Intensionality and the Gödel Theorems

Author(s): David D. Auerbach

Source: Philosophical Studies: An International Journal for Philosophy in the Analytic
Tradition , Nov., 1985, Vol. 48, No. 3 (Nov., 1985), pp. 337-351
Published by: Springer

Stable URL: https://www.jstor.org/stable/4319795

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Springer is collaborating with JSTOR to digitize, preserve and extend access to Philosophical
Studies: An International Journal for Philosophy in the Analytic Tradition

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 DAVID D. AUERBACH

INTENSIONALITY AND THE GODEL THEOREMS

(Received 4 September, 1984)

Philosophers of language have drawn on metamathematical results in varied

ways. Extensionalist philosophers have been particularly impressed with two,
not unrelated, facts: the existence, due to Frege/Tarski, of a certain sort of
semantics, and the seeming absence of intensional contexts from mathemati-
cal discourse. The philosophical import of these facts is at best murky. Ex-
tensionalists will emphasize the success and clarity of the model theoretic
semantics; others will emphasize the relative poverty of the mathematical
idiom; still others will question the aptness of the standard extensional
semantics for mathematics.
In this paper I will investigate some implications of the Godel Second
Incompleteness Theorem for these positions. I shall argue that the realm
of mathematics, proof theory in particular, has been a breeding ground for
intensionality and that satisfactory intensional semantic theories are implicit
in certain rigorous technical accounts. One moral to be drawn is that inten-
sionality does not, as a matter of course, involve incoherence.
The central argument will conclude that an extensional semantics would
attribute falsity to the Godel Second Incompleteness Theorem. Since we
have good reason to believe the theorem true, the correct inference is to the
insufficiency of extensional semantics. The intensional context lurking in the
Second Theorem is that of indirect discourse. This indirect discourse is,
contra Quine, not "at variance with the characteristic objectivity of science".
In fact, rigorous generalizations of the Second Theorem provide the proper
non-extensionalist treatment.
Each of the matters just broached involves significant subtleties - even the
matter of justifying the truth of the Second Theorem. Moreover, the relevant
technical work is not well reported in the philosophical lterature and has
only recently been assimilated into logic texts.2 In what follows, therefore,
the methodological and metamathematical surveys are necessary preliminaries.

Philosophical Studies 48 (1985) 337-351. 0031-8116/85.10

? 1985 by D. Reidel Publishing Company

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 338 DAVID D. AUERBACH

II

This section illustrates some methodological aspects of the interpretation of

metamathematical results, the Godel Theorems in particular. Among the per-
sistent concerns of philosophers of mathematics are the implications of
various technical results. This can be overt, as in 'What are the philosophical
implications of A?', where A is some technical result. Such questions raise
the further question: what theses are needed to get from A, a purely mathe-
matical proposition, to B, a non-mathematical proposition? This latter ques-
tion arises covertly when theorems are rephrased or glossed. The following
sketch of an answer to one such question is only intended to highlight the
special features of my subsequent treatment of the Godel Second Incomplete-
ness Theorem.
In 1930 Godel proved that a certain formal system, which he called P, is
either incomplete or co-inconsistent. This is not what is generally referred to
by 'Godel's First Incompleteness Theorem'. The 1930 result gains importance
because P is important and because the proof of the result is clearly general-
izable. A theorem to the effect that a large and important class of formal
systems shares the property of incompleteness with P would seem of greater
import. A refinement of Rosser's is needed to yield the familiar Incomplete-
ness Theorem.

(1) There is no consistent complete axiomatizable extension of Q.

(1)3 expresses an up-to-date generalization of the results of the 1930s and

certainly obtains for us the 'large' class of formal systems we asked for.

(1) is a provable mathematical result.

(2) Any sufficiently strong consistent formal system of arithmetic

is incomplete.

(2) is often used as an expression of the Godel result. Since (1) and (2) are
not prima facie synonymous, nor does (2) look wholly mathematical, what
warrants both the assertion of (2) and the claim that it is an expression of the
First Incompleteness Theorem?
Getting from (1) to (2) is a special case of the problem I characterized
above - getting from mathematical A to nonmathematical B. One thesis
needed is: In this context ix is an extension of Q' implies 'x is sufficiently
strong'.4 In fact the result extends to theories into which Q is interpretable,
but this is a refinement which need not detain us.

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 INTENSIONALITY AND THE GODEL THEOREMS 339

Additionally, some account of 'of arithmetic' is required. Such an account

should yield the thesis: Q is a formal system of arithmetic and all its exten-
sions in the language of arithmetic are formal systems of arithmetic.5
Each such thesis, dredged up to create a non-enthymematic valid argument
from (1) to (2), needs to be justified. Why is it true that Q and its extensions
are 'of arithmetic'? I shall not pause over this interesting, and not so simple,
question. The kind of theses I have been talking about (call them 'connective
theses') often assert that a particular formalism is an adequate formalization
of some notion. Church's Thesis is a good example. Other examples include
claims that a formal semantics corresponds in a certain way to an intended
interpretation. Semantics of first-order predicate logic/logical validity is such
a pair, Kripke semantics/Leibnizian possibility another.
These connective theses are the additional premises needed to produce a
valid argument from a mathematical theorem to a philosophical claim. What
has often been stressed in regard to Church's Thesis is also true of many
important connective theses - they are not mathematical truths and do not
partake of mathematics' clarion certainty and precision. Associated with
Godel's Second Incompleteness Theorem are connective theses of a some-
what surprising character. Consideration of the Second Theorem is perhaps
best begun by looking at the analogue to (2) - the proposition that is not a
mathematical theorem.

(3) If T is a sufficiently strong consistent formal system of arithmetic,

then any sentence of T that says that T is consistent is not
derivable in T.

(3) is a relatively careful rendering of one of the many things that Godel is
often said to have proved. Sometimes it is put: 'The consistency of (a formal
system of) arithmetic is underivable in (that formal system of) arithmetic'.
In many texts such words are offered as glosses of some technical result
(usually proved only sketchily, if at all) labeled 'Godel's Second Incomplete-
ness Theorem'. Since what are derivable, or fail to be, are the formulas of a
formalism, I take it that (3) is an acceptable rendering of the slightly looser
remarks.6
These looser remarks, or sometimes (3), are the usual material for philo-
sophical writings concerning the philosophical significance, or consequences,
of the Godel Second Theorem." Since (3) isn't the mathematically proved
Second Theorem, such writings would be helped by an argument that (3)

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
340 DAVID D. AUERBACH

is true - the Godel Second Theorem being a premise in such an argument -

as well as by evidence that the second Theorem is a special sort of premise in
such an argument. By this last I simply mean something that would justify
taking (3) as a gloss, not a mere consequence, of the Second Theorem. Thus,
in discussing (1), the First Theorem, and its relation to (2), one notes that the
additional premises are either definitional or trivial. (1) is thereby 'special',
being neither definitional nor trivial. We need to take a closer look at the
differences between the two Godel theorems, to begin to uncover what stands
to (3) as (1) does to (2).

1III

An early reference to the Second Theorem is found in Godel's 1931 paper,

in which a proof of what he calls Theorem XI is sketched. What does Godel's
sketch of a proof of Theorem XI show? Briefly, an underivable formula is
exhibited, different from the one (17Gen r) exhibited in the proof of his
First Incompleteness Theorem. It should be recalled that in the proof of the
First Theorem Godel constructs a formula that he shows, on hypothesis of
consistency of P, to be underivable (in P). That is, he shows that there is a
proof that the consistency of P implies that a certain formula is underivable.
Corresponding to this, Godel's proof, there is a derivation in P of a conditional
formula of P corresponding to Godel's implication: a conditional whose
antecedent is a sentence which is the formalization of the assertion that P is
consistent, and whose consequent is the formal sentence saying that the
Godel sentence is not derivable. By the construction of the First Theorem
this consequent is equivalent in P to the Godel sentence itself. Hence, the
conditional whose antecedent is the consistency sentence and whose conse-
quent is the Godel sentence is a theorem of P. And since modus ponens
is a rule of inference of P, the formalized statement of consistency cannot
be derivable in P if P is consistent.
This is a rough and deliberately innocent sketch of the idea of the proof of
XI. A detailed proof would involve constructing (or, at least, showing how to
construct) the crucial formal derivation. This would include making adequate
sense of the notion of formalization that infests the above sketch; it is here
that the complications of the Second Theorem reside. However, the semantic
flavor of this theorem can be preliminary appreciated by noticing that what is
underivable is a formula and that this formula is said to say that the formal-

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 INTENSIONALITY AND THE GODEL THEOREMS 341

ism is consistent (recall (3)). Unlike the situation with some of the informal
intuitive descriptions of the First Theorem (and with my description of the
'consequent' above) that are often given, this apparent semantic character is,
I shall argue, intrinsic. The constraints on the antecedent are different in kind
from those on the consequent.
To draw the contrast between the two Theorems more sharply: The First
Theorem predicates a simple syntactic property of members of a large class of
formal systems. This property, incompleteness, is simple in at least the follow-
ing respect - it is definable in terms that do not invoke anything akin to a
translation, or formalization, relation; T is incomplete just in case there is a
sentence of T such that neither it nor its formal denial is derivable in T.
The Second Theorem (as glossed by (3)) predicates a certain more com-
plex property of members of a large class of formal systems. Call this proper-
ty 'KURT'. T has KURT if and only if any sentence of T that says that T
is consistent is not derivable in T. We will be seeing what it takes to render
this definition coherent, for 'T' taken as a variable.
What is the underivable formula of the Godel Second Incompleteness
Theorem? Let us call it, restricting our attention for the moment to just one
formalism, CON(P)8. Call the underivable formula of the First Theorem 'G'.
A proof of the Second Theorem establishes that 1- CON(P) -+ G (in fact
H CON(P) e G) and so, not H CON(P). In the light of these facts what would
make CON(P) importantly different from G?
Let Pf(y, x) be an open sentence of P that numeralwise expresses9 the
proof relation (assuming some fixed Godel numbering of syntactic objects,
and sequences of them, satisfying the usual constraints) and Fm(x) an open
sentence of P that numeralwise expresses sentencehood. Using PF (and a
symbol representing a substitution function), a proof of the First Theorem
constructs G, a formula equivalent in P to Vy - Pf(y, g), where g is the
Godel number of that very formula, G. 3 x(Fm(x) & - 3y Pf(y, x)) mimics,
in quantificational structure, a standard definition of consistency. Let such a
formula be (temporarily) CON(P). Note that this does not pick out a particu-
lar sentence G nor a particular sentence CON(P) since many formulas of P
will numeralwise express the proof relation. This produces a fatal equivoca-
tion in the preceding paragraph.
This construction does not guarantee that such a CON(P) says that the
formalism is consistent. It is not essential to a proof of the First Theorem
that the Godel sentence say of itself that it is underivable. (Indeed, it is only

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 342 DAVID D. AUERBACH

the case the Godel sentence is equivalent in Q to a sentence H which asserts

something about G. But even this need not be that G is underivable.)10 That
it may seem to be one is an artifact of certain informal, motivating, semantic
accounts of the First Theorem. Certain entailments of the proposition that
G says of itself that it is underivable are all that are used and these are suf-
ficiently captured by the relation of numeralwise expressibility. The plausibil-
ity of regarding G as saying of itself that it is not derivable arises from con-
sidering the standard interpretation, the Godel numbering, and regarding
Pf as expressing the proof relation. In the case where Pf only satisfies the
entailments mentioned, however, this last is at best a pun on 'numeralwise
expressible'. A sentence so constructed is true (in the standard interpretation)
if and only if it is not derivable; but nothing stronger than this extensional
agreement is forthcoming.11 This is all the First Theorem requires, even in the
general case where P is replaced by an arbitrary sufficiently strong theory T.
Requiring only extensional agreement with respect to the proof predicate
will not, however, in the general case, yield extensional agreement with respect
to the consistency statement. That is, for some T, and some CON(T) con-
structed as above, it is false that: CON(T) is true if and only if T is consistent.
Moreover, even in the case of P there are two sentences, CON1(P) and CON2(P),
constructed as above, but such that they are not logically equivalent; and one
of them is derivable in P.

IV

I have claimed that the difference between G and CON(P) is that a G may be
constructed from an open sentence that numeralwise expresses the proof
relation and that numeralwise expressibility is the only constraint needed to
show such a G to be underivable; a CON(P), constructed as above, may be
derivable. Concerning saying that, I have pointed out that the plausibility of
regarding G as saying that G is not derivable arises, inter alia, from regarding
Pf as expressing the proof relation, but that this plays no role in establishing
(1) or (2). But suppose we take this plausibility at face value and benefit
from the heuristic value of being able to explain the workings of G1 by
adverting to the simple logic of the antinomies. Then a CON(P), constructed
as above, should say that P is consistent. But then, I have claimed just above
and will show below, either (3) is false or the account of says that is defec-
tive. Furthermore, abandoning all accounts of says that would eviscerate

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 INTENSIONALITY AND THE GODEL THEOREMS 343

the Second Theorem. I will now establish and illuminate these claims.
Rosser exploited numeralwise expressibility's inability to distinguish co-
extensive relations. By constructing a new open sentence that numeralwise
expresses y is a proof of x but which had special properties as well, Rosser
was able to improve the G6del result. Letting Pf numeralwise express the
proof relation, Rosser used

Pf(y, x) & 7 3y(y <x & Pf(y, neg(x))).

This reads 'y is a proof of x and there is no shorter proof of the negation of
x'. A moment's reflection reveals that, for a consistent formalism, this
numeralwise expresses what Pf does. An even simpler device will make the
derivability of "consistency" more blatant. Define Pf' as

(4) Pf(y, x) & -IPf(y, k)

where k is the Godel number of '0 = 1'. If the formalism is consistent (4)
numeralwise expresses what Pf(y, x) does. Consider the consistency schema

(5).
(S) 3 3y 4(y, k)

Then, although the instance of (5) involving Pf is not derivable, the instance
in which Pf ' replaces F, i 3 y(Pf(y, k) & i Pf(y, k)), is a theorem of logic.
What isn't derivable in the formalism is that Pf and Pf' are coextensive and so
numeralwise express the same relation.
In a sense, the G's constructed from such deviant Pf conditionally assert
their own underivability. The condition is that the formalism is consistent.
This condition the formalism cannot discharge. (These intuitive readings are
discussed below.) The prima facie incorrectness of 'y is a proof of x and y
isn't a proof of 0 = 1' as expressing y is a proof of x is not misleading; a
proper account of the Second Theorem must respect this fact.

The quarantining of the deviant predicates is accomplished in rigorous ac-

counts of the Second Theorem that prove generalized versions of it. It does
not follow as a matter of course that such accounts yield (3). ((3) If T is a
sufficiently strong consistent formal system of arithmetic, then any sentence
of T that says that T is consistent is not derivable in T.) A purely mathemati-

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 344 DAVID D. AUERBACH

cal result about formalisms may, in producing an underivable formula for

each formalism, clearly be a generalized Second Theorem because the result
is a rigorous version of the sketch on page 340. But it remains to justify such a
result's rejection of certain extensionally correct proof predicates.
Technical accounts of the Second Theorem may vary as to how manifest
is the relation between their machinery and our semantic concerns. It is
striking that they all bear some familiar hallmark of intensionality. What are
some of these accounts?
The earliest treatment occurs in [Hilbert-Bernays, 1939], wherein three
derivability conditions are enumerated and used to prove a rigorous version
of the Second Theorem. Any proof predicate that satisfies the derivability
conditions (one of which is essentially numeralwise expressibility) will suffice
for the Second Theorem; the labor is in showing that any particular proof
predicate does satisfy them. This approach is continued in the work of Lob
(1955) and in recent work on modal systems, where O is interpreted as
derivable in a formal system. '2
The Hilbert-Bernays derivability conditions were conditions on the formal
proof relation, stricter than numeralwise expressibility, that were sufficient
to guarantee an unequivocal Second Theorem in the following sense: Any
proof predicate meeting the three conditions would satisfy the requirements
of the proof of the Second Theorem; moreover, the proof relations con-
structed for familiar theories were seen to satisfy the conditions.
The work of Feferman (1960) and Jeroslow (1965) was stimulated by
remarks of Kreisel (1965).'3 Feferman points out that merely "numerically
correct" proof definitions are inadequate for certain results. Results for
which they are adequate he calls "extensional", the rest "intentional". The
deviant proof predicates lead to useful extensional results (e.g. Rosser)
while others have no intrinsic interest (a provable "consistency" sentence).
For Feferman the weakness of the Hilbert-Bernays approach is that verifying
whether a particular predicate satisfies the conditions is laborious.
Feferman presents a large class of formal systems and proves the Second
Theorem for them. The key to his approach is the notion of a formal system
that he employs. The consistency sentence for any system is built from the
proof predicate in some standard way (by a straight-forward transcription of
any one of the equivalent definitions of consistency); the proof predicate
is straight-forwardly transcribed from the presentation of the formal system.
The trick is in obtaining a fornal object to represent the presentation.

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 INTENSIONALITY AND THE GODEL THEOREMS 345

More precisely: If a(x) is a formula that numeralwise expresses the axioms

of T, a proof predicate can be constructed in a standard way from a. The
phrase 'a consistency sentence for a formal system' makes sense only if one
individuates formal systems more narrowly than by their theorem sets. By
narrowing the individuation it becomes possible to generalize various meta-
mathematical results by conditions on the formulas ac.
Since many ai's numerically define the same set of axioms, different formal
proof predicates will be defined for the same axioms; one for each ca. Deviant
ax's are bizarre ways of presenting the axioms - bizarre enough to carry a
trivial assurance of consistency (recall schema (5) with 1 replaced by Pf').
Feferman constructs a sentence Ga, for each formula at that numerates
an extension of P, such that Ga is underivable in the system corresponding
to ai. Ga corresponds to the usual Godel sentence. (5.6) is a simplified version
of Feferman's generalization of the Second Theorem.

(5.6) Let a be a formula that numerates in Q the axioms of a consistent

extension of P. Call this extension A. If a is an RE-formula
HA CON,a -? G and hence not HA CON,.

(5.6) differs from a blindly extensional generalization of the Godel result

only in the restriction that a be what Feferman calls an RE-formula. While
the precise definition of this does not concern us here what is crucial is
Feferman's (5.9), which establishes that the purely extensional generaliza-
tion, (5.6) without the restriction on ct, is false.

(5.9) Let A be a certain kind of extension of P. There is an a* numeral-

wise expressing A's axioms in A, such that HA CON,* 14

Not surprisingly, Feferman's a* is a close relative of the deviant proof predi-

cates described earlier. a*(x) = a(x) & Vz(z <x -* CONa 1Z) & StK(x), where
a numeralwise expresses A's axioms in Q, and Stk represents is a sentence.
It can be read: x is an a-axiom and all the finitely axiomatized subsystems
formed by the axioms shorter than x are consistent. A facile, but not essenti-
ally misleading, way of reading the corresponding 'consistency' sentence is:
The largest consistent subsystem of A is consistent. 'The largest consistent
subsystem of A' denotes A, given that A is consistent; but, of course, it is
just this fact that is not given to A itself (cf. the end of the section IV).
In fact, then, the notation 'CONp' is dangerously vague, unless one has a
particular presentation of the axioms in mind. 'Its' formalization, 'CON<>',

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
346 DAVID D. AUERBACH

makes this clear; particularly

derivable in P. The RE ci's are not an ad hoc grouping; it can be argued
(though not here) that the defmition of RE-ness meshes perfectly with our
intuitions about what formalisms are.

VI

The problem then is to rescue the plausibility of construing formulas as

making metamathematical remarks, without falsifying (3). Taking any open
sentence of arithmetic whose extension, in the standard interpretation, is
the extension of is a derivation of (via some admissible Godel numbering)
as expressing the formal proof relation marks a commitment to an exten-
sionalist semantics. The cavils of the earlier sections show that such a seman-
tics would indeed fail to do justice to KURT. A semantics based on the
narrower individuation of predicate expressions that, for example, Feferman's
treatment mandates, will do justice to KURT15. It is worth emphasizing
again that Godel's Theorem XI cannot be proved in generality without some-
how stigmatizing the deviant proof predicates.
The skeptical reader, wary of intensional semantics in general, may regard
a coherent applied intensional semantics with suspicion. I will in what follows
try to alleviate such suspicions.
It is the common meta-theoretic ground of many programs for semantic
theory, and the centerpiece of Davidsonian approaches, that the truth-value
assigned to a sentence depends only upon the semantic value of the parts. Of
course, much controversy surrounds the proper elucidation of this gnomic
remark. Intensionality is ascribed if the semantic value of a part needs be
something other than the extension of that part. It is patent here that the
notion of part is crucial to such ascription. Frege argued for such a duality
of semantic values by finding contexts that produced a difference in truth-
value despite co-extensiveness of parts. Russell's theory of descriptions can
sometimes be used to avoid ascriptions of intensionality by a change in
the ascription of parthood.
Consider the language of elementary proof theory. Evidently it contains
the predicate 'is a derivation-in-T of ', for each formalism T. (Every-
thing that follows will apply as well to any synonymous predicate, whether
couched in Japanese, German, or English.) Using such a predicate we can
express consistency - 'There is no derivation-in-T of i' where i is a favorite

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 INTENSIONALITY AND THE GODEL THEOREMS 347

contradiction of a theorem of T. Call this sentence Con(T). We are interested

in the semantics of this piece of (technical) natural language. In particular,
is the meaning of Con(T) given by the extension of its parts? Not if we wish
to express (3) in this language. How did the technical details reveal this?
Con(T) is not a candidate for being underivable-in-T. We do want to show
that it is unprovable in T by appropriately formalizing the language of ele-
mentary proof theory and producing a sentence of T, say CONa, whose
underivability permits the inference to the unprovability of Con(T). This is
precisely the problem of page 340 - establihsing (3) on the basis of a certain
technical fact. Let us look at this sort of inference from a wider point of
view.
We say that it is provable in P that 2 < 3 because a certain formula of P
is derivable and the standard interpretation for P makes the appropriate link
between the formula and the manner in which the standard model is described.
Mates 16, for example, discusses the reasons for this last clause. He points out
that in establishing instances of Tarski's schema T, the way in which the
interpretation is described is utilized. The same interpretation, I, given
differently, yields both

'La1 a2 ' is true under I iff 2 is less than 3.

and

'La, a2' is true under I iff the only even prime is less than 3.

as consequences of the definition of truth in an interpretation. For purposes

that exceed mere consideration of truth conditions in an extensional language,
the non-identity of the two displayed sentences is vital. One such purpose,
ubiquitous in logic texts, is judging whether a formal sentence is an adequate
rendering of an English sentence; and this is relative to the way in which the
interpretation is given.
The context we are considering (that 2 < 3 is provable in P) is very like
the translation context that Mates is concerned with. He points out that the
truth-value of wffs is independent of the manner of specification of the inter-
pretation, though the meaning is not. The Fregean move involves locating a
context that is sentisitive to the manner of specification, but in the language
being interpreted; this will produce a difference in truth-value, not just in
meaning. The Second Theorem provides such a context; the proof of the
Second Theorem requires significantly more than correct extension be true

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 348 DAVID D. AUERBACH

of the provability predicate. The language of elementary proof theory, in as

much as it contains (3), is intensional.
The skeptic may see the 'says that' idiom of (3) as begging the question,
and look to (6) as expressing the content of a generalized Second Theorem.

(6) If T is a consistent formal system of sufficient strength, it is not

provable in T that T is consistent.

It should be clear by now that this won't help. A cogent argument for (6)
will have recourse to a proof of a generalized Second Theorem. Viewed as a
semantics the Feferman treatment assigns true to (6) by virtue of (5.6). The
key to the linkage is the occurrence of the formulas a as part of the formal-
ized consistency sentence. The a's are assigned by the semantics as represen-
tatives of formal systems. We can think of them as fixing the reference for
proper names of the formalism. (5.9) demonstrates that a (6) whose semantics
treated co-extensive a's indiscriminately would be false.
The skeptic might now object to the implicit syntax, that is, the parts,
and say that the superficial form of (6) is misleading.

(7) If T is a formal system of sufficient strength, the sentence CON,,

is not derivable in T.

(7) might be offered. The skeptic goes on to say that an appropriate technical
account, says Feferman's supplies the sentences CON<. This account must,
of course, bow to the requirements set by (5.9); this will force the skeptic
to identify formal theories with RE a 's.
Having been forced to the bifurcation (RE/non-RE) of the class of oa's
numeralwise expressing each set of theorems, the skeptic has sacrificed two
things: 1) an explanation of the bifurcation and, more importantly for
present purposes, 2) a reason for regarding (7) as a remark about consistency.
The skeptic is unable to distinguish the First and Second Theorems in any
interesting way (cf. Section III). For if (7) is to be about consistency, by way

of CON, being a consistency sentence for formalism T, then a semantics is

required - one which (5.9) tells us makes the truth of (7) depend on some-
thing other than the extensions of the parts of CON,.
If a semantic theory for the language of elementary proof theory allowed
a non-RE ca to be equivalent to some RE a then we would see that it had to
be an incorrect theory. For then it would be part of the meaning of 'formal-
ism' that a formalism be consistent. It is part of the charm of the Godel

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 INTENSIONALITY AND THE GODEL THEOREMS 349

Second Incompleteness Theorem that it entails that such a semantic theory

would not only be empirically false, but simply inconsistent.
In sum, recalling the remarks at the end of section II, (1) is to (2) as
(5.6) is to (3). The language of elementary proof theory has a coherent
intensional semantics that certifies this ratio. An extensionalist semantics
is neither desirable nor necessary for all of mathematics.17

NOTES

' Resnik (1974) and Detlefson (1979) are recent exceptions that take account of the
special features of the Second Theorem.
2 See Boolos-Jeffrey (1974), Boolos (1978) and Monk (1976).
(1) first appears in Tarski (1960).
4 The converse implication supports a valid deductive argument from (1) to (2). Un-
fortunately the converse implication is false; weaker theories than Q are sufficiently
strong. The suggested implication is true, but not sufficiently helpful. A complete analysis
would bring out the contextual nature of 'sufficiently strong'. Alternatively, one could
use the converse implication, replacing the predicate 'is an extension of Q' with 'has
those features of Q that are relevant to the proof of G 1', or some illuminating coextensive
predicate.
' If (2) strikes one as mathematical, note that all that is really needed for my point is
that passage from (1) to (2) is mediated by theses unsupported by mathematical evidence.
(2) seems mathematical because the mediating theses seem definitional.
6 'Derivable', 'underivable', etc. will be used in connection with formulas of a formalism;
'provable', 'proof', etc., are reserved for the ordinary notions of unformalized mathe-
matical practice. 1 occasionally violate this convention, for the sake of custom, in the
context of discussing is a proof of and the 'proof' predicate. Strict speaking would
demand 'is a derivation of' and 'derivation predicate'. Note that the looser remark cited
in the text offers little obvious guidance regarding a choice between 'underivable' and
(unprovable'.
See Note 1, above.
P p is the well-known Peano arithmetic, so-called because of its name. Q is a well-studied
theory in the language of arithmetic. It has finitely many axioms (seven, all simple and
clearly true in the standard model), all recursive functions are representable in Q, and yet
it is a rather weak subtheory of P (that addition is commutative is not a theorem of Q).
Therc are even weaker theories that will suffice for the F-irst Theorem; Q's virtue is its
finite axiomatizability. See Tarski (1960) and Boolos (1978). For now 'CON(P)' is a
simple term.
9 Numeralwise expressibility is a three place relation among formal systems, relations
among or properties of numbers, and predicates of formal systems. A formal predicate
that numeralwise expresses a relation in an arithmetically correct formal system is there-
by guaranteed to be extensionally correct with respect to that relation. 1 numeralwise
expresses R, R an m-place relation, ill

(i) if R(n ), then HI((n)

(ii) if flR(-n), then l-1(l),

where '' denotes the function from numbers to their standard numerals in the formal-
ism. All recursive relations are numeralwise expressible in Q and its extensions, including,
of course. P. 1 is said to numerate R, in consistent formalisms, if R(-n) Ho (n).

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 350 DAVID D. AUERBACH

10 Of course G implies that its for

say that G is underivable and, a fortiori, that something is underivable. Any formula
that says that something is underivable, either existentially or using a canonical name, is
a consistency sentence. That is, G and CON(P) are not importantly different provided
that G really does say that G is underivable. Kleene (1950), p. 211 has a similar remark,
but with the semantic proviso on G hidden in Kleene's use of 'intuitively' and 'intuitive'.
" A typical proof of the First Theorem enforces this extensional agreement by con-
straints on the open sentences used to build up G; in particular, the constraint that Pf
numeralwise express is a proof of. This allows such a proof to be purely syntactic, in
that it need not mention the standard interpretation. The use of standard numerals
in the definition of numeralwise expressibility is, however, crucial.
12 See Boolos (1978).
13 The cleanest presentation of the technical facts is in Monk (1976), pp. 298-307.
Feferman (1962), pp. 265-272, contains a useful precis of the technical material in
Feferman (1960). In Monk (1976) the RE-formulas are not mentioned as such, but the
remark on p. 304 concerning '`(g"A)' does as well. Monk's text is particularly nice
for its dovetailing of the Lob and Feferman material.
14 For a finitely axiomatized theory there is a unique, principled, best choice for an co.
For finitely axiomatized A, this ce is written [A] and [A] = the obvious formalization of
'x= a, V x = a2 V * V x = xn' where a,, . . ., an are the (Godel numbers of the) axioms.
A is reflexive just in case for each finite F C A, I- CON[ FJ* Feferman's G., CONo, etc.,
are specific constructions defined over a broad class of canonically presented formalisms,
schematic only in ea. No theory with induction can be finitely axiomatized, no theory
without induction can formalize formalisms - so the little fact here is of no help with
consistency sentences.
15 Jeroslow's approach, though intertranslatable with Feferman's, is more direct.
Jeroslow avoids the standard encodings of the usual primitive recursive syntactic rela-
tions and functions, whereas Feferman presents a generalized theory of those functions
and relations. Jeroslow specifically represents formal systems as Post Canonical Systems:
"Formal logics are not usually understood as Post Canonical Systems, but there is a
natural, uniform procedure for viewing them as such, provided that all the mechanical
rules which constitute the formal logic are specified, even the inductive rules for gener-
ating the terms, formulas, etc. The idea here is that the predicates of proof theory are
always inductively defined, and Post Canonical Systems are the language of inductive
definitions par excellence." Post Canonical Systems thus formalize the presentations of
formal systems given in logic books. As the F1eferman approach is certified as semantical-
ly correct by arguing that the RE/non-RE distinction is principled, the Jeroslow account
is certified by arguing for Jeroslow's Thesis: PCS's are the best representations of formal
systems. The details of such argumentation I leave to another paper.
It is in the spirit of the origins of formal systems as an object of study that they be
regarded as systems for generating syntactic objects, in categories, independently of
their intended meaning. Note that, from this point of views, axiom schemata have no
place, as such; non-finitely axiomatized theories are to be identified with "the finite
number of rules which describe the generation of the infinite number of axioms"
(Jeroslow, 1971). Not only is this accord with the view of formal systems as combina-
torially secured producers of theorems, but also can be connected to the epistemological
motives behind a Hilbert-style program. Kreisel (1965) adduces some additional con-
ceptual grounds for Jeroslow's Thesis in pointing out that we often wish to distinguish
formal systems by their rules, and not by their theorems or even their set of proofs.
Typical contexts that require such a fine-grained distinction of theories are evidentiary
ones. One formulation of a set of theorems may be evident (i.e., evidently true) and
hence foundationally sound and another not. Moreover, the establishment of their
(extensional) equivalence may not be evident.

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms
 INTENSIONALITY AND THE GODEL THEOREMS 351

16 Mates (1972), pp. 75-78.

17 An ancient ancestor of this paper existed in 1976. I h
from the comments of George Boolos, Harold Levin, Michael Resnik and most especially
Mark Richard.

BIBLIOGRAPHY

Boolos, G.: 1978, The Unprovability of Consistency: An Essay in Modal Logic (Cam-
bridge University Press, Cambridge).
Boolos, G. and R. Jeffrey: 1974, Computability and Logic (Cambridge University Press,
Cambridge).
Detlefson, M: 1979, 'On interpreting Godel's Second Incompleteness Theorem', Journal
of Philosophical Logic 8, pp. 297-315.
Feferman, S.: 1960, "Arithmetization of Metamathematics in a General Setting",
Fundamenta Mathematicae XLIX, pp. 35-92.
Feferman, S.: 1962, "Transfmite Recursive Progressions of Axiomatic Theories", JSL
XXVII, pp. 259-316.
Hilbert, D. and Bernays, P.: 1939, Die Grundlagen der Mathematik, 2nd ed. (Springer-
Verlag, Berlin).
Jerolsow, R.: 1971, 'On Godel's consistency theorem', unpublished manuscript.
Jeroslow, R.: 1973, 'Redundancies in the Hilbert-Bernays derivability conditions for
Godel's Second Incompleteness Theorem', Journal of Symbolic Logic XXXVIII,
pp. 359-367.
Kleene, S. C.: 1950, Introduction to Metamathematics (Van Nostrand, Princeton).
Kreisel, G.: 1965, 'Mathematical logic', in T. L. Saaty (ed.), Lectures on Modern Mathe-
matics III (Wiley, New York).
Lob, M. H.: 1955, "Solution of a Problem of Leon Henkin", Journal of Symbolic Logic
XX,pp. 115-118.
Mates, B.: 1972, Elementary Logic, 2nd ed. (Oxford University Press, New York).
Monk, J. D.: 1976, Mathematical Logic (Springer-Verlag, New York, Heidelberg, Be
Resnik, M.: 1974, "The Philosophical Significance of Consistency Proofs", Journal of
Philosophical Logic 3, pp. 133-147.
Tarski, A., A. Mostowski, and R. Robinson: 1960, Undecidable Theories (North-Holland,
Amsterdam).

Department of Philosophy and Religion,

North Carolina State University,
Box 8103,
Raleigh, NC 2 7695-8103,
U.S.A.

This content downloaded from

193.0.116.21 on Wed, 07 Feb 2024 12:56:07 +00:00
All use subject to https://about.jstor.org/terms