[[This paper appears in my anthology Philosophy of
Mind: Classical and Contemporary Readings (Oxford University
Press, 2002), pp. 608-633. It is a heavily revised version of a paper
first written in 1994 and revised in 1995. Sections 1, 7, 8, and 10
are similar to the old version, but the other sections are quite
different. Because the old version has been widely cited, I have made
it available (in its 1995 version) at http://consc.net/papers/content95.html.
What follows is an application of a framework I have developed in
other papers, to issues about the contents of thought. The discussion here often passes over details that are
explored in more depth in those papers. The gentlest introduction is
Sense and Intension" (Chalmers 2002b); Chalmers forthcoming a and
b give full details. The framework presented here has much in common
with existing ideas in the philosophy of mind and language, especially
Kaplan's (1989) and Stalnaker's (1978) two-dimensional analyses of
language, Lewis's (1979) analysis of the contents of thought, and
various proposals that have been made about the nature of narrow
content. For some connections between these ideas, see section 9 and
"The Foundations of Two-Dimensional
What follows is an application of a framework I have developed in other papers, to issues about the contents of thought. The discussion here often passes over details that are explored in more depth in those papers. The gentlest introduction is in "On Sense and Intension" (Chalmers 2002b); Chalmers forthcoming a and b give full details. The framework presented here has much in common with existing ideas in the philosophy of mind and language, especially Kaplan's (1989) and Stalnaker's (1978) two-dimensional analyses of language, Lewis's (1979) analysis of the contents of thought, and various proposals that have been made about the nature of narrow content. For some connections between these ideas, see section 9 and "The Foundations of Two-Dimensional Semantics".]]
Here are six puzzles about the contents of thought.[*]
*[[For background material on the six puzzles, see: (1) Putnam 1975, Burge 1979; (2) Frege 1892; (3) Kripke 1979; (4) Perry 1979; (5) Schiffer 1990; (6) Kripke 1980.]]
(1) Is content in the head? Oscar believes that water is wet. His twin on Twin Earth, which is just like Earth except that H2O is replaced by the superficially identical XYZ, does not. His twin's thoughts concern not water but twin water: Oscar believes that water is wet, but Twin Oscar believes that twin water is wet. This suggests that what a subject believes is not wholly determined by the internal state of the believer. Nevertheless, the cognitive similarities between Oscar and his twin are striking. Is there some wholly internal aspect of content that they share?
(2) Frege's puzzle. In thinking that Hesperus is Hesperus, I think about the same objects as in thinking that Hesperus is Phosphorus. But the first thought is trivial and the second is not. How can this difference in cognitive significance be reflected in a theory of content?
(3) Kripke's puzzle. In France, Pierre is told (in French) that London is pretty, and he believes it. Later, he arrives in London and thinks it is ugly, never suspecting that "Londres" and "London" name the same city. It seems that Pierre simultaneously believes that London is pretty and that London is not pretty. Pierre is highly rational, however, and would never believe a contradiction. What is going on?
(4) The problem of the essential indexical. When I believe that I am in danger, I will take evasive action. This belief seems to be essentially indexical, or self-directed; if I merely believe that x is in danger, where (unbeknownst to me) I am x, I might do something else entirely. How can we square this indexical aspect with an account of the contents of thought?
(5) The mode-of-presentation problem. If Jimmy says "Lois believes that Superman can fly", he speaks truly. If he says "Lois believes that Clark Kent can fly", he speaks falsely. But on many accounts, the proposition that Clark Kent can fly is the same as the proposition that Superman can fly. If so, it seems that to believe that Clark Kent can fly, it is not enough to believe the corresponding proposition; one must believe it under an appropriate mode of presentation. What is a mode of presentation, and how can these be integrated into an account of belief ascription?
(6) The contingent a priori. Say it is stipulated that one meter is the length of a certain stick in Paris. Then it seems that one knows a priori that the stick is one meter long, if it exists. But it seems contingent that the stick is one meter long, as it might have been that the stick was longer or shorter than one meter. How can one have a priori knowledge of a contingent truth?
These puzzles are not unrelated. All of them suggest incompleteness in a familiar view of thought content, on which the content of a thought is tied to the external objects one is thinking about. In particular, most of them raise questions about how well such an account of thought content reflects rational or cognitive aspects of thought. Because of the dependence on external factors, this sort of content often seems to be dissociated from the rational relationships between thoughts (as witnessed by puzzles 2, 3, and 6), and from their role in guiding cognition and action (as witnessed by puzzles 1 and 4).
To resolve these and other puzzles, many have postulated a separate dimension of content — so-called "narrow content" — that depends only on the internal state of a thinker, and that is more closely tied to cognition and action.[*] The road from intuition to theory has been a difficult one, however, and no account of narrow content has yet gained widespread acceptance. It is widely held that because narrow content is internal, it lacks the sort of relation to the external world that is required to qualify as content. For example, many have thought that narrow content is not the sort of thing that can be true or false, as the Twin Earth cases show us that truth-conditions are not determined internally.
*[[Arguments for narrow content can be found in Dennett 1981, Fodor 1987, Lewis 1994, Loar 1988, Segal 2000, and White 1982.]]
I think that these problems are illusory, and that there is a robust and natural notion of narrow content such that narrow content has truth-conditions of its own. This can be seen by developing the idea that content has two dimensions. On the account I will give here the content of a thought can be decomposed into two components: its epistemic and subjunctive content. Subjunctive content is a familiar external variety of content. Epistemic content has the following properties: (1) it is determined by the internal state of a cognitive system; (2) it is itself a sort of truth-conditional content; (3) it reflects the rational relations between thoughts. The first property ensures that epistemic content is a variety of narrow content. The second ensures that it is a truly semantic variety of content. The third ensures that it is central to the dynamics of cognition and action. Because of these three properties, epistemic content can help to resolve many problems in the philosophy of mind and language.
In what follows, a thought is a token propositional attitude that aims to represent the world: for example, a belief, an expectation, or a hypothesis. Thoughts have truth-values (truth, falsity, and possibly others), and are often expressed in language by sentences. Thoughts are often (perhaps always) composed of concepts. Concepts are mental tokens that are often expressed in language by terms. Where thoughts have truth-values, concepts have extensions: for example, individuals, classes, or properties. The truth-value of a thought typically depends on the extension of the concepts involved: for example, the truth value of my thought Hesperus is Phosphorus depends on whether the object that is the extension of Hesperus is the same as the object that is the extension of Phosphorus.
It is a familiar idea that concepts and thoughts can be associated with an intension: a function from possible worlds to extensions or truth-values. The intension of a concept maps a possible world to the concept's extension in that world: in a given world, the intension of my concept renate picks out the class of creatures with a kidney in that world. The intension of a thought maps a world to the thought's truth-value in that world: in a given world, the intension of my thought all renates are cordates will be true if every creature with a kidney in that world also has a heart. In effect, a concept's intension captures the way that its extension depends on the nature of the world, and a thought's intension captures the way that its truth-value depends on the nature of the world.
It is a somewhat less familiar idea that a concept or thought can be associated with two intensions. First, there is an epistemic intension, picking out a thought or concept's extension across the space of epistemic possibilities. This intension captures the epistemic dependence of extension or truth-value on the way the actual world turns out. Second, there is a subjunctive intension, picking out a thought or concept's extension across the space of subjunctive or counterfactual possibilities. This intension captures the subjunctive dependence of extension or truth-value on counterfactual states of the world, given that the character of the actual world is already fixed. On the two-dimensional picture I will develop, a thought's epistemic intension is narrow content, while a thought's subjunctive intension is often wide content.
To give a quick illustration: for my concept water, the epistemic intension picks out H2O in our world (the Earth world), and XYZ in a Twin Earth world. This reflects the fact that if I accept that my actual world is like the Twin Earth world (i.e., if I accept that the liquid in the oceans is and always has been XYZ), I should accept that water is XYZ. drinkable liquid). By contrast, the subjunctive intension of my concept water, picks out H2O in both the Earth world and the Twin Earth world. This reflects the fact that given that water is H2O in the actual world, the counterfactual Twin Earth world is best described as one in which water is still H2O, and in which XYZ is merely watery stuff. As a rough approximation, we can say that the epistemic intension of water picks out a substance with certain superficial characteristics (e.g. a clear drinkable liquid) in any given world, while the subjunctive intension of water picks out H2O in all worlds. A similar pattern exists for many other concepts; the basis of the pattern is discussed in what follows.
Let us say that a thought is epistemically necessary when it can be justified a priori: that is, when there is a possible reasoning process that conclusively justifies the thought with justification independent of experience. A thought is epistemically possible (in a broad sense, related to but distinct from the usual philosophical sense) when it cannot be ruled out by a priori reasoning: that is, when its negation is not epistemically necessary. Intuitively, this holds when the thought does not involve an a priori contradiction. On this understanding, my thought water is H2O is epistemically possible, as is my thought water is XYZ. No amount of a priori reasoning can lead to the justified rejection of either of these thoughts. For all I can know a priori, the world might be such as to make either of these thoughts true.[*]
*[[A few fine details here: (1) We can say that a thought is conclusively justified with it has the sort of justification that carries a guarantee of truth: the sort of justification carried by deduction or analysis, for example, as opposed to the non-conclusive justification carried by induction or abduction. The restriction to conclusive a priori justification is required to exclude, e.g., false mathematical thoughts that might be a priori justified by induction from true thoughts. (2) A priori indeterminate thoughts, if there are such, are neither epistemically possible nor epistemically necessary; the same goes for their negations. To handle such cases, one should say that a thought is epistemically possible when its determinate negation is not epistemically necessary. (3) Certain theoretical view (e.g. Salmon 1986) hold that it is knowable a priori that water is H2O, on the grounds that 'water is H2O' expresses roughly the same proposition as 'H2O is H2O', where this proposition is knowable a priori. Even on these controversial views, however, it is clear that the token thought water is H2O is not epistemically necessary as defined above: that is, there is no reasoning process that can justify this thought a priori.]]
When a thought is epistemically possible, it is natural to hold that there are various specific scenarios compatible with the thought. A scenario can be thought of as a maximally specific epistemic possibility: one with all the details filled in. For example, the mere thought that water is XYZ is compatible with many epistemically possible hypotheses about the precise distribution of XYZ in my environment and about everything else that is going on in the world. Each of these maximally specific epistemically possible hypotheses corresponds to a scenario.
To flesh out this intuition further, it seems reasonable to say that some scenarios (those involving XYZ in certain distributions in my environment) verify my thought that water is XYZ: if I accept that the scenario obtains, I should accept that water is XYZ. Other scenarios (e.g. those involving H2O in my environment) falsify my thought that water is XYZ: if I accept that the scenario obtains, I should deny that water is XYZ. Equivalently, we can say that my thought that water is XYZ endorses scenarios in the first class, and excludes scenarios in the second class.
In effect, the space of scenarios constitutes my epistemic space: the space of specific epistemic possibilities that are open to me a priori. If I had no empirical beliefs, all of epistemic space would be open to me. As I acquire empirical beliefs, my epistemic space is narrowed down. Any given belief will typically divide epistemic space into those epistemic possibilities that it endorses and those that it excludes. The basic idea I will pursue is that the narrow content of a thought is given by the way that the thought divides epistemic space.
Scenarios are much like possible worlds. For now, we can represent scenarios using possible worlds, though I will consider potential differences later. For example, let the H2O-world be a world like ours with H2O in the oceans and lakes, and let the XYZ-world be a specific Twin Earth world with XYZ in the oceans and lakes. Then for all I can know a priori, my world might be qualitatively just like the H2O world, or it might be just like the XYZ-world. So these two worlds each represent highly specific epistemic possibilities for me. We can put this by saying it is epistemically possible (in the broad sense) that the H2O-world is actual, and that it is epistemically possible that the XYZ-world is actual.
For any given scenario, one can in principle consider the hypothesis that the scenario is actual. For any given world W (on the possible-worlds understanding), it is epistemically possible that W is actual: that is, it is epistemically possible that one's own world is qualitatively just like W. When one considers a world as an epistemic possibility in this way, one is considering it as actual: that is, one is thinking of it as a way one's own world may be.[*]
*[[The phrase "consider a world as actual" is due to Davies and Humberstone (1980), developing ideas presented by Evans (1979). The explication given here differs from that given by Davies and Humberstone, who do not talk explicitly about epistemic possibility, but it is in much the same spirit.]]
Any given scenario verifies some thoughts and falsifies others. Here, we can say that a scenario falsifies a thought T when the hypothesis that the scenario is actual is rationally inconsistent with T. A scenario verifies T when the hypothesis that the scenario is actual is rationally inconsistent with the negation of T. For our purposes here, it is natural to say that a thought and a hypothesis are rationally inconsistent when their conjunction is epistemically impossible: that is, when this conjunction can be ruled out by a priori reasoning. On this interpretation, a scenario verifies a thought when acceptance of the hypothesis that the scenario is actual implies the thought: that is, when this acceptance can lead by a priori reasoning to acceptance of the thought.
Take my thought water is H2O. The H2O-world verifies my thought: if I accept that the H2O-world is actual, I must rationally conclude that water is H2O. It would be rationally inconsistent to accept that the H2O-world is actual (i.e., that the liquid surrounding me with a certain appearance and distribution is and always has been H2O, and so on) but deny that water is H2O. By contrast, if I accept that the XYZ-world is actual, I must rationally conclude that water is not H2O. It would be rationally inconsistent to accept that the XYZ-world is actual (i.e. that the liquid surrounding me with a certain appearance and distribution is and always has been XYZ, and so on), and at the same time to accept that water is H2O. So the XYZ-world falsifies my thought water is H2O, and verifies thoughts such as water is not H2O, or water is XYZ.
There is nothing here that contradicts the claim by Kripke and Putnam that water is necessarily H2O. Kripke and Putnam are dealing with what is often called "metaphysical" possibility and necessity, which is usually sharply distinguished from epistemic possibility and necessity. Even if it is not metaphysically possible that water is XYZ, it is epistemically possible that water is XYZ: we cannot rule out the hypothesis that water is XYZ a priori. If Kripke and Putnam are right, then when the XYZ-world is considered as a metaphysical possibility (in effect, considered as a counterfactual world different from ours), it is best described as a world where XYZ is not water. But it is clear that when it is considered as an epistemic possibility (i.e. considered as a way our own world may be), and when verification is defined as above, it verifies the hypothesis that water is XYZ.
The indexical character of some thoughts forces us to refine the possible-worlds understanding of scenarios. If we consider an objective world W as actual, this does not yield a fully determinate epistemic possibility. Take a world W containing both H2O and XYZ, in the oceans and lakes of separate planets. If I consider W as actual, I am not in a position to determine whether water is H2O or XYZ, since I do not know which planet I am on. In effect, a fully determinate hypothesis must include information about my location within a world. To handle this, we can represent a scenario by a centered world: a world marked with an individual and a time at its "center".[*] A centered world corresponds to a world from a perspective, marked with a viewpoint at its center. In the case above, there will be many centered worlds corresponding to W, some of which are centered on individuals on the H2O planet, and some centered on individuals on the XYZ planet. Now, when I consider the hypothesis that a centered world W' is actual, I consider the hypothesis that my world is qualitatively just like W', that I am the individual marked at the center of W', and that the current time is the time marked at the center of W'. Given that sort of information in the case above, I will be in a position to determine which planet I am on, and I will be in a position to determine whether water is H2O or XYZ.
*[[This notion is introduced by Quine (1968), who defines a centered world as a world with a marked space-time point. The definition above is due to Lewis (1979).]]
Given the above, we can naturally define a thought's epistemic intension as a function from scenarios to truth-values. The epistemic intension of a thought T is true at a scenario W when W verifies T, and is false at a scenario W when W falsifies T. As before, W verifies T when it is rationally inconsistent to accept that W is actual and deny T, and W falsifies T when it is rationally inconsistent to accept that W is actual and accept T.
For a more precise definition of epistemic intensions, we would need to be more precise about what it is to consider a scenario as actual. The informal understanding above suffices for many purposes, but a more detailed account can be given as follows. To consider a scenario W as actual is to consider the hypothesis that D is the case, where D is a canonical description of W. When scenarios are understood as centered worlds, a canonical description will conjoin an objective description of the character of W (including its physical and mental character, for example), with an indexical description of the center's location within W.[*] The objective description will be restricted to semantically neutral terms: roughly, terms that are not themselves vulnerable to Twin Earth thought experiments (thus excluding most names, natural kind terms, indexicals, and terms used with semantic deference). The indexical description will allow in addition indexical terms such as 'I' and 'now', to specify the center's location. We can then say that W verifies a thought T when a hypothesis that D is the case implies T.[*] Or equivalently, where S is a linguistic expression of T, W verifies T when a material conditional 'if D, then S' is a priori. These matters are explored in much more depth in Chalmers (forthcoming a).
*[[On a full account, a canonical description of a centered world is
required to be an epistemically complete statement in an idealized
language that uses only semantically neutral terms and indexicals. A
statement D is epistemically complete when D is epistemically possible
and there is no S such that both D∧S and D∧¬S are epistemically
possible. A semantically neutral term is one that behaves in the same
way under epistemic and subjunctive evaluation. The indexicals
allowed include 'I', 'now', and any others required to characterize
the center of the world. This treatment requires that for every
centered world, there exists an epistemically complete description
using only semantically neutral terms and indexicals. This claim can
be supported by noting (i) that there will be an epistemically
complete description for every world (a consequence of the
idealization of the language), and (ii) that semantic non-neutrality
does not in itself add expressive power in characterizing epistemic
possibilities (at most, it affects the description of metaphysical
possibilities). See Chalmers (forthcoming a) for more on this.
Note that there is no requirement that a canonical description be
given in a purely microphysical vocabulary or in a purely phenomenal
vocabulary, or in a combination of the two. I have defended elsewhere
(Chalmers 2002a, Chalmers and Jackson 2001) the claim that a
conjunction ('PQTI') of microphysical, phenomenal, indexical, and
"that's all" truths about the actual world implies all truths about
the actual world. If so, then this sort of description will provide a
canonical description of the actual world (at least if we put the
microphysical description in semantically neutral form), and an
analogous description will plausibly suffice for many other worlds.
But this is a substantive claim, and is not built into the definition
of a canonical description. If PQTI is not epistemically complete, a
canonical description will need to include further information about a
world. Of course there may be many canonical descriptions for a
single world, but all of these will imply each other, so all will
yield the same epistemic intensions.
Note that there is no requirement that a canonical description be given in a purely microphysical vocabulary or in a purely phenomenal vocabulary, or in a combination of the two. I have defended elsewhere (Chalmers 2002a, Chalmers and Jackson 2001) the claim that a conjunction ('PQTI') of microphysical, phenomenal, indexical, and "that's all" truths about the actual world implies all truths about the actual world. If so, then this sort of description will provide a canonical description of the actual world (at least if we put the microphysical description in semantically neutral form), and an analogous description will plausibly suffice for many other worlds. But this is a substantive claim, and is not built into the definition of a canonical description. If PQTI is not epistemically complete, a canonical description will need to include further information about a world. Of course there may be many canonical descriptions for a single world, but all of these will imply each other, so all will yield the same epistemic intensions.]]
In the case of water is H2O, the thought's epistemic intension will be true at the H2O-scenario (a scenario centered on Oscar surrounded by H2O, say), and will be false at the XYZ-scenario (a scenario centered on Twin Oscar, surrounded by XYZ). On a first approximation, one might suggest that the thought's epistemic intension will be true in a scenario when the dominant clear, drinkable liquid around the center of that scenario has a certain molecular structure. This seems to capture roughly what it takes for us to judge that water is H2O in the actual world, depending on how things turn out. But this sort of approximation is no replacement for the real intension. The intension itself is best evaluated by considering specific scenarios and determining the consequences for the truth-values of our thoughts.
The existence of epistemic intensions is grounded in the fact that given sufficient information about the actual world, we are in a position to know whether our thoughts are true. For example, given sufficient information about the appearance, behavior, composition, and distribution of objects and substances in my environment, I am in a position to determine whether water is H2O. Even if the information had turned out differently, I would still have been in a position to determine whether water is H2O. So given enough relevant information about a scenario, I am in a position to determine whether, if that information is correct in my own world, water is H2O. The same goes for all sorts of other thoughts. It may be that in some cases (involving vague concepts such as tall, for example), a complete specification of a scenario does not settle a thought as true or false. In that case, we can say that the thought's epistemic intension is indeterminate at that world. But otherwise, the thought's epistemic intension will be true or false at the world.
To help evaluate an epistemic intension at a world, one can use various heuristics. One useful heuristic for evaluating the epistemic intensions of one's own thought T, expressible by a sentence S, is to intuitively evaluate an indicative conditional: 'if W is actual, is it the case that S?' As with other indicative conditionals, one evaluates this conditional epistemically (by the "Ramsey test"): one hypothetically accepts that W is actual, and uses this to reach a rational conclusion about whether or not S is true.[*] If yes, then W verifies T; if not, then W falsifies T. To stress the epistemic character of the conditional, one can also appeal to "turns-out" conditionals such as the following: 'if W turns out to be actual, will it turn out that S'? For example, it seems reasonable to say that if the XYZ-world turns out to be actual, then it will turn out that water is XYZ.
*[[Note that this heuristic invokes the intuitive correctness conditions (or "assertibility conditions") of an indicative conditional, which are given by the Ramsey test, rather than the truth-conditions, whose nature is disputed. We might think of the intension defined by the indicative conditional heuristic as a "Ramsey intension". Ramsey intensions and epistemic intensions are very similar, but it is arguable that they come apart in some cases (see Yablo 2002, Chalmers 2002a). Ramsey intensions have the advantage that they do not invoke the notion of apriority, and so are available even to Quineans and others who reject this notion; this makes it clear that even a Quinean can accept a version of the general framework here. Still, my own view is that the definition in terms of apriority is more fundamental.]]
Some thoughts have a very straightforward epistemic intension. For example, it is plausible that the epistemic intension of my thought I am a philosopher will be true at precisely those scenarios where the individual at the center is a philosopher. The identity of the individual at the center does not matter: it might be David Chalmers, and it might be Immanuel Kant. After all, my knowledge that I am not Immanuel Kant is a posteriori, so a scenario centered on Kant represents an epistemic possibility for me in the broad sense. It seems clear that if I accept that the Kant scenario is my actual scenario (i.e., that I am Kant philosophizing at the center of that scenario), then I should conclude that I am a philosopher.
As for a thought such as Hesperus is Phosphorus: it is plausible that this thought will be verified by roughly those scenarios where the bright object visible in a certain position in the evening sky around the individual at the center is identical to the bright object visible in the morning sky around the individual at the center. Again, this captures roughly what it takes for us to judge that Hesperus is Phosphorus in the actual world, given sufficient empirical information.
With a mathematical thought such as 2+2 = 4, or pi is irrational, the thought's epistemic intension will be true in all worlds. This reflects the fact that these thoughts can be justified a priori, so that the negations of these thoughts will not be rationally consistent with any a posteriori hypothesis (the conjunction will itself be epistemically impossible). The same goes even for complex mathematical thoughts whose truth we are not in a position to know ourselves. The notion of epistemic possibility and necessity involves a rational idealization away from our contingent cognitive limitations. By definition, if it is even possible for a thought to be conclusively justified a priori, then the thought is epistemically necessary. If so, the thought has a necessary epistemic intension.
(This idealization also helps to deal with a naural worry: that in practice human thinkers are too limited to entertain the complete hypothesis that a scenario W is actual. A thought T is verified by W if a possible thought that W is actual implies T; that is, if it is possible for a material conditional thought if D is the case, then T to be conclusively justified a priori, where D is a canonical description of W. This possibility may idealize away from the cognitive limitations of the thinker. Intuitively, we can think of the thinker as engaging in ideal reasoning.[*] In practice one can often avoid this sort of idealization by appealing to a relevant partial description D of W, where the thinker can entertain the hypothesis that D is the case, and where this hypothesis rationally settles the status of T.)
*[[As an alternative aid to the imagination, we might suppose that the thinker is assessing T with the aid of a supercomputer that stores the relevant information about W and carries out necessary a priori calculations. See Chalmers and Jackson 2001, section 4, for more on this idea.]]
It is tempting to say that conversely, when a thought has a necessary epistemic intension, it is knowable a priori. On the centered-worlds understanding of scenarios, this claim is equivalent to the thesis that when a thought is epistemically possible, it is verified by some centered world. I think this thesis is correct, and have argued for it elsewhere, but it is nontrivial, at least when the worlds in question are understood as metaphysically possible worlds. Some philosophical views entail counterexamples to this claim. For example, on some theist views it is metaphysically necessary that a god exists, but the existence of a god cannot be known a priori. If so, then a god exists is not a priori, but its epistemic intension will be true in all metaphysically possible worlds. In effect, there are not enough possible worlds on this view to represent all epistemic possibilities. A similar result follows from some views on which the laws of nature in our world are the laws of all worlds: there will be no worlds with different laws to represent the epistemic possibility of different laws. The same goes for some materialist views on which zombies are epistemically possible but not metaphysically possible: on some such views, no possible world will correspond to the zombie epistemic possibility.
All of these views are controversial, and I have argued elsewhere (Chalmers 2002a) that they rest on an incorrect conception of metaphysical necessity. Sometimes these views are presented as drawing support from Kripkean a posteriori necessities such as 'Hesperus is Phosphorus' and 'water is H2O', but the Kripkean examples are all compatible with the thesis that every epistemic possibility is verified by a centered possible world. So these views require a posteriori necessities of a sort much stronger than those discussed by Kripke, and there is reason to doubt that "strong necessities" of this sort exist,
Still, one who accepts any of these views will deny the thesis that every epistemically possible thought is verified by a centered world. Such a theorist could nevertheless preserve the thesis that every epistemically possible thought is verified by a scenario, by understanding scenarios as something other than centered metaphysically possible worlds. For example, one can define a space of maximal epistemic possibilities in purely epistemic terms (perhaps using a construction from epistemically consistent thoughts or sentences), and one can then make the case that every epistemically possible thought is verified by a scenario of this sort. (For example, the theist view above entails that even if there is no godless world, there is still a godless scenario.) I have taken this purely epistemic approach to scenarios elsewhere (Chalmers forthcoming b), as it is more neutral and arguably more philosophically fundamental. For reasons of simplicity and familiarity, I will usually stay with the centered-world approach to scenarios in this paper, but it should be kept in mind that the alternative understanding is available.[*]
*[[A terminological point: I generally use "epistemic intension" for the intension defined over the space of maximal epistemic possibilities (whether or not these coincide with centered metaphysically possible worlds), while I use "primary intension" for the same sort of intension defined over the space of centered metaphysically possible worlds (whether or not these exhaust the maximal epistemic possibilities). In applications of the two-dimensional framework to metaphysical questions, the restriction to metaphysically possible worlds is often crucial, so in these contexts the issues are cast in terms of primary intensions. A central issue here is the thesis that a thought is a priori iff it has a necessary primary intension; this thesis makes a plausible but substantive claim about metaphysical possibility (one that is false if a god exists is metaphysically necessary but not a priori, for example). In applications of the framework to epistemic and semantic questions (such as the current discussion), the restriction to metaphysically possible worlds is less crucial, so in these contexts the issues are cast in terms of epistemic intensions. Here the claim that a thought is a priori iff it has a necessary epistemic intension can be seen as closer to definitional (if a god exists is necessary but not a priori, we can construe scenarios epistemically so that there will be a godless scenario). Of course if the plausible but substantive thesis above is correct, then every maximal epistemic possibility corresponds to a centered world, and primary and epistemic intensions will come to much the same thing.]]
One important note: It is tempting to suppose that the epistemic intension of a thought T can be evaluated in a scenario W by asking: what is the truth-value of T, as thought in W? But this is not so. On the present proposal, T's epistemic intension can be evaluated in scenarios containing no copy of T; even when a copy of T is present, it usually plays no special role. For example, my thought I am a philosopher is verified by a scenario regardless of whether I think I am a philosopher there. To take a more extreme example, the epistemic intension of my thought someone is thinking is false in a scenario that contains no thoughts. In these cases, all that matters is the epistemic relation between the hypothesis that W is actual and the thought T. Nothing here requires that T be present in W. One might define a different intension (a "contextual intension"; see section 9 and Chalmers forthcoming a) using the heuristic above, but such an intension behaves in a quite different way, and will not have the same sort of epistemic properties as an epistemic intension. This will be important later.
One can define epistemic intensions for concepts as well as for thoughts. A concept's epistemic intension picks out its extension in a scenario considered as actual. A precise definition involves some tricky details (see Chalmers forthcoming a), so here I will simply illustrate the idea intuitively. Let us take a singular concept C expressible by a term B. To evaluate C's epistemic intension in a scenario W, one considers the hypothesis that W is actual, and uses B to ask: 'what is B?' (Here B is used rather than mentioned.) One can appeal to the indicative conditional 'if W is actual, what is B?'. Alternatively, one can appeal to the rational consistency of judgments of the form C is such-and-such with the hypothesis that W is actual.
For example, in the XYZ-scenario, the epistemic intension of my concept water picks out XYZ. As before, I can say: if the XYZ-scenario is actual, then water is XYZ. In the H2O-scenario, on the other hand, the epistemic intension of my concept water picks out H2O. More generally, one might say as a first approximation that in a given scenario, the epistemic intension of my concept water picks out the dominant clear, drinkable liquid found in the oceans and lakes around the individual at the center. As before, however, this is just an approximation, and the true intension corresponds to the results of considering and evaluating arbitrary scenarios as epistemic possibilities.
One can do something like this for an arbitrary concept. Even for a seemingly nondescriptive concept, such as Gödel, it will still be the case that given full information about a scenario and given the hypothesis that this information obtains in the actual world, one will be in a position to make a rational judgment about the identity of Gödel under that hypothesis. This mirrors the fact that given relevant information about the actual world, one is in a position to determine the identity of Gödel, and more generally is in a position to determine the extension of arbitrary concepts. This rational dependence of judgments about extension on information about the character of the actual world can be encapsulated in an epistemic intension.[*]
*[[For this reason, the current framework can be seen as neutral between "causal" theories of reference (on which reference is determined by a causal chain) and "descriptive" theories of reference (on which reference is determined by a description). Even a causal theorist should allow that relevant information about the actual world dictates rational judgments about our concept's extension. This methodology underlies Kripke's own arguments for the causal theory: in effect, he considers epistemic possibilities that we could discover to obtain (e.g., that a man called 'Gödel' stole the proof of the incompleteness of arithmetic from a man called 'Schmidt'), and reaches judgments about a term's extension on that basis (here, we judge that 'Gödel' will pick out the stealer, not the prover). So even on the causal theory, a term will plausibly have an epistemic intension: it is just that this epistemic intension may have a causal element. For example, for the epistemic intension of my concept Gödel to pick out a given individual in a scenario, it may be required that that individual stand in the right sort of causal relation to the subject at the center of the scenario. See Chalmers 2002b for more here.]]
These epistemic intensions are often difficult to characterize in independent terms, but for some concepts this characterization is straightforward. If we take a quasi-descriptive concept such as Hesperus (where we assume this functions to rigidly pick out the evening star in the actual world), we can say that the epistemic intension of Hesperus picks out the evening star around the center of an arbitrary scenario. Or if Julius functions to rigidly pick out the inventor of the zip, the epistemic intension of Julius will pick out the inventor of the zip in a given scenario.
The epistemic intension for an indexical concept is also very simple. The epistemic intension of my concept I picks out the individual at the center of a scenario. The epistemic intension of now picks out the time at the center. The epistemic intension of here picks out the location of the individual at the center, at the time at the center. The epistemic intension of today picks out (roughly) the day that includes the time at the center. And so on.
When a thought is composed from concepts, its truth-value typically depends on the concepts' extensions. In such a case, the thought's epistemic intension will equally be determined by the concepts' epistemic intensions. For example, the epistemic intension of A is B will be true at a world when the epistemic intensions of A and B pick out the same individual there. One will find a similar compositionality of epistemic intensions wherever one finds compositionality of extensions.
In contemporary philosophy, epistemic intensions are less familiar than what I will call subjunctive intensions. To evaluate a thought's subjunctive intension, one evaluates it in a world considered as counterfactual. To consider a world as counterfactual, one considers it as a subjunctive possibility: as a way things might have been, given that the character of the actual world is already fixed. In our world as it actually is, the liquid in the oceans and lakes is H2O. Nevertheless, the liquid in the oceans and lakes might have been XYZ. So we can say that the XYZ-world might have obtained, and that the XYZ-world represents a subjunctive possibility.[*]
*[["Subjunctive" because this sort of possibility is grounded in the semantically subjunctive notion of what might have been the case (Kripke is explicit about this), and because the evaluation of such possibilities reflects the use of subjunctive conditionals. See Chalmers 2002b here.]]
The subjunctive intension of a thought T in a world W picks out the thought's truth-value in W when W is considered as counterfactual. Here, we grant that the character of the actual world is already fixed and ask what would have been the case if W had obtained. If T is expressible by a sentence S, we can evaluate T's subjunctive intension at W by using S to ask: 'if W had obtained, would it have been the case that S?' If yes, then T's subjunctive intension is true at W; if no, then T's subjunctive intension is false at W. When T's subjunctive intension is true at W, we can say that W satisfies T.
For example, if the XYZ-world had obtained — that is, if the liquid in the oceans and lakes had been XYZ — then (if Kripke and Putnam are correct) XYZ would not have been water.[*] XYZ would merely have been watery stuff, and water would still have been H2O. If so, then the XYZ-world satisfies my thought water is H2O, and the subjunctive intension of my thought is true at the XYZ-world. More generally, if Kripke and Putnam are correct, then the subjunctive intension of my thought water is H2O is true at all possible worlds.
*[[I think that it is not obvious that Kripke and Putnam are correct about this, and a case can be made that it might have been that water was XYZ. For the purposes of this discussion, however, I will go along with the common view that Kripke's and Putnam's intuitions are correct. I also think that even if Kripke and Putnam are right about language, it is not obvious that this extends to thought. But again, for the purposes of this discussion, I will go along with the common view that the modal properties of a term such as 'water' mirror modal properties of the underlying concept water that the term expresses.]]
It is clear that subjunctive intensions can behave quite differently from epistemic intensions. We have seen that the XYZ-world verifies water is not H2O, but it satisfies water is H2O. This difference is rooted in the difference between epistemic and subjunctive possibility, and the corresponding difference between considering a world as actual and as counterfactual. This is mirrored in the different behavior of indicative and subjunctive conditionals: it seems reasonable to say indicatively that if the liquid in the oceans and lakes is XYZ, then water is XYZ; but if Kripke and Putnam are right, it is not reasonable to say that if the liquid in the oceans and lakes had been XYZ, then water would have been XYZ. In considering a world as counterfactual, empirical facts about the actual world make a difference to how we describe it. In considering a world as actual, they do not.
Something similar goes for an indexical thought such as I am David Chalmers. If Kripke is right, it could not have been that I was not David Chalmers. If so, then I am David Chalmers is true in any world considered as counterfactual (or at least in any world where I exist). Note that there is no special need for a center in the world here: once we know all the objective facts about a counterfactual state of affairs, we know all that we need to know, even to settle indexical claims. So subjunctive possibilities can be represented by ordinary uncentered worlds, and subjunctive intensions are defined over uncentered worlds.
We can associate subjunctive intensions with concepts in a similar way. A concept's subjunctive intension picks out its extension in a world considered as counterfactual. For a concept C expressible by a term B, we can use B to ask: 'if W had been actual, what would B have been?' For example, in the case of water, we can say that if the XYZ-world had been actual, then water would still have been H2O. So the subjunctive intension of water picks out H2O at the XYZ-world, and plausibly picks out H2O in all possible worlds.
For many concepts, the concept's subjunctive intension picks out its actual extension in all possible worlds. This applies in particular to rigid concepts: those expressible by rigid designators, such as names or indexicals, picking out the same object in all worlds. For example, Kripke argues that 'Hesperus' is a rigid designator: if Hesperus is actually Venus, then it could not have been that Hesperus was other than Venus. If so, then the subjunctive intension of Hesperus picks out Venus in all possible worlds. Similarly, given that 'I' is a rigid designator, the subjunctive intension of my concept I picks out David Chalmers in all possible worlds.
For a purely descriptive concept such as circular or the inventor of the zip, by contrast, the subjunctive intension is plausibly very similar to the epistemic intension. For example, both the epistemic and subjunctive intensions of the inventor of the zip plausibly pick out whoever invented the zip in a given world. Note the difference with Julius, which has the same epistemic intension but whose subjunctive intension picks out the actual inventor in all worlds. The difference reflects the intuition that if (for example) Ned Kelly had invented the zip, he would have been the inventor of the zip, but he would not have been Julius. (Compare: if Ned Kelly actually invented the zip, then he is Julius.) Some concepts behave in an intermediate manner: for example, the subjunctive intension of the discoverer of water does not pick out the actual extension in all worlds, but it is nevertheless quite different from the epistemic intension, due to the presence of the rigid concept water as a constituent.
The subjunctive intension of a concept or thought always depends in some way on the concept's epistemic intension and the actual world. For a purely descriptive concept, the subjunctive intension may simply be a copy of the epistemic intension, across uncentered worlds. For a rigid concept, the subjunctive intension will correspond to the value of the epistemic intension at the actual world, projected across all possible worlds. In other cases, the dependence may be somewhat more complex, but it will still exist.
We can encapsulate this dependence by associating concepts and thoughts with a two-dimensional intension. This intension maps an ordered pair (V, W) consisting of a scenario and a world to an extension or a truth-value in W. When a thought T is evaluated at (V, W), it returns the truth-value of T in the counterfactual world W, under the assumption that V is actual. (If a sentence S expresses T, we can use this heuristic: 'if V is actual, then if W had obtained, would it have been the case that S'?) Like an epistemic intension, a two-dimensional intension can plausibly be evaluated without relying on empirical knowledge, since all the empirical knowledge one needs is given in the first parameter V. To evaluate a thought's subjunctive intension at W, one evaluates its two-dimensional intension at (A, W), where A is the actual scenario. If we understand scenarios as centered worlds, the value of a thought's epistemic intension at a scenario W will coincide with the value of its two-dimensional intension at (W, W'), where W' is an uncentered version of W.[*] This two-dimensional intension is useful for certain purposes, but most of the time we need only appeal to a thought's epistemic and subjunctive intensions.
*[[In this way the epistemic intension can be seen as equivalent to the "diagonal" of the two-dimensional intension, in a manner reminiscent of the "diagonal proposition" of Stalnaker 1978. But see section 9 for reasons why epistemic intensions are not fundamentally diagonal intensions.]]
Within this framework, we can analyze the Kripkean "necessary a posteriori". Let us say that a sentence S is subjunctively possible when it is possible in the familiar Kripkean sense: that is, when a modal sentence such as 'it might have been the case that S' is true. A thought is subjunctively possible when it is expressible by a subjunctively possible sentence. Subjunctive necessity is defined correspondingly. Then it is easy to see that when a thought is subjunctively necessary, its subjunctive intension is true in all worlds, and vice versa. Cases of the Kripkean "necessary a posteriori" (e.g., water is H2O) arise when a thought has a necessary subjunctive intension (the thought is true in all worlds considered as counterfactual) but a contingent epistemic intension (the thought is false in some world considered as actual). Cases of the Kripkean "contingent a priori" (e.g. Julius invented the zip) arise when a thought has a contingent subjunctive intension but a necessary epistemic intension.
There should be no question of whether the epistemic or the subjunctive intension is the intension associated with a given concept. The full story can only be given two-dimensionally. One or the other may be more useful for various specific purposes. In matters of linguistic content across a community, the subjunctive intension often plays a central role: different users of a name or natural kind term often have quite different associated epistemic intensions while sharing the same subjunctive intension. For questions about the rational properties of thought and its role in governing action, however, we will see that the epistemic intension is central.
Let us call a thought or concept's epistemic intension its epistemic content, and a thought or concept's subjunctive intension its subjunctive content. Let us say that when a thought or concept's content depends only on the intrinsic state of the thinker (that is, when every possible intrinsic duplicate of the thinker has a corresponding thought or concept with the same content), the content is narrow. Let us say that when content does not depend only on a thinker's intrinsic state (that is, when an intrinsic duplicate could have a corresponding thought or concept with different content), the content is wide. One can make the case that epistemic content is narrow, while subjunctive content is often wide.
It is clear that subjunctive content is often wide. For example, Oscar (on Earth) and Twin Oscar (on Twin Earth) are more or less intrinsic duplicates (abstracting away from differences due to the presence of H2O and XYZ in their bodies), and have corresponding concepts that they express by saying 'water'. But the subjunctive intension of Oscar's concept water picks out H2O in all worlds, while the subjunctive intension of Twin Oscar's concept water picks out XYZ in all worlds. Something similar applies to most rigid concepts, including Hesperus and even I. Here, a subjunctive intension depends on a concept's extension, which usually depends on a subject's environment, so two intrinsic duplicates can have different subjunctive intensions. In other cases, subjunctive content will not depend on the environment. For example, purely descriptive concepts such as circular and the inventor of the zip will plausibly have subjunctive intensions that are shared between duplicates. But in cases where a concept or thought's subjunctive intension depends not just on its epistemic intension but on the way the actual world turns out, we can expect that subjunctive content will be wide content.
This environment-dependence does not extend to epistemic content. A concept's epistemic content is usually quite independent of its actual extension, and of the way the actual world turns out more generally. An epistemic intension encapsulates the way in which our rational judgments about extension and truth-value depend on arbitrary empirical information; so the intension can be evaluated without knowing which epistemic possibility is actual. The factors that make subjunctive content wide content appear to be irrelevant to epistemic content.
This can be illustrated by looking at familiar cases. Take Oscar's and Twin Oscar's respective thoughts T1 and T2, expressed by saying 'there is water in my pool.' Let W1 be the Earth scenario centered on Oscar, with H2O in the oceans and lakes and H2O in Oscar's pool. Let W2 be the Twin Earth scenario centered on Twin Oscar, with XYZ in the oceans and lakes and XYZ in Twin Oscar's pool. Then clearly, W1 verifies T1 and W2 verifies T2. But also, W2 verifies T1: if Oscar hypothetically accepts that W2 is actual, he must rationally accept T1. Equally, W1 verifies T2: if Twin Oscar hypothetically accepts that W1 is actual, he should rationally accept T2. So the epistemic intensions of T1 and T2 are on a par with respect to these worlds.
Something similar applies to other scenarios. Let W3 be a Twin Earth scenario centered on Twin Oscar with XYZ in the oceans and lakes, but an isolated amount of H2O in Twin Oscar's pool. Then W3 falsifies both T1 and T2. If Oscar accepts that W3 is actual, he should reject T1; if Twin Oscar accepts that W3 is actual, he should reject T2. The same goes for any other world: if W verifies T1, it will also verify T2, and vice versa. The same goes equally for any intrinsic duplicate of Oscar. We can even imagine Vat Oscar, who is a brain in a vat receiving artificial stimulation. Vat Oscar can entertain the hypothesis that his environment is just like W1 (or W2 or W3) and can reach rational conclusions on that basis, and the conclusions that he reaches will mirror those of Oscar and Twin Oscar. So Vat Oscar has a thought with the same epistemic intension as Oscar's.[*] The same holds for intrinsic duplicates in general; so the epistemic content of Oscar's thought is narrow.
*[[Thus even a brain in a vat might have thoughts with epistemic content. This can be used to address Putnam's (1981) anti-skeptical argument that if he were a brain in a vat, he could not think I am a brain in a vat. A brain in a vat could think a thought with the appropriate epistemic content, if not the appropriate subjunctive content. It could also think a thought such as I am in a skeptical scenario, which has more or less identical epistemic and subjunctive content. The epistemic contents of such a thought seems sufficient to express a significant skeptical possibility, true only in worlds in which the individual at the center lacks the usual sort of epistemic contact with the surrounding world.]]
The same goes for other thoughts and concepts. For example, even though I may have a twin whose concept expressed by 'Hesperus' has a different extension and subjunctive intension, his concept nevertheless has the same epistemic intension as mine, picking out roughly the evening star near the center of any scenario. Similarly, although the I concepts of my twins will have an extension and subjunctive intension that differs from mine, they will have the same epistemic intension, picking out the individual at the center of any scenario.
One can even apply this analysis to the cases used by Burge (1979) to argue for the social nature of mental content. Bert has a belief that he expresses by saying 'arthritis sometimes occurs in the thighs.' In fact, arthritis is a disease of the joints and cannot occur in the thigh, so it seems that Bert has a false belief about arthritis. Twin Bert, an intrinsic duplicate of Bert, also has a belief that he expresses by saying 'arthritis sometimes occurs in the thighs'. But Twin Bert lives in a community in which the word 'arthritis' is used for a different disease, one that affects the muscles as well as the joints: we might call it 'twarthritis'. It seems that Twin Bert has a true belief about twarthritis. Where Bert believes (falsely) that he has arthritis in his thigh, Twin Bert does not: Twin Bert believes (truly) that he has twarthritis in his thigh. Burge concludes that in this sort of case, belief content is not in the head.
Here, the crucial factor is that Bert uses the term 'arthritis' with semantic deference, intending (at least tacitly) to use the word for the same phenomenon for which others in the community use it. We might say that this term expresses a deferential concept for Bert: one whose extension depends on the way the corresponding term is used in a subject's linguistic community. It is clear that for deferential concepts, extension can depend on a subject's environment, as can subjunctive intension. The subjunctive intension of Bert's concept arthritis picks out arthritis in all worlds, while the subjunctive intension of Twin Bert's concept picks out twarthritis in all worlds.
Let T1 and T2 be the thoughts that Bert and Twin Bert express by saying 'arthritis sometimes occurs in the thighs.' Let W1 be Bert's own centered world, with a surrounding community that uses the term `arthritis' to refer to a disease of the joints. Let W2 be Twin Bert's centered world, with a surrounding community that uses `arthritis' to refer to a disease that can occur in the thigh. Then clearly W1 falsifies T1 and W2 verifies T2. At the same time, W2 verifies T1: if Bert accepts that W2 is actual — that is, if he accepts that his linguistic community uses 'arthritis' for a disease that can occur in the thighs — then (since his concept is deferential) he should rationally accept that arthritis can occur in the thighs, and so should accept T1. Similarly, W1 falsifies T2: if Twin Bert accepts that W1 is actual — that is, if he accepts that his community uses 'arthritis' only for a disease of the joints — then he should reject his thought T2. So the epistemic intension of T1 is false at W1 and true at W2, and exactly the same is true for T2.
Something similar applies to any other scenarios that Bert and Twin Bert evaluate. In general, the epistemic intension of their arthritis concepts in those scenarios will pick out the extension of the term 'arthritis' as used in the linguistic community around the center of those scenarios. (In worlds where the term is not used, the epistemic intension will arguably be empty or indeterminate.) And the same goes for any intrinsic duplicate of Bert. Any such duplicate can entertain the hypothesis that a given scenario W is actual, and will rationally reach conclusions similar to Bert's.
One can apply the same reasoning to Putnam's case of 'elm' and 'beech', in which a subject can use the terms with different referents despite users having no substantive knowledge to differentiate the two. In this case, the terms are being used deferentially: the epistemic intension of the subject's concept elm picks out roughly whatever is called 'elm' around the center of a scenario, and the epistemic intension of her concept beech picks out roughly whatever is called 'beech' around the center of a scenario. Here again, the epistemic intension is independent of the environment. So we can see that semantic deference and "the division of linguistic labor" is quite compatible with thoughts and concepts having internally determined epistemic content.
Putnam suggests that terms such as 'water' and 'elm' show that if a concept's intension is internally determined, it cannot determine the concept's extension. The current analysis shows that this is only half-true. The epistemic intension of a concept determines its extension, and the epistemic intension is internally determined. Of course, the epistemic intension is a centered intension, taking a centered world as argument, at least on the possible-worlds understanding of scenarios. So Putnam's claim still holds for uncentered intensions. But any intension requires facts about the actual world to determine extension, and it is most natural to regard the actual environment of a thinker as centered, so an internally determined centered intension is very useful here.
Why is epistemic content narrow? On the surface, this is because a thought's epistemic content is rationally prior to any knowledge of a subject's environment: it captures the way a thought's truth-value depends on the character of the environment, and so is independent of the environment itself. More deeply, it may be because epistemic content is defined in terms of the rational properties of thoughts, where the relevant rational properties are internally determined. For example, if one subject has a thought that is justifiable a priori, a corresponding thought in any intrinsic duplicate of that subject will also be justifiable a priori; if so, a thought's epistemic necessity is determined by the internal state of the thinker. This observation can be combined with the observation that when one subject entertains the hypothesis that a scenario W is actual, a duplicate of that subject must also be entertaining the hypothesis that W is actual. This second observation can be grounded in the fact that these hypotheses involve semantically neutral descriptions of scenarios, so there is no possibility of a "Twin Earth" difference between thinkers here. Putting these two observations together, it follows that if the hypothesis that W is actual epistemically necessitates a thought in one subject, it will also epistemically necessitate the corresponding thought in any duplicate subject. So epistemic content is narrow.
(Of course, the epistemic content of a thought will almost always depend causally on the external world, but it will not depend constitutively on the external environment. Whenever the external environment affects the epistemic contents of our thoughts, it will do so by affecting the internal state of the thinker.)
As promised, this sort of narrow content is truth-conditional. The epistemic content of a thought delivers conditions that one's actual centered world must satisfy in order for one's thought to be true. We might think of these as a thought's epistemic truth-conditions, as opposed to a thought's subjunctive truth-conditions, which govern truth across counterfactual worlds. Of course these truth-conditions can come apart at a given world: at the XYZ-world, the epistemic truth-conditions of water is XYZ are satisfied, but the subjunctive truth-conditions are not. This is to be expected, given the different functions of epistemic and subjunctive evaluation. One might worry that because of this, a thought could turn out to be both true and false in the actual world, but this is impossible. When evaluated at the actual world, epistemic intensions and subjunctive intensions always give the same results.
In recent times, the "content" of a thought has usually been identified with something like its subjunctive content;[*] but the epistemic content seems to be an equally good candidate. As before, there is no need to decide which is the content. That being said, there are a number of ways in which the epistemic content of a thought is responsible for the explanatory work that we would expect a notion of content to do.
*[[Alternatively, content is often identified with a structured proposition composed from either subjunctive intensions of the concepts involved, or from the extensions of the concepts involved (when the concepts are rigid). This sort of structured content is more fine-grained than a subjunctive intension, but it has the same truth-conditions, and depends on the environment in a similar way. What I say below about subjunctive intensions applies equally to structured propositions. Likewise, what I say about epistemic intensions can easily be adapted to a view on which the contents of thoughts are structures composed from epistemic intensions of concepts.]]
First, epistemic content reflects the rational relations between thoughts. If one thought implies another thought a priori, the epistemic intension associated with the first entails the epistemic intension associated with the second (that is, in all scenarios in which the first intension is true, the second intension is also true). If I know that it is hot where I am now, I know that it is hot here, and vice versa; this is reflected in the fact that the epistemic contents of the two thoughts are the same. The subjunctive contents of the thoughts are very different, however. The subjunctive intension of the first thought is true at a world if it is hot where DJC is at time t in that world; the subjunctive intension of the second thought is true at a world when it is hot in place p in that world (where t and p are the time and place of the actual thoughts). The epistemic contents of the thoughts reflect their rational relationship, but the subjunctive contents do not.
It is straightforward to see why this is so. If one thought entails another a priori, then any scenario that verifies the first will verify the second. Conversely, it is plausible that 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. (As before, if scenarios are identified with centered worlds and if strong necessities exist, this converse claim will be false. For example, the epistemic intension of a god exists might be entailed by any intension without the thought being a priori. Again, on such a view the claim can be preserved by moving to the purely epistemic understanding of scenarios.) This is not so for subjunctive intensions: entailments between these may turn on facts about the external world that are not accessible to the thinker.
This can be applied straightforwardly to explain the informativeness of a thought such as Hesperus is Phosphorus. Although its subjunctive intension is equivalent to that of the trivial Hesperus is Hesperus, its epistemic intension is quite distinct, so it is not cognitively trivial. In effect, epistemic intension here plays the role of Fregean sense. Again, it is epistemic intensions that reflect the rational properties of thoughts.
We can also invoke epistemic content in the case of Kripke's Pierre, who paradoxically seems to believe that London is pretty and that London is not pretty, without any breakdown in rationality. Pierre's concepts Londres and London have quite different epistemic intensions: in a given scenario, the first picks out (roughly) the famous city called 'Londres' that the individual at the center has heard about, whereas the second picks out (very roughly) the grimy city in which that individual has been living. The subjunctive intensions are identical, picking out London in every world. So Pierre's two beliefs Londres is pretty and London is not pretty have contradictory subjunctive intensions, but their epistemic intensions are quite compatible. Rational relations are determined by epistemic content, so contradictory subjunctive intensions support no charge of irrationality.
Intuitively, Pierre's two beliefs are rationally compatible because there are specific ways the actual world could be that are consistent with both. There is a scenario in which 'Londres' names a faraway, beautiful city (maybe it is in India), and in which the individual at the center inhabits an entirely distinct ugly city called 'London'. For all Pierre knows and believes, such a scenario could be actual: this scenario verifies both of Pierre's beliefs. As long as there is such a scenario, satisfying the epistemic intensions of all of Pierre's thoughts, the epistemic contents of these thoughts will be compatible, and Pierre's rationality will not be in danger.
This brings out the relation between this account and Dennett's (1981) suggestion that the narrow content of a thought is reflected in the notional world of the thinker. We can take the notional world to be a scenario (really a class of scenarios) that verifies all of a subject's beliefs, or at least as many as possible.[*] Pierre's notional world is a world in which there is a beautiful faraway city called 'Londres', and a grimy city close at hand called 'London'. If Pierre really lived in his notional world, he would be right about everything and rarely surprised.
*[[Dennett suggests that the relevant worlds are "the environment (or class of environments) to which the organism as currently constituted is best fitted." This class may be quite different from the class of worlds that verify all of a subject's beliefs: subjects are sometimes better fitted to worlds that falsify their beliefs (when they are pessimistic or altruistic, for example); they often have beliefs about distant matters that are irrelevant to fitness; and their fitness often turns on matters about which they have no beliefs. See also the criticisms in Stalnaker 1989 and White 1991, and White's more refined account. Dennett's and White's suggestions might be seen as a first attempt at giving a naturalistic reduction of something in the vicinity of epistemic content. Such a reduction is likely to be a major project in its own right.]]
On similar grounds, one can make the case that epistemic content reflects the cognitive relations between thoughts. Here there is an important qualification: epistemic content as I have defined it does not distinguish the various cognitive relations that might hold between thoughts that are deductively equivalent. From the point of view of epistemic content, a complex mathematical proof is as trivial as modus ponens; the fine-grained cognitive dynamics of deduction lies beyond the reach of epistemic content as I have defined it here. I think a more fine-grained variety of epistemic content can handle these cases better (see Chalmers forthcoming b). I will set these issues aside here, as subjunctive content does not handle them any better, and they are largely independent of the issues at play in this paper.
A qualified thesis would be the following: insofar as epistemic content or subjunctive content reflect the cognitive relations between thoughts, the contribution of epistemic content screens off the contribution of subjunctive content. That is, in cases where two thoughts are cognitively related, then (1) in related cases where the epistemic content of the thoughts is held constant but the subjunctive content is varied, the cognitive relations are preserved (except insofar as cognitive relations can be affected by varying factors independent of both epistemic and subjunctive content, as in the deductive case); and (2) in related cases in which the subjunctive content is preserved but epistemic content is not, the cognitive relations are damaged. One can make this case straightforwardly by examining cases; the details parallel those of the discussion of the explanation of behavior, below, so I will not duplicate them here.
A third advantage of epistemic content is its suitability for a role in the explanation of behavior. It is often noted that subjunctive content seems slightly out of synchrony with what one would expect of an explanatory psychological state. To use an example of Kaplan's (1989), if you are watching me and my pants catch fire, our respective beliefs that my pants are on fire now will have the same subjunctive content (true in all worlds in which DJC's pants are on fire at time t), but will lead to very different actions (I might jump into a river, while you just sit there). The difference between our actions does not seem to be something that a characterization in terms of subjunctive content alone can explain. In a similar way, belief states can produce very similar behavior for apparently systematic reasons, even when the beliefs have very different subjunctive content: witness the behavior that my twin and I produce when we think about twin water and water respectively, or the similarity between the actions of two people who think I am hungry. It seems that a whole dimension of the explanation of behavior is hard for subjunctive content to explain.
These explanations can be easily handled in terms of epistemic content. If you and I think I am hungry, the epistemic contents of our thoughts are the same, and that similarity is reflected in the similarity of our actions. When you and I both believe that my pants are on fire, on the other hand, our epistemic contents are very different, and our actions differ correspondingly. Note that this provides a straightforward solution to Perry's problem of the essential indexical: it is epistemic content, not subjunctive content, that governs action, and epistemic content, consisting in a centered intension, is a sort of indexical content.[*]
*[[Perry (1979) considers the possibility that centered ("relativized") propositions might provide a solution, but dismisses it on the grounds that believing that such a proposition P is "true for me" does not distinguish me from third parties who also believe that P is true for me, but act differently. The trouble is that Perry's locution "true for me" introduces an unnecessary extensional element. What distinguishes me from the third parties is rather that I believe P simpliciter, or better, that my belief has P as its epistemic content.]]
Epistemic content also accounts for the similarity of action between twin cases; this similarity reflects the fact that my beliefs about water and my twin's beliefs about twin water have the same epistemic content. But we need not move to the realm of science fiction to see the point. Two thoughts can share epistemic content even when two thinkers are quite different, as our thoughts I am hungry show. Even in these cases, similarities in epistemic content will lead to similarities in action, other things being equal. Suppose I think that Superman is across the road, and I want to have Superman's autograph: then other things being equal, I will cross the road.[*] If you have thoughts with similar epistemic content to mine, then you will do the same. If your thoughts share only subjunctive content with mine, while having different epistemic content — perhaps you think that Clark Kent is across the road, but want Superman's autograph — then your corresponding behavior may be quite different.
*[[To simplify the discussion, I make the happy assumption that Superman is actual and is identical to Clark Kent.]]
In general, whenever the content of a thought is causally relevant to behavior, its contribution is screened off by that of epistemic content in the following sense. If an alternative thought had the same epistemic content but different subjunctive content, the resulting behavior would have been physically indiscernible (except insofar it might be affected by changing factors independent of both sorts of content); whereas if it had the same subjunctive content but different epistemic content, the resulting behavior might have been quite different.
To see the latter point, we need only examine cases like those above. The thoughts I am hungry and The guy over there is hungry (unknowingly looking in a mirror) will lead to very different behavior, even though their subjunctive content is the same. When Lois Lane is trying to cut Clark Kent's hair, her observation "Clark's hair breaks the scissors" will prompt a reaction very different from that provoked by a corresponding thought concerning Superman. If I hear that Cary Grant is starring in a movie, I might be more likely to watch than if I hear that the movie stars Archie Leach. In all these cases, different reactions are provoked by a difference in the epistemic content of a thought. In general, whenever the epistemic content of a thought is varied, different consequences can be expected, even if subjunctive content is preserved throughout.[*] Given that epistemic content is central to cognitive relations and that cognition governs action, this is just what we would expect.
*[[Of course, thoughts like Cary Grant is in the movie and Archie Leach is in the movie might lead to the same actions despite their different epistemic content, if I know that Cary Grant is Archie Leach. But even here, there exist circumstances under which the thoughts might play a different role: imagine someone telling me that Cary Grant is not Archie Leach after all. In general, whenever two thoughts have different epistemic content, there are at least hypothetical circumstances under which the action-governing roles of the thoughts will differ.]]
By contrast, if the subjunctive content of a thought is varied but epistemic content is kept constant, behavior stays indistinguishable throughout. Perhaps, unbeknownst to me, Cary Grant is an elaborate hoax, a co-operative construction by avant-garde animators and the mass media. In such a case, my thought about Cary Grant will have no nontrivial subjunctive content, but as long as it has the same epistemic content, my behavior will be indistinguishable from the case in which he is real. Or perhaps Cary Grant is really Ludwig Wittgenstein in disguise: if so, the thought has a very different subjunctive content, but the same behavior results. Similarly, when my twin and I think I need some more water for this pot, the subjunctive contents of our thoughts differ, but we both go to the sink.
We can make a similar point within a single system. Take Evans' example of 'Julius', which functions to rigidly designate whoever invented the zip. Then the epistemic intensions of my concepts Julius and the inventor of the zip will be the same, but the subjunctive intensions will be very different. Despite the difference in subjunctive intensions, however, it is clear that any thoughts of the form Julius is such-and-such and The inventor of the zip is such-and-such will play very much the same role in directing cognition and action. The rigidification and consequent difference in subjunctive intension is largely irrelevant. One exception: the two concepts may behave differently in subjunctive thought, as when one judges that Julius might not have been the inventor of the zip, but not that Julius might not have been Julius. But even here the difference is accounted for by a difference in the internally determined two-dimensional intension, rather than by a difference in subjunctive content per se.
Some might object that there are cases in which we individuate behavior extrinsically — for example, Oscar drinks water while Twin Oscar drinks twin water — so there is a dimension of behavior that escapes epistemic content. But even in this sort of case, subjunctive content does not usually help. Even Twin Oscar, with his different subjunctive content, would drink water if he was in Oscar's present environment. What is relevant to behavior here is not subjunctive content but current environment, as we can see by an extension of the varying-factors strategy; and I certainly do not wish to deny that current environment is relevant in the explanation of behavior.
The only cases in which there is a direct tie between subjunctive content and behavior are cases in which behavior is individuated by an intentional object: for example, Oscar searches for a glass of water, whereas Twin Oscar searches for a glass of twin water. This connection holds across all environments, as behavior only counts as water-searching if it is caused by water-thoughts. But for the same reason, this is a very weak sort of relevance for subjunctive content: as Fodor (1991) notes, in these cases the subjunctive contents of thoughts are not causally relevant to action, but instead are conceptually relevant, in effect determining the category the action falls under.[*] Subjunctive content gives us very little purchase in the explanation of action here, as we will only know that some behavior is water-searching if we already know that water-thoughts lie behind it. In a causal (as opposed to a conceptual) explanation of the action, epistemic content will still play the central role.
*[[See Fodor 1991 for a detailed argument along these lines. I note also that one can individuate this sort behavior intentionally but still narrowly if one individuates by epistemic content.]]
Why is epistemic content primary? To answer this question, it is useful to think of my belief contents as constituting a model of my world, a kind of map by which I steer. This is a model of the world as I find it, a centered world with me at the center, and my beliefs are constraints on that world. Beliefs constitute a model by constraining epistemic space: the space of epistemic possibilities that were open to me a priori. One belief rules out one group of epistemic possibilities as a candidate for the world where I am, another belief rules out another group, until only a limited class of worlds is left. I operate under the assumption that my world is one of those worlds, and if I am lucky I will not be too surprised.
My world-model is ultimately a notional world: a set of epistemic possibilities, such that none of these would overly surprise me if they turned out to be actual. The constraints on these possibilities are those of epistemic content. Any further constraints imposed by subjunctive content are not useful to me. The subjunctive content of my belief that the liquid in thermometers is mercury endorses only those worlds in which thermometers contain the element with atomic number x, but this constraint is so distant that if it turned out that the liquid has atomic number y, I would not be in the least surprised. In an important sense, this constraint is not reflected in my world-model at all. Insofar as my world-model is useful to me in guiding cognition and action, the constraints on it are entirely those of epistemic content.
In making a case for the primacy of epistemic content, I have not appealed to any a priori methodological principles such as the dictum that what governs behavior is in the head. The case for epistemic content has been made directly, independently of questions about physical realization. Indeed, nothing I have said implies that facts about a thinker's environment are irrelevant to the explanation of behavior. Facts about the proximal environment will clearly play an important role insofar as they affect a thinker;[*] facts about the current environment are crucial to explaining the success or failure of various actions; and facts about environmental history will at least be central to a causal explanation of a thinker's current cognitive state. All that follows from the present framework is that the environment is not relevant to the explanation of behavior in virtue of its role in constituting subjunctive content. The kind of content that governs behavior is purely epistemic.
*[[It may even be that in certain cases, epistemic content can itself be constituted by an organism's proximal environment, in cases where the proximal environment is regarded as part of the cognitive system: if a subject's notebook is taken to be part of a subject's memory, for example (see Clark and Chalmers 1998). Here, epistemic content remains internal to a cognitive system; it is just that the skin is not a God-given boundary of a cognitive system. This is another way in which the issue between epistemic and subjunctive content runs deeper than the issue between internalism and externalism.]]
All this raises a puzzle about the role of belief ascriptions in psychological explanation. If what has gone before is correct, the kind of content that governs cognition and action is epistemic content, which is narrow. But at the same time, there is strong evidence that the kind of content attributed by belief ascriptions is often wide. Does this mean that the common-sense framework of explanation of behavior in terms of belief ascription should be discarded? Alternatively, is the success of the common-sense framework evidence that something in these arguments has gone badly wrong?
Neither conclusion is justified. The present framework shows how it can at once be true that (1) belief ascriptions ascribe wide content, (2) narrow content governs action, and (3) belief ascriptions explain action. In short: belief ascriptions ascribe a combination of epistemic and subjunctive content. It is in virtue of the subjunctive component that the ascribed content is wide, and it is in virtue of the epistemic component that the ascribed content is explanatory.
A full justification of this answer requires two things. First, we need an analysis of what is attributed in belief ascriptions, so that we can see precisely what sorts of epistemic and subjunctive content are attributed. Second, we need an analysis of the role of belief ascriptions in psychological explanation, so that we can see that even in ordinary practice, it is the epistemic content attributed that carries the explanatory burden. I cannot provide anything like a complete treatment of these matters — the analysis of belief ascriptions deserves entire volumes of its own — but I can provide a preliminary sketch.
It is easy to see that ordinary belief ascriptions ascribe both epistemic and subjunctive content. If I say 'Ralph believes that Clark Kent is muscular', then in order for my utterance to be true, Ralph must have a belief that satisfies two sorts of constraints. First, the belief must have the subjunctive content of the proposition that Clark Kent is muscular (perhaps we can allow a certain amount of variation in the subjunctive content, if for example his concept of muscularity is slightly different from the norm). But that alone is not enough: a belief that Superman is muscular would have the same subjunctive content, but would not make my ascription true. As is often noted (e.g. Schiffer 1990), for the ascription to be true, the belief must involve a concept that refers to its object (Clark Kent) under an appropriate mode of presentation.[*]
*[[Some views (e.g. Salmon 1986) take ascriptions such as 'Lois believes that Clark can fly' to be strictly speaking true, so that modes of presentation are irrelevant to truth. Even if these highly counterintuitive views are accepted, the current account can be viewed an account of the (pragmatic) intuitive correctness conditions of belief ascriptions. Either way, we need an account of these intuitive correctness conditions to explain the function of belief ascriptions in psychological explanation.]]
In the current framework, modes of presentation are naturally seen as epistemic intensions. If Ralph refers to Clark Kent under an epistemic intension that picks out whoever is called 'Clark Kent', or one that picks out whoever is that reporter with glasses at the Daily Planet, or some more complex intension in the vicinity, my belief ascription will be true. If Ralph refers to Clark Kent under an epistemic intension that picks out the guy in the cape, or one that picks out the strongest man in the world, my belief ascription will be false. One might say that for the ascription to be true, Ralph must refer to Clark Kent under a 'Clark Kent'-appropriate epistemic intension. Here, the conditions on a 'Clark Kent'-appropriate epistemic intension are somewhat vague and unclear, and they may well be context-dependent, but it is clear from an examination of cases that they are substantive.
To take another case, if I am right in saying 'Tom believes that he is hungry', then Tom must have a belief with more or less the appropriate subjunctive content, true of all those worlds in which Tom is hungry at time t. But there is also a strong constraint on epistemic content. In particular, Tom must refer to himself via the epistemic intension that picks out the individual at the center in every scenario. If he sees someone in the distance clutching their belly, without realizing that he is in fact looking into a mirror, then a thought that that person is hungry has the right subjunctive content, but on the most natural reading it does not make my ascription true. The ascription will only be true if Tom's belief refers to himself under a self-concept, which requires a very specific sort of epistemic content. One might say that here, Tom must refer to himself under a 'he'-appropriate epistemic intension, where in context the only 'he'-appropriate epistemic intension is the purely indexical intension.
The general principle here is something like the following. A belief ascription 'x believes that S' is true when the ascribee has a belief with the subjunctive intension of S (in the mouth of the ascriber), and with an S-appropriate epistemic intension. Here, the epistemic intension is usually much less strongly constrained than the subjunctive intension. The conditions on S-appropriateness may well be complex and context-dependent; their precise nature is one of the hardest questions in the theory of belief ascriptions. Still, one can make a few generalizations. Much of the time, an epistemic intension that is not too different from the ascriber's will be S-appropriate, and much of the time, an epistemic intension that involves the terms in S itself will be S-appropriate. But this does not yield any sort of general condition. Rather, the appropriateness-conditions are best revealed by careful investigation of judgments of the ascription's truth in specific cases involving various different epistemic intensions.
In effect, this yields truth-conditions on belief ascriptions that parallel those of what Schiffer (1992) calls a "hidden-indexical" theory of belief ascription (although I have remained neutral on the ascriptions' logical form), with epistemic intensions playing the role of modes of presentation.[*] If something like this is correct, then epistemic intensions yield a solution to Schiffer's "mode of presentation" problem.[*] Epistemic intensions are perfectly suited to satisfy what Schiffer (1990, p. 252) calls "Frege's constraint" on modes of presentation: roughly, that a rational person may both believe and disbelieve that y is such-and-such only if the two beliefs involve different modes of presentation of y. If "rationality" is interpreted to involve idealized a priori reasoning, then the satisfaction of this constraint follows from the fact that epistemic intensions reflect a priori connections between thoughts.
*[[See also Crimmins 1991 and Richard 1990. Many of the insights of these and other philosophers on the semantics of belief ascription should be straightforwardly adaptable to the present framework.]]
*[[This sort of possibility is not mentioned in Schiffer's (1990) otherwise thorough survey of potential modes of presentation.]]
We can apply this to the case of Pierre, and the ascriptions 'Pierre believes that London is pretty' and 'Pierre believes that London is not pretty.' To satisfy these ascriptions, Pierre must have beliefs with the specified subjunctive intension, referring to London under a 'London'-appropriate epistemic intension. Pierre's London and Londres concepts have different epistemic intensions, but both intensions are 'London'-appropriate. So by virtue of his belief Londres is pretty, Pierre satisfies the first ascription, and by virtue of his belief London is not pretty he satisfies the second. Before, we explained Pierre's beliefs by noting that his two beliefs have contradictory subjunctive intensions but compatible epistemic intensions, where only the latter is relevant to rationality. Now, we can explain the apparent contradiction in the belief ascriptions by noting that two different epistemic intensions can both be 'London'-appropriate, so the two ascriptions do not in fact ascribe a rational contradiction to Pierre.[*]
*[[So Kripke's "Principle of Non-Contradiction" is false: someone can rationally believe that S and believe that not-S, as long as the beliefs involve different epistemic intensions both of which satisfy the appropriate constraints.]]
We have seen that content decomposes naturally into epistemic and subjunctive content; we now see that belief ascription puts strong constraints on both. Ideally, a full theory of belief ascription will specify the nature of these constraints for any given ascription, telling us the conditions that beliefs' epistemic and subjunctive contents must satisfy in order to make the ascription true. We can think of a belief ascription as marking out a subspace in the space of ordered pairs of epistemic and subjunctive content.
Given that epistemic content governs action, it follows that if belief ascriptions are to causally explain action, it must be in virtue of the epistemic content ascribed; the subjunctive content ascribed is redundant to the explanation. To make this case properly requires examining many specific cases, but the general point can be straightforwardly illustrated. One way to see the primacy of epistemic content is to consider belief ascriptions involving empty names, such as 'Santa Claus'. These ascribe no nontrivial subjunctive content, but ascription of beliefs about Santa Claus seem to function in precisely the same way in the explanation of action as do ascriptions of beliefs about real people. We might explain Karen's agitation on Christmas Eve in terms of her belief that Santa Claus is coming, that he will not fit down the chimney, and so on. Santa's non-existence and the corresponding absence of subjunctive content make little difference to the success of such an explanation. What governs Karen's actions are her notions of Santa Claus; and what governs the success of the explanation is the epistemic content that these ascriptions ascribe to her. This is a typical case of the role of belief ascriptions in explanation: even when non-trivial subjunctive content is ascribed (as when the referent of the name exists), it makes little difference to the patterns of explanation.
In a very wide variety of cases in which content explains action, we can see that the explanation succeeds even if the subjunctive content attributed is ignored. For instance, if we explain my opening the refrigerator in terms of my belief that there is water in the refrigerator and my desire for a glass of water, we never need to invoke the H2O-involving subjunctive content. The explanation gains sufficient purchase from the epistemic content ascribed alone — roughly, the content that there is some of the liquid with the appropriate properties in the refrigerator, and that I want some of that liquid, and so on.[*]
*[[I leave aside here the important question of the epistemic content of desires, and the semantics of desire attributions. On my view, the epistemic content of a desire cannot in general be represented by a simple intension. Rather, it is a sort of two-dimensional intension that can endorse a different set of worlds depending on which scenario is actual. This is clearest in cases such as "I wish I were two inches taller" or "I want to be over there". The moral is that the content of desires is perhaps more deeply two-dimensional than the content of beliefs.]]
It might be objected that there are cases in which the constraints on the epistemic content ascribed by a belief ascription are weak, so that subjunctive content must be doing any explanatory work. I think that ascriptions putting weak constraints on epistemic content are rare, but let us assume they can occur: perhaps an attribution of a belief about Smith constrains the relevant epistemic intension very little.[*] Even so, if we look at explanatory practice, we see that epistemic content is still doing the real work. For example, perhaps we explain why Bev goes to the pub by saying that she wants to see Smith and believes that Smith is at the pub. Leaving aside constraints in the concepts of seeing, the pub, and so on, there is a constraint on epistemic content implicit in the 'Smith' attributions. It is implicit here that the two 'Smith' concepts in Bev's thoughts have the same epistemic intensions. If her belief and her desire had very different epistemic content associated with 'Smith' — perhaps she wants to see Batman and believes that Smith is at the pub, not knowing that Smith is Batman — the inference from those states to her action would fail. So there is a strong joint constraint on epistemic content: despite a lack of constraint on the individual beliefs, Bev is implicitly ascribed the belief that a person she wants to see is at the pub. It is this ascribed belief that is doing the real explanatory work, and this ascription clearly puts a heavy constraint on epistemic content. To make the case that all such examples can be similarly analyzed requires a detailed treatment, but this illustrates the general pattern.
*[[How should one analyze so-called de re belief attributions, of the form 'S believes of x that it is F'? In the current framework, one might adapt the proposals of Kaplan 1967 and Lewis 1979 by holding that such an ascription is true when S has a belief with the appropriate subjunctive intension, true in worlds where A has property P, where A is the referent of 'x' and P of 'F', and when the belief involves a concept that picks out A under a de-re-appropriate epistemic intension. Here, a de-re-appropriate intension is one that entails acquaintance: this requires that in any scenario in which the intension yields an extension, the subject at the center is acquainted (in the contemporary non-Russellian sense) with this extension.]]
It follows that the centrality of narrow factors in the causation of action need not overthrow the role of belief ascriptions in explaining behavior, as some (e.g. Stich 1983) have suggested it should. At most we have shown that belief ascriptions are a somewhat rough-edged tool: they wrap both components of content into a single parcel, bringing the idle subjunctive content into play alongside the epistemic content that does all the work. But this should not surprise us; we cannot expect a folk theory to be maximally efficient.[*]
*[[Why is subjunctive content ascribed at all? I think the reasons are tied to language. First, we ascribe beliefs in the same language we use to describe the world, and when we use world-involving language to ascribe epistemic content, world-involving constraints come along naturally in the package. Second, subjunctive content is important to understanding the success of communication and of collective action. When I tell you that I have a cold, you acquire a thought whose epistemic content is different from mine, but whose subjunctive content is the same. Communication very frequently involves transmission of subjunctive content, and our collective cohesion (if not our individual actions) can often be understood in terms of shared subjunctive content. But both of these points deserve a much more extensive development.]]
In moving from common-sense psychology toward a developed cognitive science, we might expect that the kind of content that is invoked will become more purely epistemic, and that subjunctive content will be relegated to a secondary role or dropped entirely.[*] We might also expect that better tools will be developed to specify the epistemic contents of thoughts than the current rough-and-ready language of belief ascription. This might qualify as a revision of our folk notion of belief, emphasizing and refining the elements of epistemic content that are already present within it. But precisely because those elements are already present and playing a central role in our practices, such a development would fall well short of elimination.
*[[It is arguable that cognitive psychology is already mostly concerned with epistemic content rather than subjunctive content, insofar as it is concerned with content at all. For example, the psychological literature on concepts seems to be largely concerned with how concepts are applied to the actual world, concentrating on something like the epistemic intensions of the concepts involved. See Smith and Medin 1981 and Patterson 1991.]]
The framework outlined here is related to a number of existing proposals. There is a clear structural resemblance to other broadly two-dimensional frameworks, such as proposals by Kaplan (1989) and Stalnaker (1978) for analyzing the content of language, and proposals by White (1982) and Fodor (1987) for analyzing the contents of thought. The idea that this sort of proposal can be used to yield a sort of narrow content has been criticized by Block (1991), Stalnaker (1989; 1990), and others, and extended to an earlier version of the present proposal by Block and Stalnaker (1999). So we need to examine the relationship between these proposals, to see whether the criticisms apply. I think that on examination, the current framework differs in fundamental respects from the others, so that their problems do not arise here.
The relationship can be brought out by contrasting epistemic intensions with contextual intensions. A thought's contextual intension is defined by the heuristic discussed earlier: T is true in a centered world W (with T present at the center) if T is true as thought at the center of W. Likewise, the contextual intension of a concept C will return C's extension in worlds with C at the center. (One can define contextual intensions for sentences and other linguistic expressions similarly.) There are various possible variations here: one might have different requirements for what counts as a token of T in a world, or one might require only a token of T's type (for some relevant type) rather than T itself. But however one does things, the centered worlds here are functioning as contexts in which a thought (or concept) occurs, and a contextual intension encapsulates the context-dependence of a thought's truth-value or a concept's extension.
As we saw before, contextual intensions are quite different from epistemic intensions. An obvious difference: epistemic intensions give no special role to thought tokens within a scenario, and can be evaluated in scenarios without any such tokens at the center. Thus the epistemic intension of I am a philosopher can be true at a scenario regardless of what the being at the center is thinking. A thought such as someone is thinking has an epistemic intension that is plausibly false at some centered worlds (e.g., those without any thoughts), although its contextual intension (on a natural understanding) is true at all centered worlds in which it is defined. A deeper difference: where contextual intensions represent context-dependence, with centered worlds representing contexts of thought, epistemic intensions represent epistemic dependence, with centered worlds representing epistemic possibilities. This is a very different conception, and yields quite different behavior. The frameworks of Kaplan and Stalnaker illustrate this. Kaplan defines the character of a linguistic expression type as a function from a context of utterance to the expression's content (roughly, subjunctive intension) relative to that context. In some ways this resembles the two-dimensional intension discussed above (in effect a function from centered worlds to subjunctive intensions), but the underlying ideas and resulting behavior are quite different. For example, on Kaplan's framework names such as 'Hesperus' and 'Phosphorus' have identical characters, picking out the same content in all contexts. This happens because the referent of a name is essential to that name, so any use of the name in any context will have the same referent. For this reason, Kaplan notes that his framework cannot provide a solution to Frege's puzzle in the case of names, natural kind terms, and the like. But as we have seen, the epistemic intension associated with a subject's use of a name behaves very differently, often picking out different individuals in different centered worlds (whether a name has its referent essentially is irrelevant on a non-contextual understanding), and holds out much more hope of dealing with Frege's puzzle.
Stalnaker defines the diagonal proposition of an expression token as a function from a world containing the token to the truth-value of the proposition that the token expresses in that world, as evaluated in that world. This bears a formal resemblance to an epistemic intension, which can be seen as equivalent to the diagonal of a two-dimensional intension. But again, the underlying ideas and resulting behavior are different. On Stalnaker's framework, the diagonal proposition of an expression is defined at worlds where it has a very different meaning. At a world where 'water is solid' means that snow is white, for example, the diagonal proposition of 'water is solid' will be true if snow is white in that world. This is quite different from an epistemic intension: if my usage is nondeferential, the use of terms such as 'water' in a scenario will be irrelevant to epistemic intensions. Stalnaker (1999) notes that because of this, diagonal propositions are not closely connected to a priori truth. This seems correct, but the problem does not generalize to epistemic intensions, which have a built-in connection to a priori truth.
White (1982) and Fodor (1987) generalize these analyses to the contents of thought. Fodor defines the narrow content of a thought as a function from a context of thought to a thought's (wide) content in that context. White does something similar, although his account is slightly more complex and he requires that a functional duplicate of the original thinker be present in the relevant context. As before, these proposals are based on context-dependence, and give results that differ correspondingly: witness the intensions of I am a philosopher and someone is thinking.[*]
*[[Related proposals for understanding narrow content in broadly contextual terms are given by Brown 1986 and Loar 1988.]]
Block (1991) gives a number of objections to proposals of this sort. White's proposal is subject to a charge of holism: no two different subjects can have thoughts with the same narrow content, unless they are functional duplicates. Further, it seems that the narrow contents of a subject's thoughts all change every time the subject acquires a new belief, or indeed every time that anything happens in the mind of the subject. This problem does not apply to epistemic intensions. There is no problem with quite different thinkers having thoughts with the same epistemic intension: for example, very different people can have I am a philosopher thoughts with the same epistemic intension. Further, epistemic intensions will not usually change with the acquisition of new beliefs. A change in epistemic intension requires a change in a subject's rational pattern of judgments about scenarios considered as actual: a change in belief may change the subject's judgments about which scenarios are actual, but it will not usually change a subject's rational judgment about what will be the case if a given scenario is actual. It may be that epistemic intensions sometimes drift over time, or that corresponding thoughts of different thinkers sometimes have different epistemic intensions, but this falls well short of a general holism.
Block charges Fodor's proposal with underdetermination: it is left unclear how to evaluate the mapping across worlds. The main problem is that of "what is held constant". That is, to know which worlds fall in the domain of the intension, one needs to know just what features of the original thought must be present in the thought token at the center. If only a sort of mental syntax is held constant, the result will be an intension that delivers wildly varying results across worlds: there will be worlds where the mapping for water picks out steel, if a token with that mental syntax has a different meaning. If extension is held constant, then the intension will be trivial: the mapping for water picks out H2O in all worlds. For better results, one might suggest that the token's narrow content be held constant, but that presupposes what we are trying to explain. So it seems very difficult to set things up so that the mapping yields a notion of narrow content that behaves in an appropriate way.
Again, epistemic intensions do not have this problem. There is no issues concerning what to "hold constant" across worlds here, since there is no need for the original token to be present in different worlds. Rather, we simply appeal to the original thought, and to its epistemic relations with the hypothesis that a given world is actual. These epistemic relations are well-defined, being grounded in the idealized rational judgments of the subject. They also do not presuppose any theoretical notion of narrow content; but they can be used to ground such a theoretical notion.
Fodor himself (1987, p. 50) raises the problem that his sort of narrow content is not semantically evaluable (for truth and falsity), and so is not really content; rather, it is just potential content, delivering a content in a context. (He later rejects narrow content for this reason.) Again, epistemic content is immune to this problem. An epistemic intension is a sort of first-order content, placing direct constraints on the world, with truth-conditions of its own. Epistemic intensions can also stand in semantic relations such as entailment, and can be analyzed using semantic frameworks involving possible worlds, which allows for significant explanatory power.
Stalnaker (1990) considers the idea that some version of his diagonal proposition (or "realization conditions") might yield an account of narrow content, and raises three criticisms. First, he suggests that we cannot identity a thought independently of its content, so we cannot ask what the content of a belief would have been had it been a belief on twin earth. Second, he notes that diagonal propositions are defined only in worlds containing the relevant thought token, and cannot easily be extended to worlds without the thought token. Third, he notes that on this proposal narrow content is derivative on wide content (since a diagonal proposition is defined using a two-dimensional matrix which is defined using wide content), so it presupposes rather than explains wide content. In response, it is fairly clear that the first two objections apply only to contextually defined narrow content, and not to epistemically defined narrow content. On the epistemic proposal, we never need to ask what the content of a belief would have been if it had been a belief on twin earth, and narrow content is defined in a straightforward way at worlds that do not contain the relevant thought token.
In discussing his second objection, Stalnaker raises a case that is worth addressing. If Bert uses his semantically deferential concept to think my father has arthritis in his thigh, how can we evaluate this thought in a world in which there is no word 'arthritis' in Bert's language, and in which Bert has no thoughts about his father's health? On the epistemic framework, it is most natural to say that the epistemic intension of Bert's arthritis concept picks out nothing in this world. In effect, the use of a semantically deferential concept presupposes that one lives in a community that uses the relevant term, just as a notion such as The present king of France presupposes that there is a king of France. If I discover that those assumptions do not hold in my actual scenario, it is reasonable to judge that my thoughts involving these concepts lack truth-value. The same goes for alternative scenarios. In Bert's case, the epistemic intension of his thought is indeterminate in the relevant worlds. In effect, Bert's thought partitions the space of scenarios in which the background assumptions are satisfied, and says nothing about those worlds in which the assumptions are false.[*]
*[[Other concepts whose epistemic intensions have a limited domain of determinacy include perceptual demonstratives. When I think something like That is pretty, the referent of my demonstrative is often picked out (very roughly) as the cause of such-and-such experience. In a centered world in which there is no appropriate experience at the center, the epistemic intension may lack truth-value. This raises another subtlety: to capture the content of perceptual demonstratives, one may need to build in a "marked" experience to the center of the class of actual-world candidates, as one builds in a marked individual and time. Building this into the center will sometimes be needed to secure reference to otherwise indistinguishable experiences and their perceptual objects, as with (perhaps) a speckle in a large field, or one of the symmetrical red spots in Austin's (1990) "Two Tubes" puzzle (to which the present framework then provides a solution). In certain cases, centers may also require more than one experience, and perhaps a marked thought ("this very thought"). One might suggest that the contents of a center involve objects of "unmediated" reference: oneself, the present moment, the current thought, and perhaps certain experiences and orientations. This matter is closely connected to Russell's suggestions about direct reference; I hope to explore it in more depth elsewhere (see also Chalmers 2002c).]]
As for Stalnaker's third objection: narrow content may be derivative on wide content on the diagonal understanding, but not on the epistemic understanding. Epistemic content can be defined quite independently of subjunctive content, and our definition of epistemic intensions makes no appeal to subjunctive evaluation. For this reason, an epistemic intension is not fundamentally a diagonal intension. After the fact, one can see an epistemic intension as equivalent to the diagonal of a two-dimensional intension involving both epistemic and subjunctive notions; but this complex construction is quite unnecessary to define epistemic intensions. One can characterize the first dimension of the framework in entirely epistemic terms, independently of the second dimension.
If any sort of content is derivative in the current framework, it is wide content. We have already seen that the subjunctive intension of a concept is determined by the epistemic intension in conjunction with the environment. In some cases it is a near-copy of the epistemic intension, as for simple descriptive concepts; in other cases it is determined by rigidifying the actual-world extension of the epistemic intension. By contrast, we can tell the entire story about the epistemic intension without ever involving the subjunctive intension. It therefore seems that if either intension is more fundamental, it is the epistemic intension. Still, there is no need to make too strong a claim here: both epistemic and subjunctive content are important, and both have a role to play in different domains.
Block and Stalnaker (1999) give a number of objections to the version of this framework put forward in Chalmers (1996). Many of these objections echo the objections above, and turn on interpreting the proposal via a contextual rather than an epistemic understanding.[*] Another objection is that the formal two-dimensional apparatus alone does not yield intensions with the relevant properties. This is clearly correct; but on my approach it is a substantive characterization of the intensions, not just the formal apparatus, that yields the properties in question. Block and Stalnaker also argue that the two-dimensional approach cannot explain or ground a notion of a priori truth. I have not suggested that the framework can do this; rather, I have used the notion of apriority in defining the framework. The notion of apriority, and the specific uses of it in grounding the framework, can be defended on quite independent grounds. The use of apriority in capturing the dependence of judgments about extension and truth-value on sufficient information about the world is defended at length by Chalmers and Jackson (2001).
*[[At one point, Block and Stalnaker acknowledge (in effect) that Chalmers (1996) does not intend a contextual interpretation, but suggest that a version of the "what is held constant" problem nevertheless arises in using an actual-word thought or concept to evaluate worlds without that thought or concept. I think that when things are understood in the appropriate epistemic terms, this problem clearly disappears. In fairness, it should be noted that the discussion in Chalmers (1996) is not explicit about the difference between contextual and epistemic intensions, and although the discussion tends to suggest an epistemic intension, the precise definition is left unclear. See Chalmers (forthcoming a) for discussion.]]
The current proposal also bears a resemblance to "descriptive" accounts of narrow content. It has sometimes been suggested that the narrow content of a concept such as water corresponds to the content of an associated description such as 'the dominant clear drinkable liquid in the environment', or some such. In response, a number of philosophers (LePore and Loewer 1986; Taylor 1989; White 1982) have objected that even if such descriptions exist, terms such as 'liquid' are themselves susceptible to Twin Earth scenarios (e.g., where liquids are replaced by superficially identical masses of sand), so that the content of such a description is wide rather than narrow. One might think that this objection will apply to the present proposal, since I have used descriptions of this sort to characterize epistemic intensions. But importantly, the description merely provides a rough handle on the intension for the purposes of illustration. The real narrow content is a function from scenarios to extensions, and can be characterized fully only by specifying its value at specific scenarios. As soon as we move to a summarizing description in language, imperfections are introduced, and the narrowness of the content is impurified. But the intension itself remains narrow; we should not mistake the linguistic description for the real thing.[*]
*[[This might suggest that epistemic content is "ineffable". But the real problem is simply that it is difficult to capture the epistemic content of one expression with the subjunctive content of another. Just as one can capture the subjunctive content of a concept such as water by appealing to the equivalent subjunctive content of an expression such as 'H2O', one might capture its epistemic content by appealing to the equivalent epistemic content of an expression such as 'the clear, drinkable liquid...'. It is hard to see why the second is any more objectionable than the first, or why it makes epistemic content any more "ineffable". Thanks to Frank Jackson for discussion on this point.]]
One would obtain a more closely related sort of "description" theory of epistemic content if one abstracted away from linguistic characterizations and regarded the relevant "descriptions" simply as properties that a referent must satisfy, or better, as relations to the thinker. If we speak merely of properties and relations, the linguistic contamination is avoided. Schiffer (1978) suggests a description theory of this sort, on which there is irreducibly de re reference by a thinker to himself or herself, with reference to everything else mediated by a property or relation. If we map the irreducible self-reference here to the appeal to centered worlds, and map the properties and relations to epistemic intensions, the resemblance between the accounts is clear, although Schiffer does not appeal to a two-dimensional framework, and addresses his proposal largely to the question of accounting for de re thought.
Another closely related idea is Lewis's (1979) proposal that belief involves the self-ascription of a property. The set of individuals satisfying a property corresponds directly to a class of centered worlds, as Lewis notes. Lewis (1994) argues that this sort of content is narrow and is primary in explanation. In effect, Lewis advocates a one-dimensional view of content, where apparent wide content is an artifact of belief ascriptions. While Lewis does not advocate understanding these contents in epistemic terms, and does not give a general characterization of the set of worlds associated with a belief, his examples suggest that these sets of worlds closely resemble those of an epistemic intension. So the present proposal appears to be highly compatible with Lewis's framework.
A residual problem for the present account is the problem of hyperintensionality. It seems that two beliefs — mathematical beliefs, for example — can have the same epistemic and subjunctive intensions, while nevertheless having intuitively different content, and playing quite different roles in cognition and action. To handle these issues, one needs a more fine-grained sort of epistemic content that goes beyond epistemic intensions as I have defined them. One might appeal to intensions over a more fine-grained space of epistemic possibilities, defined using a more demanding epistemic necessity operator that requires more than mere apriority (see Chalmers forthcoming b for some ideas here).[*] One might also appeal to a more basic sort of content that lies behind and determines an epistemic intension. Epistemic and subjunctive intensions are aspects of the contents of thoughts, but I have not suggested that they exhaust these contents. The nature of a complete characterization of thought contents, if such a thing can be given, remains an open question.
*[[For example, one might hold that a thought is (non-ideally) epistemically necessary when it is trivial, in a sense to be elucidated. One could then use this notion to set up a more fine-grained epistemic space of non-ideal epistemic possibilities, and could then associate concepts and thoughts with non-ideal epistemic intensions over this space. Then two concepts or thoughts that are nontrivially a priori equivalent will have the same epistemic intension as defined in the paper, but different non-ideal epistemic intensions as defined here.]]
Another open question: is it possible to reductively explain the epistemic content of a subject's thoughts in naturalistic terms, in the way that some have attempted to explain wide content in causal or teleological terms? Certainly no such explanation is currently available. A first attempt might exploit the idea that epistemic content is mirrored in the idealized rational dispositions of the subject: perhaps a subject's actual dispositions will yield epistemic content by idealization? The normative character of the idealization may pose an obstacle to reduction, however, as will the fact that these dispositions are themselves characterized by appeal to content. My own view is that epistemic content is ultimately determined by a combination of a subject's functional organization and phenomenology.[*] If so, any attempt at explanation will need to appeal to these factors. In any case, it is arguable that if wide content depends heavily on narrow content, as the current account suggests, any adequate reductive theory of wide content will require a reductive theory of narrow content first.[*]
*[[For an argument that phenomenology is essential to the epistemic content of at least some concepts, see Chalmers 2002c. See also Horgan and Tienson (this volume, chapter 49) for arguments for phenomenally constituted narrow content that can be seen as complementing the current approach.]]
*[[Arguably, contemporary causal theories of content have been unsuccessful precisely because they attempt to account for wide content directly, without taking into account the crucial epistemic dimension involved in its determination.]]
What of the six puzzles at the start? To summarize:
(1) A thought's content decomposes into epistemic and subjunctive content, given by its epistemic and subjunctive intensions. Oscar's and Twin Oscar's thoughts differ in their subjunctive contents, and as a result ground different belief ascriptions, but their epistemic contents are the same.
(2) My thoughts that Hesperus is Hesperus and that Hesperus is Phosphorus have the same subjunctive intension but distinct epistemic intensions, as the Hesperus and Phosphorus concepts have different epistemic intensions. The triviality of the former does not imply the triviality of the latter, as it is epistemic content that governs rational relations.
(3) Pierre's two beliefs have contradictory subjunctive intensions but compatible epistemic intensions. The apparently contradictory belief ascriptions arise because of the contradictory subjunctive intensions and because his two concepts of London have distinct epistemic intensions that can each make 'London'-involving belief ascriptions true. Rationality is governed by epistemic intensions, so there is no rational contradiction here.
(4) The essential indexicality of belief reflects the fact that epistemic content, not subjunctive content, governs action, and that epistemic content, unlike subjunctive content, is an indexical centered intension.
(5) The modes of presentation central to a theory of belief ascription are epistemic intensions. Belief ascriptions specify a believer's subjunctive content, and constrain the believer's epistemic content.
(6) Instances of the contingent a priori have a necessary epistemic intension but a contingent subjunctive intension. One's cognitive world-model is constrained by epistemic content, not by subjunctive content; so a contingent subjunctive intension does not indicate a cognitive achievement.
There are many problems about the contents of thought that are not resolved by this framework. These include the problems of hyperintensionality, of a full account of belief ascriptions, and of giving a naturalistic explanation of content. Some of these matters are likely to be much more difficult than the puzzles at issue in this paper, but the two-dimensional approach at least clarifies the lay of the land.
Thanks to too many people to mention, but especially Ned Block, Curtis Brown, Frank Jackson, David Lewis, Mark Sainsbury, Robert Stalnaker, and two reviewers. Thanks also to audiences at talks between 1994 and 1997 at Arizona, Cornell, Memphis, Princeton, Rice, UCLA, UC Santa Barbara, UC Santa Cruz, Washington University, Yale, and the Australasian Association of Philosophy. For comments on the revised version, thanks to Torin Alter, Chris Evans, Brie Gertler, and Daniel Stoljar.
Austin, D. F. 1990. What's the Meaning of "This"? Cornell University Press.
Block, N. 1991. What narrow content is not. In B. Loewer and G. Rey, eds., Meaning in Mind: Fodor and His Critics. Blackwell.
Block, N. and Stalnaker, R. 1999. Conceptual analysis, dualism, and the explanatory gap. Philosophical Review 108:1-46.
Brown, C. 1986. What is a belief state? Midwest Studies in Philosophy 10:357-78.
Burge, T. 1979. Individualism and the mental. Midwest Studies in Philosophy 4: 73-122.
Chalmers, D. J. 1996. The Conscious Mind: In Search of a Fundamental Theory. Oxford University Press.
Chalmers, D. J. 2002a. Does conceivability entail possibility? In T. Gendler and J. Hawthorne (eds.), Conceivability and Possibility. Oxford University Press. http://consc.net/papers/conceivability.html.
Chalmers, D. J. 2002b. On sense and intension. Philosophical Perspectives 16. http://consc.net/papers/intension.html.
Chalmers, D. J. 2002c. The content and epistemology of phenomenal belief. In Q. Smith and A. Jokic (eds.), Consciousness: New Philosophical Essays. Oxford University Press. http://consc.net/papers/belief.html.
Chalmers, D. J. (forthcoming a). The foundations of two-dimensional semantics. http://consc.net/papers/foundations.html.
Chalmers, D. J. (forthcoming b). The nature of epistemic space. http://consc.net/papers/espace.html.
Chalmers, D. J. and Jackson, F. 2001. Conceptual analysis and reductive explanation. Philosophical Review 110:315-61. http://consc.net/papers/analysis.html.
Clark, A. and Chalmers, D. J. 1998. The extended mind. Analysis 58:7-19. http://consc.net/papers/extended.html.
Crimmins, M. 1991. Talk about Beliefs. MIT Press.
Davies, M. K. and Humberstone, I. L. 1980. Two notions of necessity. Philosophical Studies 38:1-30.
Dennett, D.C. 1981. Beyond belief. In A. Woodfield (ed.), Thought and Object. Oxford University Press.
Evans, G. 1979. Reference and contingency. The Monist 62:161-89.
Fodor, J. 1987. Psychosemantics. MIT Press.
Fodor. J. 1991. A modal argument for narrow content. Journal of Philosophy 88:5-26.
Frege, G. 1892. Über Sinn und Bedeutung. Translated in (P. Geach and M. Black (eds.), Translations from the Philosophical Writings of Gottlob Frege. Blackwell, 1952.
Horgan, T. and Tienson, J. (this volume). The intentionality of phenomenology and the phenomenology of intentionality.
Kaplan, D. 1967. Quantifying in. Synthese 19:178-214.
Kaplan, D. 1989. Demonstratives. In J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan. Oxford University Press.
Kripke, S. A. 1979. A puzzle about belief. In A. Margalit (ed.), Meaning and Use. D. Reidel.
Kripke, S. A. 1980. Naming and Necessity. Harvard University Press.
LePore, E. and Loewer, B. 1986. Solipsistic semantics. Midwest Studies in Philosophy 10:595-614.
Lewis, D. 1979. Attitudes de dicto and de se. Philosophical Review 88:513-43.
Lewis, D. 1994. Reduction of mind. In S. Guttenplan (ed.), Companion to the Philosophy of Mind. Blackwell.
Loar, B. 1988. Social content and psychological content. In R.H. Grimm and D.D. Merrill (eds.), Contents of Thought. University of Arizona Press.
Patterson, S. 1991. Individualism and semantic development. Philosophy of Science 58:15-35.
Perry, J. 1977. Frege on demonstratives. Philosophical Review 86:474-97.
Perry, J. 1979. The problem of the essential indexical. Nous 13:3-21
Putnam, H. 1975. The meaning of `meaning'. In K. Gunderson (ed.), Language, Mind, and Knowledge. Minneapolis: University of Minnesota Press.
Putnam, H. 1981. Reason, Truth, and History. Cambridge University Press.
Quine, W. V. 1968. Propositional objects. Critica 2(5):3-22.
Richard, M. 1990. Propositional Attitudes: An Essay on Thoughts and How We Ascribe Them. Cambridge University Press.
Salmon, N. 1986. Frege's Puzzle. MIT Press.
Schiffer, S. 1978. The basis of reference. Erkenntnis 13:171-206.
Schiffer, S. 1990. The mode-of-presentation problem. In C.A. Anderson and J. Owens (eds.), Propositional Attitudes: The Role of Content in Logic, Language, and Mind. CSLI Press.
Schiffer, S. 1992. Belief ascription. Journal of Philosophy 87:602-14.
Segal, G. 2000. A Slim Book about Narrow Content. MIT Press.
Smith, E. E. and Medin, D. L. 1981. Categories and Concepts. Harvard University Press.
Stalnaker, R. 1978. Assertion. In P. Cole (ed.), Syntax and Semantics: Pragmatics, Vol. 9. Academic Press
Stalnaker, R, 1989. On what's in the head. Philosophical Perspectives 3:287-316.
Stalnaker, R. 1990. Narrow content. In C.A. Anderson and J. Owens (eds.), Propositional Attitudes. Stanford: Center for the Study of Language and Information.
Stalnaker, R. 1999. Context and Content. Oxford University Press.
Stich, S. 1983. From Folk Psychology to Cognitive Science. MIT Press.
Taylor, K. 1989. Narrow content functionalism and the mind-body problem. Nous 23:355-72
White, S. 1982. Partial character and the language of thought. Pacific Philosophical Quarterly 63:347-65.
White, S. 1991. Narrow content and narrow interpretation. In The Unity of the Self. Cambridge, MA: MIT Press.
Yablo, S. 2002. Coulda, woulda, shoulda. In T. Gendler and J. Hawthorne (eds.), Conceivability and Possibility. Oxford University Press.