The Access Problem for Holes

[Hello to anyone out there!

Project: exposure therapy for my fear of posting sloppy/confusing stuff to this blog continues. So uh... sorry. I hope to return to regularly posting polished content soon.]

One of the least popular aspects of my philosophy of mathematics  is what I call Moderate Quantifier Variance. So I thought I'd explain one reason why I'm such a big fan of this view (and why I think you should be too) which has nothing to do with philosophy of mathematics.


The Access Problem for Holes

Consider our knowledge of holes. As discussed in D and S Lewis' On Holes, it's appealing take holes to be distinct from things like the air inside them, portions of the matter that `hosts' (e.g. portions of cheese with some diameter aronud a hole in Swiss cheese).

We seem to know how deeply a piece of cheese must be indented in order for there to count as being a hole in that piece of cheese. But (to a crazy philosopher, for a second) this knowledge can seem rather mysterious. For our senses tell us how the cheese is distributed through space. But how do we know where to draw the line: re how shallow a hole can be? No sensory experience seems to point out a single metaphysically special place to draw the line. So how can one explain the match between human beliefs about how shallow a hole can be and objective matters of facts?

It's appealing to say: there's no mystery about human accuracy here because if we had used `hole' differently (by taking the minimum hole angle to be larger or smaller) then the meanings of our words would have been different so these alternative hole-identifying practices would have yeilded true utterances. But since variant practices of "hole" individuation can require change in sentences whcih don't even use the word hole  (e.g. a purely logical sentence like the Fregean paraphrase for `there are >3000 things' can go from true to false). So it seems plausible that allowing such changes in meaning requires allowing variant meanings for logical vocabulary like ``there is'' as well as in the meaning of  ``hole''.

So I think a great way to solve this `access problem for holes'' is to accept

Moderate Quantifier Variance: there are multiple existential quantifier like meanings which the words `there is' can take on in different ideolects, (say English_1910, English_2010 etc.), though within any particular ideolect `there is' is univocal.

• These meanings are `quantifier like' in the sense that they obey the all instances of standard first order logical inferential rules/axioms for `exists' (within the language they belong to). 
• If there is a single maximally natural quantifier sense (as e.g. Sider thinks there is corresponding to our talk about `what there is' takes on when doing ontology), these variant quantifier senses need not be mere quantifier restrictions of this fundamental sense. 

I like to use Moderate Quantifier Variance to explain/vindicate the freedom mathematicians take themselves to have to introduce new logically coherent structures for study, by saying that when mathematicians consistently extend the axioms of pure mathematics by adding axioms describing some new mathematical structures like the complex numbers, their words can both give meaning to expressions like `complex number' and also change the meaning of `there is', so that these axioms express truths.

But whatever you say about math, I think moderate quantifier variance is already useful for understanding our knowledge of swiss cheese!

The Sheffer stroke, and the blandest antirealism ever

In college my metaphysics prof said realism was the doctrine that there is a complete true description of the world, though it might take infinitely many sentences to express this description. Do the following incredibly bland considerations about different truthfunctional connectives commit me to "antirealism"?

(1) If you want thinking about propositions to play a role in psychology, then you need to individuate propositions narrowly enough that truthfunctionally equivalent sentences express different propositions.

(2) If you individuate propositions as narrowly as required by (1) then tautologies involving the sheffer stroke will be different propositions than any corresponding sentences which use more standard truthfunctional connectives.

(3) The same argument goes for all the other infinitely many n place truthfunctional connectives.

Conclusion: no language with finitely many basic connectives can express all true propositions. No finite language can be used to give a complete true description of the world. In particular if the a language L has only n place truthfunctional connectives then there will be tautologies using n+1 place truthfunctional connectives that cannot be expressed in L.

Justifying Logic and the Normal Role of Proof in Justification

Some philosophers aim to show how we can be justified in accepting certain basic logical truths by giving "rule circular" proofs of the soundness of these basic logical truths. They admit that most people will never have considered the proofs in question, and they admit that these people still count as justified in using logic. But, they say that they showing that such proofs can in principle be given makes sense of how we can be justified in believing the basic logical claims established by the proofs right now.

That idea seems prima face implausible. In general the fact that someone 100 years later will prove P from premises that I accept (like the ZF axioms) doesn't suffice to show that I am justified in believing that P now. So why should the case be any different for the proofs of logical principles?

[I would rather say that we are prima facie justified in believing these logical principles in a way that has nothing to do with the possibility of giving further argument; coming up with more or less circular ways of proving the soundness of our logical principles is (at best) a way of improving our justification]

FOL as the language for science

Maybe I'm missing something here...

Quine suggests that we adopt first order logic as the language for science. But, first order logic can't capture the notion of 'finitely many Fs'. It can only express the claim that there are n Fs for some particular n. Yet, we do understand the notion of finite, and use it in reasoning (e.g. if there are finitely many people at Alice's party, there is one person such that no one is taller than him) and potentially in science. Hence, we should not adopt first order logic as the language for science.

[The standard way to try to get around this, is by talking about relations to abstract objects like the numbers (There are finitely many Fs if there's a 1-1 map from the set of things that are F to the some set theoretic surrogate for the numbers). This would give you the right extension, if your scientific hypothesis could say that something had the structure of the numbers. But first order logic can only state axioms, like PA which don't completely pin down the structure of the numbers. Any first order axioms which you use to characterize the numbers will have non-standard models. This is Putnam's point in his celebrated model theoretic argument against realism. So, if you take this strategy, rather than saying that there are finitely many people at Alice's party, you can only say that the number of people is equinumerous items that satisfy a certain collection of first order axioms. And this does not rule out non-standard models.]

Field on Normativity and Logic

In "What is the Normative Role of Logic" Field argues that you can't understand logic descriptively as (eg. the project of studying necessarily truth preserving syntactic manipulations), and so are forced to a more normative conception of logic (logic is the study of how one ought to reason), by the following dilemma.
-classical logics can't state a general truth predicate (if they could, we could inductively argue for the soundness of logic, and hence a consistency proof for logic L in logic L, contra Godel 2)
-non-classical logics which can state a general truth predicate, sometimes fail to preserve truth, in some degenerate cases (in places where good reasoning wouldn't lead you to in the first place).

So (Field says) the only people who can *state* the descriptive criterion for being a logic, deny that logic has to have that property.

But I think there's a gap in this argument: why should you have to be able to state your criterion for what a good logical system is, *in the formal language of that logic*? In particular, why can't the anti-normativitst about logic reply like this:

A. Classical Logic Version:

Logic is the study of formal systems of syntactic manipulation which are truth preserving for various fragments of our language (e.g. english sans any truth predicate, english sans any repeated application of the truth predicate). Practically speaking, this is all we need for almost every purpose except philosophy of logic and truth. And the moral of Tarski-Godel considerations above is that this is all we can get.

Formal, exceptionless, rules for truth-preserving reasoning are great when you can get them (i.e. for limited fragments of our language) but what Field has shown, is that we can't get any such rules that apply to the informal notion of truth (as opposed to the notion of truth-of-a-sentence-in-L, for various restricted L)

Admittedly, taking this route involves giving up the traditional and somewhat attractive Fregean idea that logical principles are fully general, and hence would apply to all possible reasoning, but - at least- this seems way less revisionary than the normative relativism about logic where Field winds up.

B. Non-Classical Logic Version:

It was indeed wrong to say that logic studies patterns of inference that are always truth preserving. Field is right that Logic studies patterns of reasoning that are truth preserving "where it counts". But "where it counts" doesn't mean something normative like 'with regard to premises that one could be justified in believing', but rather, something descriptive like 'with regard to premises that people are likely to every actually accept'.

Different Senses of the Quantifers?

Carnapians want to say that different things can be truely said to exist when speaking in different language-frameworks. So the existential quantifier "Ex" will mean different things in these different frameworks. But can there really be multiple different meanings for these different uses of Ex, which would qualify as different kinds of e.g. existential quantification?

An argument that you can't is: The meaning of Ex is determined by its introduction and elimination rules. So any putative kind of existential quantifier would need to obey them. Hence different senses E1 and E2 from different frameworks would both have to obey the standard introduction and elimination rules for Ex. But if E1 and E2 obey these rules, then you can prove E1x from E2x and vice versa. Hence there is no room for ambiguity.

This argument can't be right though, if restricted quantification ('There is nothing in the fridge'. 'All the beers are in the fridge') - something that even the most ardent anti-Carnapians accept- counts as `a kind of' quantification. And intuitively it is. Hence in order to seem like a kind of quantification, a connective need not obey the full introduction rules. It suffices if there's a more limited range of instances of the introduction schema
P(x) --> Ex P(x) that speakers accept, together with all corresponding instances of the elimination schema Ex P(x). (A^B^C..^P(z) > F) ---> F (in cases where z does not occur free in A,B, C... or F). This is what we have for beers in the fridge.

Why can't the Carnapian claim that the same thing goes on with different linguistic frameworks? The different choices for when P(x) --> E2x P(x) is acceptable will each correspond to a different meaning for the existential quantifier. We can even formally represent these different possible senses for existential quantification formally, by saying a kind of existential quantification E_i corresponds to each subset S_i of the set of predicate-expressions (i.e. to each choice of what predicate-expressions the introduction and elimination schema are supposed to hold for).

You are probably worrying that this turns the Carnapian into a kind of maximalist (all the objects in question really exist, different frameworks just correspond to different framework restrictions) but I can't actually see any argument for that. So speak up if you can!