Jon Cogburn's Blog

brief thoughts about the genesis of belief

I have this strange intuition that a lot of confusion in the philosophies of mind and language arises from the fact that representations are used to explain, first, two quite different activities: (1) planning, and (2) correction, and second, two quite different kinds of things: (3) individuals, and (4) groups.

Individual planning that leads to representational explanations occur when, for example, the crow goes offline and then suddenly performs some novel activity, for example bending the paperclip in a way that allows her to unlock the cage. The best explanation is that the crow was running out counterfactuals in her mind, counterfactuals that bear a systematic relation to causal facts in the world (see THIS POST and THIS POST for a link to a relevant discussions). Mark Okrent explains intentionality in this manner, and work on counterfactual and off-line reasoning by others can help fill in the details (my student Joel Okrent and I have just started a paper on this).

Group correction is more complicated. Of course, as Feyerabend said, when conversation breaks down, then we must resort to violence. But the enlightenment ideal that we can adjudicate differences without violence rests on what Crispin Wright calls "cognitive command," the view that when two people are disagreeing, there must be some deficiency in one of them or their environment. In Truth and Objectivity, Wright has a brilliant discussion of how this gives rise to representational pictures.

Recursive thinking in non-human animals

HERE. The book looks like it will contain too many problematic Chomskyan presuppositions for my tastes, but the review by Barbara King raises some really important points.

King calls the root of higher cognition "time travel." I know that some philosophers have been theorizing about "offline intelligence" and also the important role of counterfactual reasoning. (1) I think recursivity is a consequence of these deeper patterns (and I'm not at all convinced that all languages are recursive in Chomsky's sense, e.g. the Pirahã language), (2) I think that King's time travel examples show that higher animals do engage in kind of offline, counterfactual reasoning.

This is important to me because I think that offline counterfactual reasoning is exactly the place where it makes sense to attribut beliefs in a non-anthropomorphic way. This adds another layer to Mark Okrent's fine analysis of intentionality in Rational Animals, and my student Joel Musser and I are writing a paper about it. One of Mark Lance's colleagues is working on the importance of counterfactuals in basic reasoning, and I'm going to bug Mark for the guy's name and then e-mail him for a draft today.

For eliminativism about presupposition

I know there's an old Boer and Lycan paper about this very topic that I haven't read yet, so nothing I say here may be new.

I don't think that the linguistic phenomena of 'presupposition' really exists, but that in all cases what we're calling presupposition is really a mixture of semantic entailment, implicature, and possibly related discourse phenomena (such as the procedural knowledge that indefinites usually refer to things not introduced into the discourse yet).

First, the standard definition of presupposition doesn't work. Everybody holds that a sentence's presupposition is whatever information needs to been in the common ground in order for an assertion to be felicitous. So if I say, "Mary stopped smoking" it would only be a felicitous assertion if we both already know that Mary did smoke. 

Except when it's not. In every single case of presupposition one can imagine cases where the sentence is felicitously asserted without the "presupposed' information being in the common ground. Typically, pragmatics folks say that in such cases the hearer "accommodates" the information by adding it into the common ground. But this is nothing more than an admission that the definition does not work. Since there is no generative theory of accommodation, all it does is function as a place to throw recalcitrant counterexamples. It's as though Newtonians had said that Mercury was accommodating the sun and that light accomodated different inertial frames and left it at that.

Second, the biggest reason one needs a different theory of presupposition is because of "presupposition holes," where people can reliably infer material that is provably not entailed. So, in most cases, if "Mary stopped smoking" is the antecedent of a material or subjunctive conditional, we can still infer that she used to smoke, even though nothing is entailed by the antecedent of a conditional.

But I think this is purely a case of conversational implicature (at least with stop/start and definites; I need to look at other constructions). Asserting a material conditional implicates that you don't know if the antecedent is true. Asserting a counterfactual conditional implicates that the antecedent is false.

To see how this removes the need for presupposition, consider definite descriptions. Assume that the Russellian analysis holds. Then, inote that negation has a strong tendency to narrowly scope (and Michael and Ann Hegarty have a very good explanation involving discourse effects of why this is the case). Then the implicature of a counterfactual conditional with the antecedent that "The present king of France is bald" is that there is a present king of France and he is not bald. So there is an implicature that there is a present king of France! Likewise if the information that we don't know narrowly scopes. Then the implicature for a material conditional with the antecedent that the present king of France is bald is that there is a present king of France and we don't know if he's bald. Moreover, the way these implicatures cancel I think explains the cases that linguists call "acccomodation." This clearly works with "stop/start" type presupposition because in a rich enough lexical decomposition you will get exactly analogous places for wide and narrow scope negation.

We'd need to look at the way "presupposition" works with relative clauses, and also see if the account can handle "filters" where a presupposition is blocked. These are canonically conjunctions where the second conjunct asserts (or directly entails?) the presupposition of the first conjunct. I know in some of these cases the second conjunct is providing information about scope, i.e. "The present king of France is not bald, and there is a present King of France." This works fine if uttered in a context (Sherlock Holmes type induction scenarios) where the person might have thought you meant the first conjunct to have wide scope negation. My prediction is that in other cases the second conjunct is in a quasi meta-linguistic way working to tell us something about what implicatures the hearer should be drawing, and that the entailment still goes through anyhow (see the nice discussion in the last chapter on presupposition in Kadmon's book about this). That may be completely wrong though.

Idea I'm Pursuing

My big idea (which I'm sure must have occurred to others) is that many problems of reference just are linguistic analogs to the problem of the external world, and that Division 1 of Being and Time is structured to try to dissolve these problems. The big philosophy of language punch-line in Section 33 is “Statement is a pointing out which communicates and defines” (then you get some more stuff leading up to and including 44). Each of these three concepts (pointing out, predication, communication) is built up in earlier sections from non-linguistic modes. So 17 and 18 (on reference) provide the non-linguistic materials for characterizing pointing out, 25-27 (on the they) provide the non-linguistic materials for the characterization of communication, and 31-32 (on understanding and interpretation) provide the non-linguistic materials for the characterization of predication.

Making this work requires reading Being and Time along Harmanian lines though, in the first place because Heidegger's dissolution of the problem of the external world is thoroughly unconvincing unless you follow Harman and de-emphasize Dasein in your interpretation of the tool/broken-tool dichotomy.

The part of this that's cool is you can tie it to a lot of contemporary debates in analytic philosophy of language. The part that may not be cool is that a lot of Derrida may end up coming across as thoroughly retrograde, a linguistic Cartesian who can't get to reality and who has ignored the early Heidegger's way out of the prison.

All of this may go out the window once I've gotten a good grip on Braver's A Thing of This World, which has a very plausible interpretation of Heidegger leading up to Derrida. We're going to do an on-line reading group of it starting June 15th. I'll announce it in a seperate post soon.

3d post on Brandom/(Stalnaker?)- this one on de-dicto de-re

Everything below is only from Chapter 5 of Articulating Reasons (I've yet to read Chapter 8 of Making it Explicit), so this is going to be a bit crude.

On page 177 Brandom considers the following bit of dialogue.

A: The defense attorney believes a pathological liar is a trustworthy witness.
B: Not so; what I believe is that the man who just testified is a trustworthy witness.
A: Exactly, and I have presented evidence that ought to convince anyone that the man who just testified is a pathological liar.

This is important because it suggests what is going on with the de re reading of the first sentence. According to the distinction, the first sentence understood de dicto would be to attribute incoherent beliefs to the defense attorney, the belief that something can be both trustworthy and a pathological liar [note: that there are other possibly relevant issues with "a" functioning as a generic and not merely as a quantifier here; there is a de re reading where the person is talking about the reference of the generic type "pathological liar" and not referring to one or more; I propose for now that we set these aside and charitably take Brandom to just read it existentially.] The de re reading asserts something that may be true or false. These can be paraphrased, as Brandom does, in the following way.

de re: The defense attorney claims of a pathological liar that he is a trust-worthy witness.
de dicto: The defense attorney claims that a pathological liar is a trust-worthy witness.

Then Brandom goes on to give a very interesting account of what is going on with the "of" here as a way to distinguish between de dicto and de re readings.

My suggestion is that the expressive function of de re ascriptions of propositional attitude is to make explicit which aspects of what is said express commitments that are being attributed and which express commitments that are undertaken. The part of the content specification that appears within the de dicto 'that' clause is limited to what, according to the ascriber would (or i( a strong sense should) acknowledge as an expression of what that individual is committed to. The part of the content specification that appears within the scope of the de re 'of' includes what, according to the ascriber of the commitment (but not necessarily according to the one to whom it is ascribed), is acknowledged as an expression of what the target of the ascription is committed to.

This, according to Brandom is the reason the ascriber is free to substitute co-referring terms within the "of" clause but not within the "that" clause.

I think this is a cool idea, and provides the kernel of how one can claim that there are in fact no de re interpretations of beliefs (Silcox tells me that Stalnaker argued for something similar to that, I've got to get his paper).

Assume that the semantics only yields something that means what the "that" version of Brandom's rewrite is supposed to mean. Namely a sentence that is incoherent as it stands and not guaranteed such that co-referential substitutions preserve truth. But now supposed that both speaker and hearer can from the context tell that Sam the defendant is being picked out with the phrase "a pathological liar" in A's assertion. But then both speakers accept as a consequence of A's assertion  that B believes that Sam is a trustworthy defendant.

I think you can call this kind of thing "Kripke Implicature" because (if I remember right) it's what he says in reference to Donnellan's account of what's going on when we say of a man drinking water, "The man in the corner drinking a martini is a spy." For Kripke, we've said something false, but can still communicate something true to someone who even knows that the man is drinking water, just because "The man in the corner drinking a martini is a spy" logically entails that someone is a spy, and the pragmatics of the conversation picks out that someone to everyone's satisfaction. Likewise, "a pathological liar is a trustworthy witness" entails that someone is a trustworthy witness, and the pragmatics of the conversation pick out that someone to everyone's satisfaction. Then we can go on and debate whether or not that person is a trustworthy witness.

All I'm doing here is using Brandom's insight to argue that semantically there is only a de dicto reading of the claim, and that "de re readings" are really just sentences generated by inference from the de dicto reading plus pragmatic information. If this worked it would be a boon for semantics, getting rid of the need to deal with at least one kind of scope ambiguity (I hope something like it could be extended to more general "Every man loves a woman" type scope ambiguities to avoid going the underspecification route, the semantics generating the normal wide scope of the leftmost quantifier from the sytactic form, and pragmatics giving you the rest, but that's very, very iffy.)

Another interesting thing about Brandom's account is that the de re reading is not perfectly extensional. The only things that are substitutable in the "of" phrases are what the speaker takes to be co-extensional. I think that this is perfectly right and interesting in its own case. The last remnant of de re is a kind of relativized substitutability-within-a-given-web-of-belief-(and-acting). The inferential role of attributing beliefs (again, on this model it is always de dicto) to others is sensitive to their webs of belief.

But there is still an issue of how we get to norms concerning shared reference between webs. When we communicate we act "as if" we are all mean the same thing (and as Rorty writes, to say we disagree in meaning is nothing more than a decision to stop talking with one another). Maybe Brandom gets into that in Chapter 6.

quick thought about Brandom

I've been slogging through Brandom's Articulating Reasons: An Introduction to Inferentialism, and I think I'm just missing something with the big argument in Chapter 4.

Brandom tries to semantically differentiate singular terms from predicates solely by their substitutional behavior. His big conclusion is that valid material inferences involving substitution of singular terms are symmetric, whereas those involving substitution of predicates are asymmetric (not anti-symmetric!).

Take "X entails Y" just to mean that the in some primitive sense the inference from X to Y is licit. So "Charlie is happy" both entails and is entailed by "The small dog on Hearthstone Drive is happy." But "Frank is joyous" entails "Frank is in an emotional state," but not vice versa.

Weirdly Brandom uses definite descriptions ("the so and so") in his examples and counts them as singular terms. But indefinite descriptions ("a so and so") are not symmetric. "Charlie is happy" entails "A small dog on Hearthstone Drive is happy," but not necessarily the other way around. Their might be two small dogs on Hearthstone Drive such that the one other than Charlie is happy.

Footnote 15 on page 215 sort of addresses this, but it is really pretty opaque. In full it reads:

The restriction to substitution instances is required because one may, for instance, infer asymmetrically from the applicability of a singular term to the applicabilty of a predicate: from "The inventor of bifocals is Benjamin Franklin" to "The inventor of bifocals is an American." These do not count as substitution inferences even int he extended sense allowing replacement of frames, because they cross syntactic categorical boundaries.

Maybe I'm being completely dense, but I don't see how he can treat definite descriptions as being in the same syntactic category as names, and indefinite descriptions as not. Definite descriptions and indefinite descriptions satisfy all the same syntactic substitution tests. That is you don't go from grammatical sentence to ungrammatical sentence by replacing one by the other. And Brandom explicitly states that such tests are all that creates syntactic categories (as someone who works in computationally friendly syntactic frameworks, I like this view, though I have to note that post X bar theory Chomskyans would probably disagree).

This is actually a huge problem. In Montague Grammar proper names and noun phrases such as definitite and indefinite descriptions have the same semantic type. Each takes something of the type of a predicate (if we're extensional <e,t>) and returns something of type t (a truth value). Making this work was one of the deep insights that made a recursive syntax-semantics interface possible, because the you don't have different semantic types for things that are the same syntactic type. The fact that Brandom loses this is a big mark against his theory then.

Assume that Bradom can deal with the issue of indefinite descriptions satisfying all the same syntactic distributional tests as definite descriptions and proper names. There's a bigger problem. If we start with a language that does not include negation, then proper names can yield asymmetric inferences. Assume that our language contains five predicates and three of them are true of John and two of those three are true of Billy. Then for all basic predicates Phi, Billy Phi's entails that John Phi's, and not vice versa. You have uniform asymetric substitutivity.

If you have negation in the language and understand any open sentence to define a predicate then it does not follow. Let's say that "is happy" is the basic predicate that holds of John and not Billy. Then "Billy is not happy" does not entail "John is not happy.

Isn't that a little bit weird though? On his account, logical operators are supposed to be making explicit what is already implicit in our inferential behavior. I'm not sure this is really what's going on when "Billy" is a proper name in a language with negation but a predicate in a language without. I guess the question is whether you need that category to make sense of negation in the first place. I need to reread Sections 2 and 3 with this in mind though. Brandom might just say that this second problem shows something deep about the gloriously holistic nature of language (that might be part of what Sections 2 and 3 of the chapter are getting at; I don't yet understand them).

question for old friends about work on de dicto/ de re

Does anyone know if anybody has argued that de re readings of belief ascriptions are pragmatic rather than semantic?

That is, say a Montague style syntax/semantics interface generates only the de dicto reading (e.g. John is in the belief relation to the sense of "A unicorn runs") and that a de re reading (there exists an x such that x is a unicorn and John is in the belief relation to the sense of x runs) at best gets generated pragmatically from the de dicto reading and from information about relevant objects that are being kept track of at that point in the discourse.

I'll try to explain this better when I get some time. The whole thing would clearly be tied to Donellan's referential/attributive and Kripke's animadversions would also be relevant to motivating the use of the semantic/pragmatic disinction here.

Montague had two different syntactic derivations for de dicto and de re readings. Newer approaches use underspecification to not have to multiply syntactic forms when there is no distributional data to force this. But on this issue, my intuition is that there is only the de dicto reading in any case. I'm sure some philosopher must have had a similar intuition, but I don't know the literature at all. Before looking at the relevant formal pragmatics more closely (Michael and Anne Hegarty and Jerry Weltham and I are doing a reading group on this excellent book this summer), I want to find the relevant stuff about propostional attitudes.