25-01-2014

Skolems paradox
In this essay, I shall do three things. First I will give a proof sketch of one of the LwenheimSkolem theorems, which says that any first-order set of sentences which is satisfiable has a model with a denumerably infinite domain. The next portion of the essay is then devoted to presenting Putnams model-theoretic argument which purports to show that a particular metaphysically realist position is incoherent. In short, by employing the Lwenheim-Skolem theorem Putnam argues that any such realist will be unable fix the intended interpretation of his theory of the world. Putnam thinks this realist consequently owes us an account of how his words are able to refer to their intended referents without postulating magical in another word, non-naturalistic powers. The last section of the essay discusses three replies to Putnams argument. The first two meet the argument head on, trying avoiding its conclusion, whilst the third reply merely tries to suggest that the very formulation of the argument is problematic. I argue that the two first ones fail but suggest that the last one may have more potential as an effective response to Putnams argument.

1. The Lwenheim-Skolem theorem

Properly speaking there is more than one theorem that goes by the name Lwenheim-Skolem theorem (henceforth LST). The LST theorem which is the concern of this essay is also known as the downwards Lwenheim-Skolem theorem. Together with the upwards LST, these theorems imply that any first-order set of sentences with a model has a model of every infinite cardinality. Although the following proof sketch concerns the downwards LST for first-order logic, there is a corresponding theorem for first-order logic with identity which says that any set of sentences which is satisfiable has a model which at most is denumerable. Proof sketch of the downwards LST: take any arbitrary set of first-order sentences which has a model. By the strong soundness of first-order logic, this set of sentences is consistent. The completeness proof for first-order logic shows how to construct, for any consistent set of sentences, a model with a denumerable domain. This is done as follows: let us assume that a maximal consistent, -complete extension * of i s given. A set of sentences S is maximal consistent just in case the union S {P}, for any sentence P which isnt a member of S, is inconsistent. The set * contains denumerably infinitely many sentences: we can add denumerably infinitely many individual constants to our first-order language and form denumerably infinitely many atomic sentences using these and some 1-ary predicate F. For any such sentence, either it or its negation will be a member of * because * is assumed to

1 of 6

25-01-2014

be maximal consistent. Now we construct a model for *, which at the same time is a model for because is a subset of *. The model is an interpretation I which is an ordered pair, (D,V), consisting of a domain and a valuation function. As regards the domain, let every individual constant occurring in a sentence have itself as its extension. For the valuation function, assign predicates extensions as follows. Any predicate Fn of degree n is assigned a set of ordered n-tuples of individual constants drawn from the domain such that an ordered ntuple (a1,a2,,an) is a member of the set just in case the sentence Fna1,a2,,an is a member of *. It can be shown that such an interpretation I indeed is a model for * in the sense that any sentence is true in I iff is a member of *. Since is a subset of * the interpretation I is also a model for . Since this holds for arbitrary set of sentences , we can conclude that any set of first-order sentences has a model with a denumerable domain. Q.E.D.

2. Skolems Paradox and the model-theoretic argument

Skolem (1922) used the LST to argue for the relativity of set-theoretic notions. The first-order axiomatic formalization of set theory due to Zermelo (1908) implies a theorem according to which there are sets of denumerable cardinality. LST shows, however, that this first-order formalization of set theory has a model of denumerable cardinality. Hence what is known as Skolems Paradox: that a sentence which says there are sets of a non -denumerable cardinality can be true in a model of denumerable cardinality. Skolem resolves the paradox by noting the relativity of set-theoretic notion when these are presented in the form of an axiomatic formalized first-order theory some sets may be non-denumerable with respect to the sets within a domain yet still be denumerable with respect to entities outside the domain. As both Alexander George and Paul Benacerraf note, this argument is unlikely to impress someone committed to the thesis that axiomatic set theory alone gives the meaning of our set-theoretic notions because such an opponent would disregard appeal to the notion of infinite cardinalities outside formalized set theory. Alexander George goes so far as to say that the LST carries no philosophical significance in and of itself. In one sense Georges contention is trivial: taken just by itself, the LST carries no philosophical significance but then thats true of any theorem of logic. Supplemented by the right premises, however, it may play a vital part in an argument against metaphysical realism, as Putnam tries to show. Before outlining Putnams argument, note that Putnam uses a strengthened version of the LST which requires the axiom of choice; so I will from now onwards take ZFC as the axiomatization of set theory under discussion. The strengthened version of the LST says that for any set of first-order sentences with a model M, there will be a countable submodel M which satisfies that set of sentences. To say that M is a submodel of M is to say that the predicate symbols o f

2 of 6

25-01-2014

M stand for the same relations as they did in M and that the domain of M is a subset of the domain of M; the only difference is the cardinality of the domains over which the quantifier variables range. An important consequence drawn from this is that the membershiprelation of set theory can retain its meaning inside formal set theory and the LST will still apply. Putnams intended opponent is a metaphysical realist who claims that we grasp the notions of set-theory without postulating Platonistic powers of mind. In particular, such an opponent is committed to a correspondence theory of truth and thinks that truth is theory-independent. The commitment to the correspondence theory of truth is reflected in the semantics given for first-order logic. Our first-order formalization of set theory provides theoretical constraints on the permissible interpretations of our set-theoretic vocabulary: they make the meanings of set-theoretic notions as determinate as they possibly could be from the point of view of a formal language. The first part of Putnams argument can now be put as follows: P1. LST shows that first-order ZFC has non-intended models P2. If P1, theoretical constraints on ZFC fail to pick out an intended interpretation P3. Thus, theoretical constraints on ZFC fail to pick out an intended interpretation Putnam furthermore envisages his opponent having a formalized theory of total science providing operational constraints on the permissible interpretations of our set-theoretic vocabulary. This is to be understood, say, as the measurement of the magnitudes of various rational space-time points added to our first-order theory with the use of predicates and individual constants. This will result in adding at most denumerable infinitely many space-time points and denumerable infinitely many predicates. So the new theory is the union of ZFC and this formalized theory of total science call it ZFC+. To anticipate the response that our use of language fixes the intended interpretation of our theory, let us once again add denumerably many predicates for properties and individual constants for objects in our language; add all the axioms reflecting the truths expressed by our language and we get the theory ZFC++. This is the just more theory maneuver: any additional constraints on the permissible interpretations of our words are treated as yet more theory. The argument then continues as follows. P4. By applying the strengthened LST to our newly obtained theory ZFC++, we can conclude that ZFC++ has non-intended models where the meaning of the non-logical vocabulary remains the same in the sense described above P5. ZFC++ fails to pick out its intended interpretation

3 of 6

25-01-2014

Next, Putnam presses a number of questions. One particular question which is problematic for the metaphysical realist and is directly related to Skolems paradox is the following: if our firstorder formalization ZFC++ fails to fix its intended interpretation in the sense that there are other interpretations which provide a model for ZFC++, what then fixes it? Our grasp of ZFC++ must at the very least consist in being able to assign every expression of its language a unique reference. If our theory is not capable of doing this even whilst respecting every theoretical constraint and operational constraint, even whilst respecting the meaning of nonlogical expressions, what then enables us to grasp the meanings of our words?

3. Responding to Putnams argument

Several responses have been made to the above argument. I will discuss three responses and argue that the first two are unsatisfactory. First, one might try to avoid the consequences of LST by allowing the language of first-order theories to be non-denumerably infinite. As Boolos and Jeffrey point out, LST fails to hold if we are allowed to add, for example, non-denumerably infinitely many predicates to our language. The principal difficulty with this response, pointed out by Giaquinto, is explaining how the addition of non-denumerably infinitely many predicates helps us explain how we grasp what the intended interpretation of a theory is: it is just as mysterious how our understanding would grasp such a theory how does one go through non-denumerably infinitely many sentences? It seems that in such a case, we would only be able to access a denumerably infinite subset of sentences of our whole theory and LST can be applied to that subset, making the problem reappear. Another reply, suggested by Benacerraf and further developed by Wright, insists that Putnam is wrong to treat every addition of constraints as further theory. Benacerraf, diagnosing the problem rather than trying to solve it, rightly notes that Putnam de-interprets any addition of theoretical or operational constraints by formalizing these and expanding the original theory. He then reapplies the LST with the same result. This is undoubtedly correct, but the question is why this move is objectionable. As long as the metaphysical realist hasnt g iven another explanation of how the reference of words is fixed, or grasped, the challenge posed by the model-theoretic argument remains. Wright thinks the direction to a solution suggested by Benacerraf is along the right lines. He furthermore thinks Putn ams argument depends on the thesis, in one form or another, that meaning is use. This is a vital part of the argument, but Wright gives no indication what else meaning could consist in. Noting the similarity between Putnams argument and the rule-following paradox of Kripkes Wittgenstein, he settles for

4 of 6

25-01-2014

Wittgensteins own solution that grasping some rules is not a matter of giving an interpretation of the rule but following, or obeying, the rule. For Wittgenstein, our shared human natures are ultimately what enable us to do so. Wright similarly appeals to our subconscious patterns of rationality and making judgment. Apart from being rather obscure, such a reply seems to avoid the just more theory maneuver. Yet it doesnt deal with the problem effectively. The response acknowledges that our language fails to determine reference and then desperately reaches for something else in this case human psychology to fix reference. Not only would it be a fortuitous coincidence that human psychology managed to fix reference, it would also be inexplicable. As Putnam himself says, if our use of words is not up to the task of fixing the referents of these words, thinking that something else is simply doesnt help us. A third reply to Putnams argument tries not so much to argue that Putnams argument is incorrect but aims to show that the dialectical position between Putnam and the metaphysical realist is a philosophical stalemate. Note that by applying the LST to absolutely everything, the model-theoretic argument undermines its own talk of an intended interpretation and other notions used in the metalanguage to formulate Skolems paradox they are simply not available to us because they are incoherent. If we really are in the philosophical predicament Putnam thinks were in, every attempt to formulate the problem posed by the model-theoretic argument is impossible because accepting it implies that its very formulation is incoherent. So if Putnam accuses the metaphysical realist of postulating magical powers that determine reference whilst remaining a naturalist, the realist can only reply: tu quoque. What about the fact that there are other views in philosophy of mathematics verificationism and Platonism which arent threatened by the argument can we formulate the problem from their viewpoint, so to speak? I do not think so. If verificationism or Platonism is a correct view of philosophy of mathematics, and metaphysics more generally, then there is no problem to talk about in the first place. Similarly, one cannot accept the metaphysical realist position for the sake of argument and then refute it by reductio ad absurdum: accepting the very position means no reductio can be formulated coherently.

4. Concluding remarks
The above reply to Putnams argument is apt to seem suspicious: doesnt it imply that the refutation of any philosophical view will lapse into incoherence? No. The special feature of Putnams argument, if the reply is correct, is that it embodies a meaning-theoretical skepticism so radical that it is self-refuting. Similarly, no-one can coherently assert, for example, I dont believe anything or Everything I say is false. While this does answer Putnams argument in

5 of 6

25-01-2014

one sense, it is also unsatisfactory in another sense. The conclusion of the reply is not that there is no problem for the metaphysical realist that would be a philosophical victory and not a stalemate. Here, the conclusion must be that we cannot continue: every attempt to say something about the alleged problem is self-refuting if we accept that the problem really is a problem. The situation is that of a stalemate, not merely a draw, in the sense that we would like to be able to say something more but literally cannot do so. Where that leaves us is an interesting question, though it will have to wait for another time.

6 of 6