I've fallen behind on posting about the Soames book, due to various distractions. I've written up partial entries on Chapter 8 (on Jackson) and on the first half of Chapter 9 (on me), and hope to get back to those shortly. But for now I'll post about the second half of Chapter 9, which is about my account of attitude ascriptions. One reason is that I'm giving a talk at UCLA on Wednesday on Soames on 2D, which will probably cover some combination of the material in this entry and the entry on Chapter 7, plus maybe a bit of the ASU talk.
In this chapter Soames gives four detailed arguments against my account. There are more arguments against it in Chapter 10, but here I'll concentrate on the first four. If you haven't read it already, you might look at the entry on Chapter 7 for background on my analysis.
The basic shape of the analysis is
(A) An utterance of 'A believes that S' is true when A has a belief with the secondary intension of S (in the mouth of the ascriber) and with an appropriate primary intension.
Here, what counts as an "appropriate" primary intension may be contextually determined. Strictly speaking the analysis should appeal to structured primary and secondary intensions, but mostly that won't make a difference in what follows.
Soames starts with an exposition of the view on pp. 251-53. The exposition is reasonably accurate, except that he says that my account requires that the primary intension of A's belief be "appropriately related" to the primary intension of S. Although he puts "appropriately related" in quotes, I never use this locution or talk of a relation here. I don't think that the appropriateness constraint on the ascribee's primary intension is always derived from a constraint on the relation to the ascriber's primary intension. But this doesn't matter much in what follows.
Soames notes that the account is not fully specified in that it doesn't say what the conditions of appropriateness are for a given ascription. Of course this is correct: as I say in "The Components of Content" (section 8), this is certainly not a complete analysis of attitude ascriptions. I'm far from sure that any formal principles can be given for fully determining these conditions -- they may be determined in large part by the intentions of the ascriber on a given occasion. I'm sure one can say at least some systematic things about these constraints, though, and insofar as we can do so, that will help to flesh out the analysis above. But the account above at least provides a skeleton around which which fuller analyses can be constructed.
Soames also notes that this analysis doesn't "identify the objects of belief", and it doesn't explicitly characterize the belief predicate as relational. This is partly because I'm not sure that there's any one sort of object that deserves to be called "the object of belief", and I'm not sure that belief predicates are relational (at least in the sense of expressing relations to objects denoted by "that"-clauses). So the analysis above aims to be neutral on these things, by providing an analysis of the conditions under which an ascription is true, without making strong commitments about the underlying logical form, or about the metaphysics of belief, and so on.
The account is compatible with a variety of views on these matters. It's most obviously compatible with a logical form akin to those in "hidden-indexical" analyses of belief ascription, where the "that"-clause is taken to refer to a secondary intension, and where this relation to a secondary intension is contextually supplemented with a condition on a related primary intension. If one thinks of things along those lines, one might take the "object of belief" (i.e. the referent of a "that"-clause, to which a subject is relating in believing) to be a secondary intension, with the primary intension serving as a mode of presentation of that object. But it may well also be compatible with the account to take a "that"-clause as referring in a context-dependent way to a complex two-dimensional entity, which one might then take to be the object of belief. And the account may well be compatible with a view on which "that"-clauses do not refer at all, and/or accounts on which beliefs have multiple objects. The choice between these accounts would require a lot of further investigation (and the matter is complicated by the fact that I have sympathies with instrumentalism about logical form). But the basic shape of (A) again at least provides some constraints on such an analysis.
In any case, Soames' detailed arguments don't rest on broad worries of this sort, so I'll turn to those arguments now.
Argument 1: A Priori Knowability
Soames first argument (pp. 254-257) is closely related to points discussed in the second half of my entry on Chapter 7. He starts with two principles that he says are "central to my account".
Apriori 1: A sentence S is a priori iff S has a necessary primary intension.
Apriori 2: 'It is knowable apriori that S' is true iff S is a priori.
He then notes that there are cases roughly as follows (the example is mine, but it fits his structure). A1 uses the name 'Jack' to refer to whoever committed murders M1 and M2. A2 uses the same name to refer to whoever committed murders M1, M2, and M3. Let's say there is someone who committed all three murders. Then it looks like on my account, A1's utterance of:
(1) A2 believes that if Jack exists, then he committed M1, M2, and M3.
will be true, as A1 and A2 have similar enough primary intensions associated with the name for both to count as "appropriate" here. (That's OK with me -- there will certainly be cases that work like this, whether or not this case is one of them.) But then it looks like the same should hold if we substitute 'knows' for 'believes' in (1), and indeed if we substitute 'knows a priori', since the relevant belief of A2 does indeed constitute a priori knowledge (at least on my view). So it looks like A1's utterance of
(2) A2 knows a priori that if Jack exists, then he committed M1, M2, and M3.
It seems to follow that A1's utterance of (3) is true.
(3) It is knowable a priori that if Jack exists, then he committed M1, M2, and M3.
But A1's utterance of (4) has a contingent primary intension, and so (given Apriori 1) is not a priori.
(4) If Jack exists, then he committed M1, M2, and M3.
This entails that principle Apriori 2 is false.
My response to this argument will be pretty clear from the entry on Chapter 7. First, I'm not committed to Apriori 2. If one takes a reading of 'It is knowable a priori that' on which it is equivalent to 'It is possible that someone knows a priori that', then there may well be counterexamples to Apriori 2 along the lines above. Nothing terrible follows. As explained earlier, the apriority of a sentence isn't tied by definition to the truth of an a priori knowledge ascription.
Second, if one really wants to preserve Apriori 2, one can either (2a) embrace a reading of 'It is knowable a priori that' on which it isn't equivalent to the modal quantified attitude ascription above (maybe it should be read as a simple operator, e.g. as a way of saying 'It is a priori that', analogously to 'It is analytic that' or 'It is necessary that'); or one can (2b) retain the reading as an attitude ascription, but hold that 'It is knowable a priori that' ascriptions (or at least, the uses of such ascriptions for which Apriori 2 holds) always invoke strong "appropriateness" constraints, requiring that the believer has the same primary intension as the ascriber. Given such constraints, A1's utterance of (3) won't be true.
For what it's worth, my intuitions are that an ordinary utterance of (3) would be false, so I'd probably take strategy (2a) or (2b) at least about such utterances (there may be other true utterances, for which one can deny Apriori 2). Probably strategy (2a) best captures ordinary philosophical usage.
If one takes strategy (2b), one has the residual objection: (1) seems true, and it seems that if (1) is true (2) should be true, and on this reading (3) should follow from (2), so (3) should be true. For what it's worth, my intuitions are that ordinary utterances of (1), (2), and (3) are true, unclear, and false respectively. On the current framework, assuming strategy (2b), this tends to suggest that the three different ascriptions tend to involve different appropriateness constraints: belief ascriptions involve relatively loose appropriateness constraints, a priori knowledge ascriptions involve somewhat tighter constraints, and that 'It is knowable a priori that' claims involve very tight constraints. One can then say that the entailment from (2) to (3) holds only if the contextually-determined constraints stay constant, and that uttering (3) tend to introduce stronger constraints, so an utterance of (3) can be false even when an utterance of (2) would be true. One might supplement this by saying that when (3) is uttered after (1) or (2), it can more easily be true, as the constraints may be largely inherited from the constraints involved in uttering (1) or (2). This will lead to a somewhat atypical a priori knowledge ascription. For utterances of this sort, (2) and (3) may stand or fall together, and Apriori 2 may be false. For more typical separate utterances of (2) and (3), their truth-values can come apart, and Apriori 2 will be true.
Argument 2: Watery Stuff
The next argument is on pp. 257-260. Let's say that our use of 'water' has a primary intension that we can encapsulate in the description 'watery stuff'. On Twin Earth, XYZ is the watery stuff, but there are also a few isolated pools of H2O, which (on Twin Earth) has some but not all of the properties of watery stuff. Let's say that 'sort of watery stuff' stands for these properties. People on Twin Eath come across the pools of H2O, and say
(4) That stuff is sort of watery stuff.
It seems that we can truly say, of the agents on Twin Earth
(5) They know that water is sort of watery stuff.
But it looks like (6) is false, since their knowledge ascribed in (5) is thoroughly empirical.
(6) They are in a position to know a priori -- through proper reasoning and reflection -- that water is sort of watery stuff.
But (7) is true, at least given Apriori 1 and Apriori 2, as our utterances of 'water is watery stuff' have a necessary primary intension:
(7) It is knowable a priori that water is sort of watery stuff.
Soames' objection is that it is hard to reconcile the falsity of (6) with the truth of (7). Given that it is knowable a priori that water is sort of watery stuff, and given that the Twin Earthers can entertain the proposition that water is sort of watery stuff, then they ought to be able to know that the proposition is true.
I'm not sure what to make of this objection. Certainly, I think (7) is true (at least if one inserts an existence clause, which I'll ignore here). And (6) seems plausibly false. One immediate response is to say that (6) is only false because of the "through reasoning and reflection" qualification: Twin Earthers could come to know a priori that water is sort of watery stuff, through coming to Earth and acquiring our term 'water', but they're not now in a position to do that through reasoning and reflection. So if one drops the qualification in (6), yielding something like (6a), then (6a) is true and there's no conflict with (7).
(6a) It is possible that they know a priori that water is sort of watery stuff.
But even if one sets this response aside, I'm not sure I see the problem. Going by intuitions, it seems fairly reasonable to say that (6) is false and (7) is true. And going by theory, the 2D analysis gives a perfectly good account of why (6) is false and (7) is true. Any item of a priori knowledge (or potential a priori knowledge) has a necessary primary intension. So for a belief that water is watery stuff to qualify as a priori knowledge (or as potential a priori knowledge), it must have a necessary primary intension. But given (A), it is possible to believe that water is watery stuff without one's belief having a necessary primary intension -- 'water is watery stuff' in my mouth has a necessary primary intension, but the appropriateness constraint allows that other subjects can believe that water is watery stuff even though their relevant belief has a non-necessary primary intension. In such a case, they will believe that water is sort of watery stuff, but this belief will not even be potential a priori knowledge. The primary intension is the right sort to qualify as belief that water is sort of watery stuff, but not the right sort to allow a priori knowledge.
Soames put things in terms of "propositions", whereas I usually don't, for reasons related to those mentioned earlier. But if one really wants to put things in terms of propositions, let's pretend that propositions (understood as the referents of "that"-clauses) are secondary intensions, and that a version of the hidden-indexical account above is true. Then we can say that although the Twin Earthers stand in the belief relation to the proposition that water is wet, a priori knowledge of this proposition requires standing in a relation to this proposition that is mediated by the right sort of primary intension, and the Twin Earthers' belief is mediated by the wrong sort of primary intension. Or to put it in simpler terms, they believe the proposition, but they believe it in the wrong way, and ways of believing matter where a priori knowledge is concerned. As always, the analysis above isn't beholden to this account of propositions, though, and if we take a different view of propositions, we'll just have to put the point a different way.
Argument 3: Hesperus
The third argument is on pp. 261-263. Let's say that our use of 'Hesperus' has a primary intension that we can encapsulate in a description D, which involves in part a certain sort of visibility in the evening sky. Then on the 2D account, (8) will be a priori:
(8) Hesperus is visible in the evening sky, if there is a unique thing which is D.
Given the right-to-left direction of Apriori 2 (which I won't argue with here), it looks like (9) is true:
(9) It is knowable a priori that (Hesperus is visible in the evening sky, if there is a unique thing which is D).
Soames' now invokes an "intuitively compelling exportation principle E", which I shorten slightly here:
(E) For any name n and predicate D, if 'A believes that n is F' is true, then 'A believes that x is F', where the referent of n is assigned to the variable x. So is 'A believes that I am F' (uttered by the referent of n). And same when 'n is F' is replaced by 'n is F, if there is a unique thing which is D'.
This principle allows one to move from de dicto ascriptions involving a name to corresponding de re ascriptions. Soames uses this principle to move from (9) to (10), which we can read as a de re ascription in which Hesperus is assigned to the variable x:
(10) It is knowable a priori that x is seen in the evening sky, if there is a unique thing that is D.
Soames then says that this holds even though the primary intension of 'x is visible in the evening sky', relative to an assignment of 'Hesperus' to x, is contingent (and identical to its secondary intension). And this again violates the conjunction of Apriori 1 and Apriori 2.
Invoking some "poetic license", Soames also gives a version of the argument involving an utterance made by Hesperus (!), to her neighbor Mercury:
(11) It is knowable a priori that (I am visible in the evening sky, if there is a unique thing which is D).
He notes that (11) follows from (9) by (E). But the primary intension of 'I am visible in the evening sky, if there is a unique thing which is D' is contingent, so that this violates the conjunction of Apriori 1 and Apriori 2.
There are a number of things to say here. First, as before, I'm not committed to Apriori 2, and one might think that (11) is one of the cases where it fails (because someone could have the relevant a priori knowledge under a different mode of presentation of the speaker that differ's from the speaker's mode of presentation associated with 'I'). That wouldn't be entirely unreasonable, but I'll set it aside.
Second, I think that principle (E) is false, and is certainly not "intuitively compelling". I've already discussed this principle in the entry on Chapter 4, and in the related poll on belief ascriptions. There we saw that at least in the case of belief ascriptions involving descriptive names, the principle seems intuitively to fail (at least, the majority of respondents had the intuition, in a particular case, that a de dicto ascription is true while the corresponding de re ascription is false). I think there are plenty of intuitive counterexamples to the principles involving ordinary names, too.
E.g. Susan overhears people talking about someone called 'Fred', and asks them 'Who is Fred?'. Intuitively, she doesn't know the answer to the question. As it happens, Fred is a student in her class (whose name she hadn't otherwise heard). Intuitively, (12) seems true while (13) seems false:
(12) Susan knows that Fred is Fred.
(13) The student (Fred) is such that Susan knows that he is Fred.
And certainly, if the student overhears the conversation and utters (14), the intuitive judgment is that his utterance is correct:
(14) Susan doesn't know that I am Fred.
If (12) is true while (13) and (14) are false, then (E) is false even for ordinary names. All Soames says in defense of (E) is that it is "intuitively compelling", but that seems to be straightforwardly false. Perhaps someone could mount a theoretical defense of (E) -- I don't mean to suggest that it is so obviously false to be non-negotiable. But it's clear that (E) is not so obviously true to be non-negotiable, and there's no huge cost in denying it.
So one can straightforwardly resist Soames' argument here. The exportation principle seems particularly easy to deny in the case of a priori knowledge ascriptions. In the entry on Chapter 4, for example, I noted that de dicto a priori knowledge ascriptions involving descriptive names often seem true while corresponding de re a priori ascriptions seem false. So one can quite reasonably resist the step from (9) to (10) and (11). And indeed, this resistance is grounded by my account of de re attitude ascriptions, which I'll discuss further at some point. There is an interesting residual issue about the treatment of de re a priori knowledge ascriptions, which I hope to discuss.
Finally, there's something highly problematic about Soames' treatment of (10), which assumes that when a sentence involves a variable, its primary intension is determined by the value of the variable. But that issue recurs in the next and final objection, which I'll discuss next.
Argument 4: Pappy
This argument is presented on pp. 263-265. Say that 'Pappy' is a name for my wooden paperweight. Then Soames says that (15) is necessary a posteriori:
(15) If Pappy exists, then Pappy is not made out of metal.
He then considers (16), relative to an assignment of Pappy to the variable 'x':
(16) If Pappy exists, then x is not made out of metal and Pappy = Pappy.
Soames says that since the primary intension of (16) (relative to the assignment of Pappy to 'x') is true in all contexts (scenarios), it follows by Apriori 1 that (16) is a priori. He says that Apriori 2 then entails that (17) is true, and that the truth of (17) entails the truth of (18):
(17) It is knowable a priori that (if Pappy exists, then x is not made out of metal and Pappy = Pappy).
(18) There is an x, namely Pappy, which is such that it is knowable a priori that (if Pappy exists, then x is not made out of metal & Pappy = Pappy).
Applying principle E, we get:
(19) There are x and y (both of which are identical to Pappy), such that it is knowable a priori that (if Pappy exists, then x is not made out of metal & y = Pappy).
He says that by Leibniz's Law, this entails:
(20) There is an x, namely Pappy, which is such that it is knowable a priori that (if Pappy exists, then x is not made out of metal & x = Pappy).
He notes that, given Apriori 1 and Apriori 2, this conflicts with the observation that the primary intension of (21), relative to an assignment of Pappy to 'x', is not necessary:
(21) If Pappy exists, then x is not made out of metal & x = Pappy.
He also notes that it is very counterintuitive to say that (20) is true, while (22a) and (22b) are false, as they must be if (15) is a posteriori.
(22a) It is knowable apriori that (if Pappy exists, then Pappy is not made out of metal & Pappy = Pappy).
(22b) It is knowable apriori that (if Pappy exists, then Pappy is not made out of metal).
My response to all this is pretty straightforward. One might respond by denying (E), or by denying Apriori 2, but I think that's not necessary here. Things go wrong early, when Soames considers (16). This is a sentence with a free variable (x) that is assigned a value (Pappy). To simplify, let's consider the following simpler sentence, where 'x' is assigned the value Pappy.
(23) x is made of metal
Soames assumes that on the 2D account: (i) such sentences have primary intensions, (ii) that their primary intensions are determined by composing the primary intension of the rest of the sentence with the value assigned to x (so that the primary intension of (23) is true in all centered worlds where Pappy itself satisfies the primary intension of 'is made of metal), and that (iii) such a sentence will be a priori iff its primary intension is necessary.
In response, I think that (i) is probably false, and that (ii) is certainly false. Re (i): The primary intensions of sentences referring to external objects are always sensitive to the mode of presentation of that object, and in particular are sensitive by definition to the inferential role associated with that mode of presentation. If 'x' is supposed to be a mode-of-presentation free tag for an object, one that doesn't ground any substantial inferential role, then it can't be associated with a primary intension.
It's worth noting that utterances of sentences such as (23) don't exist in ordinary English. The closest things are embedded versions on which x is bound, and unembedded versions on which an utterance of (23) is effectively a rephrasal of 'Pappy is made of metal' (or 'That is made of metal', or some other sentence involving a mode of presentation of Pappy). But of course both of these can be treated straightforwardly without leading to anything like (ii) above. Setting aside these uses, then all that's left is an artificial sort of philosophers' "sentence" that doesn't seem to be the kind of thing that one can utter in a truth-evaluable way (any utterances in the vicinity will involve a mode of presentation of Pappy). For some purposes (e.g. embedding) one might associate (23) with an "unsaturated" primary intension, along the lines of the primary intension of 'is made of metal', that needs to be saturated to yield something truth-evaluable. Perhaps for some purposes it could be useful to associate (23) with a pair of an object (Pappy) and a primary intension (that of the rest of the sentence). But it would be misleading to call such a thing the "primary intension of the sentence".
If one accepted (ii), then (23) would be associated with a necessary primary intension (at least if we set aside difference between the primary and secondary intension of "is made of metal"). I'd suggest that this is a pretty meaningless claim. If one defined primary intensions this way, then the primary intensions of these sentences would behave entirely unlike the primary intensions of ordinary sentences. For example, one would certainly then say that the connection between apriority and necessity of primary intension doesn't apply to these sentences; i.e. if one defined primary intensions to satisfy (ii), one would then have to deny (iii). I'd say it's much better to keep (iii) and to drop (ii), which never had much going for it in the first place.
All four of Soames' arguments can be straightforwardly answered, but doing so at least brings out some interesting points. Answering the first brings out some subtleties in the behavior of 'It is knowable a priori that'. Answering the second brings out certain interesting connections between belief, mode of presentation, and a priori knowability. Answering the third brings out the implausibility of the exportation principle (E). Answering the fourth brings out some subtleties about the treatment of variables in the 2D framework. In no case, however, are we forced to say anything for which there isn't independent motivation, and it doesn't look like we're forced to incur any significant costs.