1-proposition or a 2-proposition determined by a metaphysically possible 1-proposition.  Epistemically necessary propositions are, on this account, metaphysically necessary 1-propositions.  Since being epistemically necessary may on this account be identified with being a priori, we have, as in effect already noted, a mental-content version of what David calls The Core Thesis:  A Mentalese sentence is a priori iff its 1-proposition is metaphysically necessary.

 

Problems for Theory C

 

How adequate is Theory C? 

 

a.              My first point is more of a comment than a criticism.  The comment is that Theory C is in effect a notational variant of a familiar Russellian theory.  This familiar theory starts with the distinction between “knowledge by acquaintance” and “knowledge by description.”  One has knowledge by description of x provided x is uniquely such-and-such and one knows that something is uniquely such-and-such.  One has knowledge by acquaintance of x provided one knows that x is such-and-such, and this knowledge isn’t possessed by virtue of one’s having knowledge by description of x, or by virtue of having knowledge of x under some non-descriptive mode of presentation.  One’s epistemic acquaintance with x isn’t mediated by anything.  This doesn’t preclude one from characterizing a thought’s content by reference to a singular proposition containing an object known only by description.  But that can only be done when the primary content of the thought is a proposition that contains a uniqueness property instantiated by the object in the singular proposition.  In this way, our Russellian can say that the primary content of a thought is a primary proposition constructed from objects and properties with which the thinker is directly acquainted—i.e., things that can be thought of without the aid of a distinct mode of presentation—and that a thought will also derivatively have as part of its content a singular proposition containing an object or property known only by description, provided that the thought’s primary proposition contains a uniqueness property that’s instantiated by the thing known only by description.  Our Russellian can go on from here to mimic Theory C’s claims about rationality, modes of presentation, narrow content, the contingent a priori and the necessary a posteriori, and epistemic and metaphysical modalities.

I mention the equivalence with the familiar Russellian theory not to accuse David of putting old wine in new bottles (still less to suggest that David was in any way motivated by the Russellian theory), but for two other reasons.  First, the familiar problems that plague the Russellian theory also plague Theory C, and seeing the parallel will help us to appreciate this.  And second, I think the Russellian version is actually the preferred version of the Russell-Chalmers theory, for it’s simpler and can be stated literally, without appeal to the metaphor of possible worlds.

 

b.             One familiar problem with the Russellian theory is that it requires modes of presentation to be uniqueness properties (albeit uniqueness properties that may ineliminably contain particulars, such as oneself, one’s current sense data and the present moment, known by acquaintance), and the same is in effect true of Theory C.  For according to Theory C, a 1-intension that determines different 2-intensions in different centered worlds always picks out its 1-extension in a given centered world as whatever in that world has a certain uniqueness property determined by the 1-intension.  For example, the 1-intension of ‘Hesperus’ tells us that the name’s 1-extension in any given centered world is whatever in that world is the such-and-such heavenly body that appears in the evening.  Now, it might seem that Kripke refuted the Russell-Chalmers theory, since it relies on modes of presentation being uniqueness properties, and Kripke showed, with his Gödel and Feynman examples, that the thoughts expressed by sentences containing proper names need involve no such uniqueness properties.  But Kripke’s “refutation” paid insufficient attention to the metalinguistic modes of presentation available to his subjects in the Gödel and Feynman examples.  Would the person who appears to know merely that Feynman was a famous physicist have her beliefs about Feynman if she didn’t know that there was someone who was uniquely such that he was a famous physicist named ‘Richard Feynman’ and to whom those from whom she acquired the name were referring when they used the name? 

Nevertheless, there remain apparent counterexamples to the description-theoretic account of modes of presentation even when metalinguistic modes of presentation are taken into account.  Five-year-old Johnny has a fuzzy memory image of a certain woman who visited his family, and he believes that she was funny.  Johnny has no uniqueness property available in terms of the woman’s appearance, behavior or visit.  If his belief about her is under a uniqueness property, it must evidently be the woman’s being the person uniquely causally responsible in such-and-such way for Johnny’s current image.  But it’s doubtful that the five-year-old believes that just one person is causally responsible in such-and-such way for his current memory image.  Come to think of it, it’s doubtful many adults are capable of having such a belief.  Notice that the problem here isn’t that the reference isn’t determined by some uniqueness condition.  We should expect there to be a uniqueness condition that determines reference.  The problem is that the uniqueness condition might not be available as part of a thought’s content.  An analogy may be helpful.  There is evidently a uniqueness condition, however complex and however unknown, whose satisfaction by ‘circular’ in my mouth determines it to mean circularity, but that condition isn’t available to me to do any psychological work as part of the content of any of my thoughts.  In like manner, whatever condition determines the reference of  a name, demonstrative, or any other kind of term may also be unavailable as a thought content.

It may seem that both the Russellian and Chalmers have a quick and easy fix whenever a needed uniqueness property fails to materialize.  The Russellian allows that certain particulars can be objects of acquaintance, such as oneself, the present moment, and one’s current sense data, and centers of Theory C’s centered worlds can be defined with respect to anything at all.  So, when a suitable uniqueness property fails to materialize for a certain kind of referent, why can’t the Russellian simply say that we have knowledge by acquaintance of referents of that kind, and, what comes to the same thing, why can’t David simply add those referents to the centers of his centered worlds?  But such a move can work only if the kind of object in question is one that we can plausibly think about without the aid of a distinct mode of presentation, and that would have to be motivated.  Perhaps objects of the problematic kind need modes of presentation all right, but modes of presentation other than uniqueness properties.  Perhaps they need object-dependent modes of presentation, modes of presentation that can’t exist unless the things of which they are modes of presentation exist and which are individuated partly in terms of those things.  In recent years many theorists have appreciated the need for object-dependent concepts and for object-dependent propositions other than stark singular propositions, but it’s not easy to see how such concepts and propositions can be accommodated by either Theory C or the Russellian theory.

 

c.              Another problem with the Russellian theory should be familiar to those acquainted with the travails of logical atomism: there don’t seem to be enough things at the end of the line with which we’re directly acquainted.  Modes of presentation are supposed not merely to be uniqueness properties, but uniqueness properties that don’t themselves require modes of presentation; when we think about them, we’re supposed to be able to do so directly, the mode in which such a property is presented being direct acquaintance with the property itself, not some distinct thing under which we think of the property.  Yet it’s arguably doubtful that there are enough such properties to serve as modes of presentation of all the objects and properties we’re capable of thinking about.  Even more is this so when we take Burgean Twin Earth cases into account along with the Putnamian cases.  I’d like to be told the uniqueness property constructed wholly from things with which I’m directly acquainted under which I think about Saddam Hussein.

Theory C faces the same sort of problem.  In presenting Theory C, acquaintance was secured for 1-intensions by virtue of its claims about epistemic possibility and the a priori status of metaphysically necessary 1-propositions, but there is also another route to the same point about acquaintance.  We noticed earlier that according to Theory C there are semantically neutral words, such as ‘circular’, whose 2-intensions are mere copies of their 1-intensions.  The properties ascribed by these words—e.g., the property of being circular—are, on the scheme of Theory C, properties with which one is acquainted, in the sense that they can be thought of directly, without the aid of a mode of presentation.  Likewise for properties ascribed by expressions made up of semantically neutral terms together with suitable indexicals whose referents are things in the centers of centered worlds, such as oneself or the present moment.  In the case of words whose 2-intensions aren’t mere copies of their 1-intensions—words like ‘Hesperus’ and ‘water’—the 1-intension picks out a thing as the term’s 1-extension in a given centered world by virtue of the thing’s satisfying a certain uniqueness property in that world.  Now consider the 1-intension of ‘water’.  The reference-fixing uniqueness property that intension determines is something that can itself be thought about (after all, if one knows p, then presumably one’s capable of knowing that one knows p).  Either that property can be expressed (perhaps only in an idealized language) by an expression made up of semantically neutral terms and indexicals that refer to things in the centers of centered worlds (‘acquaintance-apt expressions’, for short), or it cannot.  If so, no problem; we’ve in effect reduced the mode of presentation of water to a uniqueness property with which we’re directly acquainted.  But suppose the property is such that it can’t be expressed by an acquaintance-apt expression, but only by an expression that requires the property to instantiate a distinct uniqueness property.  Now we must ask of the uniqueness property that’s the mode of presentation of the uniqueness property that determines the reference of ‘water’ whether it can be expressed by an acquaintance-apt expression.  If not, we’re off again, the bottom line being that Theory C implies that all 1-extension-determining uniqueness properties reduce in the foregoing way to properties with which one is directly acquainted—i.e., properties expressed by acquaintance-apt expressions—or else we have an unending regress of modes of presentation of modes of presentation of modes of presentation, and so on.  Since such a regress would evidently not be benign, I assume that a proponent of Theory C must hold that at the end of the line we get modes of presentation with which we’re directly acquainted and for which, therefore, we don’t need modes of presentation.  In this way, once again, Theory C encounters the familiar problem for the Russellian theory.

 

d.             According to Theory C, 1-intensions are “narrow contents,” determined “entirely by the internal state of the thinker.”  There are two problems with this.  First, whether or not this is so depends entirely on the nature of that relation that must hold between a Mentalese term and a function in order for the latter to be the former’s 1-intension, but Theory C gives no account of that relation which earns the narrow-content claim.  How, then, can the theory confidently assert it?  Who’s to say but that a lot of externalistic stuff won’t find its way into the explication of this content-determining relation?  Consider such a basic word as ‘red’.  Is it really plausible that the word in my head could express the property of being red unless I was directly or indirectly in causal contact with red things?  Second, since virtually any concept can be subject to a Burgean Twin Earth scenario, it’s hard to see how there could possibly be enough “narrow contents” to sustain the claim that every 1-intension is a narrow content.  Appealing to deference here is to no avail, since the same problem arises for 1-intensions expressed by the terms of the deferrer.

 

e.              There are problems with Theory C’s explication of epistemic modalities.  One problem may be nothing more than a verbal quibble, so we should get it out of the way. We have a notion of epistemic possibility; it’s the notion of possibility at work when someone truly reports that it’s possible that not-p, when in fact p is metaphysically necessary.  Theory C’s account of epistemic possibility doesn’t capture our ordinary notion, because propositions that it would count as metaphysically impossible 1-propositions can clearly be epistemically possible according to the ordinary notion.  Asked whether what is in fact not a mathematical theorem is a theorem, I might reply that it’s possible that it is and be speaking truly.  Moreover, our ordinary notion of epistemic possibility doesn’t seem to be a deep phenomenon requiring much of an analysis.  As we in fact use the epistemic sense of ‘possible’, p is epistemically possible for x just in case x can rule out neither p’s being true nor p’s being false, and in this sense a metaphysically impossible 1-proposition may well be epistemically possible for x.  Nor does it seem much of a mystery how we should explain how some of us dolts might take some mathematical truths to be open questions.  But I don’t think that’s a deep objection to what Theory C is up to, since David is deliberately using the label ‘epistemic possibility’ in some quasi-technical sense, and what’s really at issue is Theory C’s substantial claim, untouched by the verbal objection just put aside, that metaphysically necessary 1-propositions are all a priori, that is, capable in principle of being known a priori.  I’m inclined to doubt this for at least the following reason.  It seems plausible that intentional facts are metaphysically necessitated by non-intentional facts, but I doubt that any such entailment can be established a priori.  In any possible world that is non-intentionally indistinguishable from the actual world, your word ‘dog’ expresses the property of being a dog, but it would be rash to suppose that there is some a priori account of meaning which, together with a complete non-intentional description of the world, would enable one to deduce that your word ‘dog’ expresses the property of being a dog.  Perhaps David would claim that he’s given us an explanatory framework whose plausibility made it plausible that a specific conditional with a non-intentional antecedent and an intentional consequent could be known a priori, but I should think that the implausibility of the a priori knowledge claim has not yet been outweighed by the plausibility garnered by Theory C.  Indeed, it seems to me that other accounts of content have more plausibility.

 

f.               Finally, suppose the foregoing objections are all valid.  Does that show epistemic contents should be modeled by things other than centered metaphysically possible worlds, or do those objections give reason to doubt that there are any epistemic contents?  I should think the latter.