Stalnaker on Lewis on Intentionality

Stalnaker's "Lewis on Intentionality" (AJP 82, 2004) is a very odd paper. The aim of the paper is to show that Lewis's account of intentional content as developed in "Putnam's Paradox" -- global discriptivism with naturalness constraints -- faces various problems and conficts with what Lewis says elsewhere.

The first thing that's odd about this is that in "Putnam's Paradox", Lewis doesn't develop an account of intentional content. Rather, he discusses Putnam's model-theoretic argument and suggests that if one holds something like global descriptivism about linguistic content, adding external naturalness constraints on the interpretation of predicates would be an attractive way to block Putnam's argument for underdetermination.

As I've mentioned previously, it is clear from Lewis's many writings on the philosophy of language -- like Convention, "Languages and Language" and On the Plurality of Worlds, pp.40--50 -- that global descriptivism (with or without naturalness constraints) is not Lewis's own account of linguistic content. On Lewis's account, the correct interpretation of our sentences is not determined by what makes some set of sentences true, but rather by the behaviour, expectations and intentions of the relevant community. Assigning the proposition that there are cats to the English sentence "there are cherries" is ruled out simply because it doesn't fit our usage. Thus Lewis notes at the beginning of "Putnam's Paradox" (pp.57f. in Papers),

I shall acquiesce in Putnam's linguistic turn: I shall discuss the semantic interpretation of language rather than the assignment of content to attitudes, thus ignoring the possibility that the latter settles the formes [as Lewis himself believes]. It would be better, I think, to start with the attitudes and go on to language. But I think that would relocate, rather than avoid, the problem; wherefore I may as well discuss it on Putnam's own terms.

Apparently, Stalnaker read this passage -- the only place in "Putnam's Paradox" where Lewis mentions the interpretation of mental states -- as saying that the account of linguistic content to be sketched later in the paper is really meant to apply to mental content. But Lewis doesn't say that. He only says that Putnam's underdetermination problem also applies to the interpretation of mental states. The last remark in the quoted text also indicates that Lewis thinks the solution will be similar in both cases. The solution suggested in "Putnam's Paradox" is to employ objective naturalness as a constraint on correct interpretations. And indeed he also employs such a constraint on the interpretation of intentional states. But it is certainly odd to infer that Lewis intended all his later remarks on linguistic content -- the descriptivist framework; the rejection of external causal constraints on reference; etc. -- to carry over to the interpretation of intentional states. Even stranger is it to say, like Stalnaker, that Lewis here develops his account of intentional content.

Stalnaker's interpretation is particularly odd because Lewis has extensively written about his views on mental content, e.g. in "Radical Interpretation", "Veridical Halluzination and Prosthetic Vision", On the plurality of worlds pp.27-40, "Reduction of Mind", and elsewhere. And even more than in the case of his theory of language, it is plain that the account of intentional content developed in these writings bears little resemblance to global descriptivism. Stalnaker himself notes that if there is no language of thought, which Lewis took to be an open question, it's hard to see how the interpretation of mental states could proceed by ramsification of mental sentences and by assigning natural classes to mental predicates. But he takes that to be a problem for Lewis, rather than a problem for his understanding of Lewis.

As indicated in "Putnam's Paradox", Lewis also sees an underdetermination problem for intentional content, and again he appeals to naturalness constraints to rule out deviant interpretations. But for mental content, he also employs various other constraints, among them causal ones: the contribution of perceptions to belief content is fixed mainly by the phenomena that cause the perception.

Stalnaker notes this as well, and suggests (on p.206) that in "Putnam's Paradox", Lewis took back the causal constraint in favour of causal descriptivism, i.e. in favour of adding the constraint as a clause into the theory under consideration. But firstly, Lewis uses the external causal constraint in "New Work for a Theory of Universals" (p.50 in Papers) and On the Plurality of Worlds, which appeared shortly before and shortly after "Putnam's Paradox", respectively, and he still uses it much later, e.g. in "Reduction of Mind" (1994) and "Naming the Colours" (1998). So at best one could claim that for a very short period of time around 1984 -- a period in which he wrote virtually nothing on intentional content --, Lewis briefly jettisoned a major part of his account, without anywhere pointing out this radical change of view.

Secondly, what would it actually mean to replace Lewis's causal constraint on intentional content by a further clause in "the theory"? Presumably it would mean that when interpreting intentional states, we shouldn't assume that persons usually form true beliefs about the observable features of their surroundings, but only that they believe that persons usually form true beliefs about the observable features of their surroundings. It's hard to see how this peculiar view could be motivated by anything like the (very good) reasons that favour causal descriptivism about names over the causal theory of reference.

Finally, there's no need to speculate what Lewis might have had in mind when he said, in the passage quoted from "Putnam's Paradox", that the underdetermination problem also (or rather, mainly) applies to mental content, and how he thought the solution for the relocated problem would look like. In a footnote to the quoted passage he writes:

For a discussion of the 'relocated' problem and its solution, see the final section of my 'New Work for a Theory of Universals' [...].

Looking at that section, we find the very same non-descriptivist account of intentional content Lewis puts forward in all other places, including the external causal constraints (there called "principles of fit on the input side").

So Stalnaker assumes that Lewis develops an account of intentional content in a paper that doesn't discuss intentional content at all; and that this account is Lewis's own preferred account, despite the fact that it contradicts all his numerous earlier and later writings on intentional content, including the presentation to which readers interested in intentionality are directed in the paper where the account is allegedly developed.

Odd, isn't it?


# on 02 November 2004, 16:01

I agree that Lewis isn't a global descriptivist, either about language or about language-and-thought. But I have some worries about the view you attribute to him. In particular, the way he spells out his proposals doesn't seem to lend itself to external causal constraints being in place.

The passage of NWTU you cite says that causal constraints on content attribution are 'principles of fit'. On the very next page he summarizes all such principles of fit in terms of a Bayesian framework: credence and value functions C,V, plus a stream of evidence E. The sole fit-constraint on interpretation is then that the behaviour that occurs when E impacts should be rational by the lights of C and V, when updated by E.

The output 'principles of fit' are encapsulated by the requirement that behaviour be rational by the lights of updated credence and value functions. The input 'principles of fit' are encapsulated by the need to update by E.

How is E (a proposition) to be picked out? If it's picked out in causal terms, the case for external causal constraints is made. But I don't think this can be the story.

Suppose I'm going to wave a red flag if I discover I'm in earth, and wave a blue flag if I discover I'm in twin-earth. I see some water, which (since I'm in fact in earth), is typically caused by H20. The proposition P: that H20 is present, excludes the twin-earth world. Therefore if I get P as evidence, I should be waving my red flag (that's the sense of 'evidence' which Lewis needs to make his story run). Clearly, I'm in no position to wave the red flag, just by virtue of seeing the water, though.

This shows that the evidence I gain from seeing water (in the sense which Lewis needs) is not the proposition that H20 is present. Hence, it is not the proposition which corresponds to the state of the world which typically causes my experience.

Moral: the content of evidence (in Lewis' sense) shouldn't be specified by means of what typically-causes experiences/beliefs. Something more subtle must be going on.

Let me add that I don't see how this fits with Lewis' earlier talk about causal constraints being 'principles of fit', and I don't see how to give an alternative story about latching onto E. (An analogy, in the jargon: we want to latch onto diagonal content, rather than the horizontal content: but how to spell this out?)

# on 03 November 2004, 12:37

(Disclaimer: this may be reinventing the wheel. I must go back and read Jackson, Stalnaker, Chalmers etc to see what they say about these issues.)

Just to follow up on the final suggestion in the above. Suppose that we can individuate experiences by 'subjective indistinguishability'. (Let's suppose also that we can set up this relation without invoking the content of experience). Moreover, suppose we also assign individual experiences content via causal connections. So my experience as of water being present, in a world where the watery stuff is h20, has the content that H20 is present. Simillarly, twin-me in twin earth has counterpart experiences, but they have the content that XYZ is present... the usual story.

Now, write up the 2D matrix generated by the contents of the experiences that I and my experiential counterparts undergo. The horizontals are different for the different counterparts; and I do not know which I am. So, by a Stalnaker-style story about communicative content, what my experience communicates to me is not the horizontal, but the diagonal proposition. Call this the evidence that my experience provides. E=diagonal of experiential content.

The evidence I get from my experience we can roughly characterize as: there is watery stuff present. This will not eliminate twin-earth worlds, so it finesses my earlier objection. Plus it grounds the content of evidence in a causal way.

But notice that the causal story is pretty indirect, even for fixing the content of evidence (which is then fed into the story about the fixing of belief/desire content). It won't give me beliefs 'about' water/H20. (It's interesting that Lewis' example in NWTU concerns a paradigmatically secondary property: roughly, if a belief is regularly caused by the experience of red things, then it is a good candidate for having the content that red things are present. )

One thing that this story certainly doesn't support is the (perhaps natural) thought that if I have a perception whose content is that water is present, then by 'endorsing the perception' I can get a belief whose content coincides with this.

To summarize: we have a little spin on wide/narrow content. Experiences only have 'wide' content, Beliefs only have 'narrow' content, and the contents of beliefs and experience are linked via the content of the evidence that experience provides. This feeds into beliefs in a Bayesian way, and is fixed by the content of experience via a Stalnakerian 2D story.

# on 03 November 2004, 17:42

I agree that the propositional content of evidence E in NWTU can't be something like the proposition that there is H2O nearby. As you say, conditionalizing on this content would mean that the subject with evidence E believes that there is H2O nearby, which should manifest itself in corresponding behaviour (waving flags, uttering "there is H2O nearby", etc.); but many subjects who are looking at water don't show that behaviour. Moreover, Lewis takes it to be part of folk psychology that I and my twin-earth twin and my vat-twin share the same beliefs ("Reduction of Mind"). He would also agree that we have exactly the same evidence: this is why, strictly speaking, I don't know I'm not a brain in a vat ("Elusive Knowledge").

But it doesn't follow, I think, that "the evidence I gain from seeing water [...] is not the proposition which corresponds to the state of the world which typically causes my experience". That only follows if the proposition which corresponds to the state of the world which typically causes my experience is something like the proposition that there is H2O nearby. But it could be the proposition that there is watery stuff nearby. This would satisfy both the causal and the Bayesian constraints. (It also satisfies the vat condition because what counts is not typical causes of the evidence in the relevant individual, but typical causes in all, or perhaps all actual, members of the relevant species.)

I think the epistemic counterparts you describe in your second comment correspond to what Lewis takes to be the object of knowledge in EK: the content of my knowledge is the class of my epistemic counterparts. But Lewis says that this class, this centered proposition, is not the content of my evidence, nor presumably of my belief. That's good, as otherwise the causal condition on belief content would have to be jettisoned: by that condition, the content of my current evidence includes that there is a table in front of me; but this is not true for all my epistemic counterparts, not e.g. for my vat-counterparts.

Similarly, it seems to me that your proposal to take the diagonal of externalist content as E's content won't satisfy the causal condition. E's content wouldn't rule out that I'm a brain in a vat.

So I believe Lewis's conditions are satisfied under a certain choice of evidence content. I'm afraid the best way to characterize that choice is to say it's the one that satisfies the conditions.
But certainly Lewis isn't very clear about all this.

BTW: Multiple comment submissions should no longer happen now. Sorry.

# on 03 November 2004, 20:38

Just a note on a couple of the things. What?s the causal constraint you mention? I thought the challenge was to work out how causality enters Lewis? picture at all, given the centrality he gives to the decision theoretic mechanisms. And all I?m claiming of the above account is that it shows one (indirect) way in which a causal accounts of content ends up fixing the content of evidence, which in turn constrains beliefs etc. Also, I?m confused why the issue of epistemic counterparts comes up? (there?s the problem of specifying who my experiential counterparts are, but that?s quite different).

# on 14 December 2004, 20:58

I shouldn't trust my memory. I've checked "On the Plurality of Worlds" and Lewis does *not* mention the causal input constraint there. He even makes it sound like the further principles that are needed besides instrumental rationality (which is not Global Descriptivism) are just principles of naturalness and normality. I'm not sure what to make of this. Maybe he meant these to include the causal constraint (he points at "New Work"), or maybe he had forgotten about it, or maybe he omitted it for the sake of brevity (he had to mention the naturalness constraint because he needs it later, in section 2.3). Or perhaps he really jettisoned that constraint for a few years around 1986, without pointing out why.

# on 04 January 2006, 21:52

Just for the record (answering my previous comment), the input constraint is there, on p.106 of Plurality: "A man [...] has a thought with a certain content in virtue of being in a state which occupies a certain functional role. This definitive functional role has to do with the causal relations of that state to the thinker's sensory input, his behavioural output, and his other states."

# on 25 January 2006, 18:19

That's an input constraint, but is it a *causal* input constraint? I.e. does it require that we have some *causal* story about what fixes the content of the perceptual state that is input to the belief system?

It looks to me like the quote just requires there be a fit between the contents of sensory input, beliefs/desires, and outputs. These things have to be causally connected to each other, in order to *be* beliefs, desires etc (that follows from a general functionalism about mental-state *types*). But that's quite different from offering some direct causal constraint on perceptual content.

# on 25 January 2006, 22:37

Hi! Well, the quote says that causal relations to sensory input (not just some kind of fit) is part of what determines that a state is a thought with a certain content. Do you read this as saying that the causal relations to sensory input only determine that the state is *a thought*, whereas something else -- either the relations to the other states or to behavioural output or both -- determines that it is a thought *with this and that content*? That sounds very weird to me.

(I have to admit that I don't really understand how one can be a functionalist about mental state types without also being a functionalist about mental content. It seems to presuppose that there is one causal role shared by all beliefs, no matter their content, so that we can say, "states with that causal role are beliefs". But I think different beliefs will usually play quite different causal roles, will, for instance, lead to quite different behaviour if combined with the same desires. If so, identifying a state's causal role will not only identify it as a belief, but also narrow down the possibilities for its content. But I've probably missed something here.)

# on 26 January 2006, 16:00

I read the quote as committing to a functionalism about content. I just thought that the qutoe didn't commit itself one way or the other on the sort of issues discussed earlier in this thread. In particular, there's no implication that the content of the sensory inputs is constrained to match the state of the world that typically causes that kind of sensory input. At most, what it requires is that the content of a belief is constrained to match the content of a perceptual state that typically causes it.

It's compatible with what's been said that the story goes in the reverse direction: that perceptual states have content in virtue of the content of the belief states that they typically cause, and that these in turn have content in virtue of rationalizing actions. (Or, if you want to be holist about it, that perceptual content belief content and desire content are together determined holistically as that overall story about mental life that makes best sense out of bodily behaviour). I don't think that that is a terribly attractive story: but the quote does nothing to rule it out.

What you need is a quote that appeals to causal connections between world and mental states, rather than causal connections between mental states and mental states, in the story about what fixes mental content.
(p.s. something odd is going on with the comments box in internet explorer: it runs off the side of the page.)

# on 26 January 2006, 19:14

oops, thanks for telling! Hope I've fixed that now.

Ok, you read "sensory input" as meaning perceptions, i.e., some kind of mental state. I thought it means stimulations of sense organs. Isn't that the more common usage? On this interpretation, Lewis would say that it is part of the folk psychological role of, say, believing that it is terribly cold that this belief is typically caused by certain sensory stimulations, namely the kind of stimulations which are typically caused by it being terribly cold. This would also match what Lewis says elsewhere (like in "Veridical Halluzination" and "Naming the Colours"), and it would attribute to him a view that makes a lot more sense than the idea that the content of perceptions has nothing to do with their typical causes.

The point of mentioning sensory input instead of more distant causes is perhaps to cancel out differences in wide content: since nearby water and nearby twater stimulate my senses in the same way, this difference will not induce a difference in my beliefs, unlike the difference between nearby water and nearby fire. That is, it is part of folk psychology that a person who is undecided whether some box contains a) water or b) twater or c) fire will, just by looking inside, come to rule out either both (a) and (b) or (c); but it is not part of folk psychology that she will be able to tell apart (a) and (b) just by looking.

# on 27 January 2006, 15:12

Ok, I see how you're thinking of it now: sounds reasonable. I had in mind the passages in "New work" (p50-51) where the functional role of belief/desire is specified in part via updating credence functions via a "stream of evidence", which is itself specified by a proposition. This stream of evidence is referred to there as "sensory evidence", and I was taking "sensory input" to be the same thing.

# on 01 February 2006, 11:12

Hmm, I think you're right. It's better to construe the sensory input as the evidence upon which we conditionalize when making observations: everything else makes no difference to our beliefs. Stimulations of sense organs that remain unnoticed by the subject, for example, will not lead to corresponding beliefs. But the relevant input proposition is not the content of the preceptual experience (which might rather be defined in terms of the beliefs it causes), but something like the proposition that this experience occurred. That means charity principles play a far more important role than I used to think, while external causal input constaints are less important. Interesting.

Sorry for just hinting at obscure directions. I'll try to write down later how exactly I now think the story goes.

Add a comment

Please leave these fields blank (spam trap):

No HTML please.
You can edit this comment until 30 minutes after posting.