Response to Scott Soames (Central APA)

At the April 2006 meeting of the Central Division of the American Philosophical Association, in an author-meets-critics session on Scott Soames' book Reference and Description: The Case Against Two-Dimensionalism, I presented a comment on Soames' book, "Scott Soames' Two-Dimensionalism". The other critic was Robert Stalnaker. Soames presented his response to critics. Below is a reply to Soames' response to me, for those who were at the session and interested others. Note that this response was mostly written before the session, except for one or two paragraphs where the discussion in the session is mentioned.

Changes in View

Soames makes something of changes in my view between 1996, when The Conscious Mind was published, and now. As I said in my comment, I think the 10-page discussion of 2Dism in the 1996 book is not entirely satisfactory, but I think the evolution in my views can be overstated. For example, the epistemic interpretation of 2Dism was present in the 1996 book, although it was not adequately distinguished from the contextual interpretation. But in any case, the relevant elements of my current view are all present in the 2002 papers "On Sense and Intension" and "The Components of Content", which have far more extensive discussion of two-dimensionalism than the 1996 book, and which are discussed by Soames at length in his book.

On the epistemic interpretation of 2Dism: Soames characterizes this as a move from semantic content to private thought. I would not characterize it this way. My conception of semantic content is closer to Frege's than to Soames's: I hold that there is an important aspect of meaning that is constitutively connected to the epistemic domain. A consequence is that (as on Frege's view) the relevant aspect of meaning can sometimes vary between utterances of the same sentence. But this does not entail that this aspect of meaning is non-semantic, unless one presupposes Soames' conception, on which semantic content is by definition non-variable between speakers.

Of course this issue turns mostly on the terminological issue of how one uses "semantic". The substantive question is whether the entities I appeal to can play many (not necessarily all) of the explanatory roles that theorists have wanted meanings to play. In particular, what matters is whether they have the various specific properties that I attribute to them, which ground their role in (among other things) the analysis of the necessary a posteriori, the connection between conceivability and possibility, of various Fregean phenomena, of narrow content, of belief attributions, and so on. The variability of these 2D semantic values does nothing to compromise their ability to play these roles.

So, although Soames suggests that elements of my view have evolved in a direction that he approves of, it's important that nothing in this evolution undermines the key purposes for which I have wanted to use the two-dimensional framework all along.

Descriptivism and Scrutability

Soames briefly discusses my discussion of the passage from Jackson that he discusses in his book. I suggested that the force of Jackson's point was not that two-dimensionalism is irrefutable, but that it cannot be refuted by a certain style of argument, the method of cases (using our judgments about cases to suggest that names and descriptions give different results in those cases), as two-dimensional intensions can always be found that capture our judgments about those caes. This applies especially to Kripke's modal and epistemic arguments against descriptivism, which I argued have no force against two-dimensionalism for this reason.

In his response, Soames says I defend the claim that Kripkean thought-experiments support two-dimensionalism, and argues that they do not, as Kripke does not describe his cases in an appropriately neutral language. However, this wasn't the point: the point was that Kripke's arguments cannot refute two-dimensionalism, not that they support it. This point is unaffected by Soames' discussion.

Now, Soames may think there are some further considerations that tend to refute two-dimensionalism, for example because the scenarios in question cannot be described in neutral language. But this would be a quite different argument that goes well beyond Kripke's arguments, and Soames does not attempt to provide that argument here. (Such arguments, if provided, would be a very worthwhile contribution to the debate.) In addition, it is worth noting that characterizations of neutral language is required only for the version of epistemic two-dimensionalism that identifies scenarios with centered worlds. For the version that constructs scenarios using the epistemic modality, only a much weaker scrutabilty thesis is required.

It is also worth noting, as in the paper, that 2Dism doesn't require scenario-specifications in a purely physical and phenomenal vocabulary. It also doesn't require the strange claim that "no reference-fixing description contains any unanalyzed term", as Soames suggests. Perhaps he meant "any unanalyzed Twin-Earthable term" or something along those lines. This would be closer to correct, though one would need exceptions for indexicals, and I would put the point in terms of scenario specifications rather than reference-fixing descriptions.

Soames' Two-Dimensionalism

Much of Soames' response is devoted to his system of epistemically possible world-states (EPWs) and metaphysically possible world-states (MPWs), which I had argued can be seen as an interesting form of two-dimensionalism in its own right, and in particular can be seen to yield many of the core theses of epistemic two-dimensionalism.

Not surprisingly, Soames resists the claim that he is a two-dimensionalist. The first few paragraphs of his discussion are devoted to arguing that EPWs and MPWs do not correspond to contexts and circumstances in his framework, as a two-dimensional interpretation would require. I think this discussion misses the point. I did not suggest that Soames' system was analogous to contextual two-dimensionalism — I suggested that it is analogous to epistemic two-dimensionalism, with epistemic possibilities on one dimension and metaphysical possibilities on the other. I don't say that this is the only way to regard his system, or that it exhausts the structure in his system. I merely argued that it is possible to recover this structure from his system, with entities and intensions that satisfy the core epistemic two-dimensional theses E1-E9.

Soames ends up rejecting this mapping on the grounds that the central principle E5 (holding that a priori truths are true at all epistemically possible worlds), although true in most cases, is false in certain special cases involving the 'actually' that I discussed in my paper. Now, I think that this rejection of E5 in some cases is quite consistent with seeing Soames' system as a sort of two-dimensionalism: after all, the leading two-dimensional systems of the 1970s (Kaplan, Stalnaker, Evans, Davies and Humberstone) themselves had a connection between apriority and 2D semantic values that holds only in certain special cases. So Soames' system (with its two basic modal operators and corresponding modal spaces, and an attenuated connection between one of these spaces and apriority) can still be seen as two-dimensional in a sense at least as strong as that in which these systems are two-dimensional.

But in any case, I think that Soames' discussion of cases involving 'actually' is interesting and problematic in its own right. This issue yields the meatiest material in the exchange, so in the remainder of this piece I'll focus on it. In my paper I raised issues about the behavior of 'actually' in Soames' system, yielding two inconsistent tetrads, such that Soames must reject at least one thesis in each. I offered some suggestions about what I thought were the best options. Soames takes different options, thus circumventing the problems of the other options, but opening several new cans of worms. I will consider these cans in turn.

(1) Maximal world-states. In the first tetrad, Soames denies thesis (i): that there exists a proposition P such that @P (i.e. [in @, P] where @ is the actual world-state) is necessary a posteriori. He holds that whenever P is true, @P is a priori: one could refer to the world-state @ in principle by giving a fully detailed specification of it, and this specification could allow a subject to infer directly that P is true, so that they could know [in @, P] a priori.

This is an interesting idea, but to assess it, we need to know more about Soames' system: in particular, we need to know just what is built into "world-states", and into the "actual world-state". Soames says that these are given by "maximal world-describing properties", but he does not say much about what these are supposed to be. Presumably, these are properties along the lines of [being such that P1 & P2 & P3...] for some class of propositions that is appropriately "maximal". But Soames does not say what it is for a property or a class of propositions to be "maximal". Some candidates:

(i) The class of all propositions true at a world — but then all sorts of paradoxes threaten.

(ii) A metaphysical supervenience base, i.e. a class of propositions whose conjunction metaphysically necessitates all propositions true at a world (e.g. if physicalism is true, this will be a specification of all true propositions in fundamental microphysics). But if this is what @ is, then Soames is explicitly committed to rejecting the claim that [in @, P] is knowable a priori for all truths P.

(iii) An a priori scrutability base: i.e. a class of propositions such that all true propositions are a priori entailed by their conjunction. [In the session Soames seemed to advocate something like this option -- see the discussion under (3) below.] This will certainly ensure the apriority of [in @, P] for all truths P. But this will be ensured merely by stipulation — nothing in Soames' discussion of "maximal world-states" suggested a role for apriority in their definition. It's also not obvious that there will be such a class. It's even less obvious that there will be a single canonical such state.

(iv) A logical entailment base: i.e. a class of propositions such that all true propositions are logically entailed by their conjunction. Similar issues arise.

(Many of these options also run into problems with point (3) below, tied to Soames' view that ~P&@P is true at some EPWs even though [P iff @P] is a priori. On option (i) above, this combination is ruled out: if the specification of a world-state W includes ~P&@P, the hypothesis that W obtains will be ruled out a priori, so W will not be an EPW. Something similar goes for options (iii) and (iv).)

An additional problem here is that for many (perhaps all) of these proposals about world-states, it seems far from clear that we can , such that we can "primitively ostend" the actual world-state, as Soames requires in his argument for the thesis that [P if actually P] is a priori. Presumably this ostension would involve something like a complex demonstrative — "that Q-maximal world-state", for some theoretical conception of "Q-maximal", and some sort of pointing out at the world. But for many of the proposals above there will be multiple Q-maximal states, so it's not clear how one ostends just on this way. And more fundamentally, it doesn't seem especially plausible to me that states like this are the sort of thing one can just ostend. What seems perhaps ostendible is that actual world itself — not a Q-maximal state, but the entity that has the state. (See Richard Chappell's nice discussion of this distinction: "The actual world is not a possible world".) But then, one will merely descriptively identifies the state as something like 'The Q-maximal state of this world'. Understood this way, Soames will then have to say that the Russellian proposition expressed by 'actually P' has the actual world as a constituent (not any of its states), and his argument for the apriority of @P will then fall away.

(2) Closure of a priori knowability. Another problem: Soames holds that for all truths P, [P iff @P] is a priori, and @P is a priori. One might think it follows that for all truths P, P is knowable a priori. But this is an unacceptable result, and one that Soames rejects. So Soames is committed to denying the closure of a priori knowability under known a priori entailment, under known logical entailment, and so on. (He is presumably even committed to denying single-premise closure, as he must also hold that the conjunction [@P & [P iff @P]] is a priori knowable, even when P is not a priori knowable.)

This is at least an awkward result. Such a result occasionally crops up for other post-Kripkean theorists, but Soames' own system is otherwise pleasingly free of it. Once we have this result, there is an immediately need to appeal to further distinctions, such as guises under which propositions are entertained, modes of presentation, and so on, to distinguish cases where closure holds from those in which it does not. Soames does not usually have much time for guises and modes of presentation, but now it appears that he owes us an account of them, or of some other theoretical apparatus that can do the relevant work.

(3) Apriority and epistemically possible world-states. In the second (and more important) inconsistent tetrad, Soames denies thesis (vi), which says that when Q is a priori, Q is true at all EPWs. This came as a surprise to me, and I think it opens up the biggest cans of worms. In the paper, I'd said that this move would be very unattractive, as breaking the connection between epistemic possibility and apriority — i.e., the connection between the epistemic possibility of world-states and the epistemic possibility of propositions — tends to undermine the motivation for the framework of epistemically possible worlds in the first place.

To say more: here's a very natural conception of the framework, suggested by Soames' own discussion at many points. When one comes to know a proposition P, one thereby rules out all world-states in which ~P. When one comes to know a proposition P a priori, one thereby comes to rule out all world-states in which ~P a priori. If P is knowable a priori, and W is a world-state at which ~P, then one can rule out the hypothesis that W obtains a priori. If so, then (by Soames' own definition of an epistemically possible world-state, as one who's obtaining cannot be ruled out a priori), W is not epistemically possible. It follows (setting aside issues about indeterminacy) that if P is a priori, P is true at all epistemically possible worlds.

Soames rejects this consequence, so he must reject the conception. In particular, it looks like he must resist the claim that when one comes to know P, then one thereby rules out all world-states in which ~P. He must even reject the weaker claim that when one comes to know P with certainty, one thereby rules out all world-states in which ~P (issues tied to fallible knowledge claims are not really relevant here, as presumably one can know [P iff @P] with certainty). But then, much of the point of the framework is undermined. At least, Soames owes us a motivated account of how the framework is supposed to work given this denial, and of what the relation between knowing a proposition (a priori or otherwise) and ruling out world-states in which the proposition is true is supposed to be.

At this point, I think one can use an argument style that Soames deploys against Stalnaker to undermine Soames' own view. In discussing e.g. our knowledge that a paperweight is made of metal, Soames objects to account which explain this knowledge in terms of ruling out world-states in which some other paperweight is made of wood — he says that in coming to know this proposition, we rule out world-states in which this paperweight is made of wood. Something similar goes knowing a priori that Q, where here Q is [P iff @P]: in coming to know Q a priori, I thereby rule out world-states in which Q itself is false, and rule in only those states where Q itself is true. In the paperweight case, it seemed incorrect to say that our knowledge of a proposition ruled out world-states where the proposition is true; in this case, it seems equally correct to say that one's a priori knowledge of Q rules in world-states in which Q is false.

Soames appeals here to certain special properties of [P iff @P] and in particular the claim that it can only be known by subjects in the actual world @. But this appeal seems to be a non sequitur. The ability of subjects in other worlds to know the proposition in question is irrelevant — the issue turns on what is known and ruled out by subjects in our world who know a priori that [P iff @P]. These considerations are quite general, and nothing about the knowability or otherwise of [P iff @P] in other worlds undermines them. So Soames hasn't really done anything to justify his rejection of the general argument for E5, and this rejection threatens to undermine the power and coherence of the system of epistemically possible worlds.

(4) Truth of propositions at world-states. Furthermore, Soames' rejection of thesis E5 raises questions about what it is, on his framework, for a proposition P to be true at a world-state W, and especially at world-states W that are epistemically possible but metaphysically impossible. Some candidates:

(i) P is true at W if [if W, then P] is metaphysically necessary [i.e. if it is metaphysically impossible for P1&P2&... to be true while P is false, where P1, P2, etc are the propositions specified in W]. Problem: when W is metaphysically impossible, both P and ~P will be true at W.

(ii) P is true at W if [if W, then P] is epistemically necessary, or knowable a priori. Problem: when P is a priori, the conditional proposition is a priori, so P will be true at all EPWs, contrary to Soames' claim.

(iii) P is true at W if [if W, then P] is a logical truth. But now, it will be very difficult for @P to be true at worlds where P is false, as there is no candidate for a relevant logical truth.

(iv) P is true at W if a subjunctive conditional proposition [if W, then P] is true. Problem: subjunctive conditionals are notoriously underdeterminate, and are especially problematic when antecedents are metaphysically impossible. Even if one holds that they are not always vacuously true in those cases, it is extremely unclear that they are determinate enough to provide a foundation for evaluation here.

(v) P is true at W if some primitive relation holds between P and W: then Soames needs to say a lot more to make the case for this relation and to explain how it works.

[In discussion at the APA session, it appeared that Soames wants to say that maximal world-states are those that are in a certain sense epistemically complete, i.e. that settle the truth-value of all propositions a priori (this is something like the third option under topic (1) earlier). This tends to suggest the understanding (ii) above. To avoid the problem raised there, Soames suggested a more complex understanding of truth of a proposition at a world: roughly, the understanding above for non-modal propositions, and a further clause for modal propositions, holding that their truth at a world-state depends not just on that world-state but on the system of world-states as a whole. I think this separate treatment of nonmodal and modal propositions opens further cans of worms, but a proper discussion awaits a fleshed-out version of Soames' view.]

(5) The apriority of [P iff @P]. A comment on Soames' argument for the apriority of [P iff @P], based on subject's ability to primitively ostend the actual world-state. First, it is worth noting that nothing in this argument turns on the properties of @. The argument suggests that for all worlds W, [P iff in W, P] is knowable a priori: a subject in W could come to know this proposition by the same a priori method of ostension. This is at least a striking and somewhat bizarre result, and brings out even more strongly the failure of a priori closure on Soames' view.

One might respond by noting that in at least some worlds, there will be no subjects (or none with appropriate capacities), so that the relevant propositions involving those worlds will not be a priori knowable. In response, one could restrict the thesis to the many worlds in which subjects are in fact entertaining and coming to know the propositions in question, and this will still be a striking result.

But closer to home: the same issue arises in our world. Most propositions P are presumably such that in @, no-one ever entertains [P iff @P], let alone comes to know it a priori. But if such a propositions is not known a priori in @ by Soames' method, it is impossible that it is known a priori at all by that method: the proposition certainly cannot be known in another world W by that method, as in that world, the method of ostension will simply yield a priori knowledge of the different proposition [P iff in W, P]. So it turns out that Soames' argument gives us no reason to think that it is possible to know [P iff @P] a priori for all P. Assuming that Q is a priori knowable iff it is possible to know Q a priori, then Soames' argument gives us no reason to hold that [P iff @P] is a priori knowable for all P. At best, [P iff @P] will be a priori knowable for that small class of propositions P such that [P iff @P] is a priori known.

Conclusion: I think it is clear that Soames' response to the dilemmas in my commentary raise many more problems than they resolve. I continue to think, as in the commentary, that it would be much better for Soames to either (i) retract his argument that [P iff @P] is knowable a priori, or (ii) allow that for all non-actual metaphysically possible world-states (world-states in which [P iff @P] is false for some P), these world-states are epistemically impossible. the resulting systems will be enormously better-behaved. Of course, there will then be no reason for Soames to deny principle E5 or its consequences. So his view will then be all the closer to epistemic two-dimensionalism.

Go to: