In the interest of completing the response to Soames' arguments against my analysis of attitude ascriptions, here's a post on Chapter 10. Unlike most of the previous chapters, this chapter is available online (in a slightly older version). In the first part of the chapter, he gives arguments against "strong" and "weak" two-dimensionalism, mostly focusing on attitude ascriptions. These arguments correspond pretty closely to the arguments I responded to in the ASU commentary, which shows how my analysis can deal with the various problem cases. In the last part of the Chapter (pp. 313-324 of the book, and pp. 30-38 of the online version), he gives some further arguments against the "hybrid" view, which is his label for the view that I hold, focusing especially on some issues about de re attitude ascriptions. I'll address these arguments here.
On p. 316, Soames asks us to consider a name or natural-kind term n whose reference is fixed by a description D (say, 'Lee' and 'the youngest Chinese spy'). He then notes that the utterances of the following four sentences have the same secondary intension:
(1a) n is F
(1b) dthat[the D] is F
(1c) x is F [where the referent of 'n' is assigned to 'x']
(1d) I am F [uttered by the referent of 'n']
He then asks us to consider the following four attitude ascriptions:
(2a) A knows/believes that n is F.
(2b) A knows/believes that dthat[the D] is F.
(2c) A knows/believes that x is F [referent of 'n' assigned to 'x']
(2d) A knows/believes that I am F [uttered by the referent of 'n']
Soames first says that on my account, the primary intensions of (1a) and (1b) are the same, so (2a) and (2b) have the same truth-value. That doesn't really follow (it's possible that utterances of (2a) and (2b) could introduce different appropriateness constraints on a primary intension), but I'll let it slide. He then says that since, "as a matter of obvious empirical fact", (2c) and (2d) are true whenever (2a) is true and 'n' has a referent, the proponent of my view has no choice but to say that (1c) and (1d) have primary intensions that are "appropriate" to make (2c) and (2d) true.
He then points to certain claims that the 2D theorists holds are a priori, such as
(3a) Hesperus is visible in the evening (if ...)
(3b) Water is watery stuff (if...)
[Soames omits the existence ("if...") clause for (3b) and gives a lengthy clause for (3a). I'll just take these to be understood from now on.]
He then says that it follows that the theorist is committed to there being cases in which (4a) and (4b) are true, and so in which (4c) and (4d) are also true.
(4a) A knows a priori that n is F
(4b) A knows a priori that dthat[the D] is F
(4c) A knows a priori that x is F [referent of 'n' assigned to 'x']
(4d) A knows a priori that I am F [uttered by the referent of 'n']
Soames then raises two problems with this result. In his own words (with sentences renumbered):
There are two problems with this result. First, whatever it is that is supposed to be known apriori in the case of (4c,d) -- whether it be the complement clause, the secondary intension of the clause, or a "two-dimensional proposition" consisting of the primary and secondary intension of the clause -- need not either itself be necessary, or have a necessary primary intension. Thus, the connection between aprioricity and necessity of primary intension -- so central to ambitious two-dimensionalism -- seems to have been lost. Second, de re knowledge of the sort reported by (4c) and (4d) is *never* apriori. For example, one simply can't know apriori, of any object, that it is visible in the evening (if anything is the brightest object visible in the evening sky at times t and places p), or, of any kind, that instances of it are clear, potable, etc. As argued in chapter 4, such knowledge is always grounded in empirical knowledge derived, ultimately, from someone's acquaintance with the object or kind. Hence -- contra the ambitious two-dimensionalist -- ascriptions along the lines of (4a,b) are (in the relevant cases) always false.
Some people -- liberals about de re a priori knowledge, such as Robin Jeshion and some slices of David Kaplan -- might respond that the second half of this passage is incorrect, and that ascriptions such as (4c) and (4d) can be true. However, I'm inclined to share Soames' intuition here. At least, I think there's a notion of de re knowledge such that one can't have a priori de re knowledge in these cases. So here I'll go along with Soames assuming that de re knowledge works this way. If so, then (4c) is certainly false (assuming it's read as a de re a priori knowledge ascription), and (4d) is false also (assuming that it entails (4c)).
It seems to me that Soames' argument above goes wrong in assuming that (4c) and (4d) follow from (4a) (and likewise in assuming that (2c) and (2d) follow from (2a)). Just speaking intuitively, we saw already in the belief ascription poll that in relevant cases involving descriptive names, the majority of people judge (2a) and (4a) to be true while judging (2c) and (4c) to be false. I would guess that intuitions about (2d) and (4d) follow those about (2c) and (4c). (I'm not sure about (2b) and (4b), as 'dthat' is a technical term that is understood by different people in different ways. Here I'll ignore (2b) and (4b) aside.) So it's at least far from obvious that there is any entailment here.
Presumably Soames is tacitly appealing to his "intuitively compelling" exportation principle (E) (see argument 3 in the post on Chapter 9), which would ground a move from (2a) to (2c) and (2d), and from (4a) to (4c) and (4d). But we've already seem, both there and in the post on Chapter 4, that this principle is far from intuitively compelling, and seems to have many intuitive counterexamples. If we deny principle (E), then this argument does not get off the ground.
Furthermore, my analysis of de re attitude ascriptions itself strongly suggests that principle (E) is false, and that (2c) and (4c) are false in the relevant cases. This analysis is discussed briefly in section 8 of "The Components of Content". Although the discussion in that paper is brief, it's odd that Soames never mentions it in his book, given the centrality of issues about de re ascriptions there.
Recall my analysis of de dicto ascriptions:
(A) 'A knows/believes that S' is true iff A has a belief with the secondary intension of S (in the mouth of the ascriber) and with an appropriate primary intension.
The analysis of de re ascriptions is different but related:
(B) 'A knows/believes of n that it is F' is true iff A has a belief with the secondary intension of 'n is F' (in the mouth of the ascriber), where this belief picks out the referent of 'n' under a de re-appropriate primary intension.
[As before, one should strictly speaking invoke structured intensions here, and one also needs an "appropriateness" constraint on the aspect of the belief's primary intension that picks out F, but I'll set those things aside here.]
This account is really a straightforward adaptation to the 2D framework of Kaplan's account of de re ascriptions in "Quantifying In". The idea is that when a believer has a belief about x (the referent of 'n'), their belief will involve a particular mode of presentation of x, which here corresponds to the primary intension of the concept (i.e. mental representation) that picks out x. Some of these modes of presentation, such as typical purely descriptive modes of presentation of x (say, one that picks out the tallest person in Australia, whoever that is), won't license a de re attitude ascription. Other modes of presentation, such as perceptual demonstrative modes of presentation, will license such ascriptions.
To a first approximation, we might think that a primary intension is de re-appropriate when it is acquaintance-involving. An acquaintance-involving primary intension is one that picks out its referent via an acquaintance relation. (Slightly more precisely, though still imperfectly: an acquaintance-involving primary intension is one such that necessarily, if an object satisfies the primary intension in a centered world, the object and the subject at the center stand in a relation of acquaintance.) That's only a first approximation, though, and it may well be that what qualifies as a de re-appropriate primary intension is pretty context-dependent, partly because what counts as "acquaintance" may be context-dependent, and partly because whether acquaintance is necessary or sufficient may vary with context.
Anyway, given this account, it's easy to see why the exportation principle fails in general. Take a descriptive name such as 'Lee', for the shortest Chinese spy. Let's say that Huey says 'Lee is the shortest Chinese spy (if anyone is)', expressing a belief. Then consider the de dicto and de re ascriptions:
(5a) Huey believes that Lee is the shortest Chinese spy.
(5b) Huey believes of Lee that he is the shortest Chinese spy.
We've already seen that a majority of people have the intuition that (5a) is true and (5b) is false. The analysis above accomodates the intuition straightforwardly. Huey's belief has a secondary intension true in all worlds where x is the shortest Chinese spy (where here x is the actual shortest Chinese spy), which is the right sort of secondary intension to make both (5a) and (5b) true, according to (A) and (B). But his belief picks out x under a descriptive mode of presentation, i.e. a primary intension picking out the shortest Chinese spy in a world. That primary intension is the primary intension of 'Lee' (as used by the ascriber). That sort of sameness in primary intension usually suffices for "appropriateness" in analysis (A) of de dicto ascriptions, so the primary intension here will be appropriate for the truth of (5a), and (5a) will be true. But because the primary intension is a descriptive one that is not acquaintance-involving, it will not be a de re-appropriate primary intension (at least in most contexts), so (5b) will be false.
This exhibits the general pattern, at least in the case of typical descriptive names. To satisfy de dicto ascriptions such as (5a), it suffices to have a belief that picks out the referent under the relevant descriptive primary intension. But to satisfy de re ascriptions such as (5b), it is necessary to have a belief that picks out the referent under an acquaintance-involving primary intension. So the de dicto ascription can be true when the de re ascription is false, and the exportation principle will fail.
The case of ordinary proper names is a bit more complicated. Unlike descriptive names, it's not implausible that ordinary proper names always have acquaintance-involving primary intensions, at least given a reasonably generous notion of acquaintance where any causal connection (including linguistically mediated connections) suffice for acquaintance. If so, one might think that the exportation principle will hold in such cases. I think that's not quite right, though. Take e.g. the Susan/Fred case in my entry on Chapter 9, in which Susan overhears two people talking about someone called 'Fred', who unbeknownst to her is a student of hers. Intuitively she knows that Fred is Fred but does not know of her student (Fred) that he is Fred. So this is a case where the exportation principle is intuitively implausible even for ordinary proper names. This can be accommodated the observation that in this sort of case, a more stringent notion of de re-appropriateness is contextually invoked (one on which the primary intension that Susan expresses with 'Fred' does not qualify), perhaps because a more stringent notion of acquaintance is required.
In any case, it was already reasonably clear that the exportation principle, and the associated inference from (1a) and (3a) to (1c) and (3c), is intuitively implausible. We can now see how its falsity is accommodated, and to some extent predicted, by the analysis of de dicto and de re attitude ascriptions in "The Components of Content". So, as everywhere else in the book, it's odd that Soames assumes the principle and makes the inference without giving any argument at all, as if they are uncontroversial. And once again, responding to these arguments simply requires denying the principle, which is something that the analysis is already committed to doing, and which does not have any obvious costs.
Most of what I've said here applies strictly speaking to (2c) and (4c), which are de re ascriptions. Soames assumes incorrectly that the "hybrid" theorist (i.e. me) will treat these in exactly the same way that de dicto ascriptions are treated, presumably because he overlooks the separate analysis of de re ascriptions. (2d) and (4d) are prima facie de dicto ascriptions, though, so one might think that something further is required to see why they are not entailed by (2b) and (4b). I think the something further is pretty straightforward. Insofar as ascriptions involving indexicals and demonstratives typically entail corresponding de re ascriptions, and insofar as de re ascriptions require acquaintance-involving primary intensions, the natural conclusion is that the ascriptions involving indexicals and demonstratives typically have "appropriateness" conditions that require acquaintance-involving primary intensions. Once we've made this observation, combined with the observation that ascriptions involving names may have looser "appropriateness" constraints, we can see why (2d) and (4d) are not entailed by (2b) and (4b).
Soames gives some further related arguments on pp. 318-319, to which a similar diagnosis applies. Here he observes that if Venus can be given a name ('Hesperus') that grounds a priori knowledge that it is visible in the evening, presumably Mercury can be given a name that grounds a priori knowledge that it is not visible in the evening (let's say 'Mesperus', which functions to pick out the heavenly body which is not visible in the evening and ...). Then it looks like an utterance of (6) will be a priori, and that an utterance of (7) (of a relevant speaker) will be true.
(6) Hesperus is not Mesperus
(7) A knows a priori that Hesperus is not Mesperus.
From here, Soames infers the truth of the following de re knowledge ascription, where Venus and Mercury are assigned to the variables 'x' and 'y':
(8) A knows a priori that x is not y.
And he says that this conflicts with the "widely recognized fact" that the nonidentity of two objects, although necessary, is knowable only a posteriori.
The response to this is obvious by now. (The same goes for the related argument on p. 319.) The truth of (8) simply doesn't follow from the truth of (7). The inference once again requires the dubious exportation principle, which we have good reason to believe is false.
It's worth noting that even if the exportation principle were true, two-dimensionalism wouldn't be seriously threatened. As someone who is pretty flexible about those things, I'm prepared to allow that there are "loose" readings of de re ascriptions for which the exportation principle is true (given normal readings of de dicto ascriptions), because an acquaintance-involving primary intension isn't required to satisfy the de re ascriptions. I might even be prepared to allow that there are "tight" readings readings of de dicto ascriptions for which the exportation principle is true (given normal readings of de re ascriptions), because an acquaintance-involving primary intension is required to satisfy the de dicto ascription (even involving descriptive names). But insofar as I accepted the loose readings of de re ascriptions, I'd just deny the claim that the relevant sort of de re a priori knowledge is impossible. And insofar as I accepted the tight readings of de dicto ascriptions, I'd deny the inference from the apriority of a sentence to the truth of a corresponding de dicto a priori knowledge ascription. So even if Soames were to make a compelling case for the exportation principle (presumably involving a case for the "tight" reading of de dicto ascriptions), core two-dimensionalism would be fine. One would just need to tweak certain claims about attitude ascription around the edges.
There remains an interesting residual issue about a de re priori knowledge ascriptions to discuss, but I'll save that for another post.