HISTORY AND PHILOSOPHY OF LOGIC, IS (1994), 211-225

Frege, Hilbert, and the Conceptual Structure of

Model Theory
WILLIAM DEMOPOULOS
Philosophy Department, The University of Western Ontario, London, Ontario N6A 3K7, Canada
Received 12 July 1992
Revised 12 April 1993
This paper attempts to confine the preconceptions that prevented Frege from appreciating Hilbert's
Grund/agi!n tkr Gi!ometrie to two: (i) Frege's reliance on what, following Wilfrid Hodges, I call a
Frege-Peano language, and (ii) Frege's view that the sense of an expression wholly determines its
reference. I argue that these two preronceptions prevented Frege from achieving the conceptual
structure of model theory, whereas Hilbert, at least in his practice, was quite close to the
model-theoretic point of view. Moreover, the issues that divided Frege and Hilbert did not revolve
around woether one or the other allowed metalogical notions. Frege, e.g., succeeded in formulating
the notion of logical consequence, at least to the extent that Bolzano did; the point is rather that
even though Frege had certain semantic concepts, he did not articulate them model-theoretically,
whereas, in some limited sense, Hilbert did.
I. Introduction
In his definitive essay, 'The main trends in the foundations of geometry in
the 19th century', Hans Freudenthal summarizes the 'forerunners and competi
tors' of Hilbert's Grundiagen der Geometrie as follows:
... The father of rigor is Pasch. The idea of the logical status of geometry
occurred at the same time to some Italians. Implicit definition was analyzed
much earlier by Gergonne. The proof of independence by counter-example
was practiced by the inventors of non-Euclidean geometry, and more con
sciously by Peano and Padoa. The segment calculus was prefigured in von
Staudt's 'throw calculus.' The purport of the 'Schliessungssatze' had been
grasped by H. Wiener (1893). Even the title Grund/agen der Geometrie is far
from original. Before Hilbert, a name like this indicated research like
Riemann's and Helmholtz's. Lie's papers of 1890 appeared under this title,
and so did Killing's book in 1893, 1897.
1
Nevertheless, Freudenthal concludes that
... in spite of aU these historical facts we are [rightly] accustomed to identify
the tum of mathematics to axiomatics with Hilbert's Grund/agen: This
thoroughly and profoundly elaborated piece of axiomatic workmanship was
infinitely more persuasive than programmatic and philosophical speculations
on space and axioms could ever be, (p, 619)
If the mathematical basis for our assessment of the place of the Grund/agen in
modem axiomatics is clear, its methodological significance is only slightly less
Freudenthal, 1962. p. 619. All parenthetical page references to Freudenthal arc to this paper.
OI..... ~ $10.00 C 19114 To)'lor A r ....... l.Id
212 William Demopoulos
straightforward, Again it is difficult to improve on Freudenthal's remark that
with Hilbert
... the bond with reality is cut. Geometry has become pure mathematics. The
question whether and how to apply it to reality is the same in geometry as it is
in other branches of mathematics. Axioms are not evident truths. They are
not truths at all in the usual sense .... Hilbert's clean cut between mathemat
ics and realistic science became the paradigm of a new,methodology. (p. 6180
r
In this paper I am concerned with a methodological aspect of the Grund
lagen that is closely related to 'Hilbert's clean cut between mathematics and
realistic science', namely Hilbert's deployment of the category of expression that
we today call non-logical constants. This use is indicated at the very beginning of
the Grundlagen when Hilbert writes, 'We imagine three kinds of things '"
called points ... called lines ... called planes ... We imagine points, lines and
planes in some relations ... called lyinf on, between, parallel, congruent. ,2
Developing a point of Wilfrid Hodges, I hope to show how, by treating
Hilbert's presentation of the standard geometric vocabulary as an anticipation of
the idea of a non-logical constant, it is possible to illuminate the central issue in
his correspondence with Frege, namely, their controversy over Hilbert's
approach to proofs of independence.
4
In arguing this thesis, I hope to make it
plausible that virtually every point of controversy raised in the correspondence is
traceable to their different perceptions of the nature and significance of the
notion of a non-logical constant. Hilbert may have misconstrued non-logical
constants in some respects-here his belief that our use of such expressions
commits us to viewing axioms as a particular sort of definition comes especially
2 It is signalled rather more dramatically in the correspondence with Frege when Hilbert writes:
... you say that my concepts, e.g. 'point', 'between', are not unequivocally fixed. 'Between' is
understood differently on p. 20 and a point is there a pair of points. But it is surely obvious
that every theory is only a scaffolding (schema) of concepts together with their necessary
connections, and that the basic elements can be thou&ht of in any way one likes. E.g., instead
of points, think of a system of love, law, chimney-sweep ... which satisfies aU the axioms;
then Pythagoras' theorem also applies to these things. Any theory can alwllys be applied to
infinitely many systems of basic elements.
Hilbert to Frege, 29.12.99, draft or excerpt by Hilbert. In Frege 1980, p. 42. Hilbert's page
reference corresponds to p. 28f, Section 9 of !be tenth edition of Grund/agi!n der Gwmnrii!.
revised and enlarged by Paul Bemays and translated as Hilbert 1971.
3 Hodges 1985/86. All reference& to Hodges are to this paper ..
4 A passage from Hilbert's second letter to Frege (29.12.1899) contains a summary of the
mathematical accomplishments of the Grun4/agi!fI which emphasizes the importance Hilbert
attached to his trCltment of questions of independence:
[The work makes] it possible to understand those propositions that (arc the) most important results
of geometrical inquiries: that the parallel axiom is not a consequence of the other axioms. and
similarly Archimedes' axiom . . (It Janswers the question whether it is possible to prove that
in two identical rectangles with an identical base line the sides must also be identical, or
whether as in Euclid this proposition is a new postulate. (It makes) it possible to understand
and answer such questions as why the sum of the angles in a triangle is equal to two right
angles and how this fact is connected with the parallel axiom.
The passage is excerpted by Frege, ct. Frege 1980, p. 38f. By 'identical rectangles' Hilbert here
means rectangles of equal area; the relevant chapter for this result in the English translation is
Cb. IV, 'Theory of Plane Area'.
Frege. Hilbert, and the Conceptual Structure of Model Theory 213
to minds -but there can be little room for doubt about his clarity concerning the
novelty and importance of the conception they embody.
In the sense in which I intend it, the emergence of the notion of a
non-logical constant involved considerations that are rather more specific than
what is achieved by the mere separation of logical and non-logical vocabulary
items. In order to have the notion of a non-logical constant, it is not sufficient
merely to have distinguished logical and non-logical vocabulary. since the notion
of a non-logical constant brings with it several others, the most important among
these being the Dotion of a structure (especially, but not exclusively, the notion
of a first-order structure) and the closely related notion of truth in a structure.
These notions are simply not required by the distinction between logical and
non-logical vocabulary. By contrast, the clarity of our conception of a non
logical constant cannot be evaluated without considering our familiarity with the
notion of a structure and of truth in a structure.
6
I am not claiming that Hilbert was clear on all of these matters.' Indeed a
brief historical overview, elegantly presented by Hodges, traces the first fully
explicit isolation of the concept of truth in a structure to the 1957 paper of
Tarski and Vaught.
s
So if we accept Hodges's account, the evolution of our
thinking about tbis family of notions, to anything like our current understanding,
spans some fifty years. I see Hilbert as standing very much at the beginning of
this development, and the very same insight which for Freudenthal embodies
Hilbert's separation of geometry from experience ('the bond with reality is cut'),
also underlies the development of the notions of non-logical constant, structure,
and truth in a structure.
2. Non-logical constants and variables
Let me begin by outlining how I propose to understand the role of
non-logical constants in axiomatization. We characterize a (first-order) formal
language by the set of its non-logical constants, and we conceive of a first-order
structure as a non-empty set together with relations, operations and distin
guished elements. A structure is a structure for the language if it consists of the
right number of distinguished elements, together with relations and operations
of the appropriate arity. How we mark the relations, operations and distin
guished elements is not irrelevant to this presentation. since the labelling
determines, in a canonical way, how the non-logical constants are interpreted in
any such structure. A standard convention uses different type-fonts together
with identical indices or subscripts. The point of such a convention is to insure
that when presented with a description of any structure for the language,
nothing more is required in order to know what, in for example the case of
geometry, are the 'points' of the structure, or what is the 'betweenness' relation,
since the convention determines all of this. The connection between the three
5 Hilbert says of die axioms of order that they 'define the notion of betweenness' (3), and that
the congruenoc axioms 'define the notion of congruence or motion' (is). He does not, however,
use the expression 'implicit definition'.
6 Thus, as I am using the tenn, non-logical constants correspond to what are sometimes called
parameters. See, for example, Enderton 1972, Ct. 2.
7 This is one of several respects in which the account presented here should be contrasted with
Jaalc:lc:o Hintilc:u's discussion in Hintilc:lc:a 1988. Another is that I do not see that metalogical
notions posed any special difficulty for Frege; on this point, see 14 below.
8 Tanlc:i and Vaucht 1957.
214 William Demopoulos
notions is that truth in a structure depends on the interpretation of the
non-logical constants, where the interpretation is given by the convention that
pairs constants of the language with distinguished elements, relations, and
operations of the structure.
This is perhaps familiar. What may be less evident is that the distinction we
mark between non-logical constants and variables-viz. the distinction between
expressions with a determinate meaning, but with no fixed reference, and
expressions lacking any meaning while admitting a conventionally stipulated
reference (within a given 'range')-underlies our conception of a formal lan
guage and our ('modern') view of axioma tics. For example, by our use of the
conventional symbol. '0', for group composition, we intend that in any structure
for the language that satisfies the group axioms. '0' will designate this operation,
even though we certainly do not always have a particular group (or therefore,
group operation) in mind. Thus while '0' is without a fixed reference. its
meaning is clear: 0 is the operation of group composition. The point to notice is
that the situation would be different if '0' were a variable. In that case, a
separate stipulation would have to be made in the case of each group before we
could say, of any formula containing '0', whether or not it is satisfied in that
group. But it is clear that sentences of the theory of groups are not indetermin
ate in truth value merely because we have, in almost all cases, failed to provide
the necessary stipulation. The fact is that it is clear what the referent of '0'
would be in any group, independently of a separate stipulation by us. The
absence of a need for a separate stipulation of this sort is constitutive of what
we mean when we say of '0' that it is a meaningful expression rather than a
variable.
Of course, the view of non-logical constants just sketched is completely
compatible with the existence of non-logical constants with 'minimal' or 'purely
formal' meaning. Typically this happens when we are not axiomatizing the
theory of a class of known structures, but are merely considering the models of
a set of sentences over some arbitrarily given vocabulary; in such a case we have
non-logical constants whose meaning consists in nothing more than that, for
example, 'R
I1
' refers to the seventeenth relation of any structure of the class.
This is all that is necessary in order for the reference of the non-logical constant
to be completely determined in any structure for the language (in the technical
sense of this expression. explained earlier). From the perspective I have been
outlining, there is nothing which would require that non-logical constants must
be without any but purely formal meaning. (Still less is there anything to require
that they be without ellen purely formal meaning). To suppose otherwise is to
blur the distinction between non-logical constants and variables. This supposition
(originating perhaps with Padoa
9
) arises from the mistaken belief that the
meaningfulness of non-logical constants somehow compromises the formal
character of the analysiS of deduction. But this is just a mistake, and one way of
verifying that deductions do not appeal to meaning. while preserving the
9 See for example the following passage from his 'Logical introduction to any deductive theory'
(Padoa 1967. p. 177):
... (Ilt is obvious that a deductive theory has no practical if its undefined symbols and
its unproved propositions do not represent (or cannot represent) ideas and facts, respectively.
Thus the psychological origin of a deductive theory is empirical: Nevertheless. its logical point
of departure can be considered to be a maner of convention.
Frege, Hilbert, and the ConceptUllI Structure of Model Theory 215
distinction between non-logical constants and variables, is to show that the
arguments go through when non-logical constants possessing a preanalytic
meaning ('line,' and 'between" for example) are replaced by 'letters' like 'RI7"
i.e. by non-logical constants whose meaning is minimal in the sense just
explained. This of course is compatible with non-logical constants having more
than minimal meaning.
On the present view, non-logical constants thus behave rather like the
indexicals of a natural language, with the space of structures for the language of
the theory playing a role analogous to the role played by space-time in fIXing the
reference of inde:llical expressions like 'yesterday' and 'today'. Just as the
reference of such an indexical is determined relative to a spatio-temporal
context, the refereoce of a non-logical constant is determined once we are given
a structure for the language. The situation is different with variables. Variables
are without meaning. In the course of evaluating the truth of a sentence in a
structure, we may stipulate a reference for the variables relative to an arbitrary
sequence of elements of the domain of the structure. But we do not regard the
sequence as an essential feature of the structure. Not so the distinguished
elements, relations and operations which interpret the non-logical constants
they are an essential feature of the structure.
It is interesting to compare our remarks on non-logical constants with Frege's
reaction to a paper of E. V, Huntington, one of the early exponents of the
'modem' axiomatic method.
10
Around the time of his correspondence with
Hilbert, Frege' wrote to Huntington about his (i.e. Huntington's) use of
non-logical constants, although Frege did not, of course, put the matter in these
terms. In the coone of the letter, Frege quotes a passage from Huntington's
paper 'A complete set of postulates for the theory of absolute continuous
magnitude': 'If the first of two given elements is denoted by a and the second by
b, then the object which they determine is denoted by a 0 b.' 11 He then
comments:
Here the definite article in 'the object' seems to me incorrect. For what does
'203' mean? We can assume different rules of combination. If we take the
sum then 5 is determined; if we take the product, 6 is determined; and so we
can assume countless further rules, each of which determines a different
number; but we cannot tell from '20 3' which rule is assumed. If the signs
contained some indication of the rule of combination, the matter might be
b
i
different; but as it is 'a 0 does not mean anything. It seems to me that we
are here preseuted with a cross between two different kinds of signs, namely
those that designate (or mean) and those that merely indicate .... [T]he sign
Indeed dun.: the period of elaboration of any deductive theory we cboose the ideas to be
"presented by die undefined symbols and the facts 10 be stated by the unproved propositions:
but, When we bl:Bin 10 formulate the theory, we can imagine that the undefined symbols are
_pletely deWJid of meaning and that the unproved propositions instead of stating facts, that
is, relations between ideas represented by undefined symbols, are simply conditions imposed
upon the undefillcd symbols.
(Padoa's emp/lases bve been omitted.)
IQ See Scanlan 1991_
11 Huntington 1902t1. p. 266, quoted in Frege 1980, p. 58. The editon of the correspondence also
cite lite papen ~ o n 1901/19{)2 and Huntington 1902b.
216 William Demopoulos
'0' seems to me to be a hybrid, for it is not a letter [i.e. a variable), and the
way it is introduced makes us take it for a designating one. But we see that it
does not designate anything, but is used more like an indicative sign [i.e. a
variable). Its hybrid nature gives rise to unclarities about the sense of the
propositions in which it occurs.
12
Hodges has given an accessible explanation as to why Frege found non-logical
constants so problematic: a language like that of the Begriffsschrift, or Peano's
Formulaire-a Frege-Peano language-has logical constants, variables, and a
rather restricted set of natural language expressions which occur with their
ordinary meanings. Such a language elucidates the logical resources implicit in
natural language by having a rigidly defined syntax, and thus constitutes a
logically refined or 'regimented' fragment of the natural language. But it is not,
and was not intended to be, a formal language of the sort with which we are
today familiar. One of the features of natural language that simply has no
analogue in a Frege-Peano language is the presence of indexicals. Non-logical
constants, being like indexical expressions in the way they function, would for
this reason alone make it difficult to express Huntington's axiomatization within
a Frege-Peano language, and Hodges's account therefore rather naturally
suggests itself: Frege rejects the notion of a non-logical constant because his
basic tool for the analysis of concepts- his 'concept-script' - ignores the category
of expression to which they are most similar. Rejecting non-logical constants,
Frege must reject Hilbert-style axiomatizations such as Huntington's, since in
this approach logical constants, variables, and non-logical constants exhaust the
vocabulary of the language of the mathematical theory being axiomatized. I now
want to elaborate this observation into an account of Frege's controversy with
Hilbert over the foundations of geometry.
3. Frege on independence results in geometry
In a letter to Hilbert dated 6.1.1900 Frege raises what he thinks is a general
difficulty for Hilbert's approach to proofs of independence, although he addres
ses only the matter of the parallel postulate:
.. there is a logical danger in your speaking of, for example, 'the parallel
axiom,' as if it was the same thing in every special geometry. Only the
wording is the same; the thought content is different in every different
geometry. It would not be correct to call the special case of Pythagoras'
theorem [that deals with isosceles right triangles] the theorem of Pythagoras;
for after proving that special case, one still has not proved Pythagoras'
theorem. Now given that the axioms in special geometries are all special cases
12 Frege 1980, pp. 58t. The letter to Huntington is undated. Frege's editors conjecture that it was
written in 1902. It is worth remarking that in his monograph, The funda_mal propositions of
modern algebra, Huntington refers to '.' as a variable (the term he uses is 'variable symbol') and
characterizes his 'postulates' as propositional functions. with a refererwe to Russell's Principles
of I"I1Qlhemolics. (The monograph appeared in J. W. A. Young, ed .. Monographs on topics of
modern mathematics, second edition. Longmans, Green and Co. (New York) 1915; d. p. 172.
The first edition, whiCh is not available to me, appeared in 1911.} However. Huntington does
nOI use this lenninology in connection with '0' in any of the papers cited in the correspondence.
The editors' parenthetical addition to fn. 4, p. 58 is misleading rather than clarifying, insofar as
it suggests that Huntington's later terminology was employed in the papen under discussion.
217
Frege, Hilbert, and the Conceptual Structure of Model Theory
of general axioms, one can conclude from lack of contradiction in a special
geometry to lack of contradiction in the general case, but not to lack of
contradiction in another special case.
13
Thus, according to Frege, the proposition Hilbert shows independent of the
other axioms concerns the 'general' concept of a line. But in the case of
Euclidean geometry, we are concerned with the 'special' case of Euclidean lines.
From the fact that two propositions dealing with the general concept are
independent, we cannot infer that propositions which concern the Euclidean
case are independent. Since every independence claim is equivalent to an
assertion of consistency, this falls under an abstract discussion presented earlier
in the letter, namely Frege's remark that
(i]f'a general proposition contains a contradiction, then so does any particular
proposition that is contained in it. Thus if the latter is free from contradiction
we can conclude that the general proposition is free from contradiction, but
not conversely. 14
This remark is clearly incorrect if we understand Frege to be comparing
universal sentences with their instances. IS However, the discussion of Pytha
goras's theorem, which directly follows the remark, suggests that Frege's talk of
particular propositions contained in a general proposition is not concerned with
instances but with propositions which make assertions about concepts of wider
and narrower scope; the former propositions are called 'general', the latter,
'particular'. The examples of right triangle and isosceles right triangle, line and
Euclidean line, support this interpretation. Reading the passage in this way, we
can put Frege's point as follows: Hilbert has shown that it is consistent to assert
the axioms of incidence, order and congruence, together with the denial of the
parallel postulate, of the general concept of a line; but from this it does not
follow that this may consistently be asserted of Euclidean lines, since were we
to spell out the meaning of 'Euclidean line', this would require the parallel
postulate.
Now it really doesn't matter whether or not we agree with F ~ e g e on the issue
of whether Euclidean line presupposes the parallel postulate, since it is clear
that he has missed a crucial point. Anachronisms aside, we can put the matter
this way: On the modern, 'Hilbertian' view, the parallel postulate is true in
some structures and false in others, and to see this, we simply don't need the
notion of a Euclidean line. But for Frege this simple answer is quite problematic.
If sense determines reference, the general proposition can be true where the
particular proposition is false only if the sense of the general proposition (the
proposition which concerns the general notion of a line) is different from the
senses of those propositions that refer to Euclidean lines. But then we are
constantly equivocating and never succeed in proving what was promised,
namely, a result about the Euclidean axiom of parallels, which concerns
Euclidean lines. But surely this just shows that we must give up anything like a
13 Frege }980. p. 48.
14 Frege 1980, p. 47.
IS Since 'Frege [then] appears to be saying that if (x)Fx is inoclDsistent. so is Fa for any a. [But)
this is obviously not sol. Ljet"Fx" be "Ox &. ..,(,)0,,''' Coffa 1986. p. 63. fn. 38.
218 William Demopoulos
strict adherence to Frege's view of sense and reference, if we are to understand
Hilbert's approach to axiomatics. Or to put the point slightly differently: in
order that Hilbert's axioms should be interpretable over Euclidean structures, it
is not necessary that 'line', for example, should have the sense of 'Euclidean
line'. The sense of 'line' is but one component in the determination of its
reference; the other component-the particular structure under consideration
and the way in which it contributes to the reference of the primitives-is simply
ignored by Frege's theory.
This interpretation of Frege's remarks to Hilbert is reinforced by Frege's own
description of the correspondence, conveyed several months later in a letter to
Heinrich Liebmann. Frege writes:
I have reasons for believing that the mutual independence of the axioms of
Euclidean geometry cannot be proved. Hilbert tries to do it by widening the
area so that Euclidean geometry appears as a special case; and in this wider
area he can show lack of contradiction by examples; but only in this wider
area; for from lack of contradiction in a more comprehensive area we cannot
infer lack of contradiction in a narrower area; for contradictions might enter
just because of the restriction. The converse inference is of course permissible.
Hilbert was apparently deceived by the wording. If an axiom is worded in the
same way, it is very easy to believe that it is the same axiom. But it depends
on the sense; and this is different depending on whether the words 'point',
'line', etc. are understood in the sense of Euclidean geometry or in a wider
sense. 16
Frege's difficulty with Hilbert's methodology thus stems from two sources: the
character of his concept-script and his conception of how the sense of a
designating expression determines its reference. Earlier we noted that the
concept-script, being a Frege-Peano language, has, aside from its logical
resources, only variables and non-indexical expressions; the latter have whatever
senses are given them by the natural language to which they belong. The notion
of sense appropriate for the expressions of a Frege-Peano language is one that
determines the reference of expressions independently of a structured context,
whether this consists of locations in space-time, or of structures for a formalized
fragment of the language. While questions concerning the theory of truth for the
sentences of such a language can be posed quite naturally, this is not the case
for the notion of truth in a structure. Even to pose the question of the correct
definition of this concept requires the prior isolation of non-logical constants
expressions which are not variables, but whose reference nevertheless varies
with changes in context. Such expressions differ fundamentally from Frege's
designating expressions, since for Frege the sense of a designating expression
determines its reference 'absolutely', i.e. independently of such contextual
considerations as are relevant in the case of non-logical constants. I' In order to
16 29.12.1900; Frege 1980. p. 91.
17 Frege developed views on the sense of indexical expressions in his laiC essay 'Thoughts'; ct.
Frege 1984, pp. 3S1-72. It has often been noted that these views are rather unintuitive if the
sense of such an expression is identified with its linguistic meaning. As far as I can judge, the
only respect in which Frege is committed to some notion of the 'absoluteness' or 'fixity' of
reference is the one given in the text; By ignoring a central feature of the linguistic meaning of
indexicals and indexical-like expressions. Frege docs not sec the utility of a notion of reference
Frege. Hilbert, and the Conceptual Structure of Model Theory
219
accommodate truth in a structure, the idea that sense completely determines
reference must be modified, and a theoretical specification of a class of
structures must be given. None of this lies at all comfortably within Frege's
conceptual horizon; indeed, it seems to have been very imperfectly understood
even by Hilbert, however closely his practice is in accordance with it.
lt is customary, when discussing the Frege-Hilbert correspondence, to focus
on their controversy over the 'definitional' character of Hilbert's axioms.
18
Our
account of the source of their differences can, I think, clarify the issues this
raises. It seems to me that there are two respects in which something close to
Hilbert's belief that axioms are definitions is correct.
First, there is the often cited fact that the axioms define a class of
models-whether or not, as Hilbert claimed, they succeed in defining the
primitives of the theory.19 Frege is sometimes credited with having perceived
this.20 The idea is that Hilbert, by specifying a class of models, is characterizing
a concept of second level in Frege's hierarchy of concepts. While Frege's
Begriffsschrift presents a characterization of the auxiliary devices and syntactic
principles required for an adequate analysis of deduction in mathematical
theories, it lacks the model-theoretic core that is characteristic of the language
of our mathematical axiomatizations. Frege's theory of concepts and objects,
especially his notion of a concept of second-level, is intended to fill this gap.
Indeed, we can view Frege's attempt to understand Hilbert's practice in terms of
his notion of a concept of second level as a partial defence of the thesis that the
combination consisting of the logical system of the Begriffsschrift together with
his theory of concepts and objects is a suitable framework for the analysis of
mathematical theories. We now know that this defence cannot succeed, since
Frege's 'semantic' sub-theory, dealing with concepts and their extensions, is
inconsistent. But even if the theory were consistent, the suggestion that Hilbert
has defined a Fregean concept of second level cannot, without some modifica
tion, be equated with the contemporary idea that Hilbert's axioms define a class
of structures, since structures, being sets, must fall under Fregean concepts of
first level.
21
The second respect in which Hilbert's idea that axioms are definitions seems
correct is more closely connected with the point we have been developing on
how the reference of a non-logical constant is fIXed. Let us go back to the case
of the symbol for group composition. Given the model-theoretic background
that is dependent on a structured context. While this has serious consequences for his view of
Hilbert's adUevement, I fail to see in it any pejorative 'metaphysical' connotations, such as those
sugested by Michael Hallett in his otherwise very informative paper Hallett 1989. For Frege's
remarks on indexicals see especially Perry 1977. For an opposing view, see Evans 1985. Michael
Dummett disc:u.sses the Evans-Perry controversy in the chapter 'The relative priority of thought
and language', of Dummett 1991.
18 For example, Otapter 3 of Resnik 1981) contains a diS(:ussion of the correspondence along these
lines.
19 Bemays 1942. See Resnik 1981). p. 113 for an account of Bernays' review. For a related
discussion see Stein 1988.
20 Beginning with Bemays 1942. The point is further developed by Dummett 1976 and by Resnik
1973/74. For Frege's attempt to bring his hierarchy of levels to bear on the issues, see the letter
to Liebmann, quoted earlier, and the letter to Hilbert of 6.1.1900. For a more extended
development. see 'On the foundations of geometry: first series', translated by Eike-Henner W.
Kluge. and collected in Frege 1984. pp. 273-284; see especially, pp. 283f.
21 To the best of my knowledge, this observation was first made by Hodges.
220 William Demopoulos
against which an axiomatization is presented, what is required to assign a
reference to '0' is not some further specification of its sense or meaning, but the
specification of the particular group over which the language is interpreted.
Thus, insofar as the reference of '0' does not depend on more theory, it is
plausible to hold (with Hilbert) that the axioms do give a complete specification
of the meaning of '0'. It would, however, be absurd to attempt to derive from
this observation a general theory of meaning, or even a theory of the meaning
of the primitives of an axiomatic theory. To begin with, the whole process of
fixing the reference of, for example, '0', succeeds only because the linguistic
context of the background model theory is presupposed by the analysis. But
since the intended subject-matter of a general theory of meaning is supposed to
include the language of the background theory, the first difficulty with such a
suggestion is that it simply does not address this topic. At best the proposal is a
theory of the reference of the primitives relative to the assumption that the
reference of the expressions of the background theory has somehow been fixed.
Since this assumption implies that we can label the group operation (and its
inverse) in any group, the achievement of such a theory of meaning is rather
modest.
22
It is in any event certainly not necessary to view Hilbert as presenting
us with a theory of the meaning and reference of the language of our
background theory. What seems rather to be the case is that Hilbert presup
poses the clarity of these (background) notions and proceeds to articulate a
technical framework for dealing with metatheoretical issues. Part of the basis for
Frege's refusal, even to acknowledge Hilbert's relative success, resided in his
resistance to the implication that the axioms of geometry are not truths in the
usual sense, since this would relegate geometry to the status of any other branch
of 'pure mathematics', and would thus deprive it of its foundational signifi
n
cance.
4. Frege on logical dependence
Several years after his correspondence with Hilbert and Liebmann, Frege
published a series of papers on the foundations of geometry.24 It is only at the
conclusion of the last of these papers that Frege gives serious attention to the
nature of what he caIls logical dependence. Frege's purpose is to clarify the
methodology of independence proofs, and, in particular, to explain what is
involved in establishing that one 'real axiom' is independent of another. What
he has to say in connection with the notion of dependence marks a major
concession to the correctness and coherence of Hilbert's claim that his independ
ence results hold of the traditional geometric axioms.2S Another point of interest
22 Cf. e.g., Coffa 1986. Although Coffa is somewhat guarded about his own views, I think the
general orientation of the paper suggests the view of meaning criticized in the text.
23 For a diS(:ussion of the geometric tradition in which Frege worked, see Wilson 1992.
24 'On the foundations of geometry: first series' in Frege 1984. pp. 273-284, and 'On the
foundations of geometry: second series' in Frege 1984, pp. 293-340.
2S That this essay marks such a concession has also been noted by Dummett 1976. In the
posthumously published lecture-text of 1914, 'l..Qgic io mathematics' (ct. Frege 1979, pp.
203-250), Frege recounts many of the polemical remarks of the series of papers on geometry,
repeats the charge that Hilben did not show the independence of the Axiom of Parallels from
the other Euclidean axioms. and fails altogether to mention the remarks on IogiCIII dependence
diS(:U$SCd below. See Frege 1979, pp. 247ft.
221
Frege, Hilbert, and the Conceptual Structure of Model Theory
is the extent to which Frege's discussion of der.:ndence is reminiscent of
Bolzano's account of 'deducibility' [Ableitbarkeitj,2 and therefore suggestive of
Tarski's 1936 formulation of the concept of logical consequence?'
In a review of the first English translation of this series of papers, Howard
Jack
son
28 presented a rather natural reconstruction of a sufficient condition for
independence; Jackson's reconstruction is based on the notion of a pseudo
proposition, which Frege introduced in the second Number of the second series
of essays.29 Taking Hilbert's axioms as pseudo-propositions (expressed by open
formulas containing individual, concept, and relation variables in place of the
geometric primitives) pseudo-axiom G is said to be independent of the (finitely
many) pseudo-axioms Q if the universal closure of the conditional, consisting of
G as consequent and the conjunction of the pseudo-axioms in g as antecedent,
is false. Jackson observed that if Frege thought the notion of independence
could not be applied to Hilbert's work via the apparatus of pseudo-propositions,
he was simply mistaken. On this Jackson is certainly right. But by focussing on
an account in terms of pseudo-propositions, Jackson's discussion might suggest
that Frege did not significantly extend Russell's notion of formal implication. In
a similar vein, John Corcoran and Susan Wood
JO
remark that, given the context
of his debate with Hilbert, Frege's failure to formulate the notion of logical
consequence is a 'curious nonevent in the history of logic'.
Now in fact, in his discussion of dependence, Frege comes very close to
formulating a notion of logical consequence. Frege is here concerned with the
independence of one 'real axiom' from another, where a real axiom is expressed
by a sentence whose sense is wholly determinate and whose reference is fixed
'absolutely'. Such an axiom is also true. Frege's idea
31
may be explained as
follows.
Let g U {G} be a set of propositions, expressed in a Frege-Peano language.
We may think of the pair, (g, G), as a 'premise-conclusion argument', with the
propositions in Q the premises, and G the conclusion, of the argument. (In
accordance with our current practice, Frege suppresses explicit mention of
logical laws.) The only difference between this contemporary usage and Frege's
is that Frege appears to be interested only in the case where g is a set of true
premises. If g consists of the axioms of some special science, then the premises
must be true, since the notion of a false axiom is an oxymoron for Frege.
'1fJ BoIzano 1837, section ISS, paragraph 2. (See also the Note to this section.) Frege is perhaps
closer to BoIzano's notion of the ground-oonsequence relation (AbfolgeJ, which, for Balzano,
holds between /I'Ulhs, rather than propositions in general.
27 Tarsld 1956. As we noted earlier, the concepts of non-logical constant, structure, and truth in a
structure make their first fully explicit appearance only in Tarski's post-war work. The distance
between Tarski's paper (first published in Polish and German in 1936) and the model-theoretic
analysis of logical consequence is discussed in Etchemendy 199(). Etchemendy's principal
historical point. centering on the divergence of Tarski's early notion of a model from our current
understanding, and the absence, in Tarski's 1936 paper, of the notion of a variable domain, was
noted in Corcoran 1972p3. For related observations on Bolzano, see Section IX of Berg's
introduction to his edition of BerMrd Bolzano: theory of science; 1973, pp. 20-22.
28 Jackson, 1981.
29 Modulo attention to use and mention, Frege's notion of a pseudo-proposition is evidently closely
related to Russell's notion of a propositional function, and may perhaps have been inspired by
it.
30 Corcoran and Wood 1972P3
31 See Frege 1984, pp. 335-9.
222 William Demopoulos
However Frege does not require that the propositions in g must be known to
be true.
For each type of non-logical expression E, i.e. for each concept or relation
expression, and for each proper name, we let E' be another meaningful
expression of the same grammatical type, designating a concept or relation of
the same logical 'level' as that designated by E. E and E' are assumed to be
expressions of the same Frege-Peano language. Lastly, we assume that the
association of E with E' is one-one. Evidently the mapping E E' induces a
mapping G G' on sentences (and thus, on sets of sentences) when, as in the
case of a Frege-Peano (or a 'logically perfect') language, the syntax is appropri
ately specified.
Frege asserts, as a 'basic law' [Grundgesetz] of dependence, that G is
dependent upon g only if there is no such mapping E E' from expressions to
expressions, such that Q' is true and G' is false under the mapping:
Let us now consider whether a thought G is dependent upon a group of
thoughts Q. We can give a negative answer to this question if, according to
our vocabulary, to the thoughts of the group g there corresponds a group of
true thoughts Q', while to the thought G there corresponds a false thought
G'. For if G were dependent upon Q, then since the thOUghts of g' are true,
G' would also have to be dependent upon g' and consequently G' would be
true.32
What suggests that Frege is here attempting an account of logical consequence is
the fact that he is explicitly concerned with cases where he must pass from the
truth of one proposition to a consideration of the logical possibility of the truth
of another proposition, which may in fact be false. This is precisely what is
effected by the introduction of the mapping E ...... E'.
Frege's presentation of his basic law of dependence follows an earlier, and
more characteristic account of logical dependence in terms of inferability:
Let Q be a group of true thoughts, Let a thought G follow from one or
several of the thoughts of this group by means of logical inference such that
apart from the laws of logic, no proposition not belonging to g is used. Let us
now form a new group of thoughts by adding the thought G to the group C.
Call what we have just performed a logical step. Now if through a sequence of
such steps, where every step takes the preceding one as its basis, we can reach
a group of thoughts that contains the thought A, then we call A dependent
upon the group g.33
By advancing his basic law of dependence as a law, Frege implicitly
acknowledges that he is not merely giving an elucidation of what was meant by
the earlier characterization in terms of inferability; otherwise he could not,
consistently with the principles set forth in this series of essays, claim to have
formulated a law of dependence. However, I have found nothing in his
32 Frege 1984, p. 429. Frege passes freely between sentences and the thoughts or propositions they
express without an explicit change of notation. leaving it to the context to make clear wbich is
intended. Our exposition follows him in this practice.
33 Frege 1984, p. 334.
Frege, Hilbert, and the Conceptual Structure of Model Theory 223
discussion to suggest that he considered the possibility that (Q, G) might satisfy
the basic law, even though G is not logically inferable from Q, without, at the
same time, attributing this possibility to a limitation of the mapping, E ..... E'. To
the extent that this is unclear, the attribution to Frege of a concept of logical
consequence must remain somewhat tentative. Nevertheless, it should be noted
that exactly the same unclarity arises in the case of Bolzano, and Bolzano is
routinely accorded priority on this matter.
34
By contrast, for Tarski there is no
such unclarity: Tarski's account of logical consequence is explicitly designed to
avoid precisely the limitations of analyses based on mappings like E ..... E'.
We have argued that while there is in his discussion of dependence an echo
of Bolzano and (a pre-echo) of Tarski circa 1936, Frege never achieved the key
elements of the model-theoretic explication of logical consequence: non-logical
constants, structures, and the concept of truth in a structure. On our analysis,
Frege failed to achieve these components of the model-theoretic picture as a
direct result of the character of his concept-script and his theory of sense and
reference; that theory was conceived for a language which, like the language of
the Begriffsschrift, does not require the notions of relative reference and relative
truth, notions that are characteristic of Hilbert's anticipation of a model
theoretic point of view. It should be noted that in arguing this thesis we have at
no point maintained that Frege lacks a metatheoretical perspective. And indeed,
that, in the essays on geometry, Frege had such a perspective is evident from his
formulation of his 'basic law' and his observation that
. .. to prove the independence of a real axiom from a group of real axioms
. . . we [must] enter into a realm that is otherwise foreign to mathematics. For
although like all other disciplines mathematics, too, is carried out in thoughts,
still thoughts are otherwise not the object of its investigations. Even the
independence of a thought from a group of thoughts is quite distinct from the
relations otherwise investigated in mathematics. Now we may assume that this
new realm has its own specifiC, basic truths which are as essential to the
proofs constructed in it as the axioms of geometry are to the proofs of
geometry; and that we also need these basic truths especially to prove the
independence of a thought from a group of thoughts.
3S
Our account accommodates these facts while explaining Frege's failure to pursue
the development of a framework for treating metatheoretical questions along
model-theoretic or 'Hilbertian! lines. If we have diverged from earlier discus
sions of the Frege-Hilbert controversy, it is by locating its basis in differences in
their grasp of the conceptual structure of model theory-non-Iogical constants,
structures, and truth in a structure-and, perhaps more importantly, byaccount
ing for Frege's failure to grasp this framework, not in terms of some vague
inability to deal with semantic or metatheoretic ideas, but in terms of the clash
34 For example, Corcoran 1973, p. 71, says of BoIzano that he 'Indy deserves credit for (the)
explication of logical consequence, if Tarsld does, beGause ... Balzano offered precisely the
~ m e idea.' But uide from the restriction to true premises, there is no difference between
Bolzano's characterization of deducibility and Frege's basic law of dependence. Bolzano's
notions of proposition-in-itself, idea-in-itself, and indeed, the whole tenor of his discussion, is
quite dearly anticipatory of Frege.
3S Frege 1984, p. 425. Among commentators, only Dummett 1976 appears to have perceived the
importance of the papers on foundations of geometry for our appreciation of Frege's grasp of
such issueS.
224
William Demopoulos
between his theory of sense and his concept-script on the one side, and the
'indexical' character of non-logical constants on the other.
Acknowledgements
I wish to thank John Corcoran, Michael Dummett, Michael Friedman,
Michael Hallett, Howard Jackson, Michael Resnik, and Steven Wagner for
comments on earlier drafts. I am especially indebted to Mark Wilson for several
conversations on the topics dealt with in this paper. Financial support from the
Social Sciences and Humanities Research Council of Canada is gratefully
acknowledged.
References
Boizano, Bernard 1837 Wissenscha/tslehre, (Sulzbach). Partial translations in Jan Berg,
ed, Bernard Bolzano: theory of science, Dordrecht, (Reidel), 1973, tr. Burnham
Terrell, and in Bernard Bolzano: theory of science, ed. & tr. Rolf George, Blackwell
(Oxford) 1972.
Bernays, Paul 1942 Review of Em unhekannter Brief von Gotdob Frege aber Hilberts
erste Vorlesung aber die Grundlo.gen der Geometrie, M. Steck, cd., Journal of
symbolic logic 7, 92-93.
Coffa, Alberto 1986 'From geometry to tolerance', in Robert Colodny, ed., From quarks
to quasars, University of Pittsburgh Press. (Pittsburgh).
Corcoran, John 1972/73 'Conceptual structure of classical logic', Phik;Jsophy and
phenomenalogical research 33,25-47.
--1973 'The meanings of implication', Dialogos, 9, 59-76 .
Corcoran, John and Susan Wood 1972/73 Review of Eike-Henner W. Kluge, ed. and tr.,
On the foundations of geometry and formal theories of arithmetic, Yale University
Press (New Haven), 1971, Philosophy and phenomenological research 33, 283-86.
Dummett, Michael 1976 'Frege on the consistency of mathematical theories', in Matthias
Schirn, ed., Studien zu Frege, Frommann-Ho!zboog (Stuttgart) 1976 Vol. I, pp.
229-42. Reprinted in Dummett 1991, pp. 1-16.
--1991 Frege and other philosophers, Oxford University Press (Oxford).
Enderton, Herbert, B. 1972 A mathematical introduction to logic, Academic Press. (New
York).
Etchemendy, John 1990 The concept of logical consequence, Harvard University Press.
(Cambridge).
Evans, Gareth 1985 'Understanding demonstratives', in his Collected papers, Oxford
University Press. (Oxford), pp. 291-321.
Frege, Gottlob 1979 Posthumous writings, ed. H. Hermes et al., tr. P. Long & R. White
with R. Hargreaves. Blackwell (Oxford).
--1980 Philosophical and mathematical correspondence, ed. G. Gabriel et al., abr.
B. McGuinness, tr. H. Kaal. Blackwell (Oxford).
1984 Collected papers on mathematics, logic, and philosophy, ed. B. McGuinness, tr.
M. Black et al. Blackwell (Oxford).
Freudenthal, Hans 1962 'The main trends in the foundations of geometry in the 19th
century', in Ernest Nagel et aI., eds., Logic, methodology, and philosophy of science,
Stanford University Press (Stanford), pp. 613-21.
Hallett, Michael 1990 'Physicalism, reductionism & Hilbert', in A. D. Irvine, ed.,
PhYSicalism in mathematics, Kluwer. (Dordrecht), pp. 183-257.
Hilbert, David 1899 Grundlo.gen der Geometrie. Teubner (Stuttgart). English translation
Foundations of geometry. Open Court (La Salle), 1971, tr. Leo Unger, from the 10th
German ed., rev. & enl. by P. Bernays, 1968.
Hintikka, Jaakko 1988 'On the development of the model-theoretic viewpoint in logical
theory', Synthese 77, 1-36.
Hodges, Wilfrid 1985/86 'Truth in a structure', Proceedings of the Aristotelian Society,
new series 86, 135-51.
225 Frege, Hilbert, and the Conceptual Structure of Model Theory
Huntington, E. V. 1901/1902 'Simplified definition of a group', Bulletin of the American
Mathel1Ultical Society, 2nd Series, 8, 296-300.
--1902a 'A complete set of postulates for the theory of absolute continuous mag
nitude', Transactions of the American Mathematical Society 3, 264-79.
--1902b 'Complete sets of postulates for the theories of positive integral and positive
rational numbers', Transactions of the American Mathel1Ultical Society 3,280-84.
--1915 The fundamental propositions of modern algebra, in J. W. A. Young, ed.,
Monographs on topics of modem I1UIthel1Ultics, second edition umgmans, Green and
Co. (New York).
Jackson, Howard 1981 Review of G. Frege, On the foundations of geometry and forl1Ull
theories of arithmetic, Eike-Henner W. Kluge, ed. and tr., Yale University Press
(New Haven) 1971, Journal of symbolic logic 46, 175-79.
Padoa, Alessandro 1967 'Logical introduction to any deductive theory', in van Heije
nool1, ed., 1967, pp. 118-23 English translation by Jean van Heijenool1.
Perry, John 1977 'Frege on demonstratives', Philosophical review 86, 474-97.
Resnik, Michael 1973/74 'The Fregf,:-Hilbert controversy', Philosophy and phenomeno
logical research 34, 386-403.
--1980 Frege and the philosophy of I1UIthel1Ultics, Cornell University Press (Ithaca).
Scanlan, Michael 1991 'Who were the American postulate theorists?' Journal of symbolic
logic 56,981-1002.
Stein, Howard 1988 'Logos, logic, and logistiki': some philosophical remarks on
nineteenth-century transformation of mathematics', in W. Aspray and Philip Kitcher,
eds., History of philosophy of modem I1UIthematics, University of Minnesota Press.
(Minneapolis), pp. 238-59.
Tarski, Alfred 1956 'On the concept of logical consequence', in John Corcoran, ed.,
Logic, semontics, metal1Ulthel1Ultics, second edition, Hackett. (Indianapolis), 1983,
pp. 409-420. English translation by J. Woodger.
--1986 Collected papers, ed. S. R. Givant and R. N. McKenzie. 4 vols. Birkhauser
(Basel).
Tarski, Alfred and R. L. Vaught 1957 'Arithmetical extensions of relational systems',
Composito mathemotica 13,81-102. Reprinted in Tarski 1986, Vol. 3, pp. 653-74.
van Heijenool1, Jan, ed. 1967 From Frege to Gi:ldel: a sourcebook in I1UIthel1Ulticallogic.
Harvard University Press (Cambridge).
Wilson, Mark 1992 'Frege: the royal road from geometry'. NoUs 26, 149-80.