You are on page 1of 390


Managing Editor:
JAAKKO HINTIKKA, Florida State University, Tallahassee
DONALD DAVIDSON, University of California, Berkeley
GABRIEL NUCHELMANS, University of Leyden
WESLEY C. SALMON, University of Pittsburgh
Essays on the Philosophical and Foundational Work
of Gottlob Frege
Edited by
Academy of Finland
Department of Philosophy, Florida State University
Library of Congress Catalogingin-Publication Data
Frege synthesized.
(Synthese library; v. 181)
Bibliography: p.
Includes indexes.
1. Frege, Gottlob, 1848-1925. 1. Haaparanta,
II. Hintikka, Jaakko, 1929-
B3245.F24F72 1986 193 8tHi523
Leila, 1954-
ISBN13: 9789401085236 eISBN13: 9789400945524
001: 10. 1007/9789400945524
Published by D. Reidel Publishing Company,
P.O. Box 17, 3300 AA Dordrecht, Holland.
Sold and distributed in the U.SA. and Canada
by Kluwer Academic Publishers,
190 Old Derby Street, Hingham, MA 02043, U.S.A.
In all other countries, sold and distributed
by Kluwer Academic Publishers Group
P.O. Box 322, 3300 AH Dordrecht, Holland
All Rights Reserved
1986 by D. Reidel Publishing Company
Reprint of the original edition 1986
and copyrightholders as specified on appropriate pages within
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner
Introduction 3
JOAN WEINER / Putting Frege in Perspective 9
J. VAN HEIJENOORT / Frege and Vagueness 31
HANS SLUGA / Semantic Content and Cognitive Sense 47
THOMAS G. RICKETTS / Objectivity and Objecthood: Frege's
Metaphysics of Judgment 65
TYLER BURGE / Frege on Truth 97
LEILA HAAPARANTA / Frege on Existence 155
MICHAEL D. RESNIK / Frege's Proof of Referentiality 177
NINO B. COCCHIARELLA / Frege, Russell and Logicism: A
Logical Reconstruction 197
ROBERT B. BRANDOM / Frege's Technical Concepts: Some
Recent Developments 253
PHILIP KITCHER / Frege, Dedekind, and the Philosophy of
Mathematics 299
G REG 0 R Y CUR R IE / Continuity and Change in Frege's
Philosophy of Mathematics 345
A. W. MOORE and ANDREW REIN / Grundgesetze, Section
10 375
Gottlob Frege's philosophical and foundational work was by any token
a major factor in the development of contemporary analytic philosophy.
Some say he was the grandfather of the whole tradition, some think of
him merely as its godfather. In either case, one might expect that his
work has been studied exhaustively. However, this turns out not to have
been the case. Someone - it was probably Burton Dreben - once said
that the worst-known period in the history of philosophy is always the
time fifty to a hundred years ago. The intensive work on Frege which
has been going on in the last decade and a half at first seems to belie this
dictum, but in a looser sense it fits the facts well. For it is only the
developments of the last couple of decades, largely of the last few years
when Frege's work has reached the hundred-years mark, that have
brought to light facts and issues which have shown that our understand-
ing of Frege was seriously incomplete. In recent literature, one can also
find a wealth of new and sometimes controversial viewpoints. For
instance, Jean van Heijenoort has called our attention to an important
but neglected aspect of Frege's attitude to logic and language that he
calls "logic as language". Hans Sluga has challenged on a large scale the
received view of Frege as a lonely figure in nineteenth-century phi-
losophy whose ancestry goes to medieval objectivists rather than his
German predecessors. Sluga wants to place Frege firmly in the middle
of the German philosophical tradition of his day. It is indeed unmistak-
able that there are, for instance, Kantian elements in his thinking that
had earlier been overlooked. Indeed, the idea of logic as language is
likely to be one of them. Another one is the sharp contrast between the
realm of thinking and understanding and the realm of sense and intui-
tion. Sluga's influence is illustrated amply in several papers in this
volume. In an attempt to reverse the traditional priorities, Jaakko
Hintikka has suggested, relying partly on van Heijenoort's interpreta-
tion, that the crucial part of Frege's work in semantics lies in his ideas
about the semantics of the familiar elementary logic (truth-functions and
quantification) rather than in Frege's theory of sense and reference,
which is merely intensional frosting on a more important extensional
L. Haaparanta and 1. Hintikka (eds.), Frege Synthesized, 3-8.
1986 by D. Reidel Publishing Company.
cake, even though it is typically given the pride of place in expositions in
Frege's semantics. As a part of this attempted reversal of emphasis,
Jaakko Hintikka has also called attention to the role Frege played in
convincing almost everyone that verbs for being had to be treated as
multiply ambiguous between the "is" of identity, the "is" of predication,
the "is" of existence, and the "is" of class-inclusion - a view that had
been embraced by few major figures (if any) before Frege, with the
exception of John Stuart Mill and Augustus De Morgan. Hintikka has
gone on to challenge this ambiguity thesis. At the same time, Frege's
role in the genesis of another major twentieth-century philosophical
movement, the phenomenological one, has become an important issue.
Even the translation of Frege's key term "Bedeutung" as "reference" has
become controversial.
The interpretation of Frege is thus thrown largely back in the melting
pot. In editing this volume, we have not tried to publish the last word on
Frege. Even though we may harbor such ambitions ourselves, they are
not what has led to the present editorial enterprise. What we have tried
to do is to bring together some of the best ongoing work on Frege. Even
though the ultimate judgment on our success lies with out readers, we
want to register our satisfaction with all the contributions.
The first paper included in this volume is an exploratory attempt to
reach a general perspective on Frege's work. Joan Weiner stresses in her
paper 'Putting Frege in Perspective' that once we start trying to under-
stand Frege's work as a reaction to Kantian epistemology, it is urgent to
pay attention to it also as an important part of the background of other
philosophical work. According to Weiner, the problem will be "how one
would connect Frege, the nineteenth-century neo-Kantian, with Frege,
the first twe.ntieth-century philosopher of language and mathematics".
Weiner suggests an answer to the question. She holds the opinion that
Frege subscribed to most of Kantian epistemology. However, since he
substituted his new logic, "the real formal rules of thought", for tradi-
tional Aristotelian logic, he had to modify his Kantian This,
according to Weiner, was the basic' content of virtually all of Frege's
subsequent work. Weiner points out that what we regard as Frege's
philosophical work was for him not a genuine philosophical theory but
could only be a. set of SUbjective clues. On Weiner's view, it was
designed to get us into the proper frame of mind "for understanding the
purpose and significance of his real philosophical work - the work we
think of today as his mathematical work." .
Weiner's paper thus essays into the same general direction Jaakko
Hintikka's reversal of the traditional view of Frege's semantical prior-
ities mentioned in the beginning of this Preface.
This volume is divided in four parts. The second part consist of
articles dealing with Frege's semantical and epistemological doctrines.
Jean van Heijenoort discusses in his paper Frege's view of vagueness.
He states that Frege's principle of completude (Grundsatz der Vollstiin
digkeit), according to which any function must be defined for all objects,
is closely connected with the requirement of sharpness of concepts, for
Frege takes both lack of completude and lack of sharpness to be a
failure of universal bivalence. Van Heijenoort concludes that Frege
could have rejected the principle of completude, and still preserved
bivalence, and that even if Frege had to ignore vagueness and other
vagaries in his own project, it is perhaps time for us to look at them
more carefully.
In his article 'Semantic Content and Cognitive Sense', Hans Sluga
compares Russell-style theories of meaning with Frege-style ones. In the
former theories, it is assumed that a satisfactory theory of meaning can
be built by means of a binary relation, while the supporters of the latter
theories regard a three-place relation (an expression e refers to an entity
r through having a sense s) as necessary. Sluga discusses Frege's view,
according to which a satisfactory theory of meaning must explain the
difference between trivially true and informatively true identity state-
ments, and attempts to show that Russell-style theories can, at least
partially, satisfy Frege's requirement after all, provided that we are con-
cerned with the difference between semantically trival and semantically
informative identity statements. Furthermore, Sluga argues in his paper
that it was the problem of the status of arithmetical truths that made
Frege take a critical stand against earlier semantic doctrines and intro-
duce a cognitive notion of sense.
Thomas G. Ricketts argues in his article that, for Frege, ontological
categories are secondary with respect to logical ones, and that under-
standing the character of Frege's contrast between objective and sub-
jective may help us see the primacy of judgment in Frege's thought.
According to Ricketts, "our grasp of the notion of an object ... is
exhausted by the apprehension of inference patterns and the recognition
of the truth of the basic logical laws in which these [first-level] variables
figure." Ricketts takes similar remarks to apply to the notion of concept.
Thus he argues that, in Frege's philosophy, the objecthood of thoughts
does not explain the objectivity of judgment but presupposes it.
lt is clear that Ricketts' starting-point is a number of observations
closely related to van Heijenoort's results mentioned in the beginning of
this Introduction.
In his paper, entitled 'Frege on Truth', Tyler Burge suggests that
Frege's odd-sounding conclusion about truth and falsity should be taken
seriously. In the first section of his article he claims that too little
attention has ben paid to the pragmatic basis of Frege's view that truth
values are objects. According to Burge, Frege is committed to the
doctrine that logic is primarily concerned with the normative notion of
truth. The second section of Burge's paper consists mainly of the
criticism of Dummett's interpretation of Frege's theses on truth values.
In section III Burge purports to show how Frege's identification of the
truth values with particular objects has its sources in "some of his
deepest philosophical conceptions". He holds the view that "in particu-
lar, it proceeds from a theory about the nature of logical objects, from a
thesis about the aim and ordering of logic, and from his conceptions of
assertion and truth."
In her article 'Frege on Existence' Leila Haaparanta emphasizes that
Frege's greatest insight was the idea of first-order language, which, to a
large extent, motivated the rest of his innovations. Haaparanta focuses
her attention on Frege's concept of existence, which receives special
attention in Frege's thought in connection with the thesis concerning the
ambiguity of such words for being as the English 'is'. The ambiguity
thesis was an important part of the Fregean paradigm of first-order
logic. Haaparanta argues that Frege does not only assume the word 'is'
to be ambiguous but that he considers 'exists', or the 'is' of existence, to
be an equivocal word. She suggests that the equivocity view has a meta-
physical and epistemological background in Frege's thought. Her paper
thus pushes a great deal further the suggestions of laakko Hintikka
mentioned earlier in this Introduction.
The third part of this volume is mainly focused on Frege's logical
theory. Michael D. Resnik's paper 'Frege's Proof of Referentiality' deals
with Frege's methodological principle according to which in a properly
constructed scientific language every name must have a reference. Frege
tries to prove in the Grundgesetze that his own system satisfies the
principle. Resnik shows that Frege's proof contains a number of serious
mistakes and that it would not prove what Frege wanted even if it were
correct. Moreover, he argues that reasonable demands of rigor do not
even require such a proof.
It is a widely held opinion that logicism is of no importance, as far as
current discussion of the foundations of mathematics is concerned. Nino
B. Cocchiarella, for his part, believes that logicism can be defended in
essentially the same philosophical context in which it was originally
presented. He formulates separate reconstructions of Frege's form of
logicism and of Russell's early form of logicism, which he takes to be
closely similar. He reconstructs both of the logicisms as second-order
predicate logics in which nominalized predicates are allowed to occur
as abstract singular terms. The basic difference between Frege's form
of logicism and Russell's early form of logicism is, according to
Cocchiarella, that for Frege there is a sharp distinction between
concepts denoted by usual predicates and concept-correlates denoted
by nominalized predicates, whereas for Russell concepts are their own
In the paper 'Frege's Technical Concepts: Some Recent Develop-
ments' Robert Brandom praises Dummett's work on Frege, for Dummett
realizes both the necessity of considering Frege's technical concepts in
the framework of contemporary philosophy and the possibility of
considering contemporary philosophical issues in relation to Frege's
technical concepts. He reviews two recent books, David Bell's Frege's
Theory of Judgment and Hans Sluga's Gottlob Frege, against the back-
ground of Dummett's work and of current interpretive controversies.
Finally, he discusses the justification of the definition of a value-range
(or a course-of-values, Werthverlauf) in the Grundgesetze, which yields a
criticism of Frege's procedure in introducing such concepts as reference,
sense, and function, as well as logical objects like numbers.
The articles included in the fourth part of this volume deal with
Frege's philosophy of mathematics. In his paper, 'Frege, Dedekind, and
the Philosophy of Mathematics,' Philip Kitchel' argues that Frege
accepts most of Kant's philosophy of mathematics, but tries to improve
it by showing how the whole of mathematics could be traced to the
sources of a priori knowledge and how arithmetic can be traced to the
sources of a priori analytic truths instead of pure intuition. Kitcher
defends the claim that, due to his philosophical conviction, Frege
bequeathed to his successors a misguided picture of the central prob-
lems of the philosophy of mathematics. Instead of the Fregean view,
Kitcher recommends us a Dedekindian view of mathematics, which is
also Kantian but which differs from Frege's approach in certain im-
portant respects. Frege tries to find a firm route from basic principles of
logic to the theorems of arithmetic, while Dedekind tries to show why
our representations are inevitably arithmetically structured. It is true
that Frege mentions in the Grundlagen that mathematics seems to be
inescapable in our representation of experience, but unlike Dedekind,
he does not develop the idea in more detail.
In his paper 'Continuity and Change in Frege's Philosophy of Mathe-
matics', Gregory Currie presents a largely historical analysis of Frege's
theory of real numbers in a framework which includes Frege's concept
of a value-range and his view of the relation between arithmetic and
geometry. More specifically, Currie discusses the role of the following
three principles in the development of Frege's thought: (1) To every
concept there corresponds an object, the extension of that concept. (2)
The applicability of the real numbers to measurement marks a theore-
tically important distinction between them and the natural numbers. (3)
There is a sharp distinction between the sources of arithmetical and of
geometrical knowledge. Currie concludes that the rejection of the first
principle brought about the rejection of the third principle, which, for its
part, brought about the rejection of the second principle.
The paper by A. W. Moore and Andrew Rein is an attempt to resolve
a problem in Section 10 of Frege's Grundgesetze, which Michael
Dummett calls Frege's permutation argument. The argument seems to
show that, if there is one assignment of objects to value-range terms
which satisfies Axiom V, then there are several. Frege suggests that we
can overcome the resulting indefiniteness by demanding that it is
specified for every function, when it is introduced, what values it takes
on for value-ranges and other objects (if such there be) as arguments.
However, this does not help us to ensure that a unique assignment of
objects to value-range terms satisfies Axiom V. By strengthening the
permutation argument, Moore and Rein establish a more radical form
of indeterminacy, which says that "there is no distinguishing, from
within a given theory, between isomorphic models of that theory". They
argue that the real import of the argument is that it cannot be
determined whether or not either truth-value is the value-range of any
given function and that, elaborated this way, the problem suggests its
own solution.
Until recently, there have been few attempts to read the work of Gottlob
Frege in historical context. Many of those who read Frege's writings
today believe that the usual reasons for reading a piece of philosophy in
historical context do not apply to those pieces Frege produced. This
may be a result of the popularity of two views concerning the nature of
Frege's work. One is that Frege's work has no serious philosophical
background. This view seems somewhat plausible given that Frege was
trained as a mathematician, corresponded with mathematicians through
out his career, and seems to have done work only in the quite specialized
areas of philosophy which are directly concerned with mathematics.
According to the story which goes with this view, Frege began with
philosophy of mathematics (investigating problems involved with his
mathematical work which were not exactly mathematical problems) and
saw that some work in the philosophy of logic (and later in the
philosophy of language) was nef'essary. Thus, almost accidentally, Frege
was pulled deeper and deeper into the problems of philosophy. If this
story were true, it would not be unreasonable to assume that his worries
and problems would not have been muddied by the sort of philosophi
cal assumptions and philosophically loaded terms whose prevalence in
other philosophical writings makes it so important that one considers
their historical context.
The tremendous influence Frege has had on contemporary analytic
philosophy has also led people to ignore historical context. Frege is not
only responsible for formulating modern logic, but also seems respon
sible for quite a number of the philosophical terms used and for the
philosophical questions asked today. Many philosophers believe them
selves to be working on Fregean projects and at least one of Frege's
philosophical papers, 'On Sense and Reference', is read in most intro
ductory courses in the philosophy of language. If we regard ourselves as
Frege's direct philosophical heirs, it might seem unlikely that we should
have to read his work in historical context in order to make sense of it.
For, if we are involved in, or reacting to, Frege's projects and argu
ments, surely we understand their significance as well as we understand
the significance and motivation of our own projects and arguments. 1
L. Haaparanta andi. Hintikka (eds.), Frege Synthesized, 9-27.
1986 by D. Reidel Publishing Company.
However, in the years since Michael Dummett's book, Frege: Philo
sophy of Language, first appeared, philosophers have begun increasingly
to realize that Frege's work and its motivation are not as transparent as
they traditionally have been taken to be. Of late, there have been a
number of attempts to understand Frege's work by giving it a reading
which is sensitive to historical context. In particular, there has been a
great deal of interest in the extent to which Frege's work can, and
should, be linked to Kant. Philip Kitcher has argued that there is an
important sense in which Frege's work should be viewed as having been
motivated by a concern with Kantian epistemology,2 and Hans Sluga
has argued that Frege's work intended as a contribution to a 19th
century German neo-Kantian movement. 3
I think this sort of reading of Frege's work must be correct. No sense
can be made of Frege's motivation unless we view Frege as responding,
in some serious way, to what he understood as Kantian epistemology.
But, while some of these readings have provided a clearer pictUre of the
philosophical background against which Frege's work should be read,
there has been less attention paid to the sense in which Frege's work,
given such a reading, might be taken as forming part of the background
of other philosophical work. This becomes a pressing issue once we
start trying to understand his work as a reaction to Kantian episte-
mology. For it seems unlikely that the considerable impact Frege has
had on twentieth century analytic philosophy can have been a complete
accident; yet it is not obvious how one would connect Frege, the 19th
century neo-Kantian, with Frege, the first 20th century philosopher of
language and mathematics. To sketch the outlines of how such a connec-
tion might be made, I want to examine certain tensions in Frege's
views and the constraints which taking Frege's concerns as epistemo-
logical (and, in some sense, Kantian) will impose on any Fregean
responses to these tensions. There is little evidence that Frege was
aware of these tensions; he certainly does not seem to have mentioned
them explicitly. However, I believe many of his later writings can be
read as responses I to these tensions which meet the epistemological
constraints. When Frege's later writings are read in this way, an entirely
different picture begins to emerge. This picture will end up coinciding
in remarkable ways with a plausible interpretation of Wittgenstein's
Tractatus. Indeed, this sort of reading of Frege's work might enable us
to understand how the view of the Tractatus could have grown out of
Wittgenstein's appreciations of tensions in Frege's work.
I will begin with a brief sketch of the sense in which I think Frege's
work should be read as having been motivated by both a dissatisfaction
with some of the details of Kantian epistemology (as Frege understood
it) and by a desire to make a version of Kantian epistemology work.
Kant, as Frege understood him, believed that traditional Aristotelian
logic set out the formal rules of all thought or, as Frege might have said,
that all inferences which follow from the necessary laws of reason alone
can be shown to be valid using Aristotelian logic. But Frege was
convinced that Kant was mistaken here. Frege thought that the formal
rules of all thought licensed more than those inferences licensed by
Aristotelian logic and that his logical notation, the Begriffsschrift, was
designed to set out the true formal rules of all thought. Frege's overall
project was to modify Kantian epistemology in order to incorporate the
real formal rules of all thought. This project might not seem to require
much more than the formulation of the formal rules of all thought and
the substitution of these rules for those of Aristotelian logic. However,
the substitution of Frege's new logic for traditional Aristotelian logic
created tensions in what Frege saw as the Kantian picture. I think that
virtually all of Frege's subsequent work can be viewed as attempts to
deal with the problems which he saw as having resulted from this
substitution. But in order to make this plausible, it is necessary to say
something about how Frege took himself to be a Kantian. I will try to do
this in the next few pages. My aim will not be to detail and defend such
a reading, but only to make clear, in broad outline, the perspective
which makes the tensions in Frege's own views so revealing.
It is not easy to pin down the sense in which Frege took himself to be
a Kantian. Many of Frege's explicit claims seem to indicate that his
pretensions were quite modest - that his aim was merely to patch up
trivial problems in the comers of Kantian epistemology. For instance,
on those questions which Frege does not discuss (e.g. the source and
justification of our knowledge of geometry) he refers us directly to
Kant's answers. He also says about Kant
I feel bound, therefore, to call attention also to the extent of my agreement with him,
which far exceeds any disagreement.
If Kant was wrong about arithmetic, this does not seriously detract, in my opinion, from
the value of his work. His point was, that there are such things as synthetic judgments a
priori; whether they are to be found in geometry only, or in arithmetic as well is of less
Since Frege claims that the project of the Grundlagen is part of an
attempt to show that arithmetic is not synthetic a priori, these comments
might be taken to indicate that Frege views his work as an attempt to
correct an almost incidental mistake of Kant's. But it would be wrong to
take this too seriously. While Frege uses many Kantian terms and
distinctions in his writings, he reinterprets them. One such distinction is
the analytic/synthetic distinction. On Frege's definition a proposition is
analytic if and only if it can be proved from definitions of the concepts
involved using only general logical laws, and this is certainly not Kant's
definition. Thus Frege's claim that Kant was wrong in saying that arith
metic was synthetic a priori requires some explanation. We must
understand the claim in light of Frege's reinterpretation of the Kantian
analytic/synthetic distinction, and this reinterpretation must be regarded
as more than a minor emendation of a detail of Kantian epistemology. It
is important to keep these considerations in mind when we attempt to
give an account of the meaning and significance of Frege's claims of
allegiance to Kant.
I think that Frege's explicit claims that he is following Kant can be
explained without much difficulty. Frege believed that our knowledge
could be divided into three categories. The first of these categories is
knowledge which is justified, if we understand the concepts involved,
merely by virtue of the rules without which thinking is impossible. The
second category is knowledge which is possible only with the aid of pure
intuition, but for which sense experience is unnecessary. Finally, the
third category is knowledge which cannot be justified without appealing
to sense experience. The purpose of Kant's a priori/a posteriori and
analytic/synthetic distinctions was, Frege thought, to mark off these
three categories. An analytic proposition was one which could be
known through the mere form of reason. But knowledge of a synthetic
proposition must have either pure intuitiort or experience as its source.
Kant's point, Frege says 5 is to show that there are synthetic judgments a
priori; i.e. that we can make judgments for whose support experience is
not necessary (or, which are a priori) but, since they are not justifiable
immediately by our understanding the concepts involved, for whose
support something additional is necessary (or, which are synthetic).
Thus when Frege says in the conclusion to the Grundlagen that Kant did
a great service in drawing the analytic/synthetic distinction
but that he
drew it too narrowly 7 and also when Frege earlier claims only to be
making clear what Kant meant by analytic and synthetic 8 he is talking
about the failure or success of Kant's distinction in marking off sources
of knowledge. Frege's point is that, assuming traditional Aristotelian
logic sets out the formal rules of all thought, Kant's distinctions do
characterize propositions according to the sources of our knowledge of
them. However, on the introduction of Frege's new logic, Kant's distinc-
tions will not be able to play this role. For, given that Frege's logic sets
out the true formal rules of all thought, Kant's distinctions have been
drawn too narrowly. Frege's redefinition of the analytic/synthetic
distinction is designed to characterize propositions appropriately
according to sources of knowledge. Frege can thereby claim to be draw-
ing the distinction so that it plays the role which Kant intended it to play
or, saying what Kant really meant by analytic and synthetic.
The most obvious objection to this sort of reading has to do with
the purported relation between Frege's and Kant's analytic/synthetic
distinctions. For Frege and, on this reading, Frege's Kant, both the
analytic/synthetic and a priori/a posteriori distinctions have to do with
justification. However, Kant's actual formulation of the analytic/synthetic
distinction (unlike that of the a priori/a posteriori distinction) concerns
content, not justification, of propositions.
There is a straightforward answer to this objection, for it is easy to find
evidence that Frege did in fact take Kant's analytic/synthetic distinction
to have something to do with justification. For instance, it is clear from
Frege's discussion of Kant is section 12 of the Grundlagen that he takes
Kant's claims that arithmetic is synthetic a priori to be a claim that pure
intuition is ''the ultimate ground of our knowledge of such judgments".
And Frege not only quotes from Kant's Critique of Pure Reason - many
of his remarks echo passages from it. Thus when Frege reworks Kant's
analytic/synthetic distinction in an attempt to draw the line between
those judgments which can be made merely on the basis of understand-
ing the concepts involved and those for which some additional support
(i.e., "truths of some special science" 9) is needed, it does not seem
unreasonable to take this as Frege's version of such passages as 10
It would be absurd to found an analytic judgment on experience. Since, in framing the
judgment, I must not go outside my concept, there is no need to appeal to the testimony
of experience in its support.
If we take this as Frege's understanding of the Kantian analytic/
synthetic distinction, his redefinition of the distinction can be motivated,
in part, by a careful consideration of some of Kant's passages which
seem to conflict when the Begriffsschrift is substituted for the traditional
logic. In a passage which Frege cites, II Kant seems to hint that analytic
truths can be seen to be true from the principle of contradiction alone
while for synthetic truths some additional support (i.e. another synthetic
proposition) is necessaryP But Kant also says that our knowledge can
be extended only by synthetic truth.
Since traditional Aristotelian logic
yielded no surprises, it may well have seemed that a truth of logic (or a
truth which follows from the principle of contradiction) cannot extend
our knowledge, and thus there would be no conflict between these two
characterizations. And, since the application of the traditional logic
could result in no new knowledge, its results could not be counted as
additional support (or synthetic propositions). Thus we could under
stand analytic truths as those which can be seen to be true from logic
alone. But while these remarks do not conflict at all if we understand by
"logic" traditional logic, Frege's new logic makes the conjunction of these
passage look odd. As Frege notes,14 some of the truths of the new logic
(which should, thereby, be analytic by the first characterization) seem to
extend our knowledge (and ought, therefore, to be synthetic by the
second characterization). As Frege formulated it, the central role of the
analytic/synthetic distinction was to separate those judgments for which
support in addition to the concepts involved was necessary from those
which can be made without additional support (or truths of a special
science). However, it is not immediately clear whether or not Frege's
new logic should be counted as additional support.
Frege would have been able to preserve many of Kant's actual words
by counting his new logic as a special science (i.e. as synthetic). How
ever, this position is not open to Frege since it conflicts with his view of
what his logic does. Frege took his logic to provide a unified account of
how the parts of a proposition contribute to determining its truth or
falsity. Such an account cannot be broken up into substantive and
nonsubstantive parts; it is meant to be nothing more nor less than a
means of setting out the formal rules of all thought, the rules which
make the use of the understanding possible. But such rules can hardly
be counted as synthetic or as truths of a special science, for they must
underlie all knowledge. Given the aims of his Begriffsschrift, Frege has
no choice but to take its results to be analytic. Thus Frege characterizes
analytic judgments as those which follow from logic and an understand
ing (or analysis by means of definitions) of the concepts involved.
Frege's characterization of analyticity introduces new complexities
into the project of categorizing a proposition as analytic or synthetic.
Thus Kant's arguments for taking arithmetic to be synthetic a priori
seem too simple, and it is not at all surprising that Frege found them
inadequate. One of the explicit motivations of the Grundlagen was to
determine the source of arithmetic knowledge, to show whether arith
metic was a priori or a posteriori, synthetic or analytic. But the explicit
project of the Grundlagen is to define the number one and the concept
of number. It is not immediately obvious that such a definition will help
in determining the source of our knowledge of arithmetic. I believe that
the first chapter of the Grundlagen should be read as an argument for
Frege's claim that definitions of the number one and the concept of
number will be necessary for determining the source of our knowledge
of arithmetic.
In his first chapter, Frege indicates that there are primitive notions
(e.g. the notion of point) which cannot be defined. And a claim that
something is not definable has epistemological consequences. In particu
lar, consider the consequences of assuming that the numbers cannot be
defined. Then arithmetic is the science of the peculiar irreducible
objects we call numbers. Now consider the claim that zero is less than
one. Since "zero" and "one" are not definable, this claim cannot be
justified by logic from the definitions of the concepts involved. Further
more, since no number words appear in Begriffsschrift, it cannot be a
straightforward logical truth. Thus, if the numbers are not definable,
arithmetical truth cannot be analytic. And Frege is quite convinced that
arithmetical truth is analytic. The reason for this seems to be that Frege
believes that everything thinkable can be numbered and thus that our
knowledge of arithmetic should not be dependent on pure intuition or
sense experience (since not all thought is). This, then is Frege's reason
for being dissatisfied with Kant's account of arithmetical truth as
synthetic a priori. If Frege could give definitions of the numbers which
showed arithmetical truth to be analytic, he would have succeeded in
correcting another of what he saw as the errors in Kantian epistemology.
It should be noted that we can understand Frege's dissatisfaction with
Kant's account without understanding Frege's Begriffsschrift. I believe
that this dissatisfaction with Kant's account of the source of our knowl
edge of arithmetical truth was one of Frege's original divergences from
Kant. We can view Frege's Begriffsschrift as an attempt to show that
analytic truths can be substantive and also as having been motivated, at
least in part, by Frege's original conviction that arithmetical truth must
be analytic.
If the above account of Frege's motivations is accurate, we can give a
very easy explanation of the purpose of Frege's arguments against Kant
and Mill. The importance of the Grundlagen project of defining the
numbers results from Frege's conviction that arithmetical truth is
analytic. Although Frege gives his reasons for thinking this in the
Grundlagen, the only conclusive argument would consist of defining the
numbers and showing that arithmetic is in fact analytic. But Frege needs
to motivate the project of defining the numbers for those who are not
antecedently convinced that arithmetic is analytic. Frege's strategy is to
use Kant's and Mill's accounts of the source of our knowledge of arith-
metic to shake his readers' confidence in the claim that the numbers are
well-understood and indefinable.
If numbers were indefinable and well-understood, one would expect
it to be easy to give an adequate account of the source of our knowledge
of arithmetic. In the first chapter of the Grundlagen Frege argues that
the available accounts of our knowledge of arithmetic are either clearly
wrong (Mill's) or inadequate (Kant's). Thus it makes sense to take
seriously the possibility that numbers are not primitive and irreducible
and to try to come up with acceptable definitions. Frege's first chapter
can be viewed as an argument that the source of our knowledge of
arithmetic has not yet been determined and that the most plausible way
of undertaking the investigation of the source of this knowledge is to
attempt to define the numbers.
Clearly all numbers must be defined for Frege's project to succeed.
Thus Frege notes in the Grundlagen that all numbers can be defined
from the number one, increase by one, and the concept of Number. And
the projects of giving definitions of these three notions are intertwined.
1 5
Hence Frege describes the project of the Grundlagen as that of defining
the number one and the concept Number. But while this is clearly a part
of the overall project of determining the source of our justification of
arithmetical propositions, the definitions alone will not suffice. Having
given these definitions in the Grundlagen, Frege's next step is to show
that all arithmetical truths can (or cannot) be derived from these
definitions by logic alone. This step was to be accomplished in the
Grundgesetze. Had no error been discovered, this work would have
completed Frege's overall project.
In the last few pages I have given an outline of the views of Frege, the
historical person. In order to understand what these views come to,
however, it is necessary to focus on some of the tensions which Frege,
the historical person, may not have recognized. I will outline some of
these below. After the Begriffsschrift, Frege's central task is to give
rigorous logical proofs of the truths of arithmetic from the primitive
truths on which these arithmetical truths depend. Once such proofs are
given, the status of the arithmetical propositions will be determined by
whether the primitive truths on which they depend are analytic or
synthetic. In order to carry out this project, we must answer two
questions. First, what are the primitive truths? Second, how do we
determine whether the primitive truths are analytic or synthetic? The
answers to these questions are deceptively simple. A primitive truth is
either a proposition all of whose terms are primitive (i.e. indefinable), or
a definition in which the only definable term is the definiendum. And
this notion of primitive truth allows an easy way of answering the
second question, for it follows that determining whether a primitive
truth is analytic or synthetic is trivial. If all its primitive terms are logical
terms (terms which are part of Begrijfsschrift), the proposition is
analytic. If some of its primitive terms are nonlogical terms, the proposi
tion is synthetic.
There are important gaps in the above description which cannot be
filled in by simple references to passages in the Grundlagen. For
instance, if we are to have general criteria for determining whether
Frege has given proofs adequate for establishing arithmetical truth as
analytic, an account of definability seems necessary. But there is no
explicit account of definability in the Grundlagen. And, given some of
the remarks in the Grundlagen, it is not entirely clear what definition is
for Frege. In particular, Frege indicates that definitions often require
Is the justification required for a definition which appears
in a proof part of that proof? In the Grundlagen Frege also gives no
explicit accounts of the notion I)f definition, the sort of justification
required for a definition, or the role the justification of a definition
might play in a proof which uses the definition.
Frege does have accounts of definition and of definability which will
serve the requisite epistemological roles, but these are not explicit in his
writings. Such an account of definition, for instance, is a consequence of
the views expressed in Frege's correspondence with Hilbert and some of
his papers, in particular, two papers entitled "On the Foundations of
Geometry" which grew out of that correspondence. I think that this
account of definition is implicit in the Grundlagen. However, it would
have been awkward to make this explicit in the Grundlagen because,
without the sense/reference distinction, it would follow that all ordinary
language is meaningless.
While it is beyond the scope of this short paper to indicate the
arguments for taking Frege to have the views on definition I attribute to
him, they can be easily summarized here. One of Frege's most explicit
statements appears in a letter to Hilbert.
Every definition contains a sign (expression, word) which previously has had no
reference and which is given a reference only through this definition. Once this has
happened, one can make out of this definition a self-evident proposition which is then to
be used like an axiom. But we must adhere to the tenet that in a definition nothing is
asserted: rather, something is stipulated. Therefore what requires a proof or some other
reasoning to establish its truth ought never to be presented as a definition. 18
There are three ways in which this seems to conflict with Frege's
Grundlagen, although there is little evidence that Frege was aware of
any of these conflicts. The first is that in this passage Frege seems to say
that definitions require no justification, while in the Grundlagen Frege
said that definitions do require justification. The second conflict is that
from the above passage it follows that, if Frege defined "one" in the
Grundgesetze, then it must have previously had no reference. But Frege
did take himself to have defined "one" in the Grundgesetze, and yet it
seems rather implausible that "one" should have had no reference
before Frege did this. Finally, the third conflict is that Frege does not
give an arbitrary definition of "one" in the Grundlagen, yet the views on
definition expressed in the above passage do not seem to leave room for
any constraints on definitions. In the next few pages I will briefly
indicate how these problems are to be answered and how the answers to
these problems lead to a completely new way of viewing Frege's work.
The first question concerns the justification necessary for a definition.
As I read the above passage, Frege is claiming that it is never necessary
to justify assigning a particular referent to a particular word. However,
this does not rule out requiring some justification to show that a
purported definition actually meets the criteria which Frege requires of
definitions. Frege says 19
I demand from a definition of a point that by means of it we be able to judge of any
object whatever - e.g. my pocket watch - whether it is a point.
Frege says this because he requires that any definition of a concept-
word must give us a description, in primitive terms, which either holds
or does not hold of each object. Similarly a definition of an object-word
must consist of a description, in primitive terms, which picks out one
and only one object. These are Frege's - admittedly very strict -
requirements on definitions. Thus some justification may be required for
a definition; it may be necessary, for instance, to prove that a certain
description holds of one and only one object. And Frege explicitly
applies this criterion of adequacy. In section 64 of the Grundlagen, he
rejects a definition of direction because
it will not, for in,tancc, decide for m whether England is the the direction of the
Earth's axis.
Thus it is not that the truth of a definition may require justification, but
rather that it may be necessary to show that the description used in a
definition is actually a defining description in Frege's strict sense.
Let us now turn to the second problem. Frege seems committed to
saying that "one" and "I" had no reference prior to his investigation in
the Grundlagen. How serious a problem this is clearly depends on the
nature of Frege's notion of reference. I think it is possible to argue that
Frege's notion of reference is a very peculiar one and that, given this
notion of reference, it would not be untoward to take "one" to have had
no reference before the work of the Grundlagen.
The passage I quoted earlier was written after "On Sense and Refer-
ence", and thus we cannot take the notion of reference which is used in
that passage to be the notion of Bedeutung in the Grundlagen. However,
these notions are related. In fact, I believe that the development of the
later notion of reference can be viewed as a response to tensions
between Grundlagen-Bedeutung and Grundlagen-definition. To see this
tension, we should first note that, in the Grundlagen, a definition fixes
the meaning (Bedeutung) of a word.
Furthermore, we can ask for the
meaning (Bedeutung) of a word only in the context of a proposition; to
fix the Bedeutung of a word, one must fix the meaning of the proposi-
tions in which it appears.21 Defining (i.e. fixing the Bedeutung of) the
number one entails fixing the sense of "zero is less than one". Thus it
seems that Frege may be committed to saying that "zero is less than
one" is meaningless before his work in the Grundlagen. Furthermore,
our use of most ordinary words does not determine a definition in
Frege's strict sense. For instance, there are people who are neither
clearly bald nor clearly not-bald. Thus no description of our ordinary
concept of baldness will meet Frege's criteria. It is not obvious that this
is a serious problem, for the sense in which Frege is committed to
holding that such propositions are meaningless is not clear, and he
certainly does not seem to believe that he was committed to this. But I
think this is symptomatic of an important tension in the Grundlagen.
We can regard Frege's sense/reference distinction as providing a
partial response to his apparent commitment to the view that ordinary
language is meaningless. If the notion of sense plays the intuitive role of
meaning, we can take the notion of reference to play the highly technical
role which is needed for working out Frege's project. To have reference,
then, is to be either primitive or used in a way which determines a
definition, in Frege's strict sense, from primitive terms. Thus, given the
vagueness of our ordinary language, most non-primitive terms will have
reference only if an explicit Fregean-style definition has been given.
Having recognized this distinction, we can take "zero is less than one" to
have (hence to be meaningful) but no reference before Frege's
work. It does, of course, follow that the sentence had no truth-value
before Frege's work. But this, I will argue, is a view to which Frege is
firmly committed.
The final problem has to do with the constraints to which Frege
subjects his definition of the number one. Given Frege's discussion of
definition, it would seem that any way of fixing the reference of "one" is
as good as any other. But Frege is only willing to consider a certain sort
of definition. However, this is not because there are any constraints on
what referent can be stipulated for "one", rather it is because Frege
wants to use his definition for a particular purpose.
In order to see that it is legitimate to impose constraints on defi-
nitions, it is useful to consider Frege's paradigmatic example of a
definition the mathematical definition of the continuity of a function.
Clearly we could assign any number of descriptions to "fis continuous"
which would not enable us to express the mathematical theorems which
the standard definition helps us express. But such definitions would not
have served the purpose for which the standard definition was drawn.
Thus it is perfectly reasonable for Frege to impose constraints on his
definition of "one" if he wants the definition to play a certain role. And
he does. Frege wants to give a definition of "one", from primitive terms,
which will make 'true all those propositions containing "one" which we
have already judged to be true. In this way, the definition will provide
the sort of foundation for arithmetic which Frege requires any science to
have. This is the sense in which the definition will constitute an analysis
of our concept of what it is to be the number one. Frege says 22
The real importance of a definition lies in its logical construction out of primItive
clements. And for that reason we should not do wIthout it, not even in a case like this.
The insight it pcrmlts into thc logical structure IS a condition for insight into the logIcal
linkage of truths. A definition is a constituent of the system of a sCIence.
The problem with this is that the propositions we take to be the truths of
arithmetic do not determine a definition, in Frege's sense, of the number
one. Frege deals with this problem in a 1914 paper, 'Logic and
Now we have to consider the difficulty we come up in giving a logical
analy,is when it is problematic whether this analysis is correct. Let us a,sume that A IS
the long established sign whose sense we have attempted to analyse
logically by constructing a complex expression which the analysis. Since we are not
certain whether the analysis is we are not prepared to pre,ent the complex
expression as one whieh can be replaced by the simple sign A. If it is our intention
to put forward a definition proper. we are not entitled to choose the sign A, which
already has a but we must choose a fresh sign B, say, which has the sense of the
complex expression only in virtue of the definition. The question now is whether A and
B have the same sense. But we can bypass thi, question altogether if we are constructing
a new system from thc bottom up; in that case we shall make no further use of the
A - wc shall only use B ... If we have managed in this way to con<;truct a system for
mathemalles without any need for the sign A, we can leave the matter there; there is no
need at all to answer the question concerning the sense in whIch - whatever It may be -
this sign had been used earlier ... However it may be felt expedient to use sign A
instead of sign B. But if we do this, we must treat it as an entirely new sign whIch had no
sense prior to the definition.
This tells us how we are to understand the project of the Grundlagen.
Before the Grundlagen, "one" did not have a referent, in Frege's strict
sense, Frege wants to build arithmetic up from primitive terms. Thus he
is required to construct a description of an object, in primitive terms,
which can replace "one" or "I" in the propositions of arithmetic without
appreciably altering mathematical practice. If this can be done, there is
no harm in using that description as a definition of "one". However,
assigning this referent to "one" rather than to some newly formulated
word should be viewed as a heuristic, not as an important elucidation of
our ordinary use of "one".
The view outlined above contains a significant gap - the lack of an
account of primitiveness. While we can recognize and give definitions,
Frege has given no explicit criteria for distinguishing between definable
and indefinable terms. And here, it may seem, we hit a real dead end.
Frege does not seem to have said what it is for a term to be primitive,
although he sometimes gives examples of primitive terms or hints about
what primitiveness comes to. On my reading of Frege's work, there is an
important reason for this. It is that the notion of primitiveness is itself
primitive, that is, indefinable. In order to understand fully the signifi
cance of this gap, it is important to consider the relation between
primitiveness and Frege's notion of objectivity.
Frege says very little about objectivity. His most sustained discussion
of the subject seems to be the discussion in section 26 of the Grundlagen.
In that section, he seems to offer three explicit criteria for objectivity.
He says that what is objective is what is expressible in words, what is
subject to laws, and what is independent of sensation, intuition, and
imagination, but not what is independent of the reason. While the
connection between these criteria may not be immediately apparent, I
have argued elsewhere that they are tightly bound together and that they
all delimit the same realm of objectivity.24 I have argued that, for Frege,
to be objective is to be subject to the laws of Begriffsschrift and that it is
this view which binds the above criteria.
If what is objective is what is subject to the laws of Begriffsschrift, then
a good deal of work is required to show that a proposition is objective.
In particular, if a proposition is subject to the laws of Begriffsschrift, all
its terms must have fixed reference. But, as I mentioned earlier, few of
our ordinary terms have fixed reference in Frege's sense. It seems, then,
that definitions of each definable term are required. But how are we to
know when this has been done? We can only recognize the completion
of such a task if we can recognize primitive terms. Frege assumes that
we can recognize primitive terms, and it remains to be seen whether this
assumption will get him into serious trouble. However, there is a more
immediate problem. It is not clear that the task has been completed
once the original proposition has been translated into a proposition
whose terms are all primitive. We must still show that all its terms have
fixed reference. Thus we must say something about what it is for a
primitive term to have reference and how we can show that a primitive
term has reference. Until now, the only means available for showing that
a term has reference have been the use of definitions. But primitive
terms cannot be defined.
While Frege has given us no explicit account of primitiveness, this
does not mean that we know nothing about what it is to be a primitive
term. The primitive terms are the indefinable terms which underly all
terms usable in communication. While we often communicate by using
vague terms, the success of such communication is possible only
because what is communicated can be made unambiguous. Thus one of
the conditions necessary for communication is that what is communi-
cable must be expressible in terms which either have fixed reference or
can be defined from terms which have fixed reference. Primitive terms
cannot be defined thus, if they can be used in communicating, they must
have fixed reference. For instance, the terms "function" and "object" are
primitive and are necessary for the explication of the logical regimenta-
tion of propositions in Begriffsschrift. Furthermore, there is an answer to
whether or not a proposition has been correctly regimented. The
regimentation must be objective and hence the terms "function" and
"object" must have reference. But what does this sort of reference come
It is important to note that the above does not amount to saying that
the words "function" and "object" refer to whatever they must refer to if
Begriffsschrift regimentation is to be objective. One of Frege's most
fundamental assumptions is that Begriffsschrift regimentation is objective
- this is the foundation of Frege's very understanding of objectivity. It
follows that "function" and "object" must have reference. But there is no
way of specifying, describing, or giving structure to what these words
refer to. The closest we can come is to give hints, i.e., to use these words
in propositions. The claim that these terms have reference is nothing
more than the claim that regimentation is objective. Thus, in the end, the
notion of reference drops out altogether.
One of the most striking features of this notion of objectivity is that it
seems that everything is objective. Certainly, for instance, we communi-
cate about the character of our mental images and if what is objective is
what is communicable, it would seem that these claims concerning our
mental states must be objective. But this is especially surprising since
Frege says that the character of our mental images is subjective. This
is best explained by carefully considering the one discussion in the
Grundlagen which deals with the subjective realm.
Frege gives an example in the Giundlagen of two beings who intuit
only projective properties and relations. One of them intuits as a plane
what the other intuits as a point, etc .... Thus what one will intuit as a
line joining two points, would be intuited by the other as a line of
intersection of two planes. Now it is important that the two beings intuit
projective, not Euclidean, properties and relations. In projective geome
try, what is true of planes is precisely what is true of points. Thus these
two beings would agree on all axioms and theorems of geometry.25 Since
they obey the same geometrical axioms and could not discover any
difference in their intuitions through language, the objective meaning for
both beings of "point" and "plane" is the same. Thus, although there
seems, from Frege's description of these beings, to be a difference in the
characters of their intuitions, they are both intuiting the same things.
The point of this example was to convince us that pure intuitions have
both objective and subjective sides. The objective side is what is express
ible and the subjective side is the character of the intuitions which is
not expressible. How is this supposed to work? Because the axioms of
geometry are known by pure intuition, they are justified by the construc
tion of mental pictures. By setting the situation out as he has, Frege has
asked us to construct the mental pictures we would imagine each of the
two beings constructing and to examine the difference between the two
pictures. I will discuss the images the two beings would construct in
order to assure themselves of the justification of the axiom which tells us
that any two points determine exactly one line. Frege's assumption, I
believe, is that we will construct two pictures which look something like
my Figure 1:
POINTS -;_' B.
i,' LINE
Fig. 1
These pictures are very different and, in some way, this difference is
supposed to exhibit the subjective difference. At first glance, however,
this might seem somewhat suspect. For the difference between these two
pictures is an objective, physical difference which can be described in
words. And one might suspect that the two beings <lescribed by Frege
could discover their difference by examining these pictures. But this
creates no problem for Frege. It is possible to imagine that one of the
beings, when looking at picture A, has the same experience the other
has when looking at picture B.
The point is that the objective difference
between A and B is a difference between the geometrical configurations
of pen marks on paper, not a difference between mental images. Frege is
not asking us to examine these objective differences, but rather to
construct these two pictures in our minds and examine the difference in
appearance (that is, the subjective difference) between the two pictures.
Now it is important to realize that it follows that Frege is not making
an objective argument. Frege's example works only if his readers
construct two mental images which are different in appearance, that is,
subjectively different. But, although the pictures he constructs might be
subjectively different, there is no way for him to guarantee that the
pictures constructed by his readers are subjectively different. This is a
case in which we must grant Frege the grain of salt he asks for in "On
Concept and Object". For Frege has nothing to say to someone who
considers his example and claims to experience no subjective difference
in the images she constructs. Subjective difference, after all, cannot be
described. Frege's example cannot be an argument, it can only be
considered a hint.
I have grawn numerous conclusions about Frege's motivations and
about underlying views which would allow Frege to respond to tensions
in his explicit picture; however, it is not entirely clear that Frege was
aware of these tensions. I will end by discussing a consequence of my
reading of the example discussed above. This consequence is something
of which, I believe, Frege was almost certainly unaware.
I have argued that, for Frege, it is not possible to do more than hint at
the difference between the subjective and objective realms. It is certainly
not possible to describe this difference objectively. But it follows from
this that the difference is not objective. This may not seem too prob-
lematic. The subjective realm is, in an important sense, empty and it does
not seem all that strange to say that everything is objective for, since we
cannot talk about the subjective, everything we can discuss must be
objective. Furthermore, Frege has said very little about objectivity, and
very little seems to be at stake for him. We can regard what Frege does
say as a few innocent hints. But this problem does not apply only to
Frege's comments about objectivity. Frege's discussions of the differ-
ence between objects and concepts, and of why number statements are
about concepts - in fact almost all of what we might be tempted to call
Frege's theoretical framework - also must have the status of hints.
If we read this consequence back into Frege's project, the perspective
shifts dramatically. I began by arguing that Frege's mathematical work
was philosophically motivated and that Frege intended us to draw
philosophical morals from it. What we think of as his philosophical
work was intended to show how we are to draw philosophical morals
from the mathematical work. None of this has really changed. However,
the implication of reading Frege's work in that way seemed to be that
the most important work was what we call his philosophical work, for
this is where Frege set out his philosophical theories. But it now appears
that Frege cannot have any philosophical theories. Far from being the
presentation of the philosophical theory within which Frege's mathe-
matical work was to play a role, what we think of as his philosophical
work can only have had the status of subjective hinting designed to get
his readers into the proper frame of mind for understanding the purpose
and significance of his real philosophical work - the work we think of
today as his mathematical work.
University of Wisconsin - Milwaukee
I would like to thank Burton Dreben, Warren Goldfarb, Mark Kaplan, Hilary Putnam,
and Thomas Ricketts for criticism and advice.
1 It is not easy to find support for this sort of view in writing, since people who believe
this tend to be pursuing Frege's questions,.not worrying about what he meant. However,
I have found that quite a number of people will defend such a view in conversation.
2 Philip Kitcher, 'Frege's Epistemology', The Philosophical Review 88 (1979), 235-
3 Hans Sluga, Gottlob Frege, (London: Routledge & Kegan Paul, 1980).
4 Gottlob Frege, The Foundations of Arithmetic, translated by J. L. Austin (Evanston:
Northwestern University Press, 1978), pp. 10 1-102.
5 Frege, Foundations, p. 102.
6 Frege, Foundations, p. 101.
7 Frege, Foundations, p. 99.
8 Frege, Foundations, p. 3.
9 Frege, Foundations, p. 4.
10 Immanuel Kant, Critique of Pure Reason, translated by Norman Kemp Smith (New
York: St. Martin's Press, 1929), A 7 /B12.
II Frege, Foundations, p. 100.
12 Kant, Critique of Pure Reason, B14.
13 Kant, Critique of Pure Reason, A8, B12.
14 Frege, Foundations, p. 101.
15 Frege, Foundations, p. 25.
16 See, for instance, Frege, Foundations, pp. 4, 77.
17 See the Frege-Hilbert correspondence in On the Foundations (!If Geometry and
Formal Theories of Arithmetic, translated by Eike-Henner W. Kluge (New Haven and
London: Yale University Press, 1971), p. 7.
18 While Frege is talking about mathematical definitions here, it is important to note
that his definition of the number one is intended to be used in proofs and hence is
19 Frege, 'On the Foundations of Geometry' in On the Foundations of Geometry, p. 63.
20 See, for instance, Frege, Foundations, p. 9.
21 Frege, Foundations, p. 73.
22 Frege, 'On the Foundations of Geometry' in On the Foundations of Geometry, p. 60.
23 Frege, 'Logic and Mathematics' in Posthumous Writings, edited by Hans Hermes et
al., translated by Peter Long et al. (Chicago: The University of Chicago Press, 1979) p.
24 Joan Weiner, Putting Frege in Perspective (Doctoral Dissertation, Harvard Univer-
25 A projective plane can be constructed from a Euclidean plane by adding, for each
line in the original plane, exactly one point at infinity. Lines which were parallel in the
original plane share a point at infinity in the projective plane. Furthermore, the points at
infinity in this plane determine a single line at infinity. Two planes which are parallel in
Euclidean space share a line at infinity when the points at infinity are added to each.
Frege's doctoral dissertation, 'Uber eine geometrische Darstellung der imaginaren
Gebilde in der Ebene', is concerned with providing the kind of foundation for these
infinitely distant (or imaginary) points which he later attempts to provide for the
26 This is undoubtedly what Frege would have us imagine if his two beings could see.
Of course, since Frege follows Kant in including sensation under intuition and since, by
hypothesis, these beings intuit only projective properties and relations, it follows that
they cannot see.
If it is Yes, say Yes; if it is No, say No. Anything else
comes from the EVil One.
Matthew 5, 37
Whoever undertakes to scrutinize logical arguments conducted outside
mathematics cannot but wince at the flightly conduct of words. Their
meanings have no exact lines of demarcation. What an adjective conveys
may depend on the noun that follows ('white wine'). We often have
difficulty in grasping the exact meaning of a verb if we do not know its
subject and complement(s). When we look up a word in a dictionary, we
see how the word's uses have come to diverge ever more from its
primary meaning and have branched out in many directions. Metaphors
lose their sparkle and, like dead stars, come to lead an obscure existence
('source of grief). Words in ordinary language are far from having neat
definitions, like that of 'even' in arithmetic. Every predicate is vague, in
the sense that there are individuals for which it is intrinsically indeter
minate whether the predicate holds or not.
In his first logical work, Frege had to point out how the vagueness of
a predicate can wreck a logical argument. Theorem 81 in 27 of his
1879 can, in modern language, be stated as follows: Let R * xy be the
closed iterate (to use Quine's expression) of a binary predicate Rxy; that
is, R*xy is to Rxy what 'x is an ancestor of y' is to 'x is a parent of y'.
Then, for any unary predicate Px which is inductive over Rxy, that is,
such that
VxVyPx& Rxy)::::l Py),
we have
VxVyPx& R*xy) ::::l Py).
Let Px be interpreted as: x is a heap of beans; and Rxy as: y is
obtained from x by the removal of one bean. Then Theorem 81 would
imply that a single bean, or even none at all, would be a heap of beans.
L. Haaparanta and J. Hintikka (eds.), Frege Synthesized, 31-45.
1986 by D. Reidel Publishing Company.
To overcome the difficulty, Frege claims that the predicate Px is not
inductive over Rxy because, u being one of certain objects, the sentence
Pu 'cannot be taken as being a judgment' (,ist unbeurtheilbar') on
account of the 'indeterminateness' (' Unbestimmtheit') of the predicate
Px. The definition of the closed iterate of Rxy contains a universal
quantifier ranging over unary predicates. If a certain unary predicate is
not included in that range, then Theorem 81 does not apply to that
predicate. The criterion that Frege uses for excluding a predicate
from the range is its' Unbestimmtheit', which leads to 'unbeurtheilbare'
sentences. With these few remarks Frege puts vague predicates outside
the pale of logic.
Let us specify our terms. For us, a predicate is a linguistic expression;
thus, if unary, it corresponds to Frege's Begriffswort and, if binary, to his
Beziehungswort. Since Frege maintains that a concept whose delimita-
tion is 'blurred' ('verschwommen') is in fact no concept at all (1969, p.
133), we shall speak of vague predicates, and not of vague concepts.
Let us briefly recall what Frege's ontology is. He has a fixed universe,
which comprehends all objects whatsoever. An object may be abstract,
like a natural number, or concrete, like the Moon. Over these objects
there are first-level functions of one or two arguments and, since he has
to quantify over these functions, he has first-level variables. The impor-
tant point for us here is that Frege does not divide his universe into
various sorts and he requires that any function (in particular, any
concept) be defined for all objects. This is what he calls his principle of
completude (,Grundsatz der Vol!standigkeit') , presented and defended
in 56-65 of 1903 and subjacent in all his work.
In his presentation of the principle Frege takes exception to what he
calls piecemeal definitions (,stuckweise Definieren'). An example of such
a definition would be addition, originally defined for natural numbers,
then successively extended to signed integers, rational numbers, and so
on. Weare dealing here with different operations: we should in each
case use a different sign and speak of the isomorphism between a system
of numbers and a certain proper part of the new system. There never
was any misunderstanding on that score in the mind of qualified
mathematicians. But some expositors chose to stress how the definitions
of addition for various kinds of numbers successively grew out of one
another, sometimes speaking of a genetic definition. Frege severely
criticizes such procedure: 'Concept-like constructions that are still in a
state of flux, that have not yet received final and sharp limits cannot be
recognized by logic as concepts; and that is why it must spurn all
piecemeal defining' (1903, p. 71). And: 'Thus nowhere do we have firm
ground under our feet. Without final definitions, no final theorems. One
would not come out of imperfection and vacillation' (1903, p. 74). If
these strong words were directed only against some occasional defective
presentations, there would be little to say, except that they perhaps
overshoot the mark. It turns out, however, that what Frege condemns in
the piecemeal definition is its being restricted to a certain domain of
mathematical objects. Every predicate, every operation has to be
defined for all objects in the Universe. If addition were not defined for
the Moon and the Moon, 'the question whether the sum of the Moon
and the Moon is 1 or whether the Moon falls under the concept
something that added to itself gives 1 could not receive an answer; in
other words, what we called a concept would in fact not be a genuine
concept, because the sharp delimitation is lacking' (1903, p. 76).
A second consequence of Frege's principle of completude is his
rejection of conditional definitions. Consider, for example, for a binary
predicate Rxy of natural numbers, a definition that would read as
(1) VxVyNx& Ny) (Rxy= . .. )),
Nx being interpreted as: x is a natural number. Peano had called such
definitions 'deJinizioni condizionate' or 'deJinizioni con ipotesi'. Frege
calls them 'bedingte Erkliirungen' and rejects them because, in the case
of definition (1), if a and b are not natural numbers, the sentence
(2) (Na& Nb) (Rab == .. )
has the truth value t for whatever syntactic object we put at the place of
the three dots; hence, for objects a and b that are not natural numbers,
(1) does not specify anything for Rab.
In a letter to Peano dated September 29, 1896, Frege writes (1976,
pp. 182-183, or 1980, p. 114; I modified the translation): 'A condi-
tional definition of the sign for a concept decides only for some cases,
not for all, whether an object falls under the concept or not; it does not
therefore delimit the concept completely and sharply. But logic can
recognize only sharply delimited concepts. Only under this presupposi-
tion can it set up precise laws. The logical law that there is no third case
beside "a is b" and "a is not b" is really just another way of expressing
our requirements that a concept (b) must be sharply delimited. The
fallacy known by the name of "Acervus" rests on this, that words like
"heap" are treated as if they designated a sharply delimited concept
whereas this is not really the case'. Frege's ban of conditional definitions
directly springs' from his principle of completude.
In his defence of that principle, Frege asks whether 'x + y' may have
a Bedeutung only when the two objects considered as arguments are
natural numbers, and he answers negatively. If the answer were posi
tive, the sentence 'The sum of the Moon and the Moon is l' would be
neither true nor false and, he claims, 'no scientific investigation can
culminate' in such a sentence (1903, p. 76). In that whole discussion
Frege's arguments are somewhat repetitious and circular. Why would
there be something unscientific in the sentence? Because some truth
value would be left indefinite. We are back to completude. Frege also
argues that, if addition were not defined for all objects in the Universe,
we would not know the number of solutions of the equation x + x = 1;
some a such that a + a = 1 might pop up in a comer where we do not
expect it. A strange and circular argument! Once we have defined the
function x + yover a definite domain, all solutions of x + x = 1, if any,
are in that domain.
Frege finally presents a more elaborate argument. From (2), by
sentential logic, we obtain
(3) Rab =f: ) & Na):::> - Nb,
hence we are led to consider an object, namely b, which is not a natural
:pumber, or, as Frege writes, 'here it is impossible to maintain the restric
tion to the domain of numbers. The state of affairs has a force that
irresistibly acts so as to break down such barriers' (1903, p. 78). But the
argument remains inconclusive; it certainly does not establish the
necessary existence of a universal domain comprehending all objects. In
(1) the quantifiers may very well range over a sufficiently large domain
of mathematical objects, and in that domain a subdomain is cut out by
the predicate Nx. In fact, when natural numbers are defined as certain
sets, the quantifiers in (1) range over sets. A mathematician would not
use definition (1) unless a definite mathematical domain has been
introduced for the quantifiers; if no such domain is under consideration,
he would replace (1) by
(4) VxVy(Rxy == ),
with the understanding that the quantifiers now range over the natural
numbers. With (4), Frege's argument cannot get started.
Frege's requirement of completude is intimately connected with that
of sharpness. For him, in fact, the two requirements seem to fuse into
one. Countless times in his writings, we find the words 'complete' and
'sharp' conjoined. Discussing completude, all of a sudden he brings in
examples of vagueness (lack of sharpness), like the word 'heap' in his
letter to Peano quoted above or the word 'Christian' at the end of 56
of 1903. For Frege, lack of completude and lack of sharpness are both a
failure of universal bivalence and thus seem to merge one into another.
Let us try to distinguish them. For an atomic sentence Pa, universal
bivalence may fail in three cases:
(1) Although 'a', by its form and its grammatical role, appears to
name an object, it actually does not; this is the well-known problem of
empty descriptions raised by Bertrand Russell. We shall say that here
the individuation has failed; as no individual is apprehended by 'a', the
question whether the predicate 'P' holds or not does not arise, and Pa
has no truth value. (If 'a' is a definite description, we may, as Russell
does, look at Pa as a relation between predicates and not analyze it as a
subject-predicate sentence.)
(2) An object is denoted by 'a', but this object is not one of which the
question whether the predicate' P' holds can be raised. Is the number 7
blue? A good dictionary says of an adjective 'is said of ... " thus
indicating its domain of applicability. This pigeon-holing is constantly
questioned by metaphoric ('sharp mind') or poetic (Rimbaud: A black,
E white, ... ) uses of words, but it certainly is an integral part of the
working of any natural language.
(3) A definite individual is apprehended by 'a' and for that individual
the question whether the predicate 'P' holds is meaningful. Nevertheless,
the criteria connected with the application of 'P' are not such that a
truth value is assigned to Pa. For example, I see this vase perfectly well,
the light is good, but is the vase blue or green? I don't know how to
answer; I grope for a new word: 'bluish', 'blue-green'.
Failure of individuation may occur in mathematical language ('the
largest natural number') or in ordinary language ('the present king of
France'). In mathematics the proper use of the singular definite article
requires an existence and uniqueness proof. In ordinary language failure
of individuation is likely to stop communication. If I start a sentence
'The present king of France is ... ',my interlocutor is liable to stop me:
'What are you talking about? France has no king now'. As for Frege, he
proscribes nondenoting individual constants from any properly con-
structed language. In 1893, p. 19, he introduces the slanted boldface
stroke for the description operator; applied to (the Werthverlauf of) a
concept, it yields the object falling under the concept in case there is one
and only one such object; otherwise, it yields the Werthverlauf itself.
Mathematics knows no completude in Frege's sense. A universe
embracing 'everything' is repugnant to mathematicians. Hilbert (1904,
p. 175, or van Heijenoort 1967, p. 130) already reproached Frege for
imposing no restriction on the range of his universal quantification, and
he introduced a system in which the quantifiers range over objects
recursively constructed from two basic objects; another mathematician,
Poincare (1906, p. 18, or 1908; pp. 181-182), was prompt to stress the
difference between the well-delimited domain thus introduced by Hilbert
and the all-embracing universe of Frege and Russell. The closest
approximation that mathematics would have to a universal domain
would be the (a?) universe of sets. But one cannot speak today of a
fixed, well-delimited universe of sets in which every mathematician
would embed the entities he is working with. Moreover, such a universe
would still be far from Frege's universe since it would not comprehend
nonmathematical objects. As to natural language, it respects sorts and
natural kinds. We do not feel obliged to ascribe a truth value to the
sentence 'The number 7 is blue'. Even in Frege's system, it is not clear
why completude, in his sense, is required. He wants bivalence, but
bivalence, in the sense that each predicate has a definite domain of
applicability and either holds or does not hold of each object in that
domain, can obtain in a many-sorted logic. No, Frege is adamant, a
predicate or a function not defined for all objects whatsoever in the
Universe would somehow be deficient. It would not be impossible, it
seems, to reconcile Frege's ontological assumptions with a many-sorted
bivalent logic. In imposing his brand of complete bivalence, Frege seems
to have fallen prey to some kind of extremism.
We can, of course, assign a truth value to the sentence '7 is blue'.
'v' x( x is blue :::J x is a physical object),
where the quantifier ranges over all objects, and
7 is not a physical object,
we obtain
7 is not blue.
This argument may be in need of some improvement. Although the sky
is hardly what we would call a physical object, it is blue. The problem is
to circumscribe a fixed and well-delimited universe of discourse con-
sisting of all objects, that is, of all ground-level entities. Frege assumed
the existence of such a universe, without paying much attention to the
problem hidden in this notion. Once a universe is postulated, predicates
can, of course, be defined, or considered to be defined, for each object
in it. We make the blanket decision that, if a is not in the domain of
applicability of'P', Pais false.
The situation, though, is somewhat different with functions, or opera-
tions. Consider the predicate Sex, y, z), taken to mean: x, y, z are natural
numbers and z = x + y. Take the predicate to be false whenever an
argument is not a natural number. Then, for every z, S(the Moon, the
Moon, z) is false, and the equation z = the Moon + the Moon has no
solution. Hence, though the predicate S is defined over the whole
universe, the function + is not. Frege, who wants functions, just as well
as predicates, to be defined over the whole universe, must, to the
predicate associated with a function, assign the value t for some objects
that are outside what we would consider to be its domain of applicabil-
ity, and the blanket solution of equating meaninglessness to falsity is not
open to him.
What are the criteria that govern the assignment of truth values
outside the domain of applicability? For Frege, it seems, the assignment
is arbitrary. The point I want to make here is not that such an enterprise
cannot be carried out, but rather that neither mathematics nor ordinary
language proceeds thus.
Since vagueness pervades natural language, Frege is led to postulate,
behind each vague predicate of ordinary language, an exact 'objective'
predicate, so that logic can operate without a hitch. In 1884, 26, or
1953, pp. 36 and 36e, he writes: 'The word "white" ordinarily makes
us think of a certain sensation, which is, of course, entirely subjective;
but even in everyday speech it often bears, I think, an objective sense.
When we call snow white, we intend to express an objective quality,
which we recognize, in ordinary daylight, by a certain sensation. [ ... 1
Even a color-blind man can speak of red and green, in spite of the fact
that he does not distinguish between these colors in his sensations.
[ ... 1 By being objective I understand being independent of our sensa-
tion, intuition [Anschauen] and imagination [Vorstellen], and 9f all
construction of mental pictures out of memories of earlier sensations,
but not independent of reason'. Frege is saying that what an empirical
word of our natural language means for us is not constituted by the
training acquired through ostension learning ('memories of earlier
sensations'), but that, behind the word, we comprehend some objective
property. A bit earlier in the same section (1953, pp. 35 and 35e; I
modified the translation), Frege had written: 'Objective here is what is
subject to law, conceptual, adjudicable, what can be expressed in words'
('Objektiv ist darin das Gesetzmiipige, BegrijJliche, Beurtheilbare, was
sich in Warten ausdriicken liipt'). Frege's words ('subject to law', 'con-
ceptual', 'can be expressed in words', 'not independent of reason')
suggest that the objective property he assumes behind every empirical
term has a definition not unlike that of 'even' in arithmetic. If there were
such a definition of 'red' or 'green', then, of course, a color-blind man
would comprehend these words. Frege has to adopt this panrationalism
because he wants to fit the natural language to his bivalent logic. His use
of the word 'beurtheilbar' already suggests that the objective property
behind the empirical term is bivalent. This is reinforced by the use of
'objective' at other places in Frege's writings (for instance, in 1892, p.
34, or 1952, pp. 63-64, where objectivity is connected with bivalence).
What Frege is in fact suggesting is that one can disregard the vagueness
prevalent in a natural language because many words are learned by
ostension. But this conception of the objective property is so contrary
to the actual working of a natural language that he has to hedge ('often',
'I think').
In this letter to Peano quoted above, Frege writes (1976, p. 183, or
1980, p. 115; I modified the translation): 'The task of our vernacular
languages is essentially fulfilled if the persons having verbal intercourse
with one another connect, with one sentence, the same thought, or
approximately the same thought. For this it is not at all necessary that
the individual. words should have a Sinn and a Bedeutung of their own,
provided only that the whole sentence has. a Sinn. Whenever inferences
have to be drawn, the case is different; it is then essential that the same
expression occurs in two sentences and that it has exactly the same
Bedeutungin both.'
Here Frege deals with a pressing problem, namely the statUs of
logical arguments conducted in ordinary language; but we may wonder,
since the lines are taken from a letter written with a running pen, how
much exegesis each word can bear. When some sentences are used in a
logical argument, each part of a sentence is to have the same Bedeutung
at each of its occurrences, and each sentence has a Bedeutung, that is, a
truth value. Speaking a bit more precisely, we could say that, if we
consider the premiss( es) and the conclusion of a rule of inference, a
predicate symbol is to occur in more than one of these formulas; and the
correctness of the rule would vanish if, to two different occurrences of
the predicate symbol, we were to assign different subsets of (a Cartesian
product of) the universal domain when we interpret the formulas. On
the other hand, when ordinary language is used only for communication,
parts of a sentence may have no Sinn and no Bedeutung, although the
sentence has a Sinn. Since no Bedeutung is mentioned for the sentence,
does it mean that the sentence is truthvalueless?
How can a word have no Sinn and/or no Bedeutung? This could be
because the word is used syncategorematically. But, first, not all words
of a sentence can be syncategorematic. And, second, this would not fit
well with Frege's systems, in 1879 and in 1893-1903, since in these
systems no sign is syncategorematic; each has a Bedeutung (and a Sinn);
the sign for negation, for example, denotes a certain function. We would
seem to be more justified, in the light of Frege's writings, in assuming
that a word has no Bedeutung because it is vague. Remember, a pre
dicate whose delimitation is 'blurred' denotes no concept. So Frege
seems to think that words of ordinary language have neither Sinn or
Bedeutung because they are vague, but that, somehow, the whole
sentence has a Sinn (though perhaps no Bedeutung, that is, no truth
value). All that remains quite schematic, and the manner in which a
sentence composed of words having no Sinn has itself a Sinn is certainly
In a manuscript written two or three years after his letter to Peano
and left unpublished, Frege wrote (1969, p. 168, or 1979, p. 155; I
modified the translation):
Inference from two premisses often, if not always, rests on there being a concept
common to both. If a fallacy is not to occur, not only must the sign for the concept be
the same, but this sign must also have the same Bedeutung [dasselbe bedeuten]. It must
have a Bedeutung independent of the context, and not first obtain one in the context,
something that is, however, very often the case with words of [ordinary] language.
Here Frege repeats what he has said in our quotation from his letter to
Peano, but only in part. He no longer speaks of words without Sinn
constituting a sentence that has Sinn. The initial remark, that in an
inference from two premisses a predicate symbol is to occur in each,
reminds us of the role of the middle term in an Aristotelian syllogism
and is not correct for modern systems; if, however, we consider all the
premisses and the conclusion of any rule of inference in any modern
system, we can safely say, I think, that every predicate symbol occurs at
least twice. And, in any interpretation, the same set has to be assigned to
each occurrence of the symbol.
Frege's view is that vagueness does not prevent communication, but
wrecks logical inferences. A word, at its two occurrences in two
different sentences, has different Sinne, that is, strictly speaking, has no
Sinn. Within a sentence, each word has a local Sinn, and these Sinne
succeed in combining in such a way that the sentence as a whole has a
Sinn; but, as we go from sentence to sentence, the Sinn of a word varies,
and these changes play havoc with logical inferences, because these
necessarily involve more than one sentence.
Frege's conception would involve, it seems, a considerable amount of
elaboration and justification. But it may be right. Vagueness is omni-
present in ordinary language, in the sense that every word has a fringe of
indeterminacy. The word 'man' is vague (and that in at least three
directions: 'adult' versus 'boy', 'Homo sapiens' versus 'ape', 'male' versus
'female'), but the word is nevertheless a serviceable component of our
language. Ordinary language is quite nimble at taming vagueness. It
makes use of all the information that the context can provide in order to
curb the vagaries of words. If that does not succeed and the indefinite-
ness of a word threatens the mutual understanding between speaker and
hearer, we drop the word and shift to another way of speaking.
Logical inference puts more severe constraints on language than does
the mere impartment of information; it tolerates less wobbling in the
senses of words, perhaps no wobbling at all. The Greeks began to worry
about the Heap and the Bald when their language became an instrument
of argumentation, whether mathematical or philosophical. Since the best
we can do, in ordinary language, is to locally reduce vagueness, but not
eliminate it, and since logic has been built on the same lines as mathe-
matics, that is, with initial bivalent predicates plus rules that maintain
bivalence (set-theoretic semantics), what is the status of logical argu-
ments conducted in ordinary language, with vague words? Shady, some
would say. Behind a vague word we perceive (imagine? assume?) a
bivalent predicate, and the argument is actually conducted with that
predicate in mind; when the words are again taken as belonging to
ordinary language, the conclusion, it is to be hoped, will remain valid.
The enterprise has an 'as if' character. In the syllogism
All Athenians are men,
All men are mortal,
Therefore, all Athenians are mortal,
the predicates ' ... is an Athenian', ' ... is a man', ' ... is mortal' are vague;
but nobody would question the validity of the argument. On the other
hand, when we try to apply mathematical induction to the predicate ' ...
is a heap', we reach an absurd conclusion, and precisely because of the
vagueness of the predicate. Here, if we would simply argue as if the
predicate ' ... is a heap' were exact, our convention would seem
glaringly artifical and leave us with a feeling of dissatisfaction. The first
Greek who drew our attention to the Heap wanted us to reflect on how
logical arguments can at all be conducted with vague words.
When we introduce logic to students, we are prone to gloss over
vagueness and we will not demur, perhaps, at giving the sentence 'It is
raining or it is not raining' as an example of the law of the excluded
middle. We carry out, without always saying so, Quine's regimentation,
that is, we assume a universe of discourse with bivalent predicates.
Quine, however, is more sophisticated than Frege about this regimenta-
tion. For him (1977, p. 195), 'the regimentation is not a matter of
eliciting some latent and determinate content of ordinary language'.
Eliciting a latent and determinate content is precisely what Frege is
doing when, behind the word 'white', he discovers the bivalent 'objec-
tive' whiteness. Quine continues: 'It is a matter rather of freely creating
an ontology-oriented language that can supplant ordinary language in
serving some particular purpose one has in mind'.
Although Frege and Quine give a different ontological status to the
universe of discourse, both agree that ordinary language has to be
supplanted if logic is to be able to function. Ordinary language is
somehow too weak to stand the stress of bivalence and should not be
asked to bear up against the requirements of logical rigidity. A logical
argument conducted in ordinary language is to be successful only
whenever we can discern, behind ordinary words, bivalent predicates, as
we perceive the watermark behind the surface of a sheet of paper.
Russell comes to a similar conclusion. According to him, we are to
imagine a precise meaning for the words of ordinary language, and logic
is 'not applicable to this terrestrial life, but only to an imagined celestial
existence' (1923, pp. 88-89). And some years later (1937, p. xi) he
writes: 'none of the raw material of the world has smooth logical
properties, but [ ... ] whatever appears to have such properties is
constructed artificially in order to have them'.
Russell's artificial construction of a bivalent world by imagination is
closer to Quine's free creation of an ontology-oriented language than to
Frege's objective realism. But the three of them, Frege, Russell, Quine,
agree on one point, namely that ordinary language has to be supplanted
by a bivalent regimented discourse if logic is to function properly.
Regimentation comes in various forms. Frege's universal domain is a
fixed Universe that comprehends all objects, the number 7 as well as
Julius Caesar, with functions on top. One could claim, as Bachmann did
in his 1975, that the individuals whose existence is imposed by the
axioms of Frege 1893 and 1903 constitute a denumerable universal
domain of purely logico-mathematical objects. But, first, even in
1893-1903 Frege constantly adduces persons or heavenly bodies as
examples of objects and uses them as counterexamples to theses that he
wants to refute; and, second, his discussion of a number of key notions,
Wirklichkeit for example, clearly indicates that he puts all objects,
whether abstract or concrete, in the same ground domain of individuals.
Hence the denumerable domain isolated by Bachman constitutes a
minimal core, but it is difficult to maintain that Frege restricts himself to
that core. (The point deserves further discussion and I hope to come
back to it on another occasion.) Russell has an infinitely stratified
hierarchy, which is not, it seems, ontologically anchored. to a fixed
ground domain. As he writes (Whitehead and Russell 1910, p. 169, or
1925, p. 161), 'lilt is unnecessary, in practice, to know that objects
belong to the lowest type, or even whether the lowest type of the
variable occurring in a given context is that of individuals or some other.
For in practice only the relative types of variables are relevant; then the
lowest type occurring in a given context may be called that of individ-
uals, so far as that context is concerned'. As for Quine, he has theories,
changed according to our needs, each of which having, it seems, a
universal domain comprehending everything that there is (in the theory).
For Frege, regimentation is global, in the sense that the supplanting
of ordinary language is done en bloc. Perhaps also for Russell, with the
difference that Frege's universe is unique and fixed, while Russells'
hierarchy can, so to speak, move up and down. But Russell's quotation,
with its reference to a given context, opens a path that we may enter and
along which we may go farther than Russell. Logic is to be used locally
to test the correctness of arguments in ordinary language. Only the few
sentences that constitute the argument are considered. We pick the
universal domain that suits us locally, introduce predicates and connec-
tives, and check the argument. We now have a logic utens rather than a
logic magna. Logic has no longer any ontological import. The question
whether the number 7 and Julius Caesar are to be put in a common uni-
versal domain of objects belongs perhaps to ontology or metaphysics,
but not to logic. Frege, no doubt, would have considered such a use of
logic to be mere bricolage. But his global supplanting of ordinary
language by a system of bivalent predicates everywhere defined over a
fixed universal domain takes us far away from the problems of ordinary
words. These words are inherently vague because the process through
which we learn them involves a finite number of instances. We have
learned how to live with that vagueness, and we may want to understand
how ordinary language works, in particular how it works in spite (or,
perhaps, because) of vagueness. But this is not a task that Frege
welcomes. 'A large part of the philosopher's task consists - or at least
should consist - in a struggle with language', he writes (1969, p. 289,
or 1979, p. 270). Frege carries on that struggle, and brushes aside
What he does is to introduce an ontology that would allow his new
logical laws to function in a way that should be unobstructed and,
moreover, as simple as possible; it should be so simple that he is even
reluctant to divide his universe into sorts. He could not have tackled the
problems of vagueness at the moment he was introducing these logical
laws without extremely complicating his ontological view of the world.
We can even say more: he would not have been able to formulate these
laws without leaving vagueness out of the picture. The only way to
proceed at an early stage in the development of a science is by bold
simplification and abstraction. As Kreisel reminds us, this was Galileo's
successful strategy, he 'found immensely manageable laws by simply
excluding the vagaries of friction and air resistance in suitable experi-
mental setups' (1984). So Frege's disregard of vagueness and other
vagaries was, in a way, inevitable. But his logical laws have been
formulated more than hundred years ago, and it is now perhaps time to
look at the vagaries.
Stanford University
Bachmann, Friedrich
1975 'Frege als konstruktiver Logizist', in Thiel 1975, 160-168.
Frege, Gottlob
1879 Begriffsschrift, eine der arithmetischen nachgebildete Formelsprache des reinen
Denkens; English translation by Stefan Bauer-Mengelberg in van Heijenoort
1967, 1-82.
1884 Die Grundlagen der Arithmetik, eine logisch-mathematische Untersuchung
uber den Begriff der Zahl; see Frege 1953.
1892 '(fuer Sinn und Bedeutung,' Zeitschrift fUr Philosophie und philosophische
Kritik, new series, 100,25-50; English translation in Frege 1952,56-78.
1893 Grundgesetze der Arithmetik, begriffsschriftlich abgeleitet, vol. 1.
1903 Grundgesetze der Arithmetik; begriffsschriftlich abgeleitet, vol. 2.
1952 Translations from the Philosophical Writings of Gottlob Frege, edited by Peter
Geach and Max Black; 2nd ed., 1960.
1953 The Foundations of Arithmetic, German text and English translation of Frege
1884 by J. L. Austin, 2nd revised edition.
1969 Nachgelassene Schriften und wissenschaftlicher Briefwechsel, vol. 1, Nachge-
lassene Schriften.
1976 Nachgelassene Schriften und wissenschaftlicher Briefwechsel, vol. 2, Wissen-
schaftlicher Briefwechsel. .
1979 Posthumous Writings.
1980 Philosophical and Mathematical Correspondence.
Hilbert, David
1904 'Uber die Grundlagen der Logik und der Arithmetik', Verhandlungen des
Dritten 1nternationalen Mathematiker-Kongresses in Heidelberg vom 8. bis 13.
August 1904; English translation by Beverly Woodward in van Heijenoort
1967, 129-138.
Kreisel, Georg
1984 'Frege's Foundations and Intuitionistic Logic', The Monist 67, 72-91.
Merrill, Kenneth R.
See Shahan, Robert W., and Kenneth R. Merrill.
Poincare, Henri
1906 'Les mathmatiques et la logique', Revue de metaphysique et de morale 14,
1908 Science et methode.
Quine, Willard Van
1977 'Facts of the Matter', in Shahan and Merrill 1977, 176-196.
Russell, Bertrand
1923 Vagueness, The Australasian Journal of Psychology and Philosophy 1, 84-92.
1937 The Principles of Mathematics, 2nd ed.
See 'Whitehead, Alfred North, and Bertrand Russell.
Shahan, Robert W., and Kenneth R. Merrill
1977 American Philosophy from Edwards to Quine.
Thiel, Christian (ed.)
1975 Frege und die moderne Grundlagenforschung.
Van Heijenoort, Jean (ed.)
1967 From Frege to Godel, A Source Book in Mathematical Logic, 1879-193l.
Whitehead, Alfred North, and Bertrand Russell
1910 Principia Mathematica, vol. 1.
1925 Principia Mathematica, 2nd ed.
We can characterize the disagreement between adherents of Russell-
style theories of meaning and those of Frege-style theories as follows:
RS theorists assume that a satisfactory theory of meaning can be
built with the binary relation - e refers to r - whereas FS theorists
maintain that a three-place relation - e through having sense s
refers to r - is required.
It should be clear from this characterization that anything an RS theory
can do can also be done by an FS theory, since the crucial notion of RS
theories, the binary reference relation, is available in any FS theory; for
we can define
e refers* to r= df' 3 s (e through s refers to r).2
The question in the dispute between the two theories is then whether
there are conditions of adequacy only FS theories can satisfy.
On my account the disagreement is not over the issue of how definite
descriptions are to be analysed. According to Russell a semantic theory
should not be expected to assign meaning directly to expressions of the
from "the f," but only to sentences in which such phrases occur. A
sentence of the form "the f is g" is then to be taken to have the same
meaning as a sentence of the form "there is one and only one thing
which is both f and g." On Frege's analysis, on the other hand, expres-
sions of the form "the f" are to be taken as proper logical constituents of
propositions which say, in effect, that the semantic theory should assign
meaning directly to such expressions.
It might seem at first sight that there is a logical link between FS
theories of meaning and the Fregean analysis of definite deSCriptions
and likewise between RS theories and the Russellian analysis of definite
descriptions. That is, however, not the case. For fairly obvious reasons
an FS theorist could consistently adopt the Russellian analysis since it is
structurally equivalent to Frege's analysis of expressions of the form "all
L. Haaparanta and J. Hintikka (eds.), Frege Synthesized, 47-64.
1986 by D. Reidel Publishing Company.
f" and "some f" which, in tum, is presumably compatible with an FS
But the claim that a Fregean analysis of definite descriptions is
incompatible with RS theories of meaning is initially more plausible. If
we treat expressions of the form "the f" as referring expressions and not
merely as having contextual meaning, we must, it seems, distinguish
between their reference and how they characterize that reference. The
expression "the teacher of Alexander the Great" refers to Aristotle and
characterizes him as a teacher of Alexander whereas the expression ''the
author of the Prior Analytic" refers to that same person, but charac
terizes him as an author rather than a teacher. Such a distinction
between the reference of an expression and how that reference is
characterized seems to demand an FS account in terms of the notions of
the reference and the sense of expressions. I will try to show that this is
not so.
A satisfactory theory of meaning in the sense in which I have spoken
of it in characterizing the disagreement between the two parties is a
theory which satisfies certain conditions of adequacy. Any such condi
tion met by an RS theory will obviously be satisfiable by an FS theory.
But Frege argues in the essay "On Sense and Reference" that there are
conditions of adequacy which only his style of theory can satisfy. One of
them, he claims, is the following:
Condition A: A satisfactory theory of meaning must explain the
(intuitively obvious) difference between trivially true and informa
tively true identity statements.
For the reasons given by Frege himself I assume that theories of this
style satisfy the condition. My question is whether RS theories also can
satisfy it. If they can, condition A cannot serve as a criterion for
determining which of the two styles of theory is preferable.
I will argue that RS theories can, in fact, satisfy condition A, but only
partially so. We can then ask whether the condition is valid only to the
extent to which both styles of theory can satisfy it. Reasons for that
claim can be produced. Even if those reasons are ultimately rejected, we
can learn something about the basic difference between RS and FS
theories by considering them.
There are other conditions of adequacy which Frege claims only his
style of theory can satisfy. Thus, he argues, that only an FS theory can
account for the logic of belief contexts and that only an FS theory can
explain the use of names which lack reference. For the purpose of the
present discussion we can set their consideration aside.
In order to show that RS theories can, at least, partially satisfy condition
A, I introduce a notion of semantic content. The claim is that this notion
can be explained by means of syntactic terms and the binary relation "e
refers to r" but can do much of what Frege's notion of sense is meant to
do. I will first illustrate the notion informally.
Given two mathematical functions f and g which we explain as
f(x) = x + 7 and g(y) = 12 - y
we can form the true equation
f(2) = g(3).
It is here obvious that the two functional expressions "f(2)" and "g(3)",
while both referring to the number 9, do- so in different ways. But in
order to describe the difference in their way of specifying the number 9,
we have no need to talk about the supposed sense of the two functional
expressions. All we need to consider is the reference of the two func-
tional terms as laid down in the initial specification, the reference of the
expressions "2" and "3" and the way the functional terms and the
number terms are combined into the complete functional expressions
"f(2)" and "g(3)". There is then a notion of semantic content, specifiable
in referential and syntactic terms alone, which allows us to say that in
the equation "f(2) = g(3)" the functional expressions have the same
reference, but differ in semantic content.
In order to make the notion of semantic content precise we need to
determine both when two expressions have the same semantic content
and when they have different ones. We must do so first for simple
expressions (proper names and functional terms) and then for complex
ones formed from them.
Since intuitively we mean by semantic content the information con-
veyed by the composition of an expression given the reference of its
constituents, it seems plausible to say that for simple expressions, i.e.,
those without composition, semantic content varies with reference. In
other words:
Two simple expressions have the same semantic content if and
only if they have the same reference.
And for complex expressions we can lay down:
Two complex expressions have the same semantic content if and
only if they are constructed in the same way out of constituents
with the same semantic content.
Thus, if "f" and "g" have the same semantic content and so do "a" and
"b", then "f(a)" and "g(b)" have the same semantic content. But if "a"
and "b" have different semantic content, then "f(a)" and "g(b)" also
have different semantic content, though they may have the same refer-
ence. Given a two-place relation "h", the two expressions "h(a, b)" and
"h(b, a)" have different semantic content since they are built up in
different ways, though they are built from the same constituents and
may have the same reference.
It should be obvious how we can give a Fregean analysis of definite
descriptions in terms of this notion of semantic content. Treating
definite descriptions as expressions of the form ''f(g)'' we can distinguish
between the reference and the semantic content of the definite descrip-
tion. Thus, ''The teacher of Alexander the Great" and "the author of the
Prior Analytic" are expressions with the same reference, but with
different semantic content.
It should also be immediately clear how we can use the notion of
semantic content to distinguish between trivial and informative identity
statements. An identity statement is informative if the two expressions
connected by the identity sign have the same reference, but different
semantic content. Thus, the sentence "The teacher of Alexander the
Great is identical with the author of the Prior Analytic" is informative
because the first expression refers to Aristotle as a teacher and the
second refers to him as an author.
It remains to be shown that the distinction between trivial and informa-
tive identity statements is only partially characterizable by means of the
notion of semantic content. That will become clearer, if we consider
what different kinds of identity statement there are.
Assuming "a" and "b" to be simple names and "m" and "n" complex
referring expressions, we can distinguish three kinds of identity state-
(1) thoseoftheform "a= b"
(2) those ofthe form "a = m" or "n = b" and
(3) those of the form "m = n".
When we test for each of those three kinds of statement how well the
notion of semantic content accounts for a distinction between trivial and
informative ones, we discover that it seems to fit the third kind of
statement, but that there are problems with the first and the second.
According to our characterization of the notion of semantic content,
all true identity statements of the first kind must be trivial. And similarly
no true identity statements of the second kind can be trivial.
That seems not to conform to our intuitions and it is here where an FS
theory appears to have an advantage. According to Frege every simple
name has a sense as well as a reference. An identity statement of the
first kind is informative for him, if the two names "a" and "b" have the
same reference but different sense. Similarly an identity statement of the
second kind is trivial for him, when the name "a" in it has the same
sense as the complex expression "m" occurring in the statement. Such
distinctions are not available in the account built on the notion of
semantic content as we have defined it, since we have laid down that
two simple names have the same content if and only if they have the
same reference.
We can remove part of the problem by changing our original defini-
tion to say:
When an expression has been explicitly introduced by means of a
definitional specification, the semantic content of the simple
expression will be the same as that of the original specification.
Two simple expressions not introduced in that way have the same
content if and only if they have the same reference.
For all identity statements of kinds (1) and (2) which contain names to
which the first part of this characterization applies the distinction can
now be made in terms of the notion of semantic content. Consider the
sentence "Cicero is identical with Tully." According to the original
characterization of the notion of semantic content the statement is
trivial; but according to the new one it is informative, if one of the
proper names has been introduced by a definite description. If both
have been introduced by definite descriptions, the statement may again
tum out to be trivial, namely when the two descriptions have the same
semantic content.
There are other possible and plausible modifications of the notion of
semantic content that make it possible to draw the trivial/informative
line of distinction closer to the intuitive one. We might, for instance,
specify that two complex expressions have the same semantic content if
one is a transformation of the second according to specifiable rules and
if the corresponding constituents in the two expressions have the same
semantic content. Such rules might state that expressions of the form "m
and n" and "n and m" have the same semantic content and likewise for
"m or n" and "n or m", as well as for "m" and "not-not-m".
Such modifications are, however, unlikely to change the fact that we
cannot completely capture the intuitive distinction between trivial and
informative identity statements in terms of such a (modified) notion of
semantic content. The case seems to be different for an FS theory, and it
is easy to see why. Frege holds that every expression with a reference
must also have a sense, for the sense is meant to determine the way the
object is given. He argues with some plausibility that in order for us to
identify or re-identify an object, we must recognize it as a something or
other. It is this mode of recognition that constitutes the sense of the sign
referring to the object. In terms of this notion of sense we can explain
why an identity statement of the first kind can be informative, and one
of the second trivial. If the two kinds of identity statement contain
proper names not introduced through definitional specifications, the
account of,the trivial/informative distinction in terms of the notion of
semantic content will not coincide with the Fregean account. And since
it is the Fregean account that is closer to the intuitive distinction we
seem to have a criterion for choosing between FS and RS theories of
meaning. The former can satisfy condition A more fully than the latter
and on that account seems to be preferable.
But this conclusion follows only if we uphold condition A without
restriction. One might argue instead that the condition is, in fact, not a
completely suitable criterion for choosing between semantic theories.
One might argue that the notions of triviality and informativeness are, in
fact, not at all unified concepts, that there are hidden semantic and
cognitive elements which must be separated when we consider require-
ments of semantic theories.
Consider the propositions "Aristotle is Aristotle" and "Aristotle was
the teacher of Alexander the Great." The former is trivial because it is
formally an instance of the principle of identity; we need not consider
how Aristotle is to be identified in order to recognize the truth of the
proposition; all we need to know is that the two occurrences of the
name refer to the same man. The second sentence, on the other hand, is
not at all trivial in this sense. It can convey information; but to an
Aristotle scholar or a Greek historian the sentence may still be a
I call the first sentence semantically trivial and the second semanti
cally informative. But what is semantically informative may still turn out
to be cognitively trivial for a particular speaker, as illustrated above. The
sentence "Aristotle is Aristotle" is, for most of us, both semantically and
cognitively trivial; but there are also sentences which are semantically
trivial though cognitively informative; "war is war" comes to mind. It is
less clear whether there can also be sentences which are semantically
informative and cognitively trivial. I will argue below that Frege, at least,
thought that there could and that this conviction motivated his introduc
tion of the theory of sense and reference.
Having distinguished the two sets of notions we are now in a position
to say why condition A may not be a suitable criterion for judging
semantic theories; for we may reasonably hold that a semantic theory
should have to explain only semantic distinctions and not cognitive
ones. Condition A should then give way to:
Condition A': a satisfactory theory of meaning must explain the
difference between semantically trivial and semantically informa
tive identity statements.
This condition, we can argue, is in fact satisfied by both RS and FS
theories and can therefore not serve as a criterion for choosing between
What reason have we for considering the notion of sense a cognitive
rather than a semantic notion? Consider again the sentence "Aristotle
was the teacher of Alexander the Great." According to Frege the
sentence is trivial, if we connect with the name "Aristotle" the sense "the
teacher of Alexander the Great." There is nothing in the expression
itself nor in the fact that this expression is a name of a particular person
which determines the sense in which the name is to be taken. If semantic
notions are those which can be explicated in terms of the nature and
structure of linguistic expressions and what they stand for, the notion of
sense is obviously not a semantic notion at all. Frege's argument for
holding that names must have sense is this: that for speakers to be able
to refer to objects they must be able to identify and re-identify objects,
objects must be given to them in particular ways. Such considerations
seem epistemic in character, rather than semantic.
It might be useful to distinguish more sharply than the philosoplftcal
literature has done so far between the semantics and the epistemology
of language. Recent work in the theory of meaning has shown how
complex issues can become in that field and for that reason alone it
might be helpful to keep its boundaries as narrowly defined as possible.
I formulate this suggestion in hypothetical terms, because I have
certainly not shown beyond doubt that a sharp distinction between
semantic and cognitive issues can be drawn in the way I have suggested.
My intention here was simply to argue that condition A is not an
uncontroversial criterion for choosing between semantic theories. If we
consider it unrestrictedly valid we commit ourselves to a view that does
not distinguish sharply between semantic and cognitive aspects of
language. My conclusion is simply that FS theories cater to such an
amalgamation, whereas RS theories demand a separation of the two
We can see how FS theories lead to an amalgamation of semantic and
cognitive issues by considering how Husserl used the Fregean notion of
sense and by generalizing it to the notion of noema generated a whole
cognitive theory from it.5 In recent work John Searle and Michael
Dummett illustrate two alternative directions in which such an amalga-
mation can lead. Searle, travelling a path parallel to Husserl's, has now
generalized the theory of sense into a theory of intentionality; the theory
of meaning has become embedded for him in a cognitive theory.6
Dummett, on the other hand, has made the theory of meaning absorb
the theory of knowledge.
In either case the boundary between the
theory of meaning and the theory of knowledge has been traversed.
The notion of semantic content appears to me of obvious interest for
systematic work in the theory of meaning. But I have introduced it
primarily for another purpose - to clarify the origin of Frege's doctrine
of sense and reference. That doctrine is closely tied to Frege's under
standing of the logical character of identity statements which, in turn, is
related to his concern with the nature of arithmetical truths.
Frege's conviction that arithmetical truths are reducible to logic, a
conviction that shaped so much of his work, is almost certainly derived
from Hermann Lotze's Logik of 1874.
There Lotze develops the
reducibility thesis in conjunction with a theory of identity that is closely
akin to Frege's own early theory in the Begriffsschrift. In order to
understand the latter it is useful to go back to Lotze's account.
Lotze, in contrast to Kant and the Kantian tradition, holds that "all
calculation is a kind of thought and that the fundamental concepts and
principles of mathematics have their systematic place in logic." (p. 26)
At the same time he agrees with Kant that arithmetical equations do not
"rest simply upon the principle of identity." (p. 504) He agrees with
Kant against Hume that such truths are necessary, but not tautologous.
Lotze, like Kant, takes the principle of identity to say no more than "a =
a", and it is obvious that the usual arithmetical truths cannot be derived
from this principle alone. Nevertheless, Lotze asserts:
We must not forget that calculation in any case belongs to the logical activities, and it is
only their practical separation in education which has concealed the full claim of
mathematics to a home in the universal realm of logic (p. 110).
For Kant the principle of identity is, of course, the one on which all
logic rests. If arithmetical truths are not derivable from that principle
alone they cannot be logical truths for him. Lotze, on the other hand,
takes a somewhat different line by introducing a distinction between the
form and the content of expressions in identity statements. And it is this
distinction which he believes will reconcile the claims that arithmetical
truths are logical, but no mere instances of the principle of identity.
If two different expressions refer to the same thing, they have for
Lotze the same content, but different form. As a matter of fact Lotze's
distinction of form and content corresponds precisely to the previously
given distinction between semantic content and reference.
The distinc
tion allows Lotze to say that an arithmetical equation like "7 + 5 = 12"
has the same content as the principle of identity, "a = a", but a different
form. He says of this equation that we have in it "a perfectly identical
judgment as regards its matter, and it is only synthetical formally
because it exhibits the number 12 first as the sum of two quantities and
then as determined by its order in the simple series of number" (p. 64).
As far as its content is concerned every identity statement (and every
arithmetical equation) is, thus, a necessary truth which says the same as
"a = a." But this necessary truth can be represented under different
forms. According to its form an identity statement can be empirical or a
priori. There can thus be empirically necessary and a priori necessary
identity statements.
Lotze says that as far as the form is concerned all
nontrivial identity statements are synthetic, some synthetic a posteriori
and some synthetic a priori. Arithmetical equations are characteristically
both identical (and hence necessary) propositions and synthetic a priori.
That seems to bring Lotze close to the Kantian view of arithmetic,
but the impression is misleading, since he does not hold that arith-
metical truths are based on an intuition of space and time. He says
rather of the exemplary equation "7 + 5 = 12":
For that which all turns upon is in fact nothing more than the assertion which is
contained in the sign of addition - viz. that quantities can be summed so as to compose
another and a homogeneous quantity; a proposition the importance of which we may
once more be tempted to ignore, because it seems to us self-evident and a mere identical
proposition defining the nature of numerical quantity as such. And so it undoubtedly is,
but how do we arrive this piece of self-evident knowledge? (p. 508).
And the answer to that question is that,
This very fact, that there is such a thing as quantity to be found in the world of ideas
... is a fact of immediate perception . .. The proposition therefore that quantities can be
summed is undoubtedly an identical proposition; but that the subject and predicate of
that proposition appear as valid in the world of ideas . . . does not follow from the
principle of identity (p. 509).
The realm of objective ideas is for Lotze the realm with which logic is
concerned. The truths that obtain in it are the truths of logic. But our
knowledge of those truths, the forms under which they appear, is
synthetic a priori knowledge. It depends on our direct apprehension of
what is to be found in the world of ideas. In order to understand the
necessity of the arithmetical truths we must apprehend that summable
quantities belong to that world. Such apprehension is synthetic a priori
in so far as it is intuitive and immediate rather than conceptual and
With this argument Lotze has, in fact, concluded that the truths of
arithmetic cannot be derived from the principle of identity alone, that
additional logical principles, a richer logic that Kant's are required in
order to carry out the derivation. He writes:
Turning to the discovery of mathematical truth, we shall not dispute the validity nor yet
the importance of the principle of identity, but we must dispute its fruitfulness; we must
insist that if it were the only principle we had to start from, mathematical truth could
never be discovered at all (p. 505).
Lotze is convinced that mathematics is an "independently progressive
branch of universal logic" (p. 26), but he never tries to spell out the
additional logical principles that he considers necessary for the logical
derivation of the mathematical and, in particular, the arithmetical truths.
That is a task Frege takes upon himself.
In so far as we can speak of a conception of meaning in Frege's early
work, it was an RS conception built on the binary relation of a name to
its content. Following Lotze's usage Frege employed the term "content"
in this period to mean as much as the reference of a name, that which
the name stands for. The account of identity statements which he offers
in his earliest publication, the Begriffsschrift, presupposes such an RS
conception and is essentially equivalent to the account given here in
terms of the notion of semantic content.
There are, to be precise, two aspects to the Begriffsschrift account of
identity which it is important to separate. The first is that identity
"applies to names and not to contents ... for it expresses the circum
stance that two names have the same content."l1 The second is that
statements can be informative because "the same content can be
completely determined in different ways." The informative identity
statement says that two ways of determining the content can yield the
same result. Frege writes:
Before this judgment can be made, two distinct names corresponding to the two ways of
determining the content must be assigned to what these ways determine. The judgment,
however, requires for its expression a sign of identity of content, a sign that connects
these two names. From this it follows that the existence of different names for the same
content is not always merely an irrelevant question of form; rather, that there are such
names is the very heart of the matter if each is associated with a different way of
determining the content (p. 21).
There is no question that Frege's distinction between the content of a
name and its way of determining a content is derived from Lotze's
account of identity and like it corresponds to our distinction between
reference and semantic content.
At the same time there is one important difference between Frege's
and Lotze's account. For Frege an identity statement is a statement
about signs, whereas Lotze makes no such assumption. One reason for
Frege's departure from Lotze's account suggests itself immediately. The
latter has the peculiar consequence that identity statements can turn out
to be both identical (and thus analytic in their content) and synthetic in
form. In the Begriffsschrift Frege may have considered it possible to
simplify the story by considering identity statements as statements about
signs. In that case, he can say that statements of the form "a = a" are
analytic and that identity statements in which two expressions of differ
ent form stand for the same content are synthetic. This is, in fact, what
he indicates when he says about the latter:
In that case the judgment that has the identity of content as its object is synthetic, in the
Kantian sense (ibid.).
But this simplification has the unfortunate consequence that Frege is no
longer able to explain how any but the most trivial arithmetical equa
tions can turn out to be logical truths. This, however, was a point he
came to appreciate only later.
While Frege allowed in the Begriffsschrift that signs can sometimes
stand for what they signify and sometimes for themselves, he later came
to reject such a "bifurcation" violently. It was, presumably, the confron
tation with mathematical formalists who assumed that number state
ments are statements about numerical signs which made Frege aware of
the need to distinguish sharply between the sign and the signified. Given
the fact that he believed that his early account of identity statements
depended on taking them as statements about the signs, we can see, at
least, one reason why he may have wanted to modify his early theory.
He writes in the essay 'On Sense and Reference':
What is intended to be said by a = b seems to be that the names "a" and "b" designate
the same thing, so that those signs themselves would be under discussion; a relation
between them would be asserted.i2
It is clear that he finds this no longer plausible; but the reason he offers
for why the earlier account is implausible is peculiar. He continues:
But this relation would hold between names and signs only in so far as they named or
designated something. It would be mediated by the connexion of each of the two signs
with the same designated thing. But this is arbitrary. Nobody can be forbidden to use
any arbitrarily producible event or object as a sign for something. In that case the
sentence a = a would no longer refer to the subject matter, but only to its mode of
designation; we would express no proper knowledge by its means (ibid.).
This explanation of what is wrong with the earlier account is peculiar
because if consists in a mere repudiation of what he had previously
affirmed without really adding new arguments. In the Begriffsschrift he
had said that "the existence of different names for the same content is
not always merely an irrelevant question of form," but he now asserts
blandly that "nobody can be forbidden to use any arbitrarily producible
event or object as a sign for something."
It seems at this point that the earlier account is, in fact, more
plausible. Every simple word in the language could, of course, have a
reference different from the one it has, but once the reference of simple
words is fixed, the reference of complex terms formed from them seems
no longer arbitrary. Why then did Frege feel himself forced to abandon
his earlier account? Surely not just because he had come to reject the
idea the identity statements are statements about the signs.
There is, in fact, no way in which we can understand the motivations for
Frege's shift by looking exclusively 'On Sense and Reference.' Instead
we must turn to the essay 'Function and Concept,' published shortly
before 'On Sense and Reference,' where the new semantic theory is
discussed for the first time.
'Function and Concept' is clearly part of Frege's work on the thesis
that arithmetic is reducible to logic. In it he introduces the notion of the
value-range of a function, meant as a generalization of the notion of
class. And he argues for a principle which he considers essential for his
reductionist program which says that statements of the form "the value-
range of the function f is identical with that of the function g" are
equivalent to "the functions f and g have the same value for identical
arguments." He writes:
The possibility of regarding a general statement to the effect that the values of functions
are equal as an equality, namely as an equality of value-ranges is, I think, not capable of
proof, but must be considered a basic law of logic (p. 26).
In the Basic Laws of Arithmetic, Frege's main work, the principle recurs
as one of the logical axioms (Axiom V) from which arithmetic is to be
derived. In accordance with his practice of regarding equivalences as
identity statements, it is stated there as:
(x/(x) = yg(y = 'VZ(j(Z) = g(Z14
Since Frege was aware of the fact that this axiom was essential for his
derivation of arithmetic, and since his project was to show that arith-
metic is nothing but an extended logic, he must have asked himself
immediately upon formulating the axiom what reasons there were for
considering it a logical principle at all.
The Begriffsschrift account of identity statements can, in fact, provide
no such reasons. The expressions on the two sides of the identity sign
obviously have different semantic content and, according to the
Begriffsschrift, statements of that sort are synthetic rather than analytic.
A new account of identity statements and a new account of meaning is
required, if Axiom V is to come out as a logical (and analytic) truth.
Frege begins the essay 'Function and Concept' by recalling the
Lotzean distinction between form and content (p. 22). But this termi-
nology is now set aside in favour of another one, the sense-reference
terminology, which occurs for the first time in the context of the
discussion of the supposed basic law of logic concerning value-ranges.
Considering the step from a statement of the form x/ex) = yg(y) to one
of the form 'Vx(j(x) = g(x, Frege writes that the latter "expresses the
same sense, but in a different way" (p. 27). It is at this point that the
new, technical notion of sense is made to do work for the first time.
The point seems clear. If we are to regard a statement of the form
"a = b" as a logical truth, we must assume that the two expressions "a"
and "b" do not only stand for the same thing, they must also have the
same meaning. But the notion of sameness of meaning that is here
required is not one that can be spelled out in terms of the notion of
semantic content. The whole point of Axiom V is to say that statements
of different form (and hence of different semantic content) can be
logically equivalent. Axiom V is meant to be semantically informative,
but cognitively obvious.
A different notion of meaning, one not explicable in purely semantic
terms, is required in order to assure us that Axiom V is indeed a logical
principle. The notion of sense is that new concept. According to Frege,
Axiom V shows us how the same content can be apprehended in two
different ways. This apprehension of sameness, which guarantees the
status of the axiom as a logical principle, does not express itself in the
formal structure of the axiom. It is an immediate, intuitive apprehension.
It is clear then that the notion of sense serves here as a cognitive notion;
it concerns our conception or apprehension of a particular truth.
Michael Dummett has recently said that Frege never repeated the
claim made in 'Function and Concept' that the two sides of Axiom V
have the same sense. He writes that
in the Grundgesetze, where much space is devoted to justifying the introduction of value-
ranges, the assertion is not repeated .... The remark in Function und Begriffis surely a
residue from his earlier style of thought, before sense and reference had been distin-
guished; when he had reflected further on the new distinction, he realized that he could
not sustain the claim.15
Dummett's assertion is, in fact, open to doubt. In the Basic Laws of
Arithmetic Frege writes:
I use the words "the function f(x) has the same value-range as the function g(x)" as
gleichbedeutend with the words "the functions f(x) and g(x) have the same value for
identical argument (p. 36).
The word I have left untranslated seems, at first sight, appropriately
rendered by "denoting the same as" or by "standing for the same thing
as."16 But there is a question whether such renderings are, in fact, right.
In 27 of the Basic Laws Frege explains:
We introduce a name by means of a definition by stipulating that it is to have the same
sense and the same reference as some name composed of signs that are familiar.
Thereby the new sign becomes gleichbedeutend with that being used to explain it (pp.
Here the word "gleichbedeutend" obviously means as much as "synony-
mous" and not just "having the same reference." And if the former is the
correct translation, then Frege is clearly still convinced that the two
sides of Axiom V have the same sense.
Even if ultimately he changed his mind on this point, the fact remains
that he initially introduced the notion of sense in order to explain why
Axiom V is not a synthetic truth and not because of the unsatisfactory
argument at the beginning of the essay 'On Sense and Reference.'
Dummett is, of course right that in the Basic Laws Frege did not justify
the introduction of the sense-reference distinction in terms of the needs
of his philosophy of mathematics. In the introduction to that book he
gives a different account of what he considers to be the most important
aspect of the new semantic theory. He writes:
Formerly I distinguished two components in that whose external form is a declarative
sentence: (1) the acknowledgment of truth, (2) the content that is acknowledged to be
true .... This last has now split for me into what I call "thought" and "truth-value," as a
consequence of distinguishing between sense and reference of a sign .... How much
simpler and sharper everything becomes by the introduction of truth-values, only
detailed acquaintance with this book can show (pp. 6f).
It may at first seem strange that Frege puts no emphasis at all in this text
on the new semantics of names and definite descriptions which he had
described in the essay 'On Sense and Reference'; but then it becomes
clear from looking more carefully at the Basic Laws that this part of the
sense-reference theory is of no significance for the logic he lays out in
that book or for the attempted derivation of arithmetic from the logic.
Going back to the essay our attention is drawn to the fact that the
largest portion of it is devoted to the defence that declarative sentences
under normal circumstances have truth-values as references, and not to
the semantics of names and definite descriptions.
The question remains why Frege should have considered the introduc-
tion of truth-values such a significant achievement. Michael Dummett
once said that Frege's doctrine that sentences refer to truth-values
"proves to have very implausible consequences."18 He was thinking then
of the assimilation of sentences to names and of predicates to functions
that goes with that doctrine. Frege obviously considered the possibility
of such an assimilation one of the great advantages of the new theory.
While he had made a distinction between semantic content and refer-
ence in his early work (or mode of designation and content, as he said in
the Begriffsschrift), he had not applied that distinction to the semantics
of complete declarative sentences. In the case of such sentences he had
simply spoken of their content. In the 1890s when he introduced his
new semantic theory he came to think that various technical simplifica-
tions would be possible, if the distinction of sense and reference was
systematically applied to all meaningful expressions, names, definite
descriptions, functional expressions, and sentences alike. In retrospect
he treats that discovery as the important breakthrough of the doctrine of
sense and reference.
But it should be clear that it was only a historical accident that had
stopped Frege from adopting the same kind of assimilation in the
Begriffsschrift. He could have distinguished there between the semantic
content and the reference of declarative sentences and he could have
taken all true sentences to have the same reference and all false ones,
too. The new semantic theory was, thus, not at all necessitated by the
introduction of truth-values. It was rather, as I have shown, the question
of the status of arithmetical truths that made the break with the earlier
semantic doctrines inevitable for him and forced him to introduce a
cognitive notion of sense.
University of California
I The fact that I am describing here a conflict between RS theories and FS theories
does not mean that I assume all theories of meaning to be of one or the other of the two
kinds. The verificationist theory of the Vienna Circle is an example of a totally different
sort of theory.
2 I introduce the term "refers*" here in order to obviate possible objections from FS
theories that the newly defined term is not the same as the binary term used in RS
theories. It is sufficient for my purpose to show that RS theories can be modelled in FS
3 I make the Fregean assumptions here that all expressions are either names or
functional expressions and that either kind can have reference. If we allow functional
terms to refer to partially defined functions (Le., functions that lack values for certain
possible arguments) we can account for the fact that expressions can have semantic
content but no reference.
4 Michael Dummett writes that "to give an account of the sense of an expression is ...
to give a partial account of what a speaker knows when he understands that expression."
('Frege's Distinction of Sense and Reference' in Truth and Other Enigmas, Cambridge,
Mass.: Harvard University Press, 1978, p. 122.) I know, in fact, of no better arguments
against taking the notion of sense as a semantic notion than the ones Dummett produces
in this essay in support of that idea.
Arguments for taking the notion of sense as a cognitive notion are also provided by
Tyler Burge, 'Sinning Against Frege,' Philosophical Review, vol. 88, 1979, pp.
398-432. While I agree with much of what Burge says, I do not quite agree with him
that Frege never intended to give a semantic theory in terms of the notion of sense. It
was just that his reflections drove him to amalgamate semantic and epistemi<; questions.
5 The best account of this is given in Dagfinn F0llesdal, 'Husser!'s Notion of Noema,' in
B. Dreyfus, ed., Husserl, Intentionality, and Cognitive Science, Cambridge, Mass: MIT
Press, 1982, pp. 73-80.
6 John Searle, Intentionality, Cambridge University Press, forthcoming.
7 Michael Dummett, 'Can Analytical Philosophy Be Systematic and Ought it to Be?' in
Truth and Other Enigmas, pp. 437-458; as well as in many other places, Vol. 65,
8 Hermann Lotze, Logic, trans!. B. Bosanquet, Oxford: Clarendon, 1884. All refer-
ences will be to this edition. I try to justify the claim that Frege was influenced by Lotze
in Hans Sluga, Gottlob Frege, London: Routledge & Kegan Paul, 1980.
9 It is unfortunate that my term "semantic content" is so close to Lotze's (and Frege's)
term "content," when the terms mean quite different things. I retain my use here for the
sake of terminological continuity with an earlier paper discussing these issues. Cf. Hans
Sluga, 'On Sense,' Proc. of the Arist. Soc., vo!., 651964-65, pp. 25-44.
10 Lotze's account has here obvious similarities to that given recently in Saul Kripke's
Naming and Necessity.
11 Gottlob Frege, Begriffsschrift, in J. v. Heijenoort, ed., From Frege to Godel, Cam-
bridge, Mass.: Harvard University Press, 1967, p. 20. All further references will be to
this edition.
12 Peter Geach and Max Black, Translations from the Philosophical Writings of Gottlob
Frege, Oxford: Blackwell, 1977, p. 56. All references to the essay 'On Sense and
Reference' will be to this edition.
13 References will be to the Geach-Black volume.
14 G. Frege, The Basic Laws of Arithmetic, ed., M. Furth, Berkeley/Los Angeles:
University of California Press, 1967, p. 105. Notation slightly altered.
15 M. Dummett, The Interpretation of Frege's Philosophy, London: Duckworth, 1981, p.
16 Furth gives the former translation (Basic Laws, p. 36); Geach and Black give the
latter (Translations, p. 154).
17 Furth is aware of the problem of how to translate "gleichbedeutend" and renders it
here as "having the same meaning." He adds in a footnote: "In view of the previous
sentence, it seems best to translate this in the manner of the ordinary German gleich
bedeuten and not to restrict it to Frege's technical use." He does not, however, see the
relevance of this point to the earlier passage.
18 Michael Dummett, 'Frege's Philosophy,' in Truth and Other Enigmas, p. 107.
The question famed of old, by which logicians were
supposed to be driven into a comer ... is the ques-
tion: What is truth?
To know what questions may reasonably be asked is
already a great and necessary proof of sagacity and
insight. For if a question is absurd in itself and calls
for an answer where none is required, it not only
brings shame on the propounder of the question, but
may betray an incautious listener into absurd answers,
thus presenting, as the ancients said, the ludicrous
spectacle of one man milking a he-goat and the other
holding a sieve underneath.
Critique of Pure Reason A 58 = B 82
The first of three fundamental principles Frege enunciates at the
beginning of The Foundations of Arithmetic bids us "always to separate
sharply the psychological from the logical, the subjective from the
objective."! As commonly understood, this principle represents little
more than Frege's insistence on the distinction between mind-indepen-
dent objects and mind-dependent states, and so expresses his rejection
of subjective idealism. Such an ontological construal of the objective-
subjective distinction in its turn supports a very common reading of
Frege according to which he is the archetypical metaphysical platonist.
The mind-independent existence of things is for Frege a presupposition
of the representational operation of language: it explains how our state-
ments are determinately true or false apart from our ability to make or
understand them. On this reading, Frege's novelty lies in his theory
of language, a theory that offers an impressively general and precise
account as to how the truth-value of a sentence is determined by the
reference of its well formed parts. Frege is thus canonized the father of
formal semantics. Furthermore, application of this theory to particular
stretches of discourse enables us to uncover the ontological presupposi-
tions of the discourse. Frege, in The Foundations of Arithmetic, provides
the paradigm for philosophical application of formal semantics in
L. Haaparanta and 1. Hintikka (eds.), Frege Synthesized, 65-95.
1986 by D. Reidel Publishing Company.
arguing that the objective truth of the statements of pure and applied
arithmetic requires the mind-independent existence of numbers.
The crucial feature of this line of interpretation is its taking ontologi-
cal notions, especially that of an independently existing thing, as prior
to and available apart from logcal ones, from notions of judgment,
assertion, inference, and truth. The explanatory priority of ontological
notions renders intelligible and inevitable the questions, "How does
language hook on to reality?" and "How do we know that the ontologi-
cal presuppositions of our discourse are satisfied?" The contemporary
consensus is that such answers to these questions as can be extracted
from Frege are patently inadequate. Frege's most promising response
might appear to lie with his doctrine that the sense expressed by a word
determine to what, if anything, the word refers. But a moment's thought
shows that Frege's doctrine, whatever other uses and motivations it has,
only splits our original question into two equally intractable ones: how
do words become associated with senses, and how do the senses, our
words express determine their referents? Reading Frege like this, we
read much subsequent philosophy of language as attempts to bridge the
gulf between language and the world that now looms so starkly. Russell's
logical atomism resting on direct acquaintance with sensory items
was an early attempt to give semantics an epistemological foundation.
Causal theories of reference are more recent attempts to do the same.
There is another philosophically more interesting and historically
more apt construal of Frege's work, one which denies to ontological
notions the independence and primacy they have on the platonist
interpretation. As I read Frege, ontological categories are wholly super-
venient on logical ones. This supervenience is the product of the
fundamental status Frege assigns to judgment. That judgment should be
the starting point for Frege's philosophy is unsurprising, given his
animus toward the naturalism and empiricism prevalent in mid-nine-
teenth century German philosophy and his corresponding sympathy
with Kant and Leibniz.
The priority of judgment is to guarantee its
objectivity, as exhibited in the linguistic practice of assertion, against any
general challenge. Thus, it is meant to render unintelligible the chasm
between thought and reality that is the consequence of the platonist
This is not to say that Frege's metaphysics of judgment is entirely
coherent. The priority of judgment shapes Frege's conception of logic
and motivates the identification of his Begriffsschrift as logic. There are,
however, deep tensions in Frege's thought that arise from the failure of
Frege's logical doctrines, harnessed as they are to the execution of his
logicist program, to be true to the conception of judgment that moti-
vates them. These tensions can even be said to open up a gap between
language and reality, though one whose character is rather different
from that just described. Moreover, I hold that Wittgenstein in the
Tractatus exploits these tensions to motivate the distinctive logical
doctrines of that work. By systematically criticizing Frege's philosophy,
Wittgenstein, unfettered by Frege's constructive interests, attempts to
rework a Fregean conceptiori of judgment into a complete and coherent
account of logic, language, and the world. We shall not, however, be in a
position to identify the difficulties in Frege's thought or to assess their
influence on subsequent philosophy without understanding the sources
of Frege's notions and doctrines in judgment. It will emerge that
anything like formal semantics, as it has come to be understood in the
light of Tarski's work on truth, is utterly foreign to Frege.
Frege's conception of judgment is best approached through an exami-
nation of his rejection of psychologism. Frege's extensive polemic is
more than the deserved dismissal of a nest of scientifically crude and
philosophically confused views. By scrutinizing the terms of criticism he
brings to bear against psychologism, we shall see that the contrast
between objective and subjective is not an ontological one. Understand-
ing the character of this contrast will prove to be the first and decisive
step toward appreciating the primacy of judgment in Frege's philosophy.
Indeed, it is in this primacy that Frege himself locates his break with
traditional logic:
For in Aristotle, as in Boole, the logically primitive activity is the formation of concepts
by abstraction .... As opposed to this, I start from judgments and their contents, and
not from concepts.
Psychologism maintains that logic is a branch of psychology: accord-
ingly, logical laws are generalizatiqns about the inferential operations
of the human mind. Frege, of course, has no quarrel with the empirical
study of human cognitive faculties, but he takes the psychologistic
logician's denial that there are non-psychological laws of logic to mani-
fest a confusion of the objective and the subjective. For this denial, we
shall see, forces an assimilation of assertions to the ventings of inner
states, to cries of pain and shouts of joy. This contrast is, I hold, the
source of Frege's conception of objectivity and logic.
To see how psychologism involves confusing the subjective and the
objective, let us consider a line of argument extractable from the intro-
duction of The Basic Laws of Arithmetic. The psychological logician
who maintains that the laws of logic are empirically established general-
izations is committed to the intelligibility of encountering and recogniz-
ing beings whose own thought is governed by laws different from those
that describe our own. So, to use Frege's example, we might encounter
beings who do not accept the principle of identity, who sometimes or
always deny the statements of the form "t = t" that we unhesitatingly
affirm. The psychological theory describing our inferential habits says,
as Frege puts it, "It is impossible for people living in the year 1893 to
acknowledge an object to be different from itself."4 The psychological
theory describing the logical aliens denies this of them. There is, of
course, no inconsistency between the two theories, as the claims of
each are suitably circumscribed so as to apply only to the appropriate
population. The possibility of logical aliens raises the question as to
which inferences are correct, ours or theirs. This seems to be an extra-
psychological question; for neither the psychologist's description of our
inferences nor his description of the aliens' addresses the question of the
self-identity of every object. Were the psychological logician to admit
the legitimacy of this question, he would concede the existence of a
nonpsychological study of inference. This concession would be lethal to
his position.
The psychologistic logician can turn aside this question only by
identifying the content of the principle of identity with that of the
psychological law that asserts the universal acceptance, by us, of the
principle. This identification is ill founded because, as Frege says:
There is no contradiction in something's being true which everyone takes to be false ....
If it is true that I am writing this in my chamber on the 13th of July, 1893, while the
wind howls out-of-doors, then it remains true even if all men should subsequently take it
to be false.
We should not, however, take this observation - one the psychologist
would find question-begging - as the resting point for Frege's criticism.
The imagined encounter with the logical aliens serves rather to highlight
the special role logic plays in discourse, a role belied by the psycholo-
gistic account.
What makes radical disagreement over logic so peculiar is that the
principles of logic provide us with a shared background, context, or
framework that enables us, first, to recognize disagreement, and second,
to arbitrate the consequent debate. Once two of us recognize that we
have made contradictory judgments, we each put forth reasons for our
opposed claims. Application of the principles of logic enables us to
measure the relevance of these further claims to the original dispute. It
also enables us to locate the source of our controversy within the bodies
of belief we each bring to bear on the dispute. Once we have done this,
characteristically, we will be in a position to use the basic laws and
methods of the particular subject matter under consideration to resolve
the difference. So, when we imagine logical aliens, we are imagining
beings with whom we cannot reason. Small wonder that Frege remarks
that to encounter logical aliens would be to discover a new type of
insanity and warns that the psychologistic logician, in both adhering to
the standards of consistency logic provides but refusing to reject the
aliens' thought as contradictory, is attempting to jump out of his own
The psycho logistic logician, if he is to maintain his monopoly on
the study of logic, must adopt a more tolerant attitude toward the
aliens. But how does the psychologist resist Frege's suggestion that he
ought to reject the aliens' thOUght as contradictory while investigating
the causal antecedents of their malady? The tolerance of the psycholo-
gist stems from his identification of logical principles with psychological
laws. From the vantage point of his theory, our appeal to the principle
of identity in rejecting the aliens' claims as contradictory amounts to
no more than an insistence that we think like this - a claim whose
content is dispassionately spelled out in the causal laws describing our
thought. This identification reveals that the psycho logistic logician is
treating disagreement in judgment as merely a species of psychological
difference. As such, the recognition of disagreement no more raises
an issue of correctness than does the acknowledgment of any other
personal idiosyncracy. Disagreement receives special treatment only
in that if I have judged p to be true and you have judged p to
be false, then, according to the laws describing our thinking, neither
of us can adopt the judgment of the other without relinquishing his
previous one.
It is this treatment of disagreement that reveals the psychologistic
logician's confusion of the objective and the subjective. Consider three
diners who have just shared a sumptuous repast. Two sigh with satiated
contentment and one groans with distended discomfort. The groan in no
way contradicts the sighs. It merely gives public expression to a state
that is causally incompatible with the the sighs express. The
incompatibility by itself raises no issue among our diners. One may
accuse another of shamming a feeling, but such an accusation does not
contradict the supposedly dishonest expression. Nor do the two diners
who sigh agree with each other. Each diner expresses his own feeling;
and even if the feelings are exactly similar, nonetheless, each in sighing
gives vent to his own. This last observation is just the reverse side of the
preceding one: the possibilities of agreement and disagreement go hand
in hand. If I cannot contradict your cry of pain, neither can I affirm it.
The psycho logistic treatment of judgment and assertion is perfectly
parallel to the preceding description of inner states and their expression.
Judgments are assimilated to inner states and assertions to the ventings
that manifest them. The price of this identification is the conflation of
contradiction with causal incompatibility. This conflation underlies the
psychologist's refusal to see any issue between the logical aliens and us.
But it also forces on us a falsified conception of judgment, one whose
misconstrual of the possibilities of agreement and contradiction im-
manent in discourse leads to a sort of solipsism of judgment. Frege
rejects this solipsism in a memorable passage:
If every man designated something different by the name "moon", namely one of his
own ideas, much as he expresses his own pain by the cry "Ouch", then of course the
psychological point of view would be justified; but an argument about the properties of
the moon would be pointless .... If we could not grasp anything but what was within our
own selves, then a conflict of opinions [based on] mutual understanding would be
lacking .... There would be no logic to be appointed arbiter in the conflict of opinions?
On the interpretation I have been presenting, the absurdity Frege is here
evoking is not the picture of the self trapped within the veil of its own
ideas unable to claw its way through to epistemic contact with the
material world. The solipsism Frege has in mind is one of communica-
tion, in which it is impossible for people to contradict each other and
hence impossible for them to agree either - the solipsism of refined
society in which every assertion is met with the demurral, "That's your
opinion." Frege thus rejects the psychologistic account of judgment for
its conflation of assertions with the ventings of inner states. These two
acts are entirely different, as the impossibility of contradicting another's
groan or affirming another's sigh shows.
It should be noted in passing that Frege thinks that the psychologist's
confusion is natural enough; for the vast portion of our uses of language
serve both to manifest judgments in assertion and also to give vent to
feelings, emotions, and ideas. Frege's tirelessly repeated admonition to
distinguish the ideas or images associated with a word or sentence from
its content, its sense and meaning, is to re-enforce our appreciation of
the radical difference between these two functions of language.
He goes
so far as to urge that logicians master several foreign languages to assist
them in readily applying this distinction:
It is true .hat we can express the same thought in different languages, but the psycho-
logical trappings, the clothing of the thought, will often be different. That is why the
learning of foreign languages is useful for one's logical education. Seeing that the same
thought can be worded in different ways, we learn better to distinguish the verbal husk
from the kernel with which in any given language, it appears to be organically bound
I have maintained that Frege's distinction between the subjective and the
objective lodges in the contrast between asserting something and giving
vent to a feeling. We now need to consider how this contrast gives rise
to Frege's conception of objectivity. Let us begin with an examination of
the interconnections Frege draws among a raft of notions - assertion,
judgment, content of judgment or thought, understanding, and inference.
None of these notions can be understood apart from the others, and it is
by attention to language and our linguistic practices that these notions
are to be collectively elucidated. For example, Frege's characterization
of assertion as the manifestation of a judgment does lend a certain
priority to the latter notion. The sentences we utter when we make
assertions, considered as series of sounds, have no communicative
powers in their own right. Nevertheless, it would be a mistake to think
that we have any understanding of what an act of judgment is apart from
the given by the formula that judgments are what assertions manifest.
Frege's characterization of judgment as the recognition of the truth of
a thought gives a structure of judgment. Let us begin with the distinction
between the act, jUdging or recognizing to be true, and the object or
content of the act. Frege introduces this distinction by observing that we
need to distinguish between understanding or grasping a thought and
recognizing it either to be true or to be false. The precedent for this
distinction lies in the intimate relation between yes-no questions and the
assertions that answer them. Frege describes this relation by saying that
an assertion that affirmatively answers a yes-no question manifests the
recognition of the truth of the thought that the question puts forward for
consideration. To distinguish between understanding or grasping a
thought and recognizing its truth is to acknowledge that we can
genuinely ask a question without being in a position then and there to
answer it, even if we know what sorts of reasoning and investigations
would answer it.
This way of thinking about judgment is bound up with Frege's insist-
ence that several people can all grasp the same thought and judge it true
or false. In this respect judging differs from feeling: no one can have my
pain. So Frege says things like the following:
Thus I can also acknowledge thoughts as independent of me. Other men can grasp them
just as much as I: I can acknowledge a science in which many can be engaged in
research. We are not owners of thoughts as we are owners of our ideas.! 0
Passages like this should not be taken to fund the platonist reading of
Frege. The platonist interpretation supposes the identification of
thoughts as abstract mind-independent objects to give an ontological
foundation to the distinction between subjective and objective. Frege's
language for talking about judgment is rather a means for systematically
redescribing selected features of our linguistic practices, those which
elucidate the various aspects of Frege's conception of objectivity. I talk
here of redescription to preclude the idea that Frege is presenting the
ontological underpinning that secures the objectivity of our judgments.
From the perspective Frege acquires in starting from judgments and
their contents, the distinction between objective and SUbjective exhibited
in our linguistic practice needs no securing and admits of no deeper
explanation. We have seen how the distinction between the act and
content of judgment is enforced by our appreciation of the relation
between questions and their answers. Talk of several people grasping
the same thought just restates the possibility of agreement that Frege
takes to be intrinsic to assertion.
We have considered the genesis of Frege's distinction between the act
and content of judgment. Let us now focus on the act, in particular on
die significance of the phrase "recognize the truth of". Frege remarks,
"A propositional [yes-no] question contains a demand that we should
either acknowledge the truth of a thought or reject it as false."!! To
grasp a thought is then to be faced with the question of whether it is to
be affirmed or denied; or, as Frege will say in order to avoid the
unnecessary introduction of two parallel species of judging, it is to be
faced with the question whether the thought or its opposite, its negation,
is to be recognized as a truth. To talk of opposition here is to observe
that the same thought may not both be affirmed and denied; one and
only one of a thought and its opposite may be recognized to be a truth.
This observation should be distinguished from the claim that a person
can not affirm and deny the same thought, that it is psychologically
impossible to do so. This second claim is an empirical claim about our
mental constitution concerning which Frege is agnostic. Frege distin-
guishes between these two claims by locating our appreciation of the
impermissibility, the incorrectness, of affirming outright contraditions in
our understanding, and hence the incorrectness itself in the opposition
between the thoughts grasped. On this picture, the contents of judgment
impose standards on our acts of judging. These "standards", unlike the
sort of standards that apply, say, to gymnastic performances, are in-
escapably applicable to any judgment; we cannot opt out of them and
still take ourselves to be making judgments. This point will receive
closer examination in Section II.
Frege takes his stand against psychologism by insisting on the
trichotomy of agreement, disagreement, and mere difference in judg-
ment. We noted how the possibilites of agreement, disagreement, and
mere difference in judgment come together. We now see the distinction
between the act and the content of judgment together with Frege's talk
of thoughts and their opposites as providing a structured way of
marking out this distinguishing feature of judgment.
There is more to our understanding of a thought than just the
appreciation that it and its opposite may not both be affirmed. A person
who understands,. for instance, both the thought that every philosopher
is wise and the thought that Socrates is a philosopher but not wise,
realizes that these two thoughts are contradictory. Here though, the
second thought is not the opposite of the first; nothing in our grasp of
them precludes their joint denial. I have been speaking of our appre-
ciation of patent inconsistencies. The other side of this appreciation
is our awareness of elementary implications, of the basis that the
recognition of the truth of one thought provides for further judments.
Frege's primary given is this awareness of obvious implications and
contradictions - I speak loosely and nontechnically here. It is this
awareness that enables us to discern agreement and disagreement, 10
reason together, and so, in the fullest sense, to communicate.
It is in this context that Frege distinguishes causal explanation of
person's beliefs from justification of those beliefs, and so separates logic
and psychology. According to Frege, there are two ways in which a
judgment may be justified: "We justify a judgment either by going back
to truths that have been recognized already or without recourse to other
judgments. Only the first case, inference, is the concern of Logic."12
Characteristically, to justify an assertion, one presents reasons for the
assertion: that is, one puts forward further assertions from which the
given assertion may be inferred. Such a procedure, of course, may only
push the question of justification back a step, to the justification of the
premises. Eventually, we arrive at primitive truths whose assertion is not
inferentially justified. These will presumably include the basic laws of
the discipline within whose purview the original assertion lies and, in the
case of empirical claims, observationally established assertions concern-
ing the properties and relations of material objects. Frege has almost
nothing to say about noninferential justification, His interest lies exclu-
sively with the first species of justification, with inference.
While we have the capacity to recognize elementary implicational
relations and with it the capacity to construct chains of reasoning that
justify various judgments, our exercise of this last ability is often
haphazard. The problem Frege sees with our argumentative practice
lies not so much in our unwitting fallacies as with enthymatic reasoning.
Often the truth of a conclusion drawn from an explicity stated premise
depends on a body of unstated assumptions. Because even our explicit
reasoning leaves so much unspoken, what passes for a demonstration of
a claim may leave us in doubt as to the judgments that justify the claim
and hence in doubt as to the ultimate epistemic status of the claim.
Frege is particularly exercised by this carelessness on account of his
desire to maintain against Kant that pure intuition plays no role in
mathematical reasoning. Frege speaks of " ... chains of deductions with
no link missing such that no step in it is taken which does not conform
to some one of a small number of principles of inference recognized as
purely logical," and goes on to complain:
To this day, scarcely one single proof has ever been conducted on these lines; the
mathematician rests content if every transition to a fresh judgment is self-evidently
correct without inquiring into the nature of this self-evidence, whether it is logical or
Logic must aim to state general principles whose application enables us
to determine if one statement, by itself, implies another, and so forces us
to identify all the premises of any proof. Application of the principles of
logic will then enforce on us the standards of explicitness, clarity, and
rigor that are the prerequisites for rational communication.
In order to play this role, the principles of logic must be applicable to
any topic whatsoever. This universal applicability of the principles of
logic constrains the form the statement of these principles may take. The
principles must not draw on the terms of any special science in such a
way as to preclude the application of logic to other subjects. Logic
cannot lie in the purview of any other subject. Nor does logic have a
restricted subject matter all its own.
Frege says, "Just as 'beautiful' points the way for aesthetics and 'good'
for ethics, so do words like 'true' for logiC."14 It might seem from this
remark that just as ethics provides us with a compilation of principles by
which to assess the moral goodness of actions, so logic enables us to
assess the truth or falsity of thoughts. Frege quickly rejects this sugges
tion, noting in the next sentence:
All sciences have truth as their goal; but logic is also concerned with truth in a quite
different way. Logic has much the same relation to truth as physics has to weight or heat.
To discover truths is the task of all sciences; it falls to logic to discoverer the laws of
Frege's negative point is clear enough. Every special science is con
cerned with truth insofar as it aims to state those general principles
which enable us to infer the truth or falsehood of further claims that lie
within the province of the discipline. How then are we to understand
logic's special and defining interest in truth?
It is tempting to read the previously quoted passage from "Thoughts"
to be suggesting that the laws of logic are about truth in much the same
way that the laws of physics are about weight and heat. The logician as
well as the physicist aims the state truths; however, the statement of
the logician's truths, unlike the physicist's, requires the use of a truth
predicate. This interpretation seems confirmed by the examples of
provisional statements of principles of inference that Frege provides in
'Compound Thoughts' for instance:
(If B, then A) is true;
Ais true.
It looks then as if the principles of logic take the form of the identifica-
tion of elementary valid inference patterns. These principles may be
conceived as asserting that any statement or thought of some given form
is true. From such a principle and the identification of some particular
thought as having the requisite form, the attribution of truth to that
thought follows. Then by use of an instance of a quasi-disquotational
paradigm like that enshrined in Tarski's Convention T, the f>tatement
itself follows. Thus, from the principles of logic together with the
auxiliary premises just mentioned, follow claims couched in the vocabu-
lary of the special sciences. The universal applicability of logic to the
provision and evaluation of argumentation in any discipline is thus
secured. The principles of logic, so conceived, generalize over the forms
of statements as regards their truth; this character is rendered patent by
the use of schematic letters and a truth predicate in their statement.
Frege's conception of the form of logical laws then appears close kin to
Quine's. 16
This impression, however, is mistaken. For Frege as well as early
Russell, the generality of logic is substantive, not schematic.
? On Frege
and Russell's view, the basic laws of logic generalize over every thing
and every property. These laws do not mention this or that thing; nor do
they generalize over things with respect to various properties or rela-
tions, at least no ordinary property - for example, being mortal - that
belongs to the subject matter of a special science. Any letters that
appear in the statement of logical laws must be understood as variables,
not as schematic letters. Moreover, the quantifiers that appear in these
laws, and in any conclusion drawn with their aid, are understood to be
unrestricted over entities of the appropriate logical type. This view of
logic contrasts sharply with the more contemporary one just canvassed.
That conception depends on a notion of a logical schema subject to
varying interpretations; logical laws are thus applicable to a range of
languages without regard to their (nonlogical) vocabulary or the range
of their quantifiers. But the notion of a logical schema that admits of
multiple interpretations is foreign to Frege's thought.
Nor is it possible, through reasonable emendations, to read the
contemporary view back into Frege.
For the contemporary view
requires the ineliminable use of a truth predicate. Such a use is anti-
thetical to Frege's conception of judgment. This conception of judgment
precludes any serious metalogical perspective and hence anything
properly labeled a semantic theory.
Frege, we have seen, draws our attention to the distinguishing
feat]Jres of judgment by use of the slogan that judgment is the recogni-
tion of the truth of a thought. He denies, however, that such remarks
constitute an informative definition of judgment.!9 If we now take truth
to be a property of thoughts, then we must take the slogan to be offering
just such a definition. So understood, it tells us that judgment is a special
kind of recognitional capacity. We have, for example, the ability, under
suitable circumstances, to identify the colors of things. Judgment now
seems to be the capacity to recognize that some thing (a thought) has
some property (truth). But this is absurd. Judgment is not itself a special
recognitional capacity, a species of some genus. Judgment is itself the
genus - to recognize anything to have a property is ipso facto to make a
judgment.2o Nor can judgment be characterized by comparing it with
our particular recognitional capacities. As these capacities are capacities
to recognize the truth or falsehood of some range of thoughts, they do
not provide an independently understandable model through which
judgment is elucidated. How, though, does taking truth to be a property
of thoughts force on us this confusion of genus and species?
The answer to this question emerges from a consideration of Frege's
curious objection to the correspondence theory of truth?! The corre-
spondence theorist treats truth as a property of thoughts and goes on to
offer a definition of this property in terms of some favored relation
between the contents of judgment and the world. Let us use the phrase
'corresponds with Reality' as a placeholder for whatever definition the
correspondence theorist offers. If truth is a matter of correspondence
with Reality, then in order, for example, to determine whether (it is true
that) Socrates is mortal, it is necessary to inquire whether the thought
that Socrates is mortal corresponds with Reality. But once again, to
determine whether (it is true that) the thought that Socrates is mortal
corresponds with Reality, one should inquire whether the thought that
the thought that Socrates is mortal corresponds with Reality itself
corresponds with Reality. And so on. A parallel regress arises for
judging as well as inquiry.
This brief argument is unsatisfying. It is obvious neither that the
regress it turns on is unavoidable, nor that the regress, even if inevitable,
is vicious. Frege's objection to the correspondence theory is an all too
casual articulation of the fundamental status he assigns to judgment. It is
cogent only in the context of the conception of judgment outlined in the
previous section. Moreover, in that context, the availability of the
objection illuminates the content of this status. Let us then consider the
genesis of the regress more closely.
Why, on the correspondence theory, must one inquire whether the
thought that Socrates is mortal corresponds with Reality in order to
determine whether Socrates is mortal? The connection between judg-
ment and truth is not casual. We saw in the previous section how our
appreciation of the trichotomy of agreement, contradiction, and mere
difference in assertion gives us a conception of standards of correctness
imposed by the contents of judgment on our acts of judging: a thought is
either to be affirmed or to be denied; it is either true or false. We cannot
take someone to be making assertions in complete disregard of the
correctness of what he asserts; such a person would be understood to be
play-acting or perhaps merely mouthing words. To take truth to be
definable forces a particular construal on this talk of standards of
correctness. For to have a definition of truth is to have a general
description of the conditions that have to be satisfied for a judgment to
be correct. So, if truth is definable, then any person who makes a
judgment must have ascertained, or taken himself to have ascertained,
whether these conditions, applied to the thought under consideration,
are satisfied. If I judge that Socrates is mortal, then I must have deter-
mined that the thought that Socrates is mortal corresponds with Reality.
But to ascertain that some condition holds is to make a judgment. So the
regress begins. Given the definition of truth, a person cannot have
judged that the thought that Socrates is mortal corresponds with Reality
unless he has judged that the content of this second judgment corre-
sponds with Reality. On the correspondence theory then, a person is
never in a position to make a judgment; for no one is ever in a position
to have satisfied, or even think of himself as having satisfied, the
standards for judgment that would ipso facto be provided by its
definition of truth.
Frege concludes his argument against the correspondence theory
with the somewhat hesitant statement, "So it seems likely that the
content [Inhalt] of the word 'true' is sui generis and indefinable."22
Frege's talk of indefinability in this context should be understood
differently from his other uses of this notion. Elsewhere Frege talks of
one concept's being definable in terms of other prior concepts. Through
such definitions the grounds for one branch of science may be seen to
lie in some other body of knowledge. Frege's definition of the concept
of number is an example of this sort of definition. Correlative to this talk
of definition, Frege speaks iIi a number of places of primitive concepts,
concepts that admit of no definition in any more basic terms.23 Truth, I
suggest, cannot be understood to be a primitive unanalyzable property
of thoughts; for this position conflicts with the argument just examined.
We observed that the argument does not turn on the details of the
correspondence theorist's definition; indeed, Frege himself generalizes
the argument to any proposed definition. In fact, nothing in the argu
ment turns on taking truth to be definable in terms of any other
concepts. The vitiating regress is generated by the assumption that truth
is a property of thoughts. The proper conclusion to the argument is not
that truth is a primitive property, but that truth is not a property at all.
Frege cautiously voices this conclusion two pages later in 'Thoughts'.
Having observed that the statement 'I smell the scent of violets' has the
same content as the statement 'It is true that I smell the scent of violets.'
Frege goes on to say:
So it seems then that nothing is added to the thought by my ascribing to it the property
of truth. . . . May we not be dealing here with something which cannot be called a
property in the ordinary sense at all? In spite of this doubt I will begin by expressing
myself in accordance with ordinary usage, as if it were a property, until some more
appropriate way of speaking is found.
Returning now to the topic that sparked this discussion of truth, let us
examine how the strong conclusion I draw from Frege's objection to the
correspondence theory motivates his identification of logical principles
with maximally general truths.
It is an upshot of the regress argument that the standards of correct
ness we think of our judgments as satisfying cannot be thought of as a
set of general conditions on the contents of judgment. The conception of
judgment in which this argument is embedded does not permit any real
metaperspective. All of this is not, however, to say that the standards of
correctness are ineffable. In denying that logical principles have a
normative status that distinguishes them from the principles of physics
or geometry, Frege observes, "Any law asserting what is can be con
ceived as prescribing that one ought to think in accordance with it."25
To think in accordance with a law is not just to affirm the law; it is also
to refrain from affirming any thought that contradicts the law. As we
observed in Section I, the other side of this prohibition is the license
that these general laws give us for making further judgments on the basis
of them. Frege thinks that the content of any developed discipline is
constituted by a body of basic laws, laws which are basic in that they
cannot be justified by appeal to other laws couched in the vocabulary of
the discipline that admit of independent justification. These laws thus
present standards for making judgments in that discipline. They consti
tute a framework to which other judgments must conform on pain of
contradiction. Frege's conception of judgment does then admit of
general standards of correctness, but the generality of these standards
does not involve any metaperspective. The general standards for the
judgments of a discipline are not provided by statements about the
discipline. They are provided by judgments within the discipline.
We are now in a position to appreciate the identification of the laws
of logic with maximally general truths. Maximally general truths are
truths that do not mention any particular thing or any particular
property; they are truths whose statement does not require the use of
vocabulary belonging to any special science. Thus, in the same way that
the basic laws of chemistry give standards for judgments in chemistry,
so maximally general truths give us standards for judgment in any
subject matter. No judgment about any individual thing and no judg
ment that generalizes over individuals with respect to a particular
property may contradict a maximally general truth. Frege clearly articu
lates this conception of logic in an unpublished writing:
How must I think in order to reach the goal, truth? We expect logic to give us the
answer to this question, but we do not demand of it that it should go into what is
peculiar to each branch of knowledge and its subject matter. On the contrary, the task
we assign logic is only that of saying what holds with utmost generality for all thinking
whatever its subject .... Consequently we can also say: logic is the science of the most
general laws of truth.
To say that the laws of logic are the most general laws of truth is to say
that they are the most general truths. The laws of physics are the laws of
truth about mass, motion, heat, and light. In mastering these laws we
come to understand physical truth. The laws of logic generalize over
every thing and every property. It is then in studying logic that we learn
whatever there is to be learned about truth simpliciter. As Frege puts
the point, "The meaning [Bedeutung] of the word 'true' is spelled out in
the laws of truth."Z7
If we thus identify the principles of logic as substantive laws in their
own right and so fundamentally similar to the laws of the special
sciences, how are we to understand the applicability of these principle::
in providing gap-free proofs to supplant commonplace enthymatic
reasoning? Frege's basic picture is clear enough. Suppose on the basis of
having judged p, I wish to prove q. In the simplest case, I might begin by
finding a logical truth whose variables may be instantiated by constants
in such a way as to yield the conditional whose antecedent is p and
whose consequent is q. The proof then begins with an assertion of this
law. On the basis of this assertion, the desired conditional is obtained.
Then p is asserted. On the basis of the assertion of the conditional and
the assertion of p, q is asserted. And characteristically, the logical law
with which the proof begins will be inferred from other simpler, more
perspicuous general truths. There is, as far as Frege is concerned,
nothing to be said about the justification for our recognition of those
basic laws of logic to be truths:
The question why and with what right we acknowledge a law of logic to be true, logic
can answer only by reducing it to another law of logic. Where that is not possible, logic
can give no answer.
Moreover, the maximal generality of these laws precludes their infer-
ence on the basis of the truths of any other discipline.
The universal applicability of the laws of logic is then secured by
their generality. This conception of generality is original to Frege and is
the product of his construal of the quantifier-variable notation of his
begriffsschrift. As traditionally conceived, logic is concerned with the
form rather than the content of judgment. A sharp distinction between
the form and content of judgment was to explicate the sense in which
logic abstracted from the content of the claims of the special sciences.
So, for example, Kant remarks:
That logic should have been thus successful is an advantage which it owes entirely to its
limitations, whereby it is justified in abstracting ... from all objects of knowledge and
their differences, leaving the understanding nothing to deal with save itself and its
Such talk of form insinuates that logic is somehow concerned with
judging itself, and this suggestion in turn can serve as the thin edge of
the wedge of psychologism. The conception of generality that Frege's
quantifier-variable notation makes available transforms the old distinc-
tion between form and content. The claims logic makes still "abstract"
from the differences that distinguish the content of the claims of the
various sciences, but this abstraction does not give logic a special
subject matter of its own - the forms of judgment. Nor does this
rob the claims of logic of content, as though they concerned
merely the "empty forms" of judgment. Frege's variables enable him to
take his analogues of the laws of traditional logic to be making substan-
tive claims in their own right, claims that are, nevertheless, applicable to
reasoning in the special sciences on account of the relation between
generalizations and their instances.
So far, the principles of logic have been identified with the basic laws
of logic. These laws get applied in the provision of proofs only through
our discerning implicational relations between logical laws and their
instances, and among the claims of the special sciences. Or to speak in a
more Fregean vein, the application of the laws of logic in proofs
depends on our making one assertion on the basis of another. If these
proofs are to be taken to be gap-free with all the premises necessary for
the proof of the conclusion explicity stated, then we will need to have
isolated a small number of principles of inference, asserting one thing on
the basis of another only when the conditions laid down in these prin-
ciples are satisfied. These principles of inference are then as necessary
as the basic laws of logic for the replacement of everyday argumentation
with proofs. Indeed, seeing how the basic laws of logic are statements
like any other that get used in proofs by our asserting further claims on
the basis of them, it appears that the inference rules rather than the
logical laws more properly deserve the title "principles of logic". For it is
the inference rules that deal with what is the defining concern of logic -
the assertion of one thought on the basis of another.
Frege's statement of rules of inference follows the pattern exhibited
by the following formulation of modus ponens:
From a conditional and the antecedent of that conditional, the consequent of th
conditional may be inferred.
What licences this permission ? We recognize the inference rule to be
valid; and to talk of recognizing the validity of an inference appears to
making a judgment. Our question then is: what thought do we affirm
when we recognize this inference rule to be valid and what basis, if any,
is there for this judgment?
It is tempting, at first blush, to identify the apprehension of the
validity of the inference rule with the recognition of the truth of the
corresponding logical law. In the case of MP this law reads something
If P and also if p then q, then q.3!
However, this identification obviates the distinction between logical laws
and rules of inference. Frege is clear on the difference here between
asserting a conditional and asserting one thing on the basis of a previous
Indeed, this difference is another of the central linguistic
precedents lying behind Frege's distinguishing the content of judgment
from the act of judging. If I assert q on the basis of having asserted p and
you deny q, you contradict me. You do not contradict me, if you deny q
in the face of my assertion of the conditional 'If p, then q'. Moreover,
were use of an inference rule to be justified by the judgment of a general
law, we would encounter the vicious regress in the provision of proofs
that Lewis Carroll pointed out. For then, in order to make a proof
complete, any use of an inference rule would have to be accompanied
by an assertion of a corresponding logical law. Only in this way would
all the premises on whose correctness the conclusion depends be
explicitly stated. But this added statement creates the need for further
inferences, each of which would need to be similarly accompanied by
assertion of justifying laws. This regress would make completed proofs
At this point, recourse to a metaperspective seems inescapable. On
the contemporary conception of logic, the acceptance of MP as a
correct rule of inference is vouchsafed by our metalogical judgment that
if a conditional is true and its antecedent is true, then so is the
consequent. This judgment is not supposed to appear as a premise in
every proof employing MP; it rather is a part of our reason for accepting
derivations in some given formal system as proofs. But a statement of
the validity of MP unavoidably involves taking truth to be flatly a
property of thoughts. This treatment of truth is precluded by Frege's
conception of judgment. The proper conclusion for him is that our
apprehension of the validity of MP is not a judgment. It is not mani
fested in any single assertion and so is, in this important sense, ineffable.
TID" apprehension is, however, manifested linguistically in the inference
we make and accept.
Once truth is excluded as a property, Frege has no nonsyntactical
metalogical vocabulary. It is then misleading to have talked earlier of
recognizing a thought to be contradictory, or of recognizing one thought
to imply another. The recognition of a thought to be contradictory is not
a judgment to the effect that a thought has some property. It is rather an
apprehension manifested by the refusal to affirm the thought while
affirming the opposite. Similarly to recognize that one thought implies
another is to be prepared to accept the second on the basis of the first.
Again, there is no judgment that gives expression to this notion of on the
basis of, implication is no more a relation between thoughts than truth is
a property of them.
We have seen how the attempt to make our apprehension of the
validity of an inference rule over into a judgment leads to the use of the
truth predicate that the regress argument excludes. This flawed attempt
at a statement is the best that can be done; and its failure shows why the
essence of inference, of logic, is ineffable. In an unpublished jotting
entitled "My basic logical Insights" Frege notes the redundancy of
attributions of truth and then says:
But it is precisely for this reason that this word ['true') seems fitted to indicate the
essence of logic ... it allows what corresponds to the assertoric force to assume the
form of a contribution to the thought. And although this attempt miscarries, or rather
through the fact that it miscarries, it indicates what is characteristic of logic. . . what
logic is really concerned with is not carried in the word 'true' at all but in the assertoric
force with which a sentence is uttered.
It remains to examine how ontological categories are supervenient on
logical ones for Frege. The treatment of this topic provides the oppor-
tunity to apply, albeit in an incomplete and only partially defended
manner, the interpretive perspective of the previous two sections to
Frege's central and celebrated distinctions and doctrines.
We noted at the beginning of Section II Frege's demand that proofs
be gap-free. Having made this demand, Frege goes on to observe two
obstacles to satisfying it. First, there is the comparatively minor barrier
posed by the sheer tedium of advancing in argument step by step in
conformity with a small number of antecedently marked out inference
patterns. Second and much more important, there is, as regards every-
day language, the difficulty of isolating out some small number of logical
laws and inferences sufficient for Frege's logicist goals. Frege complains,
"the excessive variety of logical forms that has gone into the shaping of
our language makes it difficult to isolate a set of modes of inference
which is both sufficient to cope with all cases and easy to take in at a
glance."34 He continues:
To minimize these drawbacks, I invented [erdenken] my Begriffsschrift. It is designed to
produce expressions which are shorter and easier to take in, and to be operated like a
calculus by means of a small number of standard moves.
Frege's function-argument analysis of language with its associated talk of
the truth-value of a statement being determined by the meanings of its
parts must be understood against the backdrop of this project.
How then can Frege proceed - or think of himself after the fact as
having proceeded - with the construction of a logically perspicuous
language? We observed that the starting point for Frege's conception of
objectivity is our appreciation, at least in straightforward cases, that one
statement may be asserted on the basis of the assertion of another
statement. This datum was presented at the level of the apprehension of
individual inferences. Reflection on these individually correct inferences
leads the logician to the isolation of inference patterns, patterns like
those that are eventually set forth in the inference rules. Frege is
describing the logician's reflection when he says:
Kerry holds that no logical rules can be based on linguistic distinctions; but my own way
of doing this is something that nobody can avoid who lays down such rules at all. For we
cannot come to an understanding with one another apart from language, and so in the
end we must always rely on other people's understanding words, inflexions, and
sentence-constructions in essentially the same way as ourselves ... to this end I
appealed to the general feeling for the German language.
The apprehension of inference patterns involves thinking of statements
as logically segmented into significant parts. Consider, for example, the
inference pattern set forth in Leibniz's law. Inferences that instance this
pattern lead from the assertion of a statement containing occurrences of
a term t and the assertion of an equation 'i = S', to the assertion of the
result of replacing occurences of t in the original statement by occur-
rences of s. The appreciation of the validity of this inference pattern
goes hand in hand with seeing statements to be segmented into proper
names and predicates, a predicate being the result of removing one or
more occurrences of a proper name from a statement. Of course, our
grasp of these two categories is not exhausted by Leibniz's law; reflec-
tion on inferences concerning generality plays a crucial role in imposing
this segmentation into proper names and predicates. In the end, it is not
the apprehension of a single inference pattern, but of an interlocking
series of basic inference patterns, that constitutes the recognition of
parts and features of statements as logically functioning units that
belong to various logico-syntactic categories.
This observation is indeed the point of Frege's enigmatic Context
Principle that bids us "never to ask for the meaning of a word in isola-
tion but only in the context of a statement [Satzzusammenhang)."37 It is
ultimately our appreciation that one statement may be asserted on the
basis of another and the ensuing apprehension of inference patterns that
gives us the idea of words having a meaning.
To ask after the meaning
of a word is then, first of all, to ask after its logical category; such
inquiry is answered by reflection on the implicational relations of state-
ments in which the word occurs. Thus it is that, in the body of The
Foundations of Arithmetic, Frege recurs to the Context Principle at the
conclusion of a discussion of statements of the form 'There are n P, a
discussion that argues in effect that the numeral must be reckoned as a
proper name.
By attention to selected simple inference patterns within everyday
language, Frege discerns amidst the ill-defined welter of inferences
found there, the logical segmentation that provides the scaffolding for
pruning and reforming everyday language into a Begriffsschrift. The
segmentation of statements into proper names and predicates is a
crucial aspect of this enterprise. This segmentation enables Frege to
explain the (first-level) quantifier as a second-level predicate - an
expression which, when completed by a first-level predicate, yields a
statement. Exploiting this insight, Frege eliminates the grammatically
chaotic and ambiguous ways everyday language has of expressing
generality in favor of the perspicuity of the quantifier-variable notation.
The way is then prepared for Frege's greatest achievement - the
complete and accurate depiction of the patterns of polyadic quantifica-
tional inference.
To think of statements as logically segmented is to think of the
expressions thus segregated as logical units, as playing a uniform role in
the determination of the truth or falsehood of the statements in which
logical analysis discovers them. Of course, it is not the expressions
themselves - the bare marks and noises - that so contribute. Frege is
always thinking of statements and their component expressions as
significant. Talk about the meaning of expressions of various categories
in terms of the contribution they make to the truth-value of statements
points toward the source of logical segmentation in inference patterns.
Such talk is of a piece with the attempt mentioned at the end of Section
II to state the warrant for making inferences in accordance with syn-
tactically stated rules. It tries to say what cannot be said but only shown
in the inferences we make and accept. However, such gesturing is
Frege's indispensible means for instructing us in the use of the
Begriffsschrift. By talking about truth-value determination, he obtains
our affirmation of the basic laws of his formal system, our apprehension
of the validity of his inference rules, and, generally, an understanding of
the notation that extends to those Begriffsschrift formulas for which
there are no readily intelligible analogues in everday language.
In setting forth his logical notation, Frege talks of proper names as
meaning objects and predicates as meaning concepts. Against the
backdrop of the connection just surveyed between logical segmentation
and significance, the phrase 'means an object' serves to label that feature
proper names share by which they all make the same sort of contribu-
tion of the truth-value of statements. So, as regards proper names by
themselves, ,we could use the less opaque phrase 'significant proper
name' instead. Similar remarks hold for predicates. The dichotomy
between objects and concepts comes into its own not as regards proper
names and first-level predicates but rather in introducing and distin-
guishing first-level generality (over objects) from second-level generality
(over concepts). This logical difference is masked in everyday language,
including the language of mathematics.
We observed that Frege's first-level quantifier is a second-level
predicate. The meaning of this quantifier may be specified as follows.
Completion of the empty position in the quantifier by a predicate yields
a truth just in case every object falls under the concept meant by the
predicate. In this elucidatory context, unlike those that concern just
proper names ,and predicates, 'object' is used outside the phrase 'means
an object'. Moreover, the phrase 'every object' is being used in contrast
with 'every concept' to call attention to a second sort of generality,
generality over concepts. Frege's quantifier over concepts is a third-level
predicate that stands to second-level predicates as the qUantifier over
objects stands to first-level predicates. However, the use of 'object' and
'concept' to contrast these two levels of generality can seriously mislead.
The phrases suggest the availability of a quantifier over entities of which
objects and concep\s are genera; for, in Frege's elucidating remarks, the
words 'object' and 'concept' alike appear to be first-level predicates.
Frege's conception of generality precludes any such quantifier, any such
notion of entity.
This last point is worth some elaboration, as it is essential for under-
standing how the basic laws of logic abstract from the particular content
of the statements of the special sciences. This abstraction is glossed in
terms of the relation between generalizations and their instances. A
variable may be substituted for any significant expression in a statement,
a transformation that yields a statement that speaks generally where the
given one spoke specifically. The mark of this generality lies in the
replaceability of the variable in its turn by any expression of the same
category as the original expression to obtain an instance of the general-
ization. So to each logical category of expression there corresponds a
type of variable.
Suppose ,Jlow that we attempt to get by with one level of generality.
Even restricting attention to simple statements from which generality is
absent, we observe that the replacement of the grammatical subject by a
grammatical predicate produces nonsense. This fact presumably indi-
cates that the grammatical predicate is combining two distinct roles - .
the presentation of a concept and the expression of the copula. The
understanding of the relationship between generalizations and their
instances, variables and constants, just surveyed requires that these two
roles be separately discharged in a logically perspicuous notation. To
this end, an expression, 'falls under', might be introduced to express the
copula. So, we obtain as renderings of everyday statements, statements
Socrates falls under the concept of being wise,
If x falls under the concept of being a philosopher, then x falls under the concept of
being wise.
Generalizing on the direct object position of our new expression as well
as its subject position, we can use it in the obvious way to state logical
However, our new relational predicate is an expression of a distinct
logical category; after all, no statement results from replacing it with a
proper name, with an expression that may occur in the positions on
which we are generalizing. In thus admitting an expression of a distinct
logical category, we ipso facto allow for second-level generality and with
it, further logical laws. But once second-level generality is acknowl-
edged, there is no call to treat predicates as playing the same logical role
as proper names and so forming with them a single logical category. The
acknowledgment of the distinct logical category of 'falls under' might be
resisted by maintaining that this relational predicate suffers from the
same duality of logical role it was introduced to correct - it both
presents a relation and is a copula. It should now be clear that there is
no way of disentangling these roles without reintroducing an expression
in which they are once again combined. The conclusion here is that
there are not two distinct roles played by predicates; or, as Frege says,
the concept is predicative, unsaturated.
The distinction of levels of generality is of the highest importance to
Frege. The first fruits of this distinction is his logical definition of the
ancestral of a relation. This definition enables the principle of mathe-
matical induction to be deduced from logical laws. Thus, the distinction
between generality over objects and generality over concepts is the
immediate source of the centerpiece of Frege's logicism: the reduction
of "the argument from n to (n + 1), which on the face of it is peculiar to
mathematics, to the general laws of logic."42 Small wonder then that the
third guiding principle of The Foundations of Arithmetic advises us
"never to lose sight of the distinction between concept and object."43
We are now in a position to understand how ontological categories
are, for Frege, supervenient on logical ones. The logico-syntactic source
of the notion of an object lies in first-level generality. To be an object is
to be indefinitely indicated by first-level variables. Our grasp of the
notion of an object - simply the notion of an object, not an object of
this or that kind - is exhausted by the apprehension of inference
patterns and the recognition of the truth of the basic logical laws in
which these variables figure.
We encounter at this juncture the central
elucidatory use for the phrase 'means an object' (bedeuten). This phrase
is used in contrast with the phrase 'indefinitely indicates an object'
(andeuten) to call attention to the inferential difference between state-
ments where first-level generality is present and where it is absent, and
so to distinguish first-level variables from proper names.
Similar remarks hold for the notion of a concept. The logico-syntactic
source of this notion lies in our apprehending basic inference patterns
turning on second-level variables. It is worth observing that for con-
cepts, or rather functions - for Frege this is the prior and more general
notion - there is an expression of this attitude at the outset of Frege's-
paper 'Function and Concept.' Frege there notes that mathematicians
had no use for the general notion of a function so long as they were
occupied with establishing facts about particular functions. The need for
the general notion of a function arose only when mathematical attention
turned to abstract generalizations about functions, for instance that any
function everywhere differentiable is everywhere continuous, but not
vice versa. Frege says:
The first place where a scientific expression appears with a clear-cut meaning is where it
is required for the statement of a law. This case arose as regards the function upon the
discovery of higher Analysis. Here for the first time it was a matter of setting forth laws
holding for functions in general.
And the introduction of the general notion of a function is signaled not
so much by the use of the word 'function' as by the appearance of
variables over functions. From the Fregean vantage point, the unclarity
surrounding the notion of a function in mid-nineteenth century analysis
was the product of an inadequate understanding of variables, including a
failure to distinguish sharply first-level from second-level variables,
objects from functions.
We have seen how Frege's talk of meaning serves to instruct us in the
use of the Begriffsschrift. One aspect of this talk calls for further exami-
nation - Frege's notorious doctrine that statements mean truth-values
in the same sense in which, say, 'Venus' means a planet. The position
appears outlandish. Frege seems to begin with the familiar and pre-
theoretical paradigm of the relation between an ordinary proper name
and the person, animal, edifice, etc. that bears it. The suspicion is that,
however motivated Frege's other extensions of the name-bearer relation
are, the analogy snaps when applied to the relation between statements
and truth-values.
I suggest that this line of objection to Frege's controversial thesis both
misevaluates its significance and gets the connection between Frege's
talk of meaning and the name-bearer relation backwards. We saw in
Section I how the conception of judgment that emerges from the
contrast between the objective and the subjective includes the idea of
standards of correctness. But correctness or truth cannot be thought of
as a property of the contents of judgments. How then are we to think of
it? In 'Thoughts,' Frege says that the meaning of the word 'true' is sui
In 'On Sense and Meaning,' Frege denies that the relation of
a thought to truth is the relation of a subject to a predicate and talks
instead of a difference in level between thoughts and truth-values. He
says, "Judgments can be regarded as advances from a thought to a truth
value."47 and, "We are therefore driven into accepting the truth value of
a sentence as constituting its meaning .... These two objects [the True
and the False] are recognized, if only implicitly, by everybody who
judges something to be true ... "48 Frege's talk of a difference of level
between thoughts and truth-values and of judgment as a movement from
one level to the other attempts to introduce a way of thinking about
truth in contrast to taking truth to be a property of thoughts. The
notorious doctrine should then be undet'stood as an elaboration of this
contrast; it is the more appropriate way of speaking to which Frege
alludes in 'Thoughts'.49
Nor should we take Frege's thesis to draw on our familiarity with the
name-bearer relation for its motivation. If the preceding interpretation
of the source of Frege's notion of meaning in the apprehension of
inference patterns is correct, the relation between a statement and its
truth-value is the paradigm for the meaning relation, a paradigm in
terms of which the familiar name-bearer relation is re-interpreted. Frege
acknowledges the generalization
If x = y, then Fx if and only if Fy,
to be a logical law. The inferential links between first-level variables and
proper names together with the foregoing conception of meaning practi-
cally force the identification of a colloquial proper name's having a
meaning with its having a specific bearer.
Thus understood, truth is not, properly speaking, ineffable. To treat
statements as proper names, and hence truth-values as objects, renders
the True and the False as effable as any object. The ineffability that
lurks in Frege's discussions of truth should attach instead, I suggest, to
the meaning relation thought of as pairing linguistic expressions with the
items meant by those expressions. After all, the availability of such a
relation would give us the means for converting Frege's remarks about
truth-value determination into a genuine theory containing the resources
for defining a concept of correctness or truth. It is through his talk of the
meaning relation that Frege speaks of truth-value determination for the
purpose of instructing us in the Eegriffsschrift. Once this end is achieved,
talk of meaning is to drop away; it plays no further role in logic.
There is a final topic that requires at least brief consideration -
Frege's distinction between sense and meaning. In Section I it was noted
that some of the ways Frege has of talking about thoughts lend credence
to the platonist reading - the ontological status of thoughts as atem-
poral abstract objects guarantees the objectivity of judgment. It was
claimed, however, that Frege's vocabulary for talking about assertion
and judgment - particularly the notion of thought - was introduced in
order to redescribe systematically those features of our linguistic practice
that fund Frege's conception of logic. It is through Frege's identification
of thoughts with the senses of sentences that thoughts obtain the status
of objects within Frege's system. The immediate source of Frege's
notion of sense lies in the application of his Begriffsschrift to the logical
analysis of the observations about judgment, belief, and assertion that
are behind his conception of objectivity. There is, after all, no way to
direct attention to the distinguishing features of judgment without using
locutions such as those of the form 'x asserts that p'. These observations,
couched as they are in familiar vocabulary, are clear cases of significant
statements; they are also examples of statements that contain the subor-
dinate clauses of indirect discourse. Frege takes the familiar apparent
failures of Leibniz's law here - including his extension of it to materially
equivalent statements - to show that the statements and expressions of
indirect discourse do not have the same significance they have in other
settings. The sense/meaning distinction provides a delineation of this
pervasive ambiguity that logical analysis uncovers. The details of Frege's
about sense are largely the result of his reflections on the char-
acter of truth-preserving substitutions within indirect discourse.
Frege has other uses for his notion of sense. But it is the foundational
role that observations about assertion and judgment play in Frege's
philosophy that suits the notion of sense for those uses. Of course, it is
debatable whether the notion of sense, introduced in the manner just
sketched, can discharge these other roles; and even Frege's conclusions
about truth-preserving substitutions within indirect discourse are con-
troversial. The serious problems surrounding Frege's notion of sense lie
beyond the scope of this paper.
My present point concerns the status of the senses of proper names,
including sentences, as objects. Frege's identification of thoughts as
objects is owed entirely to the logical segmentation he discerns within
indirect discourse. He takes sentences in indirect discourse to be
logically functioning units, proper names. Thus it is that Frege's concep-
tion of the objectivity of judgment and his associated view of logical
segmentation and meaning eventuate in me claim that thoughts are
objects. We find then talk of thoughts occurring in two different places
in Frege's philosophy, places he does not clearly demarcate in his writ-
ings. Talk of thoughts occurs at the very beginnings of his philosophy,
introduced to redescribe the central features of our practice of assertion.
It also occurs subsequently, once logic is in place, in discussing the
logical segmentation of indirect discourse. Understood in this way, the
objecthood of thoughts, far from explaining the objectivity of judgment,
presupposes it.
University of Pennsylvania
* The ideas in this essay reflect many conversations about Frege with Burton Dreben. I
am deeply and pervasively indebted to Warren Goldfarb, both for instruction received
from his 1977 lectures on Frege and for countless ensuing discussions. My interpreta-
tion of Frege's anti-psychologism has been influenced by an unpublished paper of Susan
Neiman. I have benefited from conversations with Michael Friedman, Peter Hylton, and
Joan Weiner. Prof. Weiner, in her dissertation Putting Frege in Perspective (Harvard
University, 1982), arrives at an understanding of the point of Frege's talk about
Bedeutung similiar to that presented in Section III.
1 Gottlob Frege, The Foundations of Arithmetic, translated by J. L. Austin (Evanston:
Northwestern University Press, 1978), p. x.
2 A discussion of Frege's philosophical antecedents that convincingly portrays Frege's
work as a kind of rationalist reaction to a naturalistic empiricism is given by Hans D.
Sluga in Gottlob Frege (London: Routledge & Kegan Paul, 1980), chapters one and two.
I am indebted to Peter Hylton for awakening me to the affinity between post-Kantian
Idealism and early Analytic philosophy that lies behind the obvious and important
3 Frege, 'Boole's Logical Calculus and the Concept-Script; in Posthumous Writings,
edited by Hans Hermes et al. (Chicago: The University of Chicago Press, 1979), pp.
4 Frege, The Basic Laws of Arithmetic: Exposition of the System, translated and edited
by Montgomery Furth (Berkeley and Los Angeles: University of California Press,
1967), p. 14.
5 Basic Laws, p. 13. See also Foundations, p. vi.
6 Basic Laws, p. 14 and p.l5.
7 Basic Laws, p. 17.
8 For instance, see Frege, 'On Sense and Meaning' in Translations from the Philo
sophical Writings of Gottlob Frege, edited and translated by Peter Geach and Max Black
(Oxford: Basil Blackwell, 1952), p. 61; see also Frege, 'Thoughts' in Logicallnvestiga
tions edited by Peter Geach; translated by Peter Geach and R. H. Stoothoff (New
Haven: Yale University Press, 1977), pp. 8-9.
9 Posthumous Writings, p. 142.
10 'Thoughts,' p. 24; but see also the material surrounding the quoted passage,
especially Frege's footnote about grasping.
11 Frege, 'Negation' in Logical Investigations, p. 31. See also Posthumous Writings, p. 7.
12 Posthumous Writings, p. 175; see also p. 3. See also Foundations, pp. 3-4, especially
the last paragraph of section three.
13 Foundations, p. 102.
14 'Thoughts,' p. 1. Similar remarks are scattered through Posthumous Writings. For
example, see p. 2, p. 126, p. 128, p. 253.
IS Frege, 'Compound Thoughts,' in Logical Investigations, p. 71.
16 See W. V. Quine, Word and Object (Cambridge: MIT Press, 1960), sec. 56,
especially p. 273; see also Quine, Philosophy of Logic (Englewood Cliffs: Prentice-Hall,
1970),pp.l0-13 and pp. 47-50.
17 The point is established in Warren D. Goldfarb, 'Logic in the Twenties,' The Journal
of Symbolic Logic 44 (1979), especially pp. 341-44 and in Jean van Heijenoort, 'Logic
as Calculus and Logic as Language,' Synthese 17 (1967), especially pp. 325-27.
18 This, I take it, is Michael Dummett's strategy in his masterful interpretation of Frege.
At this juncture then, we encounter a consequence of the difference between my
understanding of the priority Frege attaches to judgment and Dummett's. 1 sketch in this
essay an alternative to Dummett's interpretation of Frege; I have not, either at the level
of exegesis or of argument, defended my interpretation against Dummett's.
19 'On Sense and Meaning,' p. 65.
20 This observation, I take it, is behind Frege's remarking that to recognize anything to
have any property carries with it, after a fashion, a predication of truth. See 'Thoughts,'
p. 5 and Posthumous Writings, p.129.
21 'Thoughts,' p. 4. See also Posthumous Writings, pp. 128-29.
22 'Thoughts,' p. 4.
23 See Frege's letter to Hilbert of Dec. 27, 1899, letter IV/3 in Frege, Philosophical and
Mathematical Co"espondence, edited by Gottfried Gabriel et al., abridged from the
German edition by Brian McGuinness and translated by Hans Kaal (Chicago: The
University of Chicago Press), pp. 36-37. See also Frege, 'On Concept and Object,' in
Translations, pp. 42-43.
24 'Thoughts,' p. 6. See Posthumous Writings, p. 251; see also 'On Sense and Meaning,'
25 Basic Laws, p. 12.
26 Posthumous Writings, p. 128.
27 'Thoughts,' p. 2.
28 Basic Laws, p. 15.
29 Immanuel Kant, Critique of Pure Reason, translated by Norman Kemp Smith (New
York: St. Martin's Press, 1929), p. 18 (B ix); see also p. 98 (A 60 = B 85).
30 See Basic Laws Section 48, pp. 105-09 for Frege's formulation 01 the inference
rules for the formal system of Basic Laws. Rule 6 on p. 106 is Frege's version of MP.
31 Frege, thanks to his horizontal function that associates every object with a truth-
value, in the formal system of Basic Laws formulates truth-functional laws using
unrestricted first-level variables.
32 See 'Negation,' p. 34.
33 Posthumous Writings, p. 252.
34 Foundations, p. 103. Frege makes the same point a bit more sharply in 'On Herr
Peano's Begriffsschrift and My Own,' translated by V. H. Dudman Australasian Journal
of Philosophy 47 (1979), p. 2.
35 Until the fmal paragraphs of this section, Frege's distinction between sense and
meaning will be put to one side; and the term 'statement' will be used with the same
ambiguity that Frege uses 'Satz' in his pre-1891 writings. I also follow the growing
practice of using 'meaning' and its cognates as the English equivalents of Frege's uses of
'Bedeutung' and its cognates.
36 'On Concept and Object,' p. 45.
37 Foundations, p. x. Austin translates 'Satz' by 'proposition'.
38 Frege can be read as marking this point in his letter to Peano of Sept. 29, 1896, letter
XlV /7 , in Correspondence, p. 115.
39 See Foundations, p. 72 and p. 73.
40 See 'On Concept and Object,' p. 54.
41 'On Concept and Object,' p. 43. The argument of the last two paragraphs is in
essence the one Frege gives in the penultimate paragraph of this paper; see pp. 54-55.
42 Foundations, p. 93.
43 Foundations, p. x. I am indebted in this paragraph to Warren Goldbarb. The section
as a whole bears the imprint of Burton Dreben's repeated insistence on the role of
Frege's mathematical training and interests in shaping his philosophy.
44 These remarks about Frege's conception of objecthood should be distinguished from
superficially similar things Quine says. Quine puts forward his maxim "To be is to be a
value of a variable," in the context of a discussion of ontological commitment. Applica-
tion of the standard of ontological commitment it provides is made in a metalanguage
into which we may have occasion to semantically ascend in the course of certain
investigations and disputes. For present purposes, the important point is that the notion
of ontological commitment, and with it this maxim, belong to the theory of reference
and thus invoke the semantical perspective Frege eschews.
45 Frege, 'Function and Concept,' in Translation, p. 21. See also p. 41.
46 'Thoughts,' p. 4 and p. 6.
47 'On Sense and Meaning,' p. 65.
48 'On Sense and Meaning,' p. 63.
49 See 'Thoughts,' p. 6; and see above pp. 22-23.
From a natural perspective, Frege's view that sentences denote
(bedeuten) objects appears to be an irritating peculiarity. His claim that
there are only two objects denoted by sentences and that these are
Truth and Falsity has seemed to many to advance from the peculiar to
the bizarre. Indeed, a standardized form of philosophical humor has
grown up around talk of "naming the True". I think that the natural
perspective is sound and that the humor has its point. But understanding
Frege's motivations for these views provides insight into the fundamen-
tals of his philosophical standpoint and method. Such insight enriches
the natural perspective.
The importance of Frege's views on truth values in his system has
been appreciated by a number of philosophers. Michael Dummett
characterizes Frege's claim that sentences denote objects as "an almost
unmitigated disaster" for Frege's later philosophy of language (FPL,
196,643-4).1 Several authors have seen in Frege's writings the skeleton
of an a priori argument, later given by Church and Godel, that sentences
must denote only Truth or Falsity. And Frege's method of identifying
the truth values with certain courses of values has been construed as
indicating a non-realistic attitude toward numbers. I think that each of
these interpretations is mistaken. But they correctly suggest that Frege's
odd-sounding conclusions about truth and falsity should be taken seri-
ously as a key to his philosophies of language, logic and mathematics.
My aims here are historical. I shall argue in Section I that Frege's
view that sentences denote only truth or falsity has profound and
natural motivations, and that his view that truth values are objects is
more pragmatically based - and therefore less strange - than has
usually been thought. In Section II I criticize Dummett's influential
interpretation of Frege's theses on truth values and his evaluation of the
effect of those theses on Frege's philosophy of language. I also delineate
the development of Frege's views on assertion and truth between
Begriffsschrift and Basic Laws. In Section III I argue that Frege's identifi-
cation of the truth values with the particular objects he identifies them
with undergirds his realism about logical objects, and proceeds from
L. Haaparanta and f. Hintikka (eds.), Frege Synthesized, 97 -154.
1986 by Tyler Burge.
some of his deepest philosophical conceptions. In particular, it proceeds
from a theory about the nature of logical objects, from a thesis about the
aim and ordering of logic, and from his conceptions of assertion and
In order to lay the groundwork for our discussion of Frege's concep-
tions of assertion, truth, and logical objects, I will have to go over a fair
amount of familiar ground in Section I. Some readers may wish to work
through this section quickly in order to concentrate on Sections II and
III. I should caution, however, that although many of the doctrines
discussed in Section I are well-known, the ways they fit together and the
means Frege uses to motivate them are less well recognized. Under-
standing these ways and means is critical to a proper appreciation of
Frege's use of the notion of truth in his philosophy of logic and mathe-
matics - and indeed, to an appreciation of his depth as a philosopher.
Although defective in various ways, Frege's views on truth are richer
and more central to his logical theory and much of his philosophy of
mathematics than is often realized. One reason why these views are
underappreciated is that Frege refused to allow a meta-theoretic seman-
tics, as we know it, to be part of his logical theory. Another reason is
that Frege's presentation of his views has tended to encourage concen-
tration on his philosophy of language or his mathematical work as some-
what separate enterprises. The philosophy of language is expounded
largely in the great articles of the 1890's and in unpublished writing,
with little discussion of its connection to logicism. The mathematical
project is spun out in The Basic Laws of Arithmetic, which is cast in the
form of a traditional mathematical treatise - its philosophy kept to a
minimum. Underlying Frege's work is, however, a remarkably integrated
vision. We shall try to layout the central place that Frege's views on
truth have in this vision.
It is useful to separate Frege's views on truth values into several theses,
although the theses are interrelated and his arguments for them overlap.
The relevant theses are
(a) Sentences (when not defective) have denotations (Bedeu
(b) The denotation of a sentence is its truth value.
(c) Sentences are of the same logical type as singular terms.
(d) The denotation of a sentence is an object.
Frege tends to develop support for the theses in the order in which they
are listed. (See Note 6 below for a qualification.)
Frege's arguments often presuppose his distinction between sense
and denotation (which he first draws for singular terms). They almost
always presuppose or make use of his groundbreaking composition
(1) The denotation of a complex expression is functionally dependent only on the
denotations of its logically relevant component expressions.
(2) The sense of a complex expression is functionally dependent only on the senses
of its logically relevant component expressions.
(I omit certain qualifications on these principles that are irrelevant to
our concerns.) The first principle is the critical one in Frege's thinking;
the second makes important but only occasional appearances.
Thesis (a)
Frege argues that the sense of a sentence - its cognitive value, the
thought that it expresses - remains the same regardless of whether or
not the sentence's component expressions (particularly, the singular
terms) have denotations: The sense of a sentence is fixed independently
of its components' denotations [G & B, 'S & R', 63/KS 148; Cor 165/
BW 247; PPW 193-4/NS 210.] It follows from (1) and (2) and these
considerations that the sense of a sentence cannot be conceived as its
Frege further argues that one cannot reasonably hold that sentences
in general lack a denotation. In at least one passage he draws this
conclusion almost directly from the arguments of the preceding para-
It follows that there must be something associated with a sentence that is different from
the thought, something for which it is essential whether the parts of the sentence have
denotations. This is to be called the denotation of the sentence. (PPW 194/NS 210-11)
This inference clearly relies on the Composition Principle (1).
Now one might feel that the inference begs any question one might
have about whether sentences have denotations - about Thesis (a). Why
should sentences be included among the complex expressions that have
denotations? The last sentence of the passage just cited suggests some-
thing wrong with the question. Frege is not using the term 'denotation'
with a fixed, meaning in arguing for (a). Rather he is
determined to give his Composition Principle (1) a comprehensive role
in logical theory, and he is intending to fit the term 'denotation' to the
role that the principle might fruitfully play in a logical theory about
sentences. "The denotation" of a sentence is whatever is most fruitfully
seen as functionally dependent on the denotations of its parts. So far the
phrase 'the denotation' has no specific logical grammar or ontological
implications. Since the arguments for (a) do not presuppose Thesis (d),
there is so far no reason to consider the view that a sentence's denota-
tion is an object. One may at this point regard talk about sentence
denotation as potentially a de parler for an important semantical
aspect of sentences. The ontological import of such talk, if any, is left
thoroughly open.
Of course, the term 'denotation' (,Bedeutung') was not devoid of
intuitive content in Frege's arguments. 'Bedeutung' is a common word in
German, usually translated 'meaning'. In German there is no oddity in
saying that sentences have a "Bedeutung". Frege did, however, appro-
priate the term for his theoretical uses and introduced it in the essays
'Function and Concept' and 'On Sense and Denotation' through exam-
ples of singular terms ('The Evening Star', 'Odysseus' - which lacks a
denotation - '1', '2 + 2', 'the capital of England'). The examples suggest
that naming or reference - considered as relations between names and
their bearers or between a complex singular term and the object it picks
out - is one primary sort of "Bedeutung". But since Frege also used his
term to apply to a semantical relation between expressions (such as
predicates), that he emphatically did not regard as singular terms, on
one hand, and non-linguistic entities, on the other, one must view these
initial examples with some caution. They are aids in building a theory.
The point I want to press regarding Frege's quick inference to (a)
from his composition principle is that the inference is indicative of his
pragmatic attitude toward his terminology. As he repeatedly noted, the
term 'number' had expanded in its application (from the natural num-
bers, to negatives, rationals, reals,' complex numbers) under pressure
from the requirements of mathematics. Semantical terminology could be
expected to undergo similar stretching in response to the demands of
logical theory.
Frege provides a closely related, but different argument for (a). This
argument occurs in 'On Sense and Denotation' and is repeated in his
correspondence with Russell and his posthumously published writings.
The following passages suggest the argument:
The fact that we concern ourselves at all about the denotation of a part of the sentence
indicates that we generally recognize and expect a denotation for the sentence itself. (G
& B, 'S & R' 63/ KS 149)
Now it would be impossible to see why it was of value to us whether or not a word had a
denotation if the whole sentence did not have a denotation and if this denotation was of
no value to us; for whether or not that is so [whether or not the words have a
denotation] does not affect the thought. (Cor 1521 BW235)
[If a sentence had no denotation] the denotation of any part would be a matter of
indifference, for, regarding the sense of a sentence, only the sense not the denotation of
its parts comes into consideration. (Cor 1581BW 240; cf. also Cor 165, 163nlBW 247,
245n; PPW232INS250-1)
These claims are embedded in discussion of examples of nondenoting
names and in an argument for Thesis (b). But in view of the obvious
generality of their intent, I think that they are worth isolating.
Frege's argument is that we would not concern ourselves with the
denotations of sentence-parts if we were not interested in the denota-
tions of whole sentences; we clearly do concern ourselves with the
denotations of sentence-parts - we often care whether singular terms
denote something; so we are interested in the denotations of whole
The argument must again be seen in the light of the centrality of the
Composition Principle (1) and of Frege's pragmatic use of the term
'Bedeutung'. Denotations of sentences are whatever can be seen as both
central to logical theory and functivnally dependent on the denotations
of the logically relevant parts of sentences. But this argument adds a
further claim. Our interest in the denotations of words is derivative from
our interest in the denotations of sentences. That is, word denotation is
important because and only because of the importance of some feature
of sentences that is central to logical theory and functionally dependent
on word denotation.
This further claim appears to be an expression, or outgrowth, of the
context principles that Frege had enunciated earlier in The Foundations
of Arithmetic. These enunciations preceded the development of the
distinction between sense and denotation, and they took a variety of
nonequivalent forms. But they all emphasized that the analysis of the
"meaning" of a word (in retrospect, presumably, its sense and its
denotation) was to be carried out in the context of an analysis of its role
in a sentence. Frege appears to be invoking the primacy of sentences in
his argument that sentences have denotations. Our interest in the
denotations of words had to be connected in logical and linguistic
theory with some feature of sentences. Frege forged the connection by
means of the Composition Principle (1), and he called the relevant
feature of a sentence its denotation.
It is worth noting that Frege's ,reasoning is prima facie incompatible
with the idea that the notion of the denotation of a term has no other
content than that provided by an analysis of the contribution of the term
in fixing the denotation (or, truth value) of a sentence.
The argument presupposes that we have a co-equal understanding of
and application for the notion of the denotation of a term.2 Indeed it
presupposes that the notion of term-denotation is more familiar than
that of sentence denotation, though perhaps not more familiar than
that of truth value. The argument claims that whether terms have any
denotation at all is of importance to us only relative to our interest in
relevant semantical properties of sentences. It does not suggest that the
notion of term-denotation can be exhaustively defined, or characterized,
or reduced by attempting to analyze the relevant semantical properties
of sentences in total abstraction from one's ordinary understanding of
the notion of term-denotation (reference). The ordinary understanding
of term denotation is assumed to be sound. (One could produce numer-
ous passages from Frege's opposition to formalism to substantiate this
point.) The argument simply demands that such ordinary understanding
has to be connected, in one's theory, to the semantical properties of
sentences, interest in which motivates interest in the denotations of
Thesis (b)
The role of value and "interest for us" in Frege's argument for (a) needs
articulation. Frege saw logic as revealing certain norms governing ideal
thought. The sentence was the linguistic correlate of thought. We think,
according to Frege, only by means of sentences. So any logical theory
had to ground itself in an analysis of the properties of sentences that
revealed the relevant norms. Our interest in the denotations of terms,
and of functional expressions, was motivated by interest in normative
properties governing thinking, normative properties whose laws logic
sought to uncover.
This focus on the normative implications of logical theory underlies
Frege's primary argument for Thesis (b). Frege specifies what it is about
sentences that motivates our interest in the denotations of their parts.
The relevant property is the sentence's truth value.
The fact that we concern ourselves at all about the denotation of a part of the sentence
indicates that we generally recognize and expect a denotation for the sentence itself. The
thought loses value for us as soon as we recognize that the denotation of one of its parts
is lacking. We are therefore justified in not being satisfied with the sense of a sentence,
and in asking also for its denotation. But why do we then want every proper name to
have not only a sense, but also a denotation? Why is the thought not enough for us?
Because, and to the extent that, we are concerned with its truth value. This is not always
the case. In hearing an epic poem ... we are interested only in the sense of the sentences
and the images and feelings thereby aroused. In response to the question of truth we
would abandon aesthetic delight and turn to a scientific investigation. Hence also it is a
matter of no concern to us whether the name 'Odysseus', for example, has denotation so
long as we accept the poem as a work of art. It is the striving for truth that drives us
always to advance from sense to denotation. (G & B, 'S & R'63/KS 149)
When we merely want to enjoy the poetry we do not care whether, for example, the
name 'Odysseus' has a denotation ... the question first acquires an interest for us when
we take a scientific attitude - the moment we ask, 'Is the story true?', that is, when we
take an interest in the truth value .... Now it would be impossible to see why it was of
value to us whether or not a word had a denotation if the whole sentence did not have a
denotation and if this denotation was of no value to us; for whether or not that is so
does not affect the thought. And this denotation will be something that will have value
for us precisely when we are interested in whether the words have denotation (bedeu-
tungsvoll Sind), therefore when we ask after truth. (Cor 152 BW 235)
... if it is not a matter of indifference to us whether the signs that make up a sentence
have a denotation, then it is not just the thought that matters to us, but also the denota-
tion of the sentence. And this is the case when and only when we ask after truth. Then
and only then does the denotation of the sentence enter into our consideration; it must
therefore be most intimately bound up with truth. (Cor 165/BW247)
That the name ... designates is of value to us when and only when we are concerned
with truth in the scientific sense. So our sentence will have a denotation when and only
when the thought expressed in it is true orfalse. (PPW 232/NS 250-1)
Frege's argument for Thesis (b) clearly presupposes his arguments for
Thesis (a). It thus presupposes the primary importance of sentences in
logical theory. In fact, the language of the first three passages directly
echoes the first statement of context principle in The Foundations of
Arithmetic: "never to ask for the Bedeutung of a word in isolation, but
only in sentential context". The phrases I have translated "asking for its
denotation" and "ask after truth" in these three passages from 'On
Sense and Denotation' and the correspondence with Russell use the
same phrase 'zu fragen nach der Bedeutung' that occurs in the introduc-
tion to Foundations. It is almost inconceivable that Frege did not intend
to associate the passages with his slogan. The reason that the denota-
tions of words must be "asked for" only in sentential context is that the
relevantly related semantical feature of sentences - the denotation of
sentences - motivates our interest in word denotation. Our interest in
the denotation of words derives from our interest in the truth value of
sentences, or of the thoughts that they express. Truth is the relevant
norm governing our use of and interest in sentences and thoughts. The
point of logical theory should be the analysis of the most general laws
governing this norm.
Frege's argument for Thesis (b), the thesis that the denotation of a
sentence is its truth value, is not and is not intended as a deductive argu-
ment. There. is no attempt to deduce (b) from "first principles". In 'On
Sense and Denotation', he twice calls the thesis a conjecture (Vermutung
- conjecture, supposition, surmise) - (G & E, 'S & R' 64, 65/ KS 150,
151). And the remainder of article is presented as a series of "tests" of
the conjecture. After considering these tests in detail, he writes at almost
the end of the article: 'From this it follows with sufficient probability
that the cases where a subordinate clause is not replaceable by another
with the same truth value proves nothing against our view that a truth
value is the denotation of a sentence whose sense is a thought'. (G & B,
'S & R' 78/KS 162)
There is a closely related argument, proposed by Church, G6del and
others, that does take deductive form. Frege has sometimes been
constructed as giving an ellipitical, or even invalid, approximation to this
argument. I think that such a construal is very poor history. Frege's
argument rather has the form: In view of the normative aims of logical
theory and in view of the considerations that actually motivate our
interest in the denotations of terms, the appropriate feature of sentences
to connect with the denotations of the sentence's constituent parts via
the composition principle, is the sentence's truth value. We shall discuss
the Church-G6del argument shortly.
There is a supplementary argument for Thesis (b). This argument is
roughly: a sentence's truth value is dependent on the denotations of its
constituent terms in just the way that the Composition Principle (1)
requires. So given the way the notion of a sentence's denotation is
introduced, truth values are well-suited to be the denotations of sen-
In 'On Sense and Denotation' and elsewhere Frege proposes this
argument as a confirmatory consideration or an essential test of the
conclusion of the previous argument: 'If our conjecture that the denota-
tion of a sentence is its truth value is correct, the latter must remain
unchanged when a part of the sentence is replaced by an expression
having the same denotation. And this is in fact the case." (G & B,
'S & R', 64/KS 150)
Taking the appeal to the Composition Principle (1) as a confirmation
or supplement to the previous argument seems to me to be Frege's most
reasonable presentation of the relation between the two arguments for
Thesis (b). But sometimes Frege seems to place the appeal to the
composition principle in a different light. He asks, "what else but the
truth value could be found, that belongs quite generally to every
sentence, to which the denotation of its constituent parts is relevant, and
that remains unchanged by substitutions of the kind in question?"
(G & B, 'S & R', 64/KS 150; ct. also Cor158, 165/BW240, 247)
Although Frege cannot be expected to have foreseen this, his question
prompted Russell to open a semantical and metaphysical Pandora's box.
One can well imagine Russell turning over in his mind this question,
which Frege put to him more than once in their correspondence of
1902-4. For Russell was resisting the view that sentences had truth
values as denotations.
A year after the correspondence ended, Russell published his theory
of descriptions. The theory opened the possibility of maintaining alle-
giance to the Composition Principle (1), yet analyzing the logically
relevant parts of a sentence in a very different way from the way Frege
regarded as natural and appropriate. The theory simultaneously opened
the possibility of assigning a variety of different sorts of denotations to
sentential parts, and a variety of different sorts of denotations, other
than truth values, to the sentences themselves ("states of affairs", "facts",
"propositions" and so forth).
Russell demonstrated that one could do compositional semantics
without taking truth values to be the central feature functionally asso-
ciated with sentences. But it is no accident that, despite the deep
methodological interest of the theory of descriptions, Frege's approach,
not Russell's, has been the source of the mainstream development of
semantical theory in logic. No doubt one reason for the pre-eminence of
Frege's approach lies in the artificiality, from a syntactial or grammatical
point of view, of Russell's analysis of sentential constituents. But more
profound reasons are suggested by Frege's first argument for Thesis (b).
Truth (or some modalized notion of truth, like necessary truth or
validity) is the central notion of logical theory. In making truth values
the primary functional values of the Composition Principle (1), Frege
was simply uniting his formal apparatus with the conception that
motivates logical theory.
Russell's theory can, of course, accommodate the representation of
truth values, of truth-evaluations; and it maintains allegiance to a notion
of logical consequence explained in terms of truth. But formally, the
truth values enter through a side door, so to speak. The composition
principle yokes words with sentences, but it is not used primarily to
relate word-denotation to truth value. The primary semantical feature of
sentences is the "fact" they are correlated with. The denotations of
words functionally determine a ''fact'' or "proposition" composed of
attributes and (perhaps) individuals. The primary semantical feature of a
sentence is the ''fact'' that it is correlated with. Thus, from the outset,
Russell's formal theory incorporates into its subject matter entities that
evince a strong admixture of metaphysical motivation. States of affairs,
facts, and the like have a recurring attraction for the metaphysically
minded. But they have not obtained general acceptance among logicians,
and they have yet to, be shown to be indispensable for the foundations
of logic. By contrast, the more abstract notion of truth is firmly
entrenched in nearly all logical theories. Formal logical theories that
place this latter notion at their center, resting little or no weight on
arguably dispensable metaphysical entities semantically correlated with
sentences, have formed the mainline development oflogic in this century.
Frege may be seen as a certain sort of minimalist in this context. He
conceived of the fundamental part of logic - the calculus of truth values
and first and second order logic - as having an aim and subject matter
that was relatively independent of metaphysical controversy. The laws
of logic are fundamentally the laws of truth, not laws about the meta-
physical constitution of facts, propositions, or thoughts (Gedanken). (Cf.
Kl50SIKS 342; and PPW 1221NS 133.) The metaphysics of thoughts is
developed to deal with intensional contexts and with epistemic ques-
tions, which are treated only in a heuristic way in Basic Laws.
Frege took the notion of truth as a normative primitive.
He did not
leave it unexplicated, and his explications are, as we shall see, highly
controversial and involved in metaphysical commitments. But his basic
procedure is that of a good scientist in the broadest sense of the term.
He created and worked within a theory whose interpretation, for the
fundamental purposes of the science, was largely uncontroversial. Con-
troversial views were isolated and confined - to the science's heuristic
preliminaries and to its frontiers (the philosophical explication of the
notions of truth, sense, and assertion, and the application of the logic to
intensional contexts, respectively). Extensional logic, more or less as
Frege interpreted it, remains fundamental at least in the sense that it is
common ground to all logicians and in the sense that its interpretation
expresses, with a minimum of controversial accessories, that notion of
logical consequence in terms of truth which has traditionally been seen
as the central concept of the discipline.
None of this is to deny that Frege had a controversial metaphysics.
His philosophical views about truth (particularly Theses (c) and (d) and
the "redundancy" conception), his theory of sense, and his theory of
judgment and assertion are widely doubted. Indeed, one might safely
count them mistaken.
Frege's philosophical views are not, as such, a set of unfortunate
superfluities. I think that a metaphysics - or rather a set of controversial
philosophical proposals - in this area can hardly be avoided. There are
philosophical questions about truth, meaning, cognitive value, and judg-
ment that are genuinely difficult and apparently genuine. Frege
responded to - in fact, in some cases introduced - these questions.
And in order to deal with problems about informativeness, about the
commitments of propositional-attitude discourse, about the mechanisms
of word-denotation, and so forth, he postulated certain metaphysical
entities (senses, Gedanken) that are no less controversial than Russell's
facts or propositions. Russell's own theory is in part an attempt to
answer these same questions. So from a certain philosophical standpoint
it may seem that until these issues are thrashed through, Frege's position
holds no advantages over Russell's. His extensional logic owes debts
that must be paid before a balance sheet can be drawn up.
There is surely something to this standpoint. But I think that it
overlooks one of Frege's central insights and ignores the cognitive
advantages of his pragmatic method. Frege's insight is that the norma-
tive notion of truth is the central semantical feature of sentences and the
fundamental concept of the science. And his pragmatic method of
isolating controversy carries the subject a long way before philosophical
issues intrude. That has been the method of all successful sciences,
mathematical logic included; and success is perhaps our surest guide to
knowledge. These considerations tend to favor Frege's basic approach
to logic unless the philosophical issues with which he and Russell
grappled were decisively and "scientifically" decided in a way that
undermined Frege's extensional starting point. That possibility seems
The Church-Godel Argument
Here is perhaps a good place to enter into a digression on the relation
between Frege's argument for Thesis (b) and an argument proposed by
Church and G6del that is clearly inspired by Frege. (Church (1943),
G6del (1964).) The argument has a number of interesting variants, and
it has been put to even more uses. G6del's version is particularly rich in
implications. I shall, however, discuss only what has become a stan
dardized form.
The argument is supposed to show that all true sentences denote
the same thing; an analogous one would show that all false sentences
denote the same thing. The argument first presupposes that sentences
have a semantical feature that bears enough of an analogy to the central
semantical feature of terms to be given the same expression. (This
presupposition is often not made explicit. For convenience we shall,
inaccurately, call it a "premise".) Let us dub this feature "denotation" in
accord with Frege's Thesis (a). Second, the argument assumes the
Composition Principle (1). And third, it assumes that logically equiva
lent expressions have the same denotation. Take any true sentences S
and S'; S is logically equivalent with a sentence of the form '(LX)
(X = 0) = (LX) (X = 0 & S)'. So by the third premise, S and this
sentence have the same denotation. But the latter sentence yields the
sentence '(LX) (X = 0) = (LX) (X = 0 & S')' by substitution of co
denotational terms on the right side of the identity sign. So these two
sentences have the same denotation, by the second premise. But the new
sentence is logically equivalent with S'. So by the third premise, they
have the same denotation. So Sand S' have the same denotation.
Frege accepted not only the conclusion of the argument, but all three
premises. But in arguing for the conclusion, in effect Thesis (b), he did
not advance this argument. I do not find it plausible to view Frege as
giving an elliptical or invalid approximation to this argument. The
primary reason for this is the one I have already proposed. Frege
invokes the normative foundations of logic and the normative roots of
the primacy of sentences in logical theory (and in everyday language
use) in arguing for his conclusion. That is, he has a premise about the
point of logic; and he connects the notion of sentence denotation both
with this point and with his primary analytical tool, the Composition
Principle (1). The Church-Godel argument makes no such appeal to the
purpose of logic or semantics.
Another reason why Frege's argument is different can be developed
by looking ahead in our discussion. One source of plausibility for the
third premise of the Church-Godel argument derives from the com-
parison of sentences to terms - in effect, Frege's Thesis (c). Clearly, the
denotations of logically equivalent terms are the same. Insofar as
sentences are terms, or at least designators, they plausibly fall under the
same principle. But many of the considerations that led Frege to accept
Thesis (c) presuppose a prior commitment to a semantical analysis of
sentences in terms of their truth values.
A deeper version of the same sort of point can be made from another
angle. It may seem perfectly reasonable to accept the third premise of
the Church-Godel argument independently of comparison between the
semaI\tics of terms and the semantics of sentences. Suppose that we
avoid relying on the view that sentences, like terms, designate or denote
entities. The notion of the denotation of a sentence was initially intro-
duced as that notion which captured the primary semantical feature of
sentences for logical theory that could be linked up, by the Composition
Principle (1), with the denotations of terms. Should we not expect,
virtually a priori, that logical theory ought to count sentences as being
the same with respect to their primary semantical feature if it counts
them logically equivalent?
The rhetorical question packs a punch. But it still overlooks how
fundamental Frege's starting point is. The sentences that are indicated to
be logically equivalent in the Church-Godel argument are so counted
under a prior conception of logical equivalence, whether informal or
fully articulated. This notion already employs some concept of truth -
truth under all interpretations, necessary truth, or the like. From Frege's
standpoint, this notion of logical consequence (and logical equivalence)
already brings with it a commitment to truth values as the central,
logically relevant feature of sentences. So again the third premise of the
Church-Godel argument is less fundamental from Frege's standpoint
than its conclusion. Frege's syntactical analysis - Thesis (c) - his
conception of logical consequence, and the metaphysics of his logical
theory, e.g. Thesis (d), all depend on his commitment to logic's being
primarily concerned with the normative notion of truth.
It would be absurd, of course, to suggest that Frege's conception of
logical consequence in terms of (necessary) truth was somehow arbi-
trary, or merely one of many equally suitable choices. The conception
lies in the mainline tradition of logic that stretches back to its beginning.
Even those conceptions of logic prior to Frege that allowed metaphysi-
cal visions to predominate tended to maintain allegiance to the informal
conception of logical consequence from which his theory sprang. There
have been in this century a few approaches, self-consciously reacting
against the main tradition, that have departed from the standard infor-
mal conception of logical consequence, following a metaphysical, or
more often an epistemological muse. At this point, such approaches
must be regarded as secondary developments.
The preceding discussion is not intended to suggest that the Church-
Godel argument is circular. (The third premise is not equivalent to the
conclusion.) The argument is usually given in a context in which people
already have the ordinary notion of logical consequence, and in which
the notion of a denotation for sentences is open to determination. The
usual way of reading the argument is to give it the flavor of, "If you are
willing to concede that there is a notion of sentence denotation that
meets these restrictions (those of the argument's second and third
premises), I will surprise you with what the denotation of a sentence has
to be." The third premise might be bolstered by the argument I gave
above [second half, p. 109.] By contrast, Frege already knew exactly
how he wanted to use the notion of sentence denotation: it was re-
stricted by the Composition Principle (1) (second premise); but it had to
accord with the primary aim of logic, as it has traditionally been
Much of the surprise of the Church-Godel argument derives from
implicitly thinking of sentence denotation primarily in terms of the pre-
theoretical notion of naming, rather than primarily in terms of a specific
conception of its theoretical employment, as Frege did. Once one has
taken the dubious step of seeing sentences as names, or at least as desig-
nating some entities that are functionally dependent on word-denota-
tion, it is intuitively surprising to think of them as designating only one
of two entities, and odd to reify truth and falsity. Having come so far, it
is perhaps more intuitive to take sentences as "designating" possible
states of affairs, or something like that. At least many have thought so.
Seen this way, the conclusion of the Church-Godel argument is unap-
pealing. It is doubtful, however, whether anyone, except perhaps for
Church, has endorsed the argument read in this way.4 The natural and
most common response to the argument is to reject.its first "premise":
sentences do not name, refer to, or designate any entity.
As 1 have indicated, it is possible to see the argument as using a less
determinate notion of denotation that gets around this objection. One
can consider the argument without adopting Frege's Theses (c) and (d).
Then the oddity of the conclusion disappears - the better for reflecting
on the logical relationships that the argument reveals.
Theses (c) and (d) - Pragmatic Motivations
In my view, the first of Frege's arguments for (b) and both his arguments
for (a) are sound. Although (a) and (b) have often been targets of
criticism, most of the criticism stems from construing (a) and (b) in the
light of Theses (c) and (d). 1 believe that doubts about (c) and (d) are
justified. But as 1 shall try to show in the remainder of this section and in
Section II, such doubts are less interesting than has sometimes been
supposed. The discussion of these theses will serve to introduce back-
ground essential to Frege's treatment oflogical objects.
In a letter to Frege in 1903, Russell challenged Thesis (c), the view
that sentences are to be regarded as of the same logical type as singular
terms: "I have read your essay on sense and denotation, but 1 am still in
doubt over your theory of truth values, only because it seems para-
doxical to me. 1 believe that a judgment, or even a thought, is something
so completely peculiar that the theory of proper names has no appli-
cation to it." (Cor. 155-6IBW238.) The gist of Russell's challenge has
been repeated by subsequent generations, and with qualifications to
hedge against the overstatement in Russell's phrase "no application," I
would echo it. But it is easy to be led by the paradoxical ring of (c) and
(d), as 1 think Russell was led, into misunderstanding their import and
place in Frege's system.
Frege repeatedly emphasized intra-logical, pragmatic advantages for
regarding truth values as objects: "How much simpler and sharper every-
thing becomes through the introduction of truth values, only thorough
occupation with this book can show. These advantages alone already
put a great weight on the balance in favor of my conception, which
indeed may seem strange at first glance." (BL 7/ GG x.) In fact, The
Basic Laws of Arithmetic mentions only considerations involving
simplification of logical theory as motivations for (c) and (d). These
considerations are also dominant in Frege's post-paradox writing.
I should make it clear here that in calling Frege's reasoning "prag-
matic" or "intra-logical," I am not suggesting that he took the commit-
ments that he based on such reasoning to be less than absolutely serious.
Such commitments were not merely practical conveniences or technical
artifices. Frege saw himself as making objective discoveries. What I wish
to emphasize is the great extent to which Frege tried to develop his
positions from his analysis of logical structure and from observations
regarding functional analogies between different components of that
structure. In his arguments for (a)-(d), considerations that derive from
intuitions not firmly entrenched in the actual practice of logic, "meta-
physical intuitions," play a secondary role in Frege's argumentation.
Whatever conceptions most profoundly clarified and simplified logical
theory, whatever language made mathematical practice more rigorous,
more comprehensive, more fruitful, and less ad hoc, were seen as
providing insight into the most abstract features of the world.
The pragmatic cast of Frege's thought seems to have come naturally.
Only rarely did he remark on his methodology in general terms. The
effusion from Basic Laws, quoted in the previous paragraph but one,
constitutes a relatively unusual example. The following passage from a
thrice rejected manuscript of 1880-1 provides another:
All these [mathematical) concepts have been developed in science [Frege terms mathe-
matics a science) and have proved themselves fruitful. What we can perceive in them
therefore has a far higher claim on our attention thaI'l anything that everyday trains of
thought might offer. For fruitfulness is the touchstone of concepts, and the scientific
workshop is the real field of observation for logic. (PPW 331NS 36-7; cf also KS 124,
Frege's pragmatic considerations rest on analogies that are quite natural
within a formal context. In formulating the propositional calculus, it is
natural to quantify t4e letters that stand for sentences in something like
the way one quantifies into the places held by singular terms in the first-
order functional calculus. The places for sentences, like the places for
terms, stand in argument places for functional expressions; they some-
times constitute value expressions resulting from functional application;
and they never stand for functional or predicate expressions. When one
quantifies the letters that stand for sentences, the natural interpretation
of the domain of quantification is to take it as consisting of the two truth
values - as several generations of logic students have been made aware.
(Cf. Cor. 158/BE241; KS 225-6/
Another analogy, which is more debatable, is that between non-
denoting names and truth-valueless sentences in natural language.
(G & B 'S & R' 62/ KS 148; Cor. 152, BW 235). Both maintain a sense
in the absence of denotation. Frege thought that subject-predicate
sentences containing nondenoting terms always lacked truth values.
It is difficult to see to what extent he accepted this view on intuitive
grounds and to what extent he reasoned to it. If one already sees
predicates as denoting functions, then one will see a nondenoting name
as providing no argument for such a function. Functions without
argument yield no value. And if the denotations of sentences are the
values of such functions, and are counted truth values, then sentences
involving only the application of predicates to nondenoting terms (and
application of functors to sentences so obtained) will lack truth value.
This reasoning, of course, assumes the assimilation of predicates to
function signs. And this assimilation is tantamount, as we shall see, to
accepting Thesis (c). So the reasoning cannot be seen as providing much
independent support for Thesis (c).
Very likely, the view that nondenoting terms in subject-predicate
sentences yield truth-value-less sentences was also found acceptable by
Frege on intuitive grounds. The various examples he gives do elicit in
many the intuition that the sentences are neither true nor false. But
there are numerous other cases that are at best indecisive witnesses for
Frege's defense. Since this issue has been discussed at uncommon length
by others, I shall not go into it. I think that Frege's view of the intuitive
relation between nondenoting terms and truth-value-less sentences was
not very critical to his account of truth values. Since he banned non-
denoting terms from his formal theory, he rested little weight on the
An analogy of which Frege makes more is that between nonassertive
occurrences of declarative sentences (suppositions or occurrences
within other sentences) and proper names. (EL 35/GG 7.) Sentences
often occur embedded in other sentences (for example, in the antece-
dents of conditionals) in such a way as to contribute to semantical
structure, without being asserted - like terms. Moreover, whole sen-
tences can be put forward merely for consideration without carrying any
assertive force - again like terms.
These analogies between sentences and terms are, of course, not very
gripping. They take on interest when linked to Frege's larger strategy.
One of Frege's most profound contributions was to separate the notions
of predication and assertion. More generally, he distinguished the
notions of logical structure and pragmatically relevant force. The deeper
point of the present analogies is that within a formal theory that
attempts to lay bare semantical structure, one can prescind from the
primary difference between names and sentences (that only the latter
can be used to effect linguistic acts or thoughts, protoypically assertions
and judgments). The difference between names and sentences can be
taken to lie in their point, their use, not in the form of their contribution
to semantical structure. Actually, as we shall see in Section II, Frege's
formal theory did make formal distinctions between sentences and
terms. But the distinctions do not leap to the eye. Although one might
believe (as I do) that form should correspond more closely to use than
Frege's logical theory allows, subsequent formal usage has confirmed
that Frege's analogy constitutes an insight that affords at least a con-
venient alternative in setting up a logical system.
Frege's construal of predicates as functional expressions is perhaps
the most obvious and widely appreciated ground for Theses (c) and (d)
and for his view that truth values are objects. As far back as the
Begriffsschrift in 1879, Frege had interpreted predicates as function
signs (B, Section 9). Once he supplemented this initial conception with
an explicit semantical analysis, which he arrived at by the early 1890's at
latest, he was forced to think of functions as denotations for predicates.
He called such denotations "concepts". (We shall limit considerations to
1st-level concepts.) Objects, obviously, served as arguments for (lst-
level) concepts. But then there must be values for these functions. These
must be the denotations of sentences. Sentences are not themselves
functional expressions, so their denotations are not functions. Moreover,
the values of prototypical functions, the denotations of prototypical
completions of functional expressions (terms), just are objects. (We
shall, for now, regard an object as anything that is denoted by what is,
under logical analysis, a term.) Taking concepts literally to be functions
was tantamount to taking the denotations of sentences to be objects
(and the completions of predicates, sentences, to be terms). Since Frege
had independent grounds for regarding the denotations of sentences to
be truth values, this line of thought entailed that truth values were objects.
Why did Frege take concepts, denotations of predicates, literally to
be functions? One primary and lasting motivation was, yet again, prag-
matic. Seeing sentences as created by the application of functional
expressions effected a simplification in the understanding of the compo-
sition principles. The simplest construal of the Composition Principle
(1) is to take 'the denotation of sentence s' to be a singular term,
denoting an object. (Cf. Note 8.)
A closely related motive was to provide a simple formal expression
of the formal analogies between predicates and function signs. (BL 6,
34-5IGGX, 6-7/PPW 235/243-4/NS 253-4, 263). Like function
signs, predicates have open places for terms. The primary role of
predicates from the point of view of logic is functional - to take objects
into truth values.
Completeness and Incompleteness
Frege's pragmatic motives are, I think, dominant. But the analogy
between predicates and function signs is sometimes associated by Frege
with remarks that have a darker, more metaphysical hue - remarks
about similarity in their ''unsaturatedness'' or "incompleteness". He says
that the essence of a function is its making a connection between its
arguments and its values, in a specific sort of "need for completion" (BL
33-4/GG 5-6; 'C & 0' G & B 47/KS 171; 'F & C' G & B 24-5/KS
128-9). Moreover, he writes, 'An object is anything that is not a
function, so that an expression for it does not contain any empty place.
A declarative sentence contains no empty place and on that account its
denotation is to be regarded as an object' ('F & 0' GB, 32/KS 134).
Here Frege may appear to be inferring from a metaphysical thesis
about incompleteness of functions and completeness of objects and
from a thesis about how language must match reality as regards com-
pleteness or incompleteness, to the conclusion that truth values are
objects. Michael Dummett interprets the characterization of objects as
anything that is not a function in this passage (and in an equivalent one
in Basic Laws vol. I Section 2) as an ad hoc attempt to induce the reader
to accept truth values as objects (IFP, 235n). Neither the metaphysical
reading nor Dummett's attribution of desperate improvisation places
Frege in a very attractive light.
To begin with the latter interpretation, I do not find Dummett's
charge plausible. Frege's characterization of objects is independent of
Thesis (d) and precedes its adoption. The idea that objects can be
recognized as whatever is never the denotation of an incomplete, func-
tional expression goes back at least to The Foundations of Arithmetic
(Cf. pp. x, 77n, 72 - before Frege's adoption of Thesis (d).) In the latter
passage, FA 72, Frege writes that the point of counting number words
as words for objects (or self-subsistent objects) is "only to preclude the
use of such words as predicates or attributes ... " (Cf. PPW 100, 1051
NS 109.) Given Frege's view that truth values are denotations of
complete sentences, and never denotations of predicates, and given this
characterization, truth values fill the bill as objects.
Frege's characterization of an object as the denotation of any ex-
pression other than a predicate or function sign may seem either to
emasculate the notion of object or (perhaps equivalently) to commit one
to objects too easily. In discussing Frege's arguments for Thesis (a) we
attributed to him a notion of sentence denotation that does not carry
genuine ontological commitment. But now, it may seem, we are allowing
Frege to smuggle ontological commitments into his arguments for
Theses (a) and (b) by granting him an excessively liberal criterion for
ontological commitment to objects. I have argued that the relevant
criterion was not fabricated, as Dummett suggests, simply to make
palateable the view that truth values are objects. But it may seem that
Frege made illegitimate use of a criterion that was first developed in a
context in which the denotation of sentences was not an ontological
issue - resorting to a cheap means for ontological gain.
There is something to this worry. I believe, however, that it cannot be
taken at face value. Frege does argue from his characterization of
objects to Thesis (d) (in the paragraph following the relevant charac-
terization of objects in 'Function and Concept'). But he does not take
the characterization as stipulated or ungrounded. In the first place, there
are substantial analogical considerations that undedy his counting
sentences and terms "complete" and predicates and function signs
''incomplete.'' In the second place, Frege seems to have always regarded
the characterization of objects as resting on an antecedent notion of
completeness that he believed he could apply to sentences (and truth
values) as well as to terms and ordinary objects. It is the notion of
completeness that bears the weight, not the bare claim that objects are
the denotations of every sort of symbol other than function signs. The
intuitive notion of completeness underlies and motivates the syntactical,
semantical, and ontological claims.
Then isn't Frege's engrossment in the completeness-incompleteness
distinction simply a metaphysical indulgence? I would not deny that
some of Frege's uses of the distinction involve a kind of fixation that is
difficult to fathom, much less defend. But his deployment of the distinc-
tion to support his view that the denotations of sentences, truth values,
are objects, seems to me less problematic than some other uses he
makes of the distinction.
Let us lay aside Frege's view that no objects are functions and no
functions are objects. I think that this view is extremely doubtful and
that it probably does constitute an instance in which Frege allowed his
sound conceptions of logical function to harden unnecessarily into a
metaphysical doctrine. These matters are, however, intertwined with a
suprisingly large number of serious considerations (for some of them,
see Burge, 1984). I shall avoid the tangle here.
Let us consider only Frege's views that in using (what were under
logical analysis) function signs, one is committed to their denotations,
functions; and that in using (what were under logical analysis) terms,
one is committed to their denotations, objects. As we have noted,
predicates are like function signs in having empty argument places, and
in having a functional role in logical theory. Sentences are like terms in
not manifesting such formal incompleteness and in not having a func-
tional role. In concluding that the denotations of sentences are objects,
Frege may be reasonably seen not as drawing a primitively minded
inference from some pre-Socratic vision of the world as a mixture of the
complete and incomplete - but as simply summing up and embellishing
the analogies, within his logical system, between the roles of sentences
and terms, and their contrasts with predicates and function signs.
The mapping of objects and functions onto truth values - the central
semantical feature of sentences - is the primary formal role of predicate
expressions (or concepts) within formal logical theory. The deep differ-
ences between predicates and ordinary function signs, and between
sentences and terms, were largely shunted off into the theory of
force or use. Frege did not lose sight of the differences. But he
thought that he could draw ontological conclusions from a semantical
theory that abstracted from them. In regarding concepts as functions
and truth values as objects on grounds of the "incompleteness" of
signs for the latter, Frege was basing ontological commitments on
the semantical analysis of the logical forms of sentences in whose
truth he believed. Frege's methods, if not his conclusions, seem
Clarification of Extensions of Concepts
The assimilation of concepts to functions served one other large pur-
pose in Frege's system. It provided the key to his attempt to clarify the
notions of a concept and of the extension of a concept. As I have tried
to show in some detail elsewhere, Frege was unclear about and dissatis-
fied with these notions from the time he first introduced the latter in
Foundations (Section 68) (1884) up to and through the publication
of Basic Laws. (1903). (Cf. Burge, 1984.) The key to the clarifica-
tion that he attempted, until Russell unsettled him, was the notion of the
course of values of a function. Frege sought to make this notion intuitive
by appeal to the graph of a function (G & B, 'F & C' 251KS 129)
which he seemed to think of both algebraically and geometrically. He
self-consciously did not interpret the graphs as a set of ordered pairs,
for a variety of reasons deriving from his emphasis on the priority of
functions over their courses of values. We shall return to these points in
Section III.
The notion of a concept had had a long but mathematically barren
career in the logical tradition. It was not held in high esteem by
mathematicians in Frege's day. By contrast, the notion of a function was
well established in mathematics. By assimilating the denotations of
predicates to those of function signs - giving them a recognizably
mathematical role - Frege hoped to clarify the notion of a concept and
burnish its reputation. At the same time, he would be effecting a unifi-
cation of the languages of logic and mathematics in accord with his
logicist thesis. This motivation is explicit when Frege first introduces the
assimilation of concepts to functions:
... for what purpose, then, are the signs '=', '> " , <' admitted into the circle of those
that help form a functional expression? It seems that nowadays more and more sup-
porters are being won to the view that arithmetic is further-developed logic ... I too am
of this opinion, and I base upon it the requirement that the symbolic language of
arithmetic must be expanded into logical symbolism (G & B, 'F & c', 30/KS 132.)
The clarification of the notion of a concept was intended to give a firm
foundation to those objects logically associated with concepts (their
"extensions") with which in Foundations, driven by grammatical con-
siderations and his logicist goal, Frege wanted to identify the numbers.
The extension of a concept was understood as the course of values or
graph, obtained by providing all objects one by one as arguments for the
concept (function) and taking the resulting truth values as values. The
whole procedure, taken as a completed whole, was what Frege regarded
as a logical object. Since the introduction of such objects crucially
depended on Axiom V, which led to Russell's paradox, Frege's attempt
to clarify the notion of the extension of a concept by assimilating
concepts to functions failed.
The Redundancy Conception of Truth and the Notion of Object
I shall conclude our discussion of Frege's reasons for accepting Theses
(c) and (d) by considering his redundancy view of truth. In 'On Sense
and Denotation' he writes:
One might be tempted to regard the relation of the thought to the True not as that of
sense to denotation but rather as that of subject to predicate. One can, indeed, say: ''The
thought that 5 is a prime number, is true." But if one observes more closely, one notices
that really nothing more is thereby said than in the sentence '5 is a prime number.'
('S & R' G & B 641 KS 150). (Cf. also PPW 128-9, 233-4, 251-2, 255-61 NS
139-140,251-2,271-2,275-6; 'The Thought' Kl. 5141KS 347.)
Frege uses the view to ward off possible doubts about the postulation of
the truth values as objects denoted by all sentences, regardless of subject
matter. If truth were an attribute of a limited range of entities (thoughts),
it would be difficult to motivate the claim that every sentence denotes
one of the truth values and his view that (in a sense to be sharpened in
Section III) all assertive uses of sentences regardless of subject matter
are committed to the object truth.
In 'On Sense and Denotation', two large philosophical ideas emerge
in connection with the redundancy conception. One utilizes truth values
as objects in an account of assertion and judgment. The other bears on
scepticism. We shall consider these themes in turn.
Frege goes on from the passage just cited to argue that the claim or
judgment that a thought is true arises not from the predication of 'is
true' of a thought, since the sentences 'the thought that 5 is prime is true'
and '5 is prime' express the same thought regardless of whether they are
used with or without assertive force. Truth claims or judgments depend
on the combination of the form of a declarative sentence with its "usual
force." Frege thinks that such truth claims are indicative of the real
relation between a sentence or thought and its truth value. (G & B
'S & R' 65/ KS 150). Judgments are "advances from thoughts to truth
values." Since truth claims and judgments cannot be represented in
subject-predicate form, the relation of a sentence or thought to its truth
value cannot be regarded as that of subsumption of a thought (sentence)
under a property. Frege proposes that the appropriate relation is that of
a sentence or thought to its denotation. (Cf. also PPW 128-9, 233-4,
251-2/NS 139-40, 252-3, 271-2.)
One need hardly note that considered as an argument for Thesis (d),
this is pretty weak. (It is doubtful that Frege intended it as such.) One
could respond that on Frege's own account, two sentences could have
the same assertive force - both could count as assertions - while one
lacked a truth value and the other had one. So truth values' as objects
cannot be essential to the account of assertive force. Even if this reply is
not decisive, Frege does not show why it is not.
I think Frege was here again thinking analogically. Normally, the
point of using names was to secure a denotation, a bearer, to relate a
mode of presentation to an object. Normally, the point of using sen-
tences, what "matters to us," is to claim truth for a thought. The object,
in the sense of the point or objective, of sentence use was truth. It is
illuminating therefore to see truth as an object. There is more than a
suggestion of this reasoning when Frege writes:
The designation of truth values as objects may here appear as arbitrary fancy or perhaps
a mere play on words, out of which no profound consequences could be drawn. What I
call an object can be more exactly articulated only in connection with concept and
relation .... But so much should already be clear, that in every judgment, no matter how
trivial, the step from the level of thoughts to the level of denotations (the objective) has
already been taken. (G & B, 'S & R', 63-41 KS 149)
The parenthetical phrase is the key to the passage.
To many this reasoning may seem indeed to rely on a mere pun on
the word 'object'. I think that there is more to it than that. Both the
relevant objective of sentence use, truth, and objects that are denoted by
terms are for Frege mind-independent. And in some sense they are what
sentences and terms are respectively "about." (In fact, as Frege empha-
sizes in his arguments for (a) and (b), objects are of interest to us
because and only because of the objective of assertion and judgment.)
Both points resist simple or quick put-downs. Both mind independence
and being the topic of a discourse are involved in traditional explica-
tions, stemming from Aristotle, of the notion of object.
I am not suggesting that the analogies between the "objects" of terms
and the "object" (objective) of sentence use provide a sound argument
for Frege's assimilation. I believe the contrary. In fact, I believe not only
that the analogies are not compelling, but also that Frege's redundancy
view of truth,which motivates them, is untenable. (Part of the reason for
this untenability lies in the semantical paradoxes; cf. Section III.) Rather,
what I am suggesting is how Frege might have come to see the analogy
as intuitively attractive, given his view that the attribution of truth added
nothing to a thought. We shall further articulate Frege's analogy
between objects and the objective of assertion, in Section III.
Frege puts Thesis (d) and the redundancy view of truth to use as a
weapon against the sceptic about an objective world. Frege writes that
the True and the False "are recognized, if only implicitly, by everybody
who judges something to be true - and so even by the sceptic" (G & B,
'S & R 63/KS 149). (Frege assumes contrary to the legends about
Pyrrho, but probably correctly, that no sceptic suspends all judgments.)
The idea is that every act of judgment aims at truth and presupposes
some discrimination between truth and falsity. Frege explicates the point
by his redundancy thesis: truth values are not a property of thought,
where thoughts constitute one subject matter among many: "What
distinguishes [truth] from all other predicates is that it is always asserted
when anything at all is asserted." (PPW 129/ NS 140.) Since the true is
an object logically associated with the truth predicate and so with
judgment - a logical, mind-independent object - , judgment itself
presupposes an objective world. We shall sharpen Frege's point in
Section III. I think that one could probably dispense with the implausibi-
lities of the redundancy view to provide the sort of premise needed for
joining with (d) in order to defeat the relevant sceptic. If only (d) were
Godel remarked that Frege held the view that all true sentences have
the same denotation "in an almost metaphysical sense" (Godel 1964,
214). It is true that Frege puts the doctrine to use against the sceptic.
There is no question but that he thought of truth as an object. And there
are some unfortunate, but qualified and never repeated remarks in 'On
Sense and Denotation' about parts of the True (G & B, 'S & R' 65/ KS
150-1) - remarks that prompted G6del (inappropriately, I think) to
compare Frege's view with Parmenides'. But if the reasons for a view
may be seen as an index of its character, Frege's doctrine cannot
comfortably be called metaphysical. The brunt of his case for (d) rests
on the formal simplifications the view effects within his logical theory
and the clarification it was supposed to yield for the notion of the
extension of a concept. With the discovery of Russell's paradox, the
latter support was undermined, leaving only the former.
It is interesting that in his post-paradox period, Frege cites only
considerations of simplification (for example, the congeniality of the
view with the Composition Principle (1 in favor of Thesis (d). In his
epistolary responses to Russell's doubts he remains doggedly within the
elegant confines of his logical theory - repeatedly employing the
composition principle and pointing out difficulties in Russell's vague but
seminal alternatives. In 1906 at the beginning of the scrap 'What May I
regard as the Result of my Work?', he cites "a concept construed as a
function" and introduces the citation with the remark, "It is almost all
tied up with the Begriffsschrift" (PPW 184/NS 200). In his late writings
he gives up on the notion of the extension of a concept, and in 'The
Thought' (1918) the arguments against scepticism make no use of truth
Once Frege's intra-logical analogies are appreciated, there is, I think,
no need other than momentary expositional convenience to treat sen-
tences as of the same logical type as names. One may maintain in one's
semantical theory a reflection of the large differences in use between
sentences and terms. And one may return to the natural view that terms,
not including sentences, are the basic avenue of ontological commit-
The primary reason why Frege did not take this more modern view
of the matter in his great, pre-paradox writings is that he wanted to use
the truth values in his account of logical objects. Logical objects were
needed for his logicist project - the project of showing that the mathe-
matics of number is reducible to logic. For mathematics was apparently
committed to objects, the numbers; and to account for these commit-
ments Frege thought he had to generate commitments to appropriate
sorts of objects within logic. The truth values were the basic logical
objects from which all others were to be generated. (Cf. Section III.)
By roughly 1906, however, Frege seems to have given up logicism'.
So the most prominent philosophical motivation for postulating logical
objects lapsed. The doctrine that truth values are objects may have
become less important to him in his later years. He does not give up the
view, however. And I suspect that in addition to analogical or pragmatic
considerations, he retained a philosophical motive for holding it. This
motive was his desire to explicate the objectivity and informativeness of
logic - its "descriptive" as well as normative character. (Cf. Note 3 and
Section III.) Although I shall not discuss it here, I think that this motive
is profoundly conceived. But after the failure of Frege's logicist project,
the attempt to utilize the truth values as means to articulate the motive
was deprived of a coherent background theory within which to bring
together a conception of truth with a conception of logical objects. So
Frege is left without a theory within which he could argue for using
Thesis (d) to articulate his thoroughly unKantian view that logic is an
informative science of "being" (K150S1 KS 342).
We have briefly touched on Frege's view about logical objects in our
discussions of his attempt to clarify the notion of the extension of a
concept, his argument against the sceptic, and his conception of the
nature of logic. We shall return to them in Section Ill. But first, I want to
consider an influential body of thought that seems to me to have placed
Frege's views on truth values in the wrong light. This discussion will
enable us to develop further Frege's conception of truth, a conception
that will dominate our concluding reflections on logical objects.
In his two books on Frege, Michael Dummett maintains, as against
Theses (c) and (d), that sentences are not names, and truth values are
not objects. As is plain, I do not dispute this conclusion. It is the
reasoning behind Dummett's rejection of these theses, and the urgency
with which he invests that rejection, that constitute, in my opinion, a
serious misrepresentation of FregeY
We have already quoted Dummett's statement that Frege's accept-
ance of Theses (c) and (d) was an almost unmitigated disaster. For,
Dummett writes,
... it obscured the crucial fact that the utterance of a sentence, a complex term ... can
be used to effect a linguistic act, to make an assertion, give a command ... the general
notion of the sense of a word will now have to be taken to consist in the contribution
which that word makes to determining what a complex singular term, in which it may
occur, stands for, rather than what are the truth-conditions of a sentence in which it may
occur. (FPL, 7)
If sentences are merely a special case of complex proper names, ... then, after all,
there is nothing unique about sentences: whatever was thought to be special about them
should be ascribed, rather, to proper names - complete expressions - in general. This
was the most disastrous of the effects of the misbegotten doctrine that sentences are a
species of complex name ... : to rob him of the insight that sentences playa unique role,
and that the role of almost every other linguistic expression ... consists in its part in
forming sentences. (FPL, 196; cf. 643-5)
Dummett takes the adoption of Thesis (c) to underlie the relative
inconspicuousness, in Frege's later work, of statements of the context
principles, statements which had been so prominent in The Foundations
of Arithmetic (p. x, and Sections 60, 62, 106). Dummett's idea is that
since Frege's assimilated sentences to complex singular terms, he
"debarred himself from a direct statement of the context principle, since
this would have involved acknowledging a difference in logical role, of
utmost importance, between sentences and proper names of objects
other than truth-values" (IFP, 371). Dummett cites Frege's conclusion to
Section 10 of The Basic Laws of Arithmetic (cf. also Sections 29, 31-2)
as evidence that only a weakened, generalized analog of the context
principle for denotation was still adhered to: The denotations of terms
are fixed when it has been determined for every primitive function
[whether a concept or not] what the value of the function is to be for the
denotations of any terms as argument(s) (IFP, 408ft). In this principle
sentences and predication are given no special prominence over terms
and ordinary functional application. Dummett goes on to question the
coherence of the resulting doctrine.
Now there is much in Dummett's discussion that we cannot take time
to go into. The context principles form an exceedingly complex topic.
Despite my disagreement on some fundamental matters in this area, I
think that Dummett has contributed a great deal to our understanding of
the issues. Here I shall concentrate on disagreements that bear most
directly on truth values.
There is evidence that Frege did not lose sight of the "crucial fact"
that the utterance of a sentence, unlike a term, can be used to make
assertion; that he did not draw the unsound inference Dummett does
that "if sentences are merely a special case of proper names ... then,
after all, there is nothing unique about sentences ... "; and that Frege
was never robbed of the insight "that sentences playa unique role."
In the first place, there are a great number of passages throughout his
career and especially from the 1890's onward, in which Frege asserts
that the aim of logic is to understand the laws of truth (PPW 2-3,
128-9, 149, 197-8, 252, 253/NS 2-3, 139-40, 161, 212-13, 272,
273; Kl 505ff/ KS 342ff). He repeatedly characterizes these laws as
normative restrictions on judgment and assertion. Predications of truth
are not really distinguishable from the assertoric form of any declarative
sentence at all (PPW 129, 2331NS 140,251; cf. our discussion, Section
I, of the redundancy theory of truth.) Once he writes that the essence of
logic lies in assertoric, or judgmental, force (PPW 252/NS 272). The
vehicle of judgment is a thOUght and the vehicle of assertion (the
expression of a judgment) is a sentence (PPW 126, 131, 206/NS 157,
142,222-3). Thus the essence and aim of logic is repeatedly associated
with sentences and thoughts (the senses of declarative sentences) and
their logically relevant uses. The denotations of terms are almost never
discussed except in the larger context of this emphasis on the centrality
of truth, judgments, thought, assertion, and sentencehood. And in
Frege's last years, the denotations of terms receive very little attention
at all.
Moreover, there are the passages from the 1890's and later, quoted
in Section I, that occur in Frege's arguments for Theses (a) and (b)
(G & B, 'S & R' 631KS 149; Cor. 152, 158, 163n, 1651BW235, 240,
245n, 165; PPW232/ NS 250-1). These repeatedly and explicitly make
the point that the denotations of terms are of interest to us only because
of our interest in the denotations, in fact the truth values, of sentences.
Indeed, the remarks constitute a fair approximation to the slogan of
Foundations that only in the context of a sentence do words have a
Further, the implication of the same passages is that our interest and
confidence in the truth of sentences that contain terms justifies our
interest and confidence in the terms' having the denotations that they
are commonly taken to have. This implication appears to echo and
perhaps even sharpen the motivation for one of the uses to which Frege
put his contextualism in Foundations (Sections 60, 62) - defending
ontological commitment to objects (numbers) in the absence of an
intuitive, imagistic, or causal relation to them. Only the general prin-
ciple underlying this use is suggested in the argument for Theses (a) and
(b). But it is clearly indicated: ontological commitment to the denotation
of terms is justified insofar as we are justified in acknowledging the
truth of sentences that contain them. It is noteworthy that these develop-
ments of Frege's contextualist thinking occur in arguments for Theses
(a) and (b), which are in turn embedded in arguments for Theses (c) and
(d) - the very theses that Dummett holds prevented Frege from main-
taining the prominence of sentences in his contextualist principles.
What then are we to say of the considerations Dummett draws from
The Basic Laws of Arithmetic to support his view that Theses (c) and (d)
undermined Frege's commitment to the centrality of sentences in logical
theory? Let us begin with the passage Dummett cites from Section 10 of
Basic Laws that states a weakening of the context principle, one that
gives no special prominence to sentences. What Frege writes is as
With this we have determined the courses of values so far as is here possible. As soon as
there is a further question of introducing a function that is not completely reducible to
already familiar functions, we can lay down what value it is to have for courses of values
as arguments; and this can then be regarded as much as a determination of the courses
of values as of that function.
A similar passage occurs in Section 29:
A proper name has a denotation if the proper name that results from that name's filling
the argument places of a denoting name of a first-level function with one argument
always has a denotation, and if the name of a first-level function of one argument that
results from the relevant proper name's filling the of a denoting name
of a first-level function with two arguments always has a denotation, and if the same
holds also for the t-argument-places.
These remarks do indeed state a kind of context principle for fixing
term denotation - one that does not give prominence to sentences.
First-level concepts are not singled out from among the first-level
functions. (Part of the reason for this derives from a particular problem
that Frege raises in Section 10 about the interpretation of his Axiom V.
I shall not go into this point here since it would require substantial
Although the principles just quoted do not give prominence to
predication over functional application, or to sentences over terms, they
are unquestionably compatible with the view that ultimately it is the use
of a subclass of "terms," the sentences, that counts in justifying interest
in term denotation and confidence in identifying the denotations of
course-of-value terms. Dummett is right to note that Frege does not
explicitly draw this distinction in Basic Laws. He is probably also right
in holding that Frege's not doing so is partly explained by his commit-
ment to counting sentences as falling in the same syntacial category as
terms. But it does not follow that Frege had lost sight of the philosophi-
cal motivations underlying the formal system that he repeatedly stated
in other writings during the same period. The circumstance bespeaks a
lack of perspicuousness in the formal system - the price of the various
economies Frege prized. But it does not evince a major philosophical
turn away from the centrality of sentences, ultimately judgment, in
motivating logical theory.
I think that the main reason Frege gives no special prominence to
sentences over terms in Sections 10 and 29 is that to make intelligible
the primacy of concepts (or predication) in fixing term denotation, he
would have had to have entered on an excursus into his philosophy of
language. Such an excursus would have been incongruous in the context
of the book as whole, where philosophical discussion was held to a mini-
mum. The strategy of Basic Laws is ruthlessly to suppress discussion
of philosophical ideas and motivations, except where they are essential
to understanding the formal system and the proofs themselves, or where
they bear directly on mathematical practice (as in the case of the discus-
sions of definition and consistency). Where philosophical ideas intrude,
they are presented tersely and in summary fashion. Except for the
polemical introduction, the book is steadfastly mathematical.
The chief consideration that Dummett relies upon for holding that (c)
and (d) undermined Frege's commitment to the centrality of sentences
in logical theory is that "the whole thrust of [Frege's] logical doctrines"
was "to recognize no difference in the kind of logical powers that
different expressions have save as were explicable by a difference in
logical type" (IFP, 371-2). Since by Thesis (c), sentences and terms are
of the same logical type, it follows that they can have no difference in
logical power. Dummett admits that Frege never states such a principle.
But he holds, "it is implicit in his whole procedure; nothing could
illustrate it more aptly than the fact that, in the logical system of
Grundgesetze, no distinction exists between sentential and individual
variables ... " (IFP, 372).
The evidence of the numerous passages that we cited six paragraphs
back indicates that this principle must be severely qualified. Although
sentences and terms are of the same logical type, according to Frege,
some properties in which they differ are of direct and primary impor-
tance to logic. Sentences can make assertions and express judgments;
terms cannot. The semantical properties of terms are of interest to us
only because of our interest in the seman tical properties and use of
sentences and thoughts. There is no reason for thinking that Frege
wanted to deny or suppress these points in his philosophical writings.
As we have seen they are prominent in his post-Foundations work.
In fact, sentences and terms are not everywhere interchangeable even
within the formal system Frege presents in Basic Laws. So in a further
sense, they do not have the same "logical powers" despite the fact that
they are of the same "logical type." Only sentences can follow the
vertical judgment stroke in Frege's syntax; ordinary terms cannot. This
important point requires detailed explication. I shall develop it by
reference to a further consideration that Dummett adduces in favor of
his view.
Dummett notes that whereas in the Begriffsschrift there is a restriction
in the formation rules against placing the horizontal or content stroke
before anything other than an expression with judgeable content -
anything other than a sentence - , in Basic Laws this restriction is
relaxed. In the latter book, the horizontal may occur before any term (or
sentence) yielding "a name of a truth value, of the True if the original
expression named the True and of the False in all other cases" (IFP,
371). Dummett does not explain what he takes the significance of this
fact to be. But it may suggest to the unwary that Frege's system was set
up so as to allow one to "judge" (impossibly) the contents of terms. For
example, both '- 5' and 'IT 5' are grammatical expression in Frege's
logic, where the shorter vertical line in the latter expression represents
negation. (Cf. the end of Basic Laws, Section 6.)
This reasoning would be quite mistaken. (I do not claim that
Dummett employs it.) In fact, Frege's use of the horizontal in Basic
Laws constitutes one of the subtleties of the book that suggest that
Frege was keeping his philosophical motivations in mind. 1 do not see
that the use supports Dummett's view in any way. To begin with,
although the horizontal may apply to ,any name, it is itself a concept
expression: a function from objects to truth values, as Frege explains
(BL, Section 6). Informally, the horizontal means "is the True:" Concept
expressions are predicates and concepts are the denotations of senten-
tial parts (e.g. PPW 119, 1931NS 129, 210). Thus the expression' - 5' is
a sentence, though a false one. It says that 5 is the True. ' IT 5' represents
an assertion that 5 is'not the true.
Now the vertical judgment stroke can be applied only to the hori-
zontal, content stroke. So it is built into Frege's system, however
discretely, that only sentences, not ordinary terms, may be asserted.
Only the senses of sentences, thoughts, may be marked as judged. Frege
himself makes the point:
I distinguish the judgment from the thought in this way: by a judgment I understand the
acknowledgement of the truth of a thought. The presentation in the concept script
(begriffsschriftliche Darstellung) of a judgment by use of the sign" f- ", I call a statement
(Satz) of the concept script, or briefly a statement .... Of the two signs of which" f-" is
composed, only the judgment stroke contains the act of assertion (BL, Section 5).
Judgments acknowledge the truth of a thought, and thoughts are said,
over and over again throughout the period and afterward, to be charac-
teristically expressed by declarative sentences: 'The proper means of
expression of a thought is a sentence' (1897) (PPW 126, 131/NS, 137,
142-3). (Cf. also PPW 129, 138, 167, 174, 197-8,206,216, 243/NS
140,150,182,189,213-14,222-3,234,262; KI'The Tho't' 5111KS
345; G & B'S & R' 64/ KS 150 etc.)IO
The result of attaching the judgment stroke to a sentential expression,
begun by the horizontal, asserts something, but it is not a term: "The
judgment stroke cannot be used to construct a functional expression; for
it does not serve, in conjunction with other signs, to designate an object:
, I- 2 + 3 = 5' does not designate anything; it asserts something."
(G & B, 'F & C, 34/KS 137) The vertical, judgment stroke is not a
function sign, but is the sign of an act - judgment or assertion - , an act
that applies only to thoughts or sentences. (This is why one cannot
substitute a singular term denoting truth for the sentence beginning with
the horizontal in the expression' I- 2 + 3 = 5' (which would yield the
ungrammatical 'I the True').) It is here that the distinction between
sentences and terms finds its representation within Basic Laws. I I
The change regarding the grammar of the horizontal that Frege
makes between Begriffsschrift and Basic Laws is partly motivated by the
grammatical assimilation of sentences to terms. But this motivation is
less important than one might think. For in one sense the grammatical
assimilation of sentences to terms was already present in Begriffsschrift.
Insofar as this is so, the view that adoption of the position effected a
major change in Frege's later philosophy of language is rendered further
implausible. In Begriffsschrift Section 3, Frege writes:
A language is imaginable in which the sentence 'Archimedes perished at the capture of
Syracuse' would be expressed in the following way: 'the violent death of Archimedes at
the capture of Syracuse is a fact'. Here one can, if one wishes, distinguish subject and
predicate; but the subject contains the whole content, and the only purpose of the
predicate is to present this as a judgment. Such a language would have only a single
predicate for all judgments, namely 'is a fact' .... Such a language is our Begriffsschrift,
and the sign' I- ' is its common predicate for all judgments.
Here Frege is primarily intending to make the point that the subject-
predicate distinction of natural language has no comparable importance
in his logical theory. But the passage also indicates a more radical point
of view that the "content" of the first sentence can be completely
captured by the subject, a term, in the second. The sign 'f--' is seen in
Begriffsschrift as a predicate that adds nothing to the content of the term
to which it applies. This viewpoint contains more than the germ of
Frege's later commitments to the grammatical assimilation of sentences
to terms and to the redundancy conception of truth.
The changes from this position in Begriffsschrift to his later stance in
Basic Laws are fairly easy to separate out. In the first place, Frege more
clearly distinguished in the sign 'f--' an element corresponding to
jUdgmental force and an element corresponding to the expression 'is a
fact' or 'is a truth'. (The running together of force with semantical
attribution occurs elsewhere in the Begriffsschrift. Cf. for example the
semantics given in Section 5.) Thus, the vertical judgment stroke repre-
sents judgmental force, and the horizontal alone comes to represent a
semantical predicate, such as 'is a fact' or 'is true'. Presumably this
distinction is accompanied by the rejection in 'Function and Concept' of
the Begriffsschrift view that the sign' f-- ' is a predicate (G & B, 'F & C'
34/KS 137 - quoted above). On the other hand, the horizontal, taken
alone, is a predicate whose meaning is similar to that of 'is a fact'.
Distinguishing force from predication in the sign 'f--' probably
made it easier for Frege to relax the Begriffsschrift restriction against
following the sign' f-- ' with anything but a judgeable content. What was
asserted need not be just what followed the horizontal, it could be the
predication of the horizontal onto what followed it. As I have noted
earlier, Frege had already in Begriffsschrift come to view predicates as
function signs (Section 9). Given that he was also already treating the
grammar and "content" of sentences as equivalent to that of terms that
nominalize those sentences, it may have seemed a small step to allow
the horizontal to be functionally applicable to all terms, simple and
complex, clausal and nonclausal. The vertical judgment stroke could still
only apply to sentences, the expression of somethingjudgeable.
What seems to be the large step in this development, in addition to
the distinction between force and predication, is the semantical clarifi-
cation that Frege achieved. The semantical standpoint developed in
Theses (c)-(d) in effect answered the question of what the arguments
and values of the horizontal should be. The redundancy conception of
truth is the natural offspring of this semantical development and the
viewpoint expressed in Begriffsschrift Section 3. The universal predicate
'is a fact' gives way to the universal predicate 'is the true'. But in neither
case does the predicate add to the "content" (sense) of what followsP
The predicate does not change the sense (or denotation) of the results of
ordinary predication.
Thus the grammatical change in the restrictions on the horizontal
between the Begriffsschrift and Basic Laws is not really a change from
allowing only sentences to occur to allowing terms to occur. It is from
allowing only the occurrence of terms that nominalize sentences to
allowing all terms. Thesis (c) played a role in motivating this change. But
the developments associated with the changed use of ' 1-' that seem
most significant are different. The significant developments are Frege's
drawing the semantical consequences of viewing predication as func-
tional application (not the mere viewing of predication as functional
application, which is already present in Begriffsschrift), and the clear
distinction between judgmental force and predication. This latter devel-
opment, and the prominence Frege gave to truth and judgment in
motivating logical theory, undermine any claim that the grammatical
assimilation of sentences to terms deprived him of his insight into the
basic role of sentences in logical theory.
The claim that truth values are objects inevitably suggested to Frege the
question 'Which objects?' A parallel question had arisen in The Foun
dations of Arithmetic, once it had been concluded that numbers were
objects. Frege was sensitive to the initial possibility that the answer to
the latter question might be no other than 'why, the numbers - 0, 1, 2
.. .' Similarly, the truth values might tum out to be specifiable only as
truth and falsity. But Frege's belief in logicism drove him to seek a
different answer in the case of numbers. Similar forces were at work in
his views on truth values.
In The Basic Laws of Arithmetic the truth values are identified with
particular logical objects, particular extensions of concepts. The reason-
ing behind this identification is the subject of this final section. Our
discussion of this subject must be more conjectural than that of Theses
(a)-(d) because Frege wrote very little directly about it. Nevertheless,
by piecing together different strands of his views, it is possible, I think,
to weave a pattern that has some interest, and even a kind of blemished
One reason why Frege's reasoning is interesting is that it sheds light
on his conception of logical objects. Another is that it is critical to
assessing what sort of realism Frege maintained with regard to such
objects, and with regard to numbers. Each of these issues is quite
difficult and complicated. I shall begin with some very rudimentary
background for the realism issue.
A common and straightforward story about Frege's realism goes as
Frege believed that the numbers were genuine, existing abstract objects. He thought,
however, that number theory was reducible to logic. He proceeded to try to show this by
constructing a logic containing a version of set theory. He gave definitions, within the
logic, of the primitive expressions of number theory, and tried to derive the axioms and
theorems of number theory within his logic. Since he had a realist attitude toward the
ontologies of the languages of both number theory and logic, and since he regarded
numbers as particular objects, he thought that there was but one way to construct the
definitions of numerical expressions within his version of set theory. As it turned out
Frege's set theory is inconsistent; and for any viable set theory there are an infinite
number of ways of defining arithmetic within it. So even if his set theory had been
consistent and even granting that set theory is logic, Frege's logicism and his realism
about the numbers are, if not incompatible, at least deeply at odds.
There is much that is right about this familiar recitation. But it seems
to me misleading in some fundamental ways. The first derives from
Frege's attitude toward all language other than his own concept script. It
is well known that Frege thought that natural language was defective for
the purpose of expressing thOUght. But he also thought the same of
mathematics itself. Within mathematics, the problem was partly just that
the language had not been given logical form. But vagueness was also a
problem. Frege repeatedly notes that the content or sense of the term
'number' is not adequately or sharply grasped by even the most com-
petent mathematicians. Other fundamental, long-standing arithmetical
terms are afflicted by vague usage. In the first section of The
Foundations of Arithmetic, he writes:
The concepts of function, continuity, limit and infinity have been shown to stand in need
of sharper determination. Negative and irrational numbers, which had long since been
admitted into science, have had to undergo closer scrutiny of their credentials. In all
directions these same ideals can be seen at work - rigor of proof, precise delimitation of
extent of validity, and as a means to this, sharp grasp of concepts.
In the introduction to the book, referring indirectly to the concept of
number, Frege writes, 'Often it is only after immense intellectual effort,
which can continue over centuries, that a concept is successfully recog-
nized in its purity, stripped of foreign coverings that hid it from the eye
of the intellect' (p. vii). (Cf. KS 122.) Clothing, covering, veiling are
standard Fregean metaphors for the interferences of language in
thought. This theme runs throughout the book. Indeed, the book may be
fruitfully read as an attempt to remedy the inadequacies of language
(primarily mathematical language) for ideally rational conceptualization
and thought. If one substitutes 'perception' for 'language', one has the
schema for the traditional rationalist program for freeing the intellect
from non-rational factors.
Any number of senses and denotations were compatible with the
conventional significance of vague terms. It is clear that Frege thought
that conventional mathematical usage left mathematical terms vague.
That is, what a conventionally competent speaker masters does not fix a
definite sense or denotation. Frege thought that his logicist program was
required to uncover the senses and denotations of number words. So
strictly speaking, defining arithmetical terminology is not a matter of
capturirig linguistic meaning as we commoly understand it, but of
uncovering supplementations of such meaning so that the terminology
can be seen to have a definite sense and denotation. Sense lay beyond or
beneath conventional significance. (These points are discussed and
substantiated in some detail in Burge, 1984.)
It would be a mistake to infer from this point that Frege held that
defining numerical terminology involved stipulating meanings for it.
The definitions had to respect mathematical practice. Moreover, I think
it plausible that Frege thought that only one set of definitions (his)
respected all relevant philosophical considerations. Frege's point is that
by merely understanding the linguistic meaning of ordinary mathemati-
cal language, by being a competent participant in the conventions
governing it, one did not thereby attain a completely adequate grasp of
numerical concepts; one did not thereby secure completely definite
denotations for number words.
Thus if Frege did think that there was a single, correct set of logicist
definitions, it was not because the ordinary conventional meaning of
mathematical language and standard mathematical practice allowed only
one such set. Since mathematical usage was vague, it admitted of
many sharpenings. Rather the unique propriety of a set of logicist
definitions would have to depend partly on philosophical considera-
tions, considerations attendent on the logicist program.
Thus Frege's realism about the ontology of ordinary mathematics is
more subtle and qualified than the familiar narrative spun above
suggests. His realist attitude toward the language of mathematics is tied
to and supplemented by the assumption that his logical theory gives a
proper account of the objects and Junctions in the mathematical world.
Even with respect to his logical language, his realism is not com-
pletely unqualified. Not every logical constituent of the language corre-
sponds to an item in reality. For example, the function sign negation
does not in general correspond to a thought component. (PPW 149-50,
185, 1981NS 161-2, 201, 214; Cf., however, G & B "N" 131-21KS
Nevertheless, Frege thought of his logic as a tool for discovering the
nature of the mind-independent world, at least that portion of the world
with which mathematics was concerned: " ... The mathematician cannot
create something at will, any more than a geographer can; he too can
only discover what is there and name it" (FA 107-8). The theme runs
through Foundations, his correspondence with Hilbert, and his attack
on the formalists; it emerges in the introduction to Basic Laws, and it
recurs in his late writings. Functions, thoughts, and (at least until the
despair over Russell's paradox) courses of values are among the charter
members of the mind-independent world. Although this traditional
"realist" interpretation has been questioned now and again, I think it
fundamentally secure and will not argue for it in general terms here.
A second way in which the familiar account of Frege's realism that I
recited above is misleading concerns the references to set theory. I will
not go into this complicated and somewhat obscure matter here. (I have
discussed it in Burge, 1984.) I will just say bluntly that it is a
mistake to think of Frege's theories of courses of values and extensions
of concepts primarily in terms of what we now know as set theory. This
is not because Frege's theory turned out inconsistent. It is because- he
consciously and repeatedly argued against the basic intuitions that
underlie the iterative conception of sets, and because the fundamental
intuitions underlying his own theory have only scattered echoes in
mainstream set theory and even in the various nonstandard versions.
The question of whether and in what sense Frege's logicism and his
realism about numbers are affected by the multiplicity of models of
arithmetic within set-theory is clearly bound up with these matters.
This is a question that I shall not attempt to settle here.
If Frege's extensions of concepts were thought of set-theoretically,
the axiom of foundation would be everywhere violated. This is a prime
reason for feeling queasy about the set-theoretic explication. This reason
is a corollary of a more fundamental one - the primacy for Frege of
concepts over classes, or courses of values. It is better to think of
extensions of concepts visually in terms of a geometrically represented
graph, or yet better as the total (abstract) event of matching each of the
arguments with their truth values, one by one. Needless to say, these
heuristics give one only a vague start at the notion. (As logical objects,
extensions of concepts were not supposed to be dependent for their
conception on intuition or vision.) Frege never achieved a dear con-
ception of extensions of concepts that accorded with his philosophical
and mathematical preconceptions. It is arguable that no one else has
either. So we must be willing sometimes to grope along in the dim
afterglow of Russell's devastating paradox if we are to follow the course
of his reasoning.
The common thrust of the two main caveats that we have entered in
the familiar story about Frege's realism is that we need to be sensitive to
the role of his philosophical considerations, beyond what we now think
of as standard mathematical (arithmetical and set-theoretic) practice, in
assessing the character and plausibility of Frege's realism about mathe-
matical and logical objects.
How do truth values figure in all of this? They are fundamental in
Frege's notion of the extensions of concepts, a subset of which consti-
tutes the primary logical objects. Since concepts are functions from
arguments to truth values, and extensions of concepts are courses of
such values, the truth values chart the courses. Since the numbers are
certain extensions of concepts and since truth values thus figure essen-
tially in the ontology of the numbers, consideration of them is insepar-
able from consideration of Frege's realism about numbers.
But there is a more specific reason for ontological interest in truth
values. Frege identifies not only the numbers but the truth values
themselves with courses of values, extensions of concepts. Frege indi-
cates that his identification of the truth values with the specific courses
of values he chooses in Section 10 of Basic Laws is arbitrary relative to
the axioms of his logical theory. Any other choice would have been
equally consistent with those axioms. Getting straight what Frege means
in this passage, which we shall scrutinize shortly, is critical for under
standing his whole philosophical standpoint.
On its face, the passage in Section 10 of Basic Laws suggests that
Frege's theory of truth had a large stipulative component. It also
suggests that Frege's ontology of the numbers contains a massive dose
of arbitrariness, and that he was aware of this. If these suggestions are
correct, then the traditional view of Frege's realism and of the intentions
governing his logicist project must suffer substantial qualification. For
different choices as to how to identify the truth values with extensions of
concepts would yield different accounts of which objects the numbers
I believe that these initially plausible suggestions are mistaken.
Although I shall stop short of a general discussion of how Frege
regarded his definitions of the numbers, I shall argue that his identifica
tions of the truth values were, and were known to be, not in the least
arbitrary, but supported by reasons. To understand these reasons, we
must consider the philosophical context in which Frege conceived his
logicist program. The reasons are not narrowly mathematical. Their
failure to appear in Basic Laws accords with the predominantly mathe
matical emphasis of the book.
In Section 10 of Basic Laws Frege correctly argues, first, that
whether or not one or both truth values are courses of values and,
second, which courses of values they are, granted that they are courses
of values, is left completely undecided by the axioms of his logical
system (in particular by Axiom V). He concludes:
Thus without contradicting our setting 'tq,(e) = t'P(E)' equal [in denotation] to '(x)
(q,(x) = 'II (x,' it is always possible to stipulate th,!t an arbitrary course of values is to
be the True and an arbitrary different one, the False. Accordingly, let us lay down that t
(-e) is to be the True and that t(e = - (x) (x =' x is to be the False.
't4>(e) = l'(e)' is read 'the course of values of the concept 4> is
identical with the course of values of the concept lJI'; 't( -e), is read 'the
course of values of the concept is the True'; 't(e = - (x) (x = x))' is
read 'the course of values of the concept being identical with the truth
value of not all objects' being self-identical'. This passage certainly
appears to support the view that Frege's choice is a matter of stipulation.
And, of course, it is in two ways. The choice involves "stipulation"
relative to the commitments of natural language. As Frege often notes,
ordinary uses of 'true' and 'false' do not explicitly commit themselves to
truth values as objects at all. The commitment is promoted by logical
theory. Thus relative to natural language use, the identification of any
object as one of the truth values is arbitrary. Frege's choice also involves
stipulation relative to considerations of consistency with Axiom V, and
the other axioms of his system. Within a context in which mathematical
consistency is all that matters, the choice is arbitrary. But let us broaden
the context.
On numerous occasions outside of Basic Laws Frege holds that
consistency does not suffice for truth. Frege repeatedly defends this
view in opposition to early expressions of the model-theoretic viewpoint
toward mathematics. Frege's best known defenses of the view occur in
his correspondence with Hilbert (1895-1900) and in 'On the Foun-
dations of Geometry, I' (1903) (e.g. Cor 48/BW 75; FG, 25-37/KS
264-72). But the view is already quite explicit in 'On Formal Theories
of Arithmetic' (1885) (KS, 110). And it clearly guides the criticism of
formalism in the closing pages of Foundations (pp. 104-119).
These latter passages are particularly relevant to our theme.
notes that the denotation of 'the square root of -1' was not fixed by
mathematical usage prior to the advent in mathematics of complex
numbers (pp. 106-7, 11 On). He then ridicules the view that one can
simply introduce a denotation for the term by mere stipulation or
definition (pp. 107-8). One reason Frege gives is that even granted that
the purported definition is consistent, one is not thereby guaranteed that
there exists an object that satisfies the concepts used in the definition
(pp. 108ff.): in effect, consistency does not entail truth. A second reason
is that even if one succeeds in attaching the term to an object and even
if one stipulates meanings for the usual mathematical function signs
in application to this object that are compatible with those mathematical
principles that had been established prior to the introduction of com-
plex numbers, there might be philosophical considerations that militate
against the definition.
Let us elaborate the second reason in more detail since it bears
directly on the treatment of truth values in Basic Laws. Frege notes that
simultaneously with the introduction of new numbers, the meanings of
functional words like 'sum' and 'product' are extended. Suppose we
choose some object as the denotation of 'the square root of -1', say, the
Moon. So the moon multiplied by itself is -1: This explication seems to
be permitted because the [denotation] of such a product does not at all
arise from the erstwhile denotation [Bedeutung] of multiplication, and
therefore in extending this erstwhile denotation it [the denotation of the
product] can be arbitrarily determined' (p. 110).17
Frege goes on to consider multiplication and addition as applied to
imaginary numbers, and in so doing capriciously takes the time interval
of one second, instead of the Moon, as the denotation of 'the square
root of -1'. He summarizes by saying, 'one is tempted to conclude: Thus
it is quite immaterial whether i denotes a second or a millimeter or
anything else, provided only that our laws of addition and multiplication
hold good; that alone is what matters; the rest need not trouble us' (p.
Frege does not accept this position. One point he makes against it,
less interesting for our proposes, is that a contradiction may lurk
between the definitions and the rest of mathematical theory. There is no
evidence that Frege thought that the relevant definitions in fact lead to
contradiction. Frege's other objection is philosophical. By letting the
interval of a second be the denotation of 'the square root of -1',
we are bringing something quite foreign, time, into arithmetic. The second stands in
absolutely no intrinsic relation to the real numbers. Propositions proved with the aid of
complex numbers would be a posteriori judgments, or at least synthetic, unless we could
find some other proof for them, or some other [denotation) for i. We must at any rate
first make the attempt to show that all propositions of arithmetic are analytic (p. 112).
For two sections Frege develops the theme of not importing anything
foreign into arithmetic. And he ends the book by recapitulating it: by
explicitly defining the numbers as extensions of concepts, one can avoid
importing physical objects or geometrical intuitions into arithmetic (p.
These sections of Foundations provide an initial clue to understand-
ing Frege's remarks about truth values in Section 10 of Basic Laws. Like
complex numbers, truth values are seen by Frege as introduced (recog-
nized) for theoretical reasons. Their introduction also extends previous
mathematical and natural-language usage. And a variety of ontological
choices are compatible with that usage. In showing that Axiom V does
not fix the denotation of the course of value notation (or of the expres-
sions 'the True', 'the False'), Frege is indicating, as he does in his attack
on Hilbert and the formalists, that the (partially interpreted) axioms of a
theory do not by themselves fix the objects of the theory (or the senses
of its terms). One must have an understanding of the senses and denota
tions of the terms used in the axioms that is not reducible to mere
commitment to the truth of the axioms taken as linguistic objects. (Cf.
Note 2 and the accompanying text in Section I.) In stating that the
stipulations as to the identity of the truth values are arbitrary relative to
previous usage and previously stated axioms, however, Frege is not
stating that his stipulations are arbitrary, absolutely speaking.
What we need now is to understand the reasons for the particular
identifications Frege proposes in Section 10 of Basic Laws. I shall
approach his position by a process of elimination. The Foundations
passages already make it clear why the truth values could not be
identified with physical, mental, or geometrical objects. The domain of
logic is universal, whereas these objects are the topics of special
sciences, and thus their natures are explicated in synthetic propositions
(Cf. Foundations, Section 3.) Moreover, physical and mental objects
exist contingently and are known by a posteriori methods. Logic encom
passes the necessary and is fundamentally a priori.
Similar considerations seem to rule out identifying truth values with
any course of values associated with a function or concept denoted by a
term that is a primitive of, or is definable with primitives of, one of the
special sciences. Thus the extension of the concept is a cat (is an image,
is a line) is inappropriate.
Truth values could not be identified with senses because of the
arguments of Thesis (b) that we considered in Section I. A corollary of
these arguments is that such identification would be inappropriate
because senses are denoted primarily in oblique contexts, whereas a
truth value is denoted in the expression of any thought. Since, by Thesis
(d) truth values are objects, and since Frege thought no function (or
concept) is an object, truth values could not be identified with functions.
These considerations leave Frege either with identifying truth values
with the course of values of some logical function or with not identifyinK
them with any courses of values, taking them rather as primitive,
"independent" logical objects.
We have been ruling out possible identifications by appeal to what is
foreign to logic. In order to proceed further, we need to recall what
Frege saw as essential to logic. Logic essentially concerns itself with the
laws of truth. As we have seen, Frege sharpened this claim by stating
that the laws of truth yielded those norms governing ideal assertions or
judgments (PPW 2521NS 272; Kl 507-SIKS 343). Truth was the
"objective" of judgment; the most general laws governing this objective
formed the subject matter of logic. Thus the true is in a sense the most
basic logical object. We shall return to this point and sharpen it.
The conception of truth as the aim of logic informed Frege's view
that logic had an internal ordering. Logic is founded on the proposi-
tional calculus, or the calculus of truth values. Frege repeatedly empha-
sizes that one of his seminal contributions is to begin in logic with the
sentence and to derive an analysis of sentential parts and their various
semantical functions only in the context of a semantical analysis that
already features logical relations among sentences: the functional calculi
are built upon the calculus of truth values. We have cited various
passages to this effect in his arguments for the view that truth values are
the denotations of sentences (Theses (a) and (b. Frege also makes the
point in his earliest and latest writings (PPW 17, 253/NS 18-19,273).
The sense of an ordering within logic is perhaps most clearly enunciated
in a footnote Frege wrote in 1910 to Jourdain's chapter on Frege in a
history of mathematical logic:
To found the 'calculus of judgments' on the 'calculus of concepts' ... is to reverse the
correct order of things; for classes are something derived, and can only be obtained from
concepts (in my sense). But concepts are something primitive that cannot be dispensed
with in logic .... And the calculation with concepts is itself founded on the calculation
with truth-values (which is better than saying 'calculus of judgments') (Cor 192n!NS
The truth values are on the ground floor of logic - in the ontology of
the propositional calculus. They are a "subject matter" for all parts of
logic. In a sense to be explicated, truth is even more basic than falsity.
The laws of logic were for Frege "nothing other than an unfolding of the
content of the word 'true'" (PPW 3/NS 3; cf. Kl 507/KS 343).!Now
courses of values were supposed by Frege to be logical companions of
functions, and functions were denoted in all parts of logic. Prior to
discovery of the paradox, each function was thought to be accompanied
by its associated course of values. Logical objects such as courses of
values could be canonically specified only through denoting the
associated functions. Frege repeatedly emphasizes that (denoting) a
course of values is derivative from (denoting) a function. A function
sign denoted a function, but its use determined an associated course of
values. Frege seems to have regarded sentences containing function
signs as ontologically committed to their associated courses of values.
(See G & B 27, 49-50/ KS 130, 173; the point is also suggested in BL
4/001,7 and PPW123/NS 134.)
Since Frege saw logic as having a fixed order, his taking truth values
as part of the ontology of the propositional calculus meant that his
specification within logic of these objects could not depend on concepts
whose specification was conceptually derivative.
These considerations suggest an argument for taking truth, or the
True, to be specifiable in terms of a concept that is primitive within the
propositional calculus, assuming that truth is a course of values. Since all
logic is concerned with truth and is in fact the unfolding of the laws of
truth, and since truth is an object, truth must be ia the ontology of all
parts of logic - in particular the most fundamental part, the proposi-
tional calculus. For a course of values to be in the ontology of the
propositional calculus it is necessary and sufficient that it be specifiable
in terms of functions denoted in the propositional calculus. So assuming
that truth is a course of values, it must be specifiable in terms of
functions within the propositional calculus. The only such functions are
One might worry about the argument, both as a reconstruction of
Frege and as a substantive proposal, by concentrating on the second
premise. Let us assume with Frege that to be specifiable at all, a course
of values must be specifiable in terms of its associated functions. But
why iS'it necessary, in order to be in the ontology of the propositional
calculus for a course of values to be specifiable in terms of functions
denoted by expressions in the propositional calculus? And is denoting
certain functions really sufficient for being onto logically committed to
their associated courses of values?
Of course, with the hindsight that we have gained from the semantical
and set-theoretic paradoxes, both questions can be pressed. And I
would not wish to defend a Fregean answer to either. It is arguable that
sometimes the ontological commitments of a language (say, those
involved in giving a semantical theory for it) are specifiable only in a
stronger metalanguage. And since Frege's Axiom V is inconsistent, it is
sometimes the case that commitment to a given function is not sufficient
for commitment to an associated extension (a course of values). This
problem makes it plausible to deny that in using function expressions
one has dual commitments, to a function and a course of values, even in
cases where there is an associated course of values. One can make one's
commitments separately. But these problems are bound up with larger
problems concerning Frege's conceptions of truth and courses of values
- particularly with the inconsistency of his system. They do not under-
mine the argument we gave as an interpretation of Frege.
How would Frege answer the two questions about the second
premise? To the first, I think that he would reply that a language that
lacked the concepts (or could not denote the concepts) needed to
specify its ontology would be logically deficient. Such a language could
certainly not serve to express the fundamental part of logic. A logical
language that could not specify its own ontology would be dependent on
some other language to specify its primary subject matter, truth. Thus it
could not be the fundamental language for expressing the laws of truth.
But Frege regarded logic as an ideal language of thought; the fundamen-
tal part of that language should be complete for its own
This point should be seen in the light of Frege's redundancy concep-
tion of truth. Frege believed that semantical discourse about a language
could add nothing to what could already be said in the language itself. It
is obvious that he did not anticipate the sorts of considerations that lead
us to be cautious about claims that a language must be able to specify its
own ontology.
The second question about the second premise of our argument is
less interesting insofar as it bears on the interpretation of Frege. The
truth values are in fact in the ontology of his propositional calculus and
they are in fact specifiable in terms of functions (concepts) that are
denoted in the propositional calculus. Since the truth values are courses
of values in his view, and since he thought courses of values could be
specified only in terms of their associated functions, it is hard to see any
ground for denying that he thought that being so specifiable was suffi-
cient for being in the ontology of the language. (For a discussion of texts
that suggest that Frege thought that language always carried dual
commitments to concepts and their associated extensions, see Burge,
1984, Sections III-IV.)
I have argued that Frege conceived truth as the subject matter of the
most basic part of logic, and that truth had to be specifiable by means
of the primitive predicates in the propositional calculus. Frege's con-
ception of an ordering within logic motivates a corollary restriction.
Truth should not be the extension of a concept (or course of values
of a function) whose specification is in any way conceptually deriva-
tive. Truth must be specifiable in terms of a concept whose own specifi-
cation need not presuppose other types of concepts. This consideration
would seem to rule out identifying truth any course of values of a
second-level function (including second-level concepts). For the second-
level functions are introduced in logic only after, and only in terms
of, first-level functions. The consideration also seems to rule out iden-
tifying truth with any course of values of a function, of any level,
whose canonical explication presupposes specification of
The effect of these exclusions is substantial. Two large categories of
logical objects are barred as candidates for being identified with the
truth values. First, the courses of values that are not extensions of
concepts are excluded. There is only one purely logical, primitive first-
level function sign that is not a concept sign, or predicate, in Frege's
logic. This is the description operator, and its explication presupposes
specification of the course-of-values operator, which is second-level (at
least!). Of course, the course of values associated with the course-of
-values function sign 'f' is excluded. For the function it denotes is (at
least) second level. A second category of logical objects that is excluded
consists of the extensions of concepts with which the numbers are
identified. For these are extensions of second-level concepts. In fact, the
definition of 'equinumerous', which is essential to specification of the
numbers, depends on second-order quantification - i.e. third level
So, if the truth values are to be identified with courses of values at all,
and if Frege's philosophy of logic is to be respected, it appears that they
must be identified with the extensions of logical concepts that are first-
level. Such concepts ought to be denoted in the propositional calculus.
What underlies the path of exclusion that we have so far followed,
and indeed what guides us to our destination, is Frege's conception of
truth. The argument we have just given is rather speculative, considered
as an interpretation of Frege's own reasoning. The considerations that
follow are much less so.
Frege identifies truth (the True) with fe-e). '-', the horizontal,
denotes the concept that maps truth onto truth and everything else onto
falsity. Thus the horizontal denotes the concept under which only truth
falls. Truth is the extension, or course of values, of this concept. Falsity
is the extension of the concept under which only falsity falls.
Now this identification can seem to be a piece of artifice if one thinks
of courses of values purely in set-theoretic terms. From this persp'ective
truth (falsity) is identified with its own unit class. (Cf. Dummett, IFP,
404.) What could be more typical of a mere technical convenience? But
the set-theoretic gloss misrepresents Frege's view.
Frege's identification should be seen as the result of drawing out the
implications of the two ideas that we have discussed so far, and supple-
menting them with his redundancy conception of truth. The first idea
was that the truth values are logical objects. Their specification should
not "import anything foreign into logic". Further, as logical objects, their
specification must be derivative from the specification (or denotation) of
logical concepts. The second idea is that logic is an ordered unfolding of
the laws of truth, where truth is the aim of sentence use within logic.
Truth must somehow be the objective and subject matter (object) of all
parts of logic, including the most primitive part. In fact, it must some-
how be implicated in the aim of every sentence of logic. As we have
seen in earlier sections, Frege interprets the logically relevant aim of
sentence use in terms of our "striving after truth". This aim is revealed in
assertion and judgment. So truth must somehow be implicated in the
assertive use of every sentence of logic. Putting the two ideas together,
we seek a concept in terms of which we can specify truth as a logical
object, a concept that is present in the assertive use of every sentence of
It might appear that we have an approximation to the idea that truth
must be implicated in the assertive use of every sentence of logic, in
Frege's doctrine that every sentence of logic denotes truth or falsity. As
we have seen, in Section I, there was a connection in Frege's mind
between the point of sentence use and the denotation of sentences. But
truth is the aim of logic; falsity, strange to say, is not. This aim is
revealed in assertion, not simply in the grammatical form of sentences.
The concept in terms of which truth is specified is present in every
assertive use of a sentence, whereas the counterpart concept for falsity is
not. Thus the specification of truth is philosophically primary. The
specification of falsity will present itself as natural once we have under-
stood Frege's specification of truth.
The only concept that fits the requirement of being present in the
assertive use of every sentence of Frege's logic is his concept of truth,
the concept denoted by the horizontal. In unpublished writing contem-
poraneous with Basic Laws, Frege is quite explicit about the point. He
held that what distinguishes 'true' from all other predicates, and what fits
it to indicating the a'im of logic, is that "it is asserted when anything at all
is asserted" (PPW 129/NS 140). In accord with his redundancy concep-
tion, Frege held that 'it is true that _', when filled by a declarative
sentence, expresses the same sense as '_' when filled by the same
sentence. But this neutrality of sense in predications of 'true' is collateral
with the predicate's omnipresence in assertions and judgments. As we
have seen in Section II, this idea of the omnipresence of the truth
predicate traces all the way back to the Begriffsschrift (B, 3).
The horizontal expresses the notion of truth in Frege's system. It
means 'is the True' or 'is the truth' or 'is truth'.18 It is present in the
formulation of every assertion. It may accompany any declarative
sentence without adding to its sense. The concept denoted by the
horizontal is the only one within Frege's logic that meets the condition
set by his redundancy conception of truth. Frege alludes to this condi-
tion without fanfare in Section 5 of Basic Laws, where he notes the
/:l. = (-/:l.)
where '/:l.' varies over truth values. The import of the condition comes
clear if one sees sentences as substituting for '/:l.', reads '=' (as in such
cases one may in Frege's system) as the material biconditional, and
reads the horizontal as the truth predicate. The equivalence is the analog
within Frege's system of Tarki's truth schema.
Given his redundancy conception, Frege regarded the two sides of
the equivalence as having the same sense when a sentence is substituted
for'S. (Cf. G & B 63-4/KS 149-150.) This suggests that the concept
of truth may in a sense be implicated in the ontology of every (declara-
tive eternal) sentence, whether the sentence is asserted or not, and
regardless of whether a truth predicate explicitly occurs in the sentence.
This conclusion needs the assumption that a sentence is committed to
the existence of the extension of a concept C if C is a denotation deter-
mined by (a component of) the sense that the sentence expresses. Since
every sentence has the same sense as a sentence in which the concept of
truth is denoted, every sentence would by this reasoning be committed
to the existence of the extension of this concept - the truth value truth.
Although I think that there is some reason to believe that Frege
considered and accepted the relevant assumption, he did not explicitly
assert it. (Cf. G & B 27, 49-50/KS 130, 173; BL 4/GGJ 7; PPW 123/
NS 134; the numerous passages where he speaks of the senses of
sentences as being decomposable into component senses; and Burge,
1984.) I shall not defend the attribution of the assumption, however,
since I regard it as somewhat speculative.
What is indisputable and important is that Frege forces the truth
concept to be explicitly denoted by a truth predicate (the horizontal)
when a sentence is assertively used in his logic. Thus it is through
assertion that the aim and ultimate subject matter of logic are revealed.
The truth concept also expresses Frege's conception of the order
within logic. It is the concept in terms of which all others are explicated
and understood. In this regard, it is prior to any other concept canoni-
cally denoted in the propositional calculus.
In specifying truth as the extension of his truth predicate, Frege is
"deriving" a logical object from a logical concept. Obviously, it is
intuitively natural to derive the object truth from the concept of truth
(assuming that one wants such an object and that such an object must be
derived from a concept). Frege is in effect nominalizing the truth pre-
dicate by generalizing on his truth schema (in the light of his construal
of the schema in terms of Theses (b) and (d.19 But Frege's specification
of the object truth is not merely intuitively natural, granted his assump-
tion that it is a logical object and must be derived from a concept. It also
expresses his conception of the point, subject matter and order of logic,
and flows from his redundancy conception of truth. It is the only
identification that is consonant with his philosophical views.
The evidence we have been considering suggests reconsideration of a
pair of long-standing criticisms of Frege's view of truth values as objects.
One is that once truth becomes one object among others, it is difficult to
explain what it is about it that makes us want to strive after it, assert it,
acknowledge it, and so forth. (Cf. Furth, BL, pp. lii-liii. The point is
also made by Dummett in various places in FPL.) What is so terrific
about the relevant object? Frege seems to be inviting us to join a kind of
secular religion without explaining the attributes of its god that merit
our worship.
We may begin to appreciate the weakness of this criticism by
recalling Frege's contextualist arguments for Theses (a) and (b). Denota-
tion of objects with terms is of interest only because of our interest in
semantical features of sentences. Interest in sentences derives from
interest in norms governing their use in making assertions and express-
ing judgments. So the practice of using sentences to denote truth values
derived its interest from the role of assertion and judgment in our lives.
From Frege's point of view, the idea that we must explain this role in
terms of the features that certain objects, the truth values, have, would
be to put the horse behind the cart.
Frege underlines this point by specifying the object truth as what
every assertive use of a sentence (and every judgment of a thought) is
committed to. We know what the object truth is and what it is like only
through reflecting on the fact that assertion and judgment aim at it. To
be sure, Frege thinks that the laws of truth, which are laws of "being"
(Sein, Kl 507-SIKS 342-3), generate norms for assertion and judg-
ment. (Cf. Note 3.) This is because Frege assumes it as obvious that we
ought to judge in accordance with logical laws, and because he con-
strues these laws as governing an objective world of objects. But it is not
part of his view that we should be able to explain the interest for us of
the object truth, or the way that we think about it, independently from
consideration of the point of assertion and judgment - to "strive after
Frege thinks truth and judgmental force are primitive ideas. And he
does not try to explicate or philosophize about the value of "striving
after truth." But it is foreign to his system, and thus not a pseudo-
question concocted by it, to ask what it is about the object truth that
engenders our interest in it. The activity of jUdging and the practice of
assertion are primary.
The second longstanding criticism of Frege is closely related to the
first. Dummett states it in his thought provoking article 'Truth':
... it is part of the concept of truth that we aim at making true statements; and Frege's
theory of truth and falsity as the references of sentences leaves this feature of the
concept of truth quite out of account. Frege indeed tried to bring it in afterwards in his
theory of assertion - but too late; for the sense of the sentence is not given in advance
of our going in for the activity of asserting, since, otherwise there could be people who
expressed the same thoughts but went instead for denying them. (TOE, pp. 2-3)
It is certainly true that Frege said less about the roles of jUdging and
asserting in our lives than one might want in a post-Wittgensteinean
climate. But it is not true that Frege leaves our aim at making true
statements "quite out of account" in his exposition of the concept truth.
It is this aim, he says in 'On Sense and Denotation', that motivates our
asking for the denotations of terms and sentences. Late in life he writes:
" ... 'true' only makes an abortive attempt to indicate the essence of
logic, since what logic is really concerned with does not at all lie in the
word 'true' [since by the redundancy view, it contributes nothing new to
the sense of whole sentences in which it occurs as predicate] but in the
assertoric force with which the sentence is uttered" (PPW252/ NS 272).
The last main clause of the passage cited from Dummett's criticism is
difficult to interpret. But it does not seem relevant to Frege's view. Frege
nowhere, to my knowledge, writes or implicates that the sense of a
sentence is given "in advance" of our going in for the activity of
assertion. Denotation is motivated and justified by Frege in terms of our
"striving after truth." Sense is postulated as the way denotations are
presented to us in thought. Thoughts, the senses of sentences, are truth
conditions; and we are interested in truth conditions because we are
interested in arriving at truth. So our attaching senses and denotations to
sentences and sentential parts is motivated by their roles in judgment
and assertion. Frege's thesis that sentences denote objects is proposed in
the context of these motivations, not in contradiction to them.
We should now consider the possibility of taking truth and falsity as
"new" logical objects, not identical with any course of values. Frege
would not have introduced a new name for truth, an individual constant,
into his system since doing so would have created an inelegant logical
connection between his specification of truth and his truth predicate.
Moreover, such a move would controvert his view that logical objects
are derivative from logical concepts, a view bound up with his con-
textualist defense of the existence of abstract objects. But suppose that
one identified truth with \f(-e) (where the slash is the description
operator - cf. Section 11 Basic Laws) - the unique object that falls
under the concept - (is the true).
I think that Frege would answer this suggestion by asking rhetorically
what purpose a distinction between truth and the extension of the
concept is the true would serve. He would take \f(-e) to be fe-e).
(Compare the way Frege argues against postulating negative thoughts
(PPW 149-50,185, 198/NS 161-2,201,214).) Prior to the discovery
of the paradox, there were no evident logical advantages to the distinc-
tion. This point is implicated in Note 17 in Section 10 of Basic Laws.
It is arguable that Frege accepted the intuitive or metaphysical point
that a physical, mental, or geometrical object is distinct from any course
of values - on the ground that they are obviously distinct categories of
objects. (The argument would have to develop out of the central
sections of Foundations.) But such a point is evidently not applicable to
the relation between truth and courses of values. Truth is a necessary,
logical object, unlike physical, mental, and geometrical objects; and
since it is introducted on predominantly pragmatic grounds, against the
grain of prior intuition, there is no force in the claim that we make an
intuitive distinction between the object truth and courses of values.
From Frege's pre-paradox point of view, I see no purpose to the
distinction and so, for him, no point in drawing it.
I believe that the urge to draw such a distinction derives from
thinking that has already been informed by the set-theoretic and seman-
tical paradoxes. One wants to insist on a difference in "level" between
basic individuals and higher-order objects. Of course, merely taking
truth to be an individual will not suffice to deal with the semantical
paradoxes. Frege's redundancy theory of truth and his truth predicate
('is the True') are not capable of representing common uses of the
notion of truth, much less to explicating and representing the derivative,
indexical, and schematic aspects of the notion, aspects revealed by the
paradoxes. (Cf. Burge, 1979.) If there is not some pervasive provision
for levels of just the sort that Frege ignored, there can be no adequate
theory. The notion of truth cannot be adequately represented in terms
of a truth predicate that lacks any sort of stratification.
These remarks are, however, anachronistic. They presuppose knowl-
edge that Frege lacked when he wrote Basic Laws. Given his redun-
dancy conception of truth, his notion of course-of-values, his view of
truth as an object, and his logicist ambitions, the identification of truth
values that he proposed in Basic Laws Section 10 must have seemed
uniquely appropriate.
Truth is the basic logical object in Frege's system. It is the object on
which the purely logical, first level functions (including concepts) are
initially construed as operating (BL, Section 31). The logical objects
with which the numbers are identified are derivative from second and
third level concepts that operate on the first level functions. So the
numbers are less fundamental than the truth values (though all are, of
course, seen by Frege as necessarily existent). The reason truth is basic
is that it is the object (objective) of assertion and judgment. It is through
these activities that the point and the ontology of logic are revealed.
As we have noted, the truth values formed, at one time, the basis for
an argument against scepticism. They also formed the basis for Frege's
answer to Kant's dictum that there could be no knowledge of objects
without intuitions. Frege held that commitment to mind-independent
objects was inseparable from the very act of judging something to be
true. The idea is surely the quintessential distillation of the ambitions of
rationalist epistemology.
Even apart from these ambitions, the profundity and breadth of
Frege's conception must be seen as admirable. What enables Frege's
views on truth value to be of enduring importance is that they are so
largely grounded in fruitful and highly articulated insights into the
anatomy and function of logic, mathematics, language, and thought. I
think it unlikely that we have fully harvested Frege's insights.
University of California,
Los Angeles
I Abbreviations of titles listed in the bibliography, together with page numbers, will be
included in the text. Where German editions and translations of Frege's works differ in
pagination, both occurrences will by cited, separated by a slash. Responsibility for the
translations of all quotations from Frege is mine, although frequently the translations are
similar to and benefit from already published translations.
2 The fact that Frege's notion of term-denotation cannot be entirely separated from the
"name-bearer" relation has been appropriately emphasized by Michael Dummett in
FPL, Chapter 12; and in IFP, Chapter 7. For some views that proceed on the assump-
tion of the primacy of sentences (or their truth-values) and on the view that the notion
of the denotation of a term has no content other than that which is derivative from an
analysis of how the term functionally determines truth-value, see Quine (1960, Chapters
1-2; 1969, Chapters 1-2); Wallace, (1977); Davidson, (1977); Putnam (1981, Chap-
ters 1-2). Several of these authors explicitly invoke Frege's inspiration. I find the views
not only uncongenial to Frege (though unquestionably inspired by part of his doctrine),
but unpersuasive. But I shall not be able to go into these points here.
3 In calling Frege's notion of truth "normative", I am glossing over a very interesting set
of views that he held regarding the normative and descriptive aspects of logic. From the
beginning to the end of his career, Frege regarded logic as being descriptive of the laws
of logical objects, in particular those of truth. (PPW 31NS 3; K1507-81KS 342-3.) In
fact, Frege seems to have believed that in a sense logic was fundamentally "descriptive,"
fundamentally a science of "being." Normative restrictions on assertion and judgment
derived from "the way things are" regarding the laws of truth. (Kl 5081KS 342.) To
many this view of logic will seem quaint at best. I think that stripped of the particular
metaphysics with which Frege endowed it, and supplemented by Quinean and other
considerations, it can be made very powerful. In emphasizing that truth is a normative
notion, I am not ignoring the "descriptive" elements in his view. I am simply highlighting
a feature of Frege's methodology. Frege attempts to arrive at the laws of truth not by
invoking metaphysical assumptions but by concentrating on our practices of assertion,
judgment, and deductive inference and by developing his science of logic through
reflecting on the "oughts" of good intellectual practice.
4 Even he enters qualifications, (1956), p. 25 Note 66.
5 Prior to the argument's formulation, as early as 1903, Russell rejected the first
premise. He insisted that the relevant semantical notions for sentences are quite
different from those for terms. It is a bit open to question how seriously this rejection is
to be taken. Russell refused to call the relation between a sentence and a fact or pro-
position the relation of "naming" or "denoting." But he treats propositions or facts as
complexes made up entirely of entities that have traditionally been thought of as
referred to by words - properties, relations, individuals and the like. And he regards
these complexes as playing a central role in his semantical theory of sentences. Thus at
times Russell's point seems to come to little more than that sentences are not ordinarily
speaking names, a point with which Frege might well agree. Actually, of course, the issue
in Russell is quite complex. (Cf. e.g. the first lecture in 'Logical Atomism'.)
Russell remains at odds with the Church-Go del argument even in its less titillating
form - even after sentence denotation is construed as the yet-to-be-determined seman-
tical feature that is connected to the denotation of terms by means of the Composition
Principle (1). As I mentioned earlier, Russell may be interpreted as rejecting the third
premise of the argument. He accepted more or less the traditional conception of logical
equivalence and judged logical consequence in terms of the traditional modalized notion
of truth (e.g. PM Section A*I). But he took the primary semantical correlates of
sentences to be what he called "propositions" and sometimes "facts." If one interprets
Russell as accepting the first "premise" of the Church-Godel argument by granting it the
liberal conception of sentence-denotation that involves no commitment to Thesis (c) (so
sentence denotation is merely the central semantical feature of sentences in one's formal
semantics), then one must see Russell as rejecting the argument's third premise. For
then facts or "propositions" are sentence denotations; and logically equivalent facts
could, on Russell's view, differ.
What allows this position to remain compatible with the principle that exchange of
co-denotational expressions preserves truth value is, of course, the theory of descrip-
tions. This theory by itself blocks the Church-Godel argument by depriving one's
language (artifically, I think) of the definite description operator, or any comparable
device for forming complex singular terms that have denotations.
A discussion of the Church-Godel argument that is Russellian in its metaphysical
cast occurs in Barwise and Perry (1981). Unfortunately, the paper contains much that is
misleading. Frege's arguments are dismissed in two paragraphs. One paragraph charac-
terizes Frege's rhetorical question 'What else besides truth value is compatible with the
composition principle?' as a metaphysical oversight. This dismissal would perhaps
be fair if it did not ignore Frege's normative motivations and methodology. Frege's
argument from our primary interest in sentences, glossed in one sentence, is countered
by an irrelevant appeal to embedded sentences (irrelevant because the reason sentences
are interesting is that they are the vehicles of assertion and judgment). The
Church-Godel argument is discussed only on the unquestioned (but widely rejected)
assumption that sentences as wholes "designate" some entity. And it is resisted as if it
had been presented on this unquestioned assumption, and widely accepted, as a
"proof' from "virtually apriori" first principles. The incompatibility of Russell's system
with the argument's conclusion has been widely recognized. No one has tried to utilize
the argument to refute Russell, least of all Church and Godel, who were, of course,
thoroughly familiar with Russell's system. The role of the argument in the history of
semantics is more subtle than treating it as a proof from purportedly obvious first
principles could suggest.
6 The relation between Theses (c) and (d) is a complicated and subtle matter. Ontologi-
cally, of course, logical objects, such as truth values, and functions on these objects are
prior for Frege to singular terms and function signs. They exist before the signs existed,
and would have existed regardless of whether the signs did. On the other hand, Frege
held, on several occasions, that one could not engage in reasonably sophisticated
thought except by means of language. The analysis of thought relied heavily on analysis
of linguistic structure. But even in the analysis of thought, the analysis of language was
not prior in any simple sense. For thinking could and did correct the deficiencies of
language. In the light of all this, there is no simple answer to the question of whether
Frege reasoned from Thesis (c) to Thesis (d) or vice-versa. Part of why he concluded
that numbers are objects and numerals are terms was that he was able to give explicit
definitions which amounted to a criterion of identity for the numbers. On the other
hand, much of his reasoning to this conclusion was based on observations regarding the
structure of mathematical language. Similarly, his reasoning about Theses (c) and (d) is a
mixture of considerations regarding the role of objects in logic and the anatomy of the
language of logic (properly construed). I shall therefore treat Theses (c) and (d) more or
less together, without trying to sort out the various relations of relative priority that
obtain between them.
7 There are variants of the analogy between terms and sentences as regards their sense
and denotation that Frege mentions, but which I shall skim over. For example, he cites
his theory of indirect discourse as tending to confirm the introduction of truth values.
(BL 7/ GGX). The idea seems to be that, in indirect discourse, just as terms shift from
denoting their customary denotations to denoting their customary senses, so sentences
shift from denoting their truth values to denoting the thoughts they customarily express.
On Frege's conception, there need be no shift in grammatical category between the
occurrence of a sentence standing alone, and its occurrance in indirect discourse, which
is clearly a singular term position.
S One might be tempted to think that Thesis (a), or at any rate (b), already commits
Frege to taking truth values to be objects. For by our stipulation, an object is anything
that is denoted by a term. But the phrase 'the denotation of Sentence 5' (d. Thesis (b is
a term. So by Thesis (b), truth values are objects. This line of reasoning misses the fact
that the phrase 'the denotation of a sentence' need not be a term under logical analysis.
Similarly, for the phrase that begins Principle (1). 'Denotation' as applied to sentences in
the initial construal of Thesis (b) is guided by' the compositional method, loosely
expressed in Principle (1). As far as Frege's arguments for (a) and (b) are concerned,
there need be no entity that could be called under logical analysis "the denotation" of
the sentence. Of course, once Frege has committed himself to Theses (c) and (d), he can
consider Theses (a) and (b) to have proper construals, under logical analysis, that
commit him to taking truth values to be objects.
9 I do think that Dummett neglects to convey the. richness and inter-related nature of
the theoretical considerations supporting Frege's theses. His remark about Thesis (c)
that it is a "ludicrous deviation" from the forms of natural language and a "gratuitous
blunder" (FPL, 184) is, at best, immoderate. Incidentally, in FPL, Dummett spends two
thirds of his chapter 'Truth Value and Reference' on Frege's view that a sentence with a
non-denoting name thereby lacks a truth value. As I mentioned in Section I, I think that
Frege rested little weight on this view in defending Thesis (d). I think Frege regarded his
view as a consequence of the rest of his doctrine. Since the consequence accorded with
his intuitions, it had some value for him in confirming the doctrine.
10 It is true that Frege writes in Basic Laws Section 2, 'The sense of a name of a truth
value I call a thought'. This might seem to be intended to include all terms (not just
sentences and their nominalizations), as long as the term denotes a truth value. I think
that there is no independent evidence that this was Frege's intention. The remark is
illustrated only by sentences. As we have just seen, three sections later, Frege says that
thoughts are what are judged. And his system allows the act of assertion or judgment to
apply only to sentences or what they express. Thus I think that the remark in Section 2
is a slip encouraged, to be sure, by the formal assimilation of sentences to terms.
11 Furth errs in calling (rather than the vertical alone) the judgment stroke, a
distinction critical to the points we have been making. But he gives an excellent account
of the role of the notion of assertion in Basic Laws (Cf. BL, pp. xlviii-lii).
12 With one exception. When in Basic Laws the horizontal applies to expressions like '5'
or 'the course of values such that ... ', which are not sentences, the sense of the result of
the application is different from the sense of the argument expression. The former is a
thought; the latter is not. There is no analog in Begriffsschrift since the horizontal only
applied to judgeable contents.
13 The words 'determination' (for 'Bestimmung') and 'grasp' (for 'zu fassen') in this
translation replace two occurrences of the word 'definition' in J. L. Austin's otherwise
good translation of this passage from FA. Austin's choices may obscure the fact that
Frege thought not only that we needed a reduction of arithmetical terminology to other
terms, but that we needed a better grasp of the notions that such terminology expressed.
14 For a discussion that rebuts recent attacks on this interpretation, and with which I am
in broad agreement, see Michael Dumrnett, IFP.
15 This interpretation has been urged in an article, containing many interesting secon-
dary points, by Paul Benacerraf (1981). Benacerraf cites not only Section lOaf Basic
Laws but some passages in Foundations. I have discussed these latter in Burge, 1984. I
believe that I have shown them not to support the view. Here I shall concentrate on
Basic Laws Section 10.
16 I am indebted to Mary Dant for bringing home to me the kinship of these sections of
Foundations to the writings on Hilbert.
17 The bracketed substitution of 'denotation' (,Bedeutung') for 'sense ('Sinn') is specif-
ically suggested by Frege, in the light of his subsequent sense-denotation distinction, in a
letter to Husser! of May 24,1891.
18 Almost needless to say, Frege's representation of the truth predicate is not intended
exactly to reproduce ordinary language. In particular, '-t(-c)' turns out true but has no
analog in ordinary uses of 'is true'. Cf. also Note 12.
1" Frege's method of specifying or "defining" truth (in the extensional, mathematical
sense) is a primitive ancestor of Tarski's set-theoretic methods of attaining the same
objective. (Frege differs from Tarski in having no ambition to explicate, or provide a
vindication for, the concept of truth.) Both authors may be seen as summing up or
generalizing from their respective versions of the truth schema. Frege's identification of
falsity as the extension of the concept is the false is the natural counterpart of his specifi-
cation of truth.
Barwise, Jon and J. Perry: 'Semantic Innocence and Uncompromising Situations' Mid
west Studies 6, 1981.
Benacerraf, Paul: 'Frege: the Last Logicist' Midwest Studies 6, 1981.
Burge, Tyler: 'Semantical Paradox' The Journal of Philosophy 76 (1979), 169-198.
Burge, Tyler: 'Frege on Extensions of Concepts, From 1884 to 1903', Philosophical
Review 93 (1984), 3-34.
Church, Alonzo: 'Carnap's Introduction to Semantics' Philosophical Review 52 (1943),
Church, Alonzo: lntroduction to Mathematical Logic (1956) (Princeton University
Press, Princeton, 1956).
Davidson, Donald: 'Reality without Reference' Dialectica 31 (1977),246-258.
Dummett, Michael: Frege: Philosophy of Language (FPL) (Duckworth, London, 1973).
Dummett, Michael: Truth and Other Enigmas (TOE) (Harvard University Press, Cam-
bridge, Mass.; 1978).
Dummett, Michael: The Interpretation of Frege's Philosophy (IFP), (Harvard University
Press: Cambridge, Mass.; 1981).
Frege, Gottlob: Begriffsschrift und Andere Aufsiitze (B) Angelelli (ed.) (Georg OIms,
Hildesheim, 1964).
Frege, GottIob: Foundations of Arithmetic (FA), Austin trans. (Northwestern University
Press, Evanston, 1968). (The volume contains both the German and the English.
Pagination is the same as in the original.)
Frege, Gottlob: Kleine Schriften (KS), AngeIelli (ed.) (Georg OIms, Hildesheim, 1967).
Frege, Gottlob: Translations from the Philosophical Writings of Gottlob Frege (G & B)
Geach and Black (eds.), (Basil Blackwell, Oxford, 1966) 2nd edition. Obvious
abbreviations of relevant articles are also used in the text.)
Frege, Gottlob: Grundgesetze der Arithmetik (GG), (Georg Olms, Hildesheim, 1962).
Frege, GottIob: The Basic Laws of Arithmetic (BL), Furth trans. and ed. (University of
California Press, Berkeley, 1967).
Frege, Gottlob: On the Foundations of Geometry and Formal Theories of Arithmetic
(FG), Kluge ed. (Yale University Press, New Haven, 1971).
Frege, GottIob: Nachgelassene Schriften (NS) Hermes, Kambartel, Kaulbach (eds.)
(Felix Meiner, Hamburg, 1969).
Frege, Gottlob: Posthumous Writings (PPW), Long and White trans., (University of
Chicago Press, Chicago, 1979).
Frege, GottIob: Briefwechsel (BW), Gabriel, Hermes, Kambartel, Thiel, Veraart eds.,
(Felix Meiner, Hamburg, 1976).
Frege, Gottlob: Philosophical and Mathematical Correspondence (Cor), McGuinness
and Kaal (eds.) (University of Chicago Press, Chicago, 1980).
G6del, Kurt: 'Russell's Mathematical Logic' in Philosophy of Mathematics, Benacerraf
and Putnam (eds.) (prentice Hall: Englewood Cliffs, New Jersey; 1964).
Klemke, E. D. (ed.): Essays on Frege (Kl) (University of Illinois Press, Urbana, 1968).
Putnam, Hilary: Reason, Truth, and History (Cambridge University Press; Cambridge,
England; 1981).
Quine, W. V.: Word and Object (MIT Press, Cambridge, Mass.; 1960).
Quine, W. V.: Ontological Relativity (Columbia University Press, New York, 1969).
Russell, Bertrand and A. N. Whitehead: Principia Mathematica (PM) (Cambridge
University Press, New York, vol. 1,1910).
Wallace, John: 'Only in the Context of a Sentence Do Words Have Meaning' Midwest
Studies 2, 1977.
In his philosophy of language Gottlob Frege strives to present the basic
structure of language which is supposed to correspond to the structure
of what is referred to. He makes a distinction between proper names,
which refer to objects, and function-names, which refer to functions.
Function-names include concept-words and relation-words, which stand
for concepts and relations, respectively. Frege also assumes that, besides
a reference (Bedeutung), each name has a sense (Sinn), through which
the name is directed to its reference.
In his monumental work on Frege's philosophy of language Michael
Dummett lists ten theses of Frege's concerning sense and reference
(Dummett, 1981, pp. 152-203). One central principle is missing,
however, and it has likewise been ignored by most of the other Frege
scholars. This is the thesis that the word 'is' is ambiguous in a certain
way. Ignacio Angelelli comes close to attending to it when he makes
some remarks on identity and predication, and Matthias Schirn puts
special emphasis on the role of the thesis in Frege's work? but the great
majority of Frege scholars have completely overlooked the ambiguity
doctrine. Frege and Russell proposed this thesis and made it one of the
basic ingredients of modern logic. Likewise, in the Tractatus Ludwig
Wittgenstein emphasized the ambiguity of the verb 'to be' and stressed
the importance of constructing a language which prevents confusions
between different meanings of 'is'. Wittgenstein remarked that Frege's
and Russell's conceptual notation was purported to be such a language
although it did not succeed in excluding all mistakes (TLP, 3.323-
3.325). Some philosophers and linguists, including Jaakko Hintikka,
Charles Kahn, and Benson Mates, have recently discussed the ambiguity
doctrine and raised criticism against it.
The roots of the ambiguity thesis do not reach farther than the
beginning of the nineteenth century. Charles Kahn (1973a, 1973b,
1985) has argued that in the nineteenth century there was curious
interaction between the views of linguists and philosophers concerning
L. Haaparanta and J; Hintikka (eds.), Frege Synthesized, 155-174.
1986 by D. Reidel Publishing Company.
the verb 'to be', particularly concerning the notions of existence and
copula. Linguists and philologists misunderstood the ancient use of the
Greek verb einai and based their account of the verb on a mistaken
philosophical exegesis of ancient theories of being, while philosophers
relied on the work of linguists and philologists, when developing their
theories of being. In 1801 Gottfried Hermann, a German philologist,
proposed a rule which attached different accents to different meanings
of einai, thereby in effect attributing the ambiguity between existence
and copula to ancient Greek (Hermann, 1801, pp. 84-85).
Among the early opponents of the ambiguity thesis as applied to
Greek philosophy, G. E. L. Owen (1960) hinted that Aristotle's view of
being had been misinterpreted. Michael Frede (1967), for his part,
questioned the possibility of reading any sharp distinction between
existence and copula out of Plato's texts (Frede, 1967, p. 37). R. M.
Dancy (1975, 1983) has explicitly argued against efforts to apply the
ambiguity thesis to the verb einai, especially to Aristotle's verb einai,
and Jaakko Hintikka (1983, 1985) has discussed Aristotle's doctrine of
being in detail and maintained that the ambiguity thesis is completely
anachronistic when applied to Aristotle. Benson Mates (1979) has
criticized the view that Plato made a semantical distinction between the
'is' of identity and the 'is' of predication.
According to Aristotle, the realm of beings falls into different cate-
gories. Being itself is not a genus, and no single category exhausts all
beings (Met. B 3, 998b22-27, An. Post. II 7, 92b14; cf. Owen, 1965, p.
77). Aristotle assumes that to be is always to be either a substance of a
certain sort, or a qUality of a certain sort, or a quantity of a certain sort,
etc. (An. Post. I 22, 83bI3-17). However, this does not mean that
Aristotle takes 'to be' to have a completely different meaning for
different kinds of subjects. Instead, he argues:
There are many senses in which a thing may be said to 'be', but all that 'is' is related to
one point, one definite kind of thing, and is not said to be by a mere ambiguity.
(Met. r 2, lO03a33-36.)
Aristotle makes a distinction between homonymy and multiplicity of
uses. In the beginning of the Categories he states that things are
homonymous if they have only a common name but completely differ-
ent definitions. This is not what he assumes all existing things to be, but
he argues that the different uses of 'being' in the different categories
have the same focal meaning. This amounts to saying that the different
applications of 'being' are not homonymous for Aristotle but 'is' has, on
his view, only a multiplicity ofuses.
The distinction between different Aristotelian categories is, however,
quite different from the Frege-Russell distinction between different
meanings of 'is'. Fregean logic distinguishes between the following
meanings of 'is':
(1) the 'is' of identity (e.g., Phosphorus is Hesperus; a = b),
(2) the 'is' of predication, i.e., the copula (e.g., Plato is a man;
(3) the 'is' of existence,
(i) expressed by means of the existential quantifier and the
symbol for identity (e.g., God is; (3x) (g = x,
(ii) expressed by means of the existential quantifier and the
symbol for predication (e.g., There are human beings/
There is at least one human being; (3x)H(x,
(4) the 'is' of class-inclusion (e.g., A horse is a four-legged
animal; (x) (P(x):::> Q(x).
Frt(ge's conceptual notation expresses these meanings as follows:
(1) I-- (A =B)
(2) I-- <P(A)
(3) (i) a)
These formulas <?f Frege's language are judgements (Urteile), since they
are provided with the symbol' f-', which consists of the content stroke
(Inhaltsstrich) '-' and the judgement stroke (Urteilsstrich) 'I'. The
vertical stroke which connects the two horizontal ones Frege calls the
conditional stroke (Bedingungsstrich) (BS, 5). Negation is expressed in
his symbolism by a small vertical stroke which is attached under the
content stroke (ibid., 7). In the Begriffsschrift Frege uses three parallel
strokes as a sign for identity, but in the Grundgesetze he adopts the
usual sign for identity with two strokes (GGA I, 'Vorwort', p. IX).
Generality is expressed by a concavity containing a German letter plus
the same German letter holding the place of the argument (BS, 11).
Frege does not pay attention to (3) (i) in his formalism, even if he
discusses it in detail in his informal articles, such as in 'Dialog mit Piinjer
tiber Existenz' (NS, pp. 60-75). Nor has he any separate symbol for
existence, but he expresses it by means of the symbol for generality and
two negation signs.
Kahn, Hintikka, and others have mainly been worried about earlier
writers' efforts to derive the Fregean ambiguity thesis from Aristotle's
words. Since, for Aristotle, being is not a genus and to be is thus always
to be something or other, it cannot be claimed that he believed in any
pure ambiguity between existence and predication. Whether Aristotle
hinted at any of the Fregean ambiguities or not, it is at least obvious that
he gave these suggestions a minor position in his thought. This is shown
by the fact that he did not recognize any need for writing down these
different meanings of 'is' in any specific language. He was satisfied with
our natural language which does not provide us with any such
distinctions as that between identity, predication, existence, and class-
inclusion. Accordingly, even if Aristotle believed in such forms of being
as his categories, which he also found in natural language, he did not
believe in those forms of being which are shown by Frege's distinction.
Unlike Frege, Aristotle did not think that there are such relevant forms
of being as identity, predication, existence, and class-inclusion.
Frege's doctrine concerning the words 'to be' and 'is' can be gathered
from various sources, mainly from the article 'Dialog mit Piinjer tiber
Existenz', written before 1884 and published posthumously, from Die
Grundlagen der Arithmetik (1884), and from the article 'Uber Begriff
und Gegenstand' (1892). In these works Frege deals with the difference
between predication and existence, on the one hand, and the difference
between predication and identity, on the other. The distinction between
predication and class-inclusion is discussed by Frege as early as in the
Begriffsschrift (1879), where he also introduces the rest of the distinc-
tions but does not comment on them in detail. Frege's thesis not only
bears upon the different uses of the words 'to be' and 'is', but it also
concerns the different concepts that those words stand for. Frege argues
that our natural language is deficient, since it offers us one single word
for these various purposes. What we are in need of, then, is a language
which correctly reflects the distinctions between the different concepts
of being. Frege regards it as a philosopher's task to show where natural
language leads us to see things in the wrong perspective (NS, pp. 74-75
and p. 289). As Frege himself states in the BegriffsschriJt, his conceptual
notation is meant to be a language of pure thought, which is free from
It is true that Frege's principal aim was to realize his logicist program.
To carry out the program, Frege had to define arithmetical concepts by
means of logical concepts and to prove that arithmetical truths are
derivable from the axioms of logic by means of logical deduction. Frege
developed the logical implements for the derivations in the Begriffs
schriJt, and by doing so he became the foremost pioneer of modern
logic. However, it was not Frege's sole purpose to present the rules of
logical inference. Indeed, his conceptual notation was purported to be a
Leibnizian lingua characterica, from which all ambiguities were elimin-
ated and which was thus meant to be the correct semantic representa-
tion of natural language.
Frege's paradigm of first-order language was
thus essentially semantically determined. However, he did not himself
present it semantically, for, as Jean van Heijenoort (1967) and Jaakko
Hintikka (1979a, 1981b, 1981c) have argued, he believed in the ineffa-
bility of semantics. This means that he did not take it to be possible for
us to step outside the limits of language in order to consider the
relations between language and the world, because, in his view, all talk
already presupposes these semantical relations. Frege explained the
different uses of the word 'is' simply by describing his notation for a
first-order language, which was for him the only correct linguistic
representation of our concepts.
It is true many twentieth-century logicians have adopted the idea of
language as a calculus which can be freely reinterpreted in a fixed
domain of individuals.
Some logicians have even completely rejected
the claim that a Fregean first-order language - suitably supplemented,
for instance, by adding to it some higher-order logic, as Frege does - is
a universal medium of communication in the Fregean sense; meaning
both the sense that the interpretation of its names and predicates do not
vary in a fixed domain of individuals and the sense that it speaks about
one fixed domain. None the less, logicians have accepted the Fregean
quantification theory, where the ambiguity doctrine is firmly codified.
The simple reason why they have accepted this doctrine merely by
accepting the quantification theory is that the meanings of the logical
constants of the quantification theory, including those that stand for
various kinds of being, remain unchanged, even though the classes of
individuals the quantifiers range over may change. Thus adding new
elements to the basic Fregean quantification theory or relativizing the
ranges of his quantifiers does not remove a logician's commitment to the
ambiguity of 'is'.
I have here claimed that Frege has several concepts of being, without
carefully attending to Frege's own terminology, according to which
concepts are references of concept-words. We can try to avoid the
distinction between senses and references by saying that, in Frege's
logic, the word 'is' not only has a number of uses but it has various
meanings. In this paper, I shall not deal with the possibility of applying
the distinction between senses and references to such an auxiliary verb
as 'to be'.7
This paper focuses on Frege's doctrine of existence. One of the
innovations of Frege's logical theory was to construe existence as a
second-order concept, i.e., as a property of concepts.
This paper is,
however, an attempt to elucidate some lesser known aspects of Frege's
view of existence. I shall argue that Frege regards existence both as a
proper second-order concept and as an empty first-order concept and
that the distinction between the two references of 'exists' is motivated by
his metaphysical and epistemological tenets. In constructing these
tenets, we must be satisfied with Frege's brief remarks and hints. Thus,
Frege not only assumes that the word 'is' is ambiguous but he also
considers the verb 'to exist' and hence also the 'is' of existence an
equivocal word. Frege's concept of Wirklichkeit, which likewise turns
out to be a kind of concept of existence, will not be discussed in this
I have spoken of Frege's logical language as a first-order language.
Admittedly, Frege did deal with higher-order quantifiers. However, for
reasons which Will not be discussed here the higher-order component of
Frege's lingua characterica may be considered inessential, and it is in
any case largely irrelevant for the purpose of this paper.IO
The concept of existence is discussed by Frege in detail in his 'Dialog
mit Piinjer iiber Existenz', which has been published in Frege's Nachge-
lassene Schriften. The paper was written before 1884, the year Frege
completed his Grundlagen. In the Grundlagen Frege presents his doc-
trine of existence in a mature form. The argumentation put forward in
'Dialog mit Piinjer iiber Existenz' supplements the Grundlagen and is
very instructive if we are interested in the different aspects of Frege's
concept of existence. I have discussed the dialogue in my paper 'On
Frege's Concept of Being', where I tried to shed some light on how
Kant's views on existence influenced Frege's ideas. In what follows, I
shall present the main points that Frege makes in the dialogue and give
my suggestions concerning the implicit motivation of his view.
In 'Dialog mit Piinjer iiber Existenz' Frege supports the claim that such
sentences as 'Leo Sachse is' and 'Leo Sachse exists' are self-evident
(selbstverstandich), while Piinjer suggests that the word 'is' carries the
same meaning as 'is something that can be experienced' (ist erfahrbar).
For Piinjer, the set of objects of experience (Gegenstande der Erfahrung)
is a subset of the set of objects of ideas (Gegenstande der Vorstellungen).
Frege argues that Piinjer's account results in contradiction: If 'A is not'
means the same as 'A is not an object of experience', then the statement
'There is something that is not an object of experience' means the same
as 'There is an object of experience which is not an object of experience'
(NS, pp. 71-72). In an afterword to the dialogue Frege continues his
argumentation and states that if the sentence 'A is' were not self-
evident, its negation could be true in some circumstances. He concludes
that if the sentence 'There are entities which do not have the property of
being' means the same as 'Something that has being falls under the
concept of not-being (der Begriff des Nichtseienden)', it is a contradic-
tory sentence, and if the sentence 'There are B's' is equivalent in
meaning to the sentence 'Something that has being is B', the concept of
being is self-evident.
Frege seems to be driven to denying the meaningfulness of sentences
like 'A is' or 'A exists' because of his conceptions of language and the
world. He cannot say that the sentence 'Something that has being is not'
means that something for which it is possible to exist does not exist in
the actual world, since he is committed to the view that there is only one
world and that his conceptual notation is a universal language which
speaks about that world. He does not even divide his universe into various
sorts. That is indicated by his principle of completeness (Grundsatz der
VoNstandigkeit), according to which any function must be defined for all
objects (GGA II, 56-65).11 Due to his one-world view, he concludes
that the concept of being is not a determination of an object, i.e., it does
not help us to distinguish between any two objects (NS, p. 73).
We may put the same point as follows: Because of Frege's one-world
view, there can be quantifiers of one kind only, namely, quantifiers
ranging over all actually existing objects. For this reason, Frege cannot
escape the threatening inconsistency by assuming that we have two
different ranges of quantifiers in sentences like 'Something that has
being is not', which otherwise might have seemed a plausible way out for
someone who distinguished from each other the different meanings of
After having rejected the idea that existence is a real property of an
object, Frege makes an effort to convert existential statements into the
form of particular statements (NS, p. 70).'3 For instance, he turns the
sentence 'There are men' into the sentence 'Some living beings are
rational'. If the concept that occurs in a given sentence cannot be
defined by means of two concepts, Frege resorts to the concept of being
identical with itself (sich selbst gleich sein), which he takes to be the
most general concept of the hierarchy of concepts. He identifies this
concept with the concept of being. Hence, he can turn the sentence
'There are men' also into the sentence 'Something that has being is a
man', or 'Something that is identical with itself is a man' (NS, p. 71).
Consequently, Frege holds the view that we are forced to regard being
(in the sense of existence) as a concept which is superordinate to every
concept. What Frege shows here is that, according to his doctrine of
being, existence can be used as a first-order concept if one i's willing to
pay the price of its becoming an empty concept.
The concept of being
that we are here interested in is such that we predicate it of every object
of which we predicate anything. By saying that a is X, we say that a is
and that a is X Here the copula is purported to posit the object a in the
sense that it is the part of the predication which makes the predication
carry with itself the claim for existence. Copula is for Frege a concept
which applies to entities in this one world of ours as well as any other
concept, even if more generally.
The idea that existence is included in every predication apparently
brings Frege quite close to Aristotle. Frege seems to repeat Aristotle's
view that the expression 'existent man' says nothing more than 'man',
i.e., 'existent' is an empty and hence redundant word in any context
whatever (Met. r 2, 1003b27-30,I 3, 1054a16-18). What is even
more, as far as we consider Frege's concept of being as a first-
order concept, Frege does not believe in the analogy of 'is' in the sense
in which Aristotle does. For Frege, being is a 'genus'; it is an empty
concept, which has an infinitely wide extension and no comprehension.
If we limited our consideration of Frege's concept of existence to what
we have found out up to now, we could conclude that Frege's concept of
being is a univocal concept.
Why does Frege regard being as a univocal concept in this limited
sense? The reason is, again, that Frege does not divide his universe into
various sorts. Unlike that of Aristotle's, Frege's realm of beings does not
fall into different categories.
In his article 'Kritische Beleuchtung einiger Punkte in E. Schroders
Vorlesungen iiber die Algebra der Logik' Frege suggests that the
sentences 'A is' and 'A exists' could be rendered as the metalinguistic
sentence 'the name "A" has a reference' (KS, p. 208). But if Frege is
consistent in his view that we cannot step outside the limits of language,
he must consider such statements as illegitimate talk about the expres-
sions of our language.
Frege requires that in the language of science all proper names have
to be taken to be nonempty (NS, p. 135). He also assumes that if we talk
about an object we already presuppose the existence of that object
('Uber Sinn und Bedeutung', KS, p. 147). None the less, Frege admits
that we speak meaningfully about entities which do not exist. On Frege's
view, a sentence lacks only a truth-value - but not a sense - if it
contains a name which has no reference ('Uber Sinn und Bedeutung,
KS, p. 148).
Since Frege regards being as a characteristic of every concept, it
might be suggested that if we attach any concept-word to an empty
proper name, the concept brings it about that the name has a reference,
after all. That this is not the case becomes obvious if we consider
Frege's concept of concept. According to Frege, a concept-word - and
likewise a concept to which it refers - is ungesiittigt. It has a 'gap', which
can be filled with a complete expression, i.e., with a proper name (GGA
I, pp. 5-8). If we fill the gap of the concept-word with a proper
we also aim at filling the gap of the corresponding concept. If we
succeed in filling the gap of the concept, it is also true that the proper
name has a reference. If the gap of the concept is not filled, we do not
attach existence to anything, since we do not succeed in predicating
anything. If we succeed in predicating something, we succeed in predi-
cating existence at the same time. This is so because, for Frege, to be an
object implies existence. For if a is an object, then the proper name 'a'
has a reference, which amounts to saying that a exists. Consequently,
Frege's idea that we presuppose the existence of the objects we talk
about concerns only the nature of our speech acts and the pragmatical
aspects of our language, and it has nothing to do with the semantical
relation between the sentences of our language and the objects and the
functions of the world. Frege simply wants to remark that, when we say
something about an object, we do not add that the object exists.
In the Tractatus Wittgenstein argues that we arrive at nonsense if we
treat such concepts as those of object, function, number, and concept as
ordinary concepts. According to Wittgenstein, in a consistent and
perspicuous language, that something is an object, a function, etc. could
only be shown but could not be said (fLP, 4.126). David Bell (1979)
claims that Frege supported this view as far as functions are concerned,
for, on his view, that something is a function is shown by the incom-
pleteness of the sign which is used to refer to it, but it is not possible to
say that something is a function ('Uber Begriff und Gegenstand' , KS, p.
170). Bell assumes that Frege never extended this doctrine to include
the expression '() is an object' (Bell, 1979, p. 47). Frege's treatment of
existence is, however, a kind of extension, since, on his view, we do not
say that something is an object because, in using the name, we already
presuppose that there is an object to which the name refers.
There is one important and manifest reservation that Frege makes in
his argumentation in 'Dialog mit Piinjer tiber Existenz'. He concludes
there that if the sentence 'There are B's' is equivalent in meaning to the
sentence 'Something that has being is B', then the concept of being is
self-evident. His formulation suggests an alternative way of handling the
problem, in which the mentioned equiValence is denied and which thus
hints at the equivocity of 'being'. But if the expressions 'x has being' and
'there is an x' differ in meaning, Frege's argumentation for the view that
'A is' is self-evident collapses.
At the end of 'Dialog mit Piinjer tiber Existenz' Frege introduces the
doctrine that existence is a property of a concept (NS, p. 75). Frege is
inclined to maintain that existence used as a first-order concept is an
empty concept, but he insists on preserving the meaningfulness of
existence used as a second-order concept. That conviction is conspicu-
ous in his criticism of the idea that every concept is abstracted from a
number of objects. He remarks that if all concepts were elicited from
existing objects, existential statements would lose all content; once we
had a concept, we could infer that there is an object which exemplifies
the concept (GLA, 49).16
The above discussion shows that Frege assumes that 'exists' and the 'is'
of existence have two readings. They may refer either to an empty
first-order concept or to a meaningful second-order concept. In the
former case the existential statement becomes meaningful if it is trans-
formed into a metalinguistic statement which expresses that a given
proper name has a reference. In the latter case the statement tells us that
a concept is instantiated, i.e., that there is an object which has a given
property. First-order existence is formalized by means of the existential
quantifier and the symbol for identity, while second-order existence is
expressed by means of the existential quantifier and the symbol for
predication. In order to find out the philosophical motivation for Frege's
view of existeace, let us first consider the distinction between identity
and predication.
Frege discusses the problem of interpreting the concept of identity
already in the Begrijfsschrift, where he states that in an identity state-
ment a name seems to represent itself. He adds, however, that an
identity statement does not concern names only but it expresses that two
signs have the same content (Inhalt), which is determined in two
different ways by the two signs (BS, 8). He reformulates that idea in
'Uber Sinn und Bedeutung' by saying that an identity statement
expresses that two names have the same reference but different senses.
The sense of a name is the way in which the reference of the name is
presented (KS, pp. 143-144).
Frege thinks that the symbols which occur on the different sides of
the identity symbol can be replaced by each other in any context
whatsoever, and he also assumes that two objects are identical if and
only if they fall under the same o:mceptsP As we saw in the preceding
section, Frege also considers identity to be a relation of an object to
itself. These renderings of the concept of identity are subject to
Wittgenstein's criticism, according to which to say of two objects that
they are identical, is nonsense, and to say of one object that it is
identical with itself, is to say nothing at all (TLP, 5.5303). The details of
Frege's doctrine of identity and the possible changes in his view will not
be discussed in this paper.IS It is, however, worth mentioning that Frege
does not regard it as possible to define the identity of objects by the
sameness of their properties or by any other means, since any definition
is itself an identity ('Rezension von: E. G. Husserl, Philosophie der
Arithmetik 1', KS, p. 184).
In his article 'Uber Begriff und Gegenstand' Frege emphasizes that, in
order to keep objects and concepts apart, we must make a sharp
distinction between identity and predication (KS, p. 168). The principle
according to which objects must be clearly distinguished from concepts
also manifests itself in that, unlike traditional grammatical analysis,
Fregean analysis of sentences distinguishes the relation between two
concepts of the same order from predication, which, for its part,
concerns the relation between an individual and a concept or a relation
between two concepts of two different orders (NS, p. 207 and p. 210).
In the Grundlagen the distinctioJ) between objects and concepts occurs
in the list of Frege's basic principles (GLA, p. X). He also stresses the
distinction in the Grundgesetze (GGA I, p, X and p. XIV), in 'Uber die
Begriffsschrift des Herrn Peano und meine eigene' (KS, p. 233), and in
'Uber die Grundlagen der Geometrie II' (KS, p. 270).
Why does Frege stress the distinction between objects and concepts?
Frege rejects the grammatical analysis of sentences and substitutes
objects and concepts (and other functions) for subjects and predicates,
and thereby he changes the structure of universal and particular
sentences. As far as singular sentences are concerned, he does not
accept the identification of individuals with their essential properties.
For Frege, the sentence 'Plato is a man' contains the 'is' of predication,
which must be sharply distinguished from the 'is' of identity ('Logik in
der Mathematik' (1914), NS, pp. 230-231).19 This means that Frege
does not regard it as possible for us to have knowledge of what an
object is in itself by means of our concepts. For him, all properties are
on a par, whether they are called essential or accidental in traditional
philosophical literature.
Surprisingly enough, Frege's much debated distinction between Sinn
and Bedeutung witnesses the same epistemological view. Disregarding
Frege's views on the sense and the reference of a sentence, which Frege
also labels a proper name, we can present Frege's doctrine of the sense
and the reference of a proper name as follows: The sense which a
proper name expresses and which is a way of presenting the object to
which that proper name refers belongs to the object. Moreover, we
should have complete knowledge of the object, only if we knew all its
senses, which is not possible for us ('Uber Sinn und Bedeutung', KS,
pp. 144-147). Accordingly, Fregean senses of an object seem to be
complexes of the concepts under which the object falls. This interpreta-
tion of Frege's concept of Sinn is supported by his examples, according
to which 'the Evening Star' and 'the Morning Star' are senses of Venus
and 'the teacher of Alexander the Great' arid 'the pupil of Plato' are
senses of Aristotle. Frege also argues that a proper name is related to an
object via a sense and only via a sense and each proper name must
express at least one sense (NS, p. 135). Hence, according to Frege, it is
not possible to talk meaningfully of an object without thinking of the
object as falling under some concept. Frege's remarks (;oncerning senses
and references thus give us more support for the hypothesis that Frege
believes in the universality of language. They may even hint at the view
that there are no properties which belong to objects before there is a
conceptual system which attaches senses to objects.
Frege's theory of sense and reference shows that Frege not only takes
it to be impossible to find out any essential properties of objects, which
would be identical with objects themselves, but he also considers the
forming of complete concepts of objects as being beyond the abilities of
a finite human being. Frege echoes Leibniz's thOUght that a human being
is only able to form partial concepts of individuals, while God sees in
the concept of an individual all that can be predicated of that individual
(Leibniz, 'Discourse on Metaphysics', sec. 8 and sec. 9). For Frege, an
object is neither identical with any essential property nor with any
combination of the concepts under we can know the object to fall.
By stressing the distinction between objects and properties, or concepts,
Frege in effect draws the limits of human knowledge.
What has been said above shows why Frege calls special attention to
the distinction between identity and predication. Frege's conceptual
notation, which is meant to be a universal language, allows us to speak
about objects only through different configurations formed by concept-
words and other function-names. We cannot step outside these con-
figurations in order to consider the relations between our language and
objects themselves. An identity statement can only tell us that two
names have the same Bedeutung but, according to Frege, we cannot say
what this Bedeutung is. If Frege were consistent, we could not even
accept a metalinguistic sentence which tells something about the relation
between names and references. An identity statement tries to say some-
thing that cannot be said, while predication is precisely the way in which
our reason is capable of handling objects.
Even if Frege subscribed to the principle that one and only one
linguistic symbol or distinction should correspond to each reference or
distinction of the universe, he did not eliminate the identity symbol from
his language. Wittgenstein was more consistent in this respect, for he
regarded an identity statement only as a rule, which concerns the
substitutivity of names in different contexts. He assumed that we can
give up the identity symbol, when we have realized the idea of universal
language so that there are no more two different names for anyone
single object in our language. According to Wittgenstein, the identity
symbol is not an essential part of the conceptual notation (fLP, 5.53).
Frege considered the role of the identity symbol from a completely
different perspective. As he mentioned already in the Begriffsschrift, he
did not regard the identity statement only as a rule which concerns
names. At the same time, he insisted on the principle that his language
speaks about something and that each distinction and each symbol in
language must correspond to a unique sense, a unique reference and
a unique distinction in the universe. Hence, we must try to find out
what the counterpart in the world is in the case of identity statements.
What Frege found in the world was a distinction between senses and
references, which follows from the distinction between objects in
themselves and objects as we know them. Frege wanted to make a
distinction between objects as metaphysical entities and objects as we
know them, and he also wanted that this distinction is visible in his
universal language. For this reason, he distinguished the 'is' of identity
from the 'is' of predication in his conceptual notation and hence
incorporated the identity symbol in his language. Frege did not notice
that he ought to have excluded identity statements from his language
precisely because they aim - at saying something about objects as
metaphysical entities, that is, in the way in which we cannot speak about
objects in language, according to Frege.
In the previous section I argued that Frege's view of language and the
world influenced his doctrine of existence. The discussion above con
cerning identity and predication serves to clarify the details of Frege's
view that the word 'exists' is equivocal. It is not only the case that Frege
regards 'exists' and the 'is' of existence as equivocal in the sense that
there are two concepts of existence apart from all linguistic contexts but
that each context determines to which concept the words refer in each
case. What is more, Fregean analysis has the additional consequence
that 'exists' and the "is' of existence preserve their equivocity in some
contexts. That is what happens if we attach them to proper names. I
shall clarify this point in what follows on the basis of what I argued
above concerning identity and predication. That also yields an answer to
the question concerning the philosophical background of Frege's view of
If Frege's Sinne are complexes of properties of objects, the sentence
'a exists' expresses the thought that there is an object which is P, Q, R,
etc. Since the sentence 'There is a P' means, for Frege, the same as the
sentence The concept P is instantiated', likewise, the sentence 'a exists'
means that a certain bundle of concepts is instantiated. Here existence
turns out to be something that is asserted of a bundle of properties.
Frege nowhere explicitly draws this conclusion from his premisses, but
his suggestions are, however, evident. In 'Uber Sinn und Bedeutung'
Frege undertakes to show that identity statements can be meaningful
even if they seem to be either vacuously true or self-contradictory. The
solution he offers is that one may associate a different sense with 'a' and
with 'b' even if 'a = b' is true. If Frege supports this kind of analysis, he
must also admit that 'a exists' makes sense. That is because one can, of
course, attach a sense to 'a' without knowing that a exists, as easily as
one can attach a sense to 'a' and 'b' without knowing that 'a= b' is true.
Existence is not included in the Sinn expressed by a proper name. If we
take the individual properties expressed by the proper name apart and
form a judgment of each, the sentences which express the judgments
may either be true or lack a truth-value. For Frege, forming a concept or
a bundle of concepts is independent of the instantiation of that concept
or bundle of concepts.
Like the distinction between identity and predication, the thesis of
the equivocity of 'exists' and the 'is' of existence is motivated by
epistemological considerations concerning the limits of human knowl-
edge. My suggestion for construing Frege's doctrine of existence in the
case of sentences like 'a exists' or 'a is' is hence the following: If we say
that a exists and if someone asks us what it is that exists, we are not able
to answer the question in any other way than by mentioning some of the
concepts under which that objects falls. We can say that the sentence 'a
exists' means that there is an object which has the properties P, Q, R,
etc. Existence turns out to be a second-order concept, which means
instantiation of a bundle of properties. But since we cannot say what the
object a is as abstracted from our concepts, our answer to the question
concerning what a is in itself comes down to saying that a is a, which is
an empty statement. On Frege's view, we can say that an object is what it
is, that it is identical with itself, which is an empty statement, but we
cannot say what it is, i.e., what it is identical with. Therefore, in the
sentences 'a exists' and 'a is' words 'exists' and 'is' can be read either as
expressions of a meaningful second-order concept or as expressions of
an empty first-order concept. If we interpret the words as referring to
the first-order concept, the corresponding sentence can be transformed
into the sentence 'The name "a" has a reference', but that is, of course,
of no help for us in finding out what the reference is.
Thus, Frege's distinction between the two concepts of existence
ensues from his endeavour to distinguish objects in themselves from
objects considered through the descriptions that we can attach to them.
As I concluded above, Frege wants to make a distinction between
objects as metaphysical units and objects as we know them, and he also
wants that this distinction is visible in his universal language. But again,
if Frege were to be consistent, he should eliminate the existence
expressed by the existential quantifier and the symbol for identity from
his language, for it tries to say something that, on Frege's view, cannot
be said in language.
Academy of Finland
* I am very grateful to Professor Jaakko Hintikka for valuable suggestions and
My thanks are due to the Philosophical Society of Finland, which has given me the
permission to use extracts from my monograph Frege's Doctrine of Being (Acta
Philosophica Fennica 39, 1985) in this article.
1 See 'Uber Sinn und Bedeutung', KS, pp. 144-147, GGA I, p. X and 26, and 'Uber
die Grundlagen der Geometrie I-Ill', KS, p. 285.
2 See AngeleIli (1967), pp. 253-254, and Schirn I (1976), pp. 20-21.
3 See Hintikka (1979b, 1981a). See also Kahn (1973a, 1973b) and Mates (1979).
4 See Hintikka (1973,1983,1985) and Knuuttila (1985).
5 See, e.g., 'Uber den Zweck der Begriffsschrift', BS, 1964, p. 98, 'Uber die Begriffs-
schrift des Herrn Peano und meine eigene', KS, p. 227, and 'Anmerkungen Freges zu:
Philip E. B. Jourdain, 'The development of the theories of mathematical logic and the
principles of mathematics', KS, p. 341.
6 See van Heijenoort (1967), Goldfarb (1979), and Hintikka (1979a, 1981b, 1981c).
7 For further discussion of this problem, see Haaparanta (1985a), pp. 56-58.
B See, e.g., GLA, 53.
9 For Frege's concept of Wirklichkeit, see Haaparanta (1985a).
10 Note that Frege tried to treat higher-order entities by correlating with each of them a
Werthverlauf, which is always a first-order entity, for instance, a value of first-order
quantifiers. See, e.g., GGA I, 3 and 10.
II See van Heijenoort (this volume).
12 This was suggested to me by laakko Hintikka.
13 Frege makes the same suggestion in 'Aufzeichnungen fiir Ludwig Darmstaedter', NS,
14 This is also pointed out by R. Stuhlmann-Laeisz (Thiel, 1975, p. 126).
15 However, Frege anticipates the distinction between object-language and metalan-
guage in the article 'Logische Allgemeinheit', where he makes the distinction between
Hilfssprache and Darlegungssprache. See NS, p. 280. Cf. also Haaparanta (1985a), p. 35.
16 I do not want to argue that if existence is considered as a second-order concept, it is
no more a problematic concept. The expression 'There is a - ' is far from clarified by
the expression 'The concept - is instantiated'. Cf. Williams (1981), pp. 59-60.
17 See BS, 8, GLA, 65, and 'Uber Sinn und Bedeutung', KS, p. 150. See also GGA
I, 20, and NS, p. 131.
18 See Schirn (1976), Band II. See also Haaparanta (1985a).
19 Cf., e.g., Aristotle, Met. Z 6. M. J. Woods (1975) argues for the interpretation that,
according to Aristotle, the sentence 'Socrates is a man' means that Socrates is identical
with a man. See also Hintikka (1985). For further discussion, see Haaparanta (1985a),
Frege, G.: 1879, Begriffsschrift, eine der arithmetischen nachgebildete FormeZsprache des
reinen Denkens. Verlag von L. Nebert, Halle a. S.; repr. in G. Frege (1964), pp.
1-88. (Referred to as BS.)
Frege, G.: 1883, 'Uber den Zweck der Begriffsschrift', in Sitzungsberichte der lenaischen
Gesellschaft flir Medizin und Naturwissenschaft fUr das lahr 1882, Verlag von G.
Fischer, lena, pp. 1-10; repr. in Frege (1964), pp. 97-106.
Frege, G.: 1884, Die GrundZagen der Arithmetik: eine Zogisch mathematische Untersu-
chung tiber den Begrijf der ZahZ, Verlag von W. Koebner, Breslau; repro and trans!'
by l. L. Austin in The Foundations of Arithmetic! Die Grundlagen der Arithmetik,
Basil Blackwell, Oxford, 1968. (Referred to as GLA.)
Frege, G.: 1891, Funktion und Begriff, Vortrag, gehalten in der Sitzung vom 9. Januar
1891 der lenaischen Gesellschaft fiir Medizin und Naturwissenschaft, H. Pohle,
Jena; repr, in KS, pp. 125-142.
Frege, G.: 1892, 'Uber Sinn und Bedeutung', Zeitschrift fUr Philosophie und philoso-
phische Kritik 100, 25-50; repr, in KS, pp. 143-162.
Frege, G.: 1892, 'Uber Begriff und Gegenstand', Vierteljahrschrift flir wissenschaftliche
Philosophie 16,192-205; repr. in KS, pp. 167-178.
Frege, G.: 1893, Grundgesetze der Arithmetik, begrijfsschriftlich abgeleitet, 1. Band,
Verlag von H. Pohle, lena. (Referred to as GGA I.)
Frege, G.: 1894, 'Rezension von: E. G. Husser!, Philosophie der Arithmetik 1', Zeit-
schrift [iir Philosophie und philosophische Kritik 103, 313-332; repr, in KS, pp.
Frege, G.: 1895, Kritische Beleuchtung einiger Punkte in E. Schroders Vorlesungen
iiber die Algebra der Logik', Archiv [iir systematische Philosophie 1,433-456; repro
Frege, G.: 1896, 'Uber die Begriffsschrift des Herrn Peano und meine eigene', in
Berichte uber die Verhandlungen der koniglich siichsischen Gesellschaft der Wissen-
schaften zu Leipzig, Mathematisch-Physische Klasse, 48. Band, pp. 361-378; repr.
in KS, pp. 220-233.
Frege, G.: 1903, Grundgesetze der Arithmetik, begriffsschriftlich abgeleitet, II. Band,
Verlag von H. Pohle, Jena. (Referred to as GGA II.)
Frege, G.: 1903, 'Uber die Grundlagen der Geometrie II', Jahresberichte der Deutschen
Mathematiker- Vereinigung, 12. Band, pp. 368-375; repr. in KS, pp. 267-272.
Frege, G.: 1906, 'Uber die Grundlagen der Geometrie I-III', in Jahresberichte der
Deutschen Mathematiker- Vereinigung, 15. Band, pp. 293-309, 377-403, 423-430;
repr. in KS, pp. 281-323.
Frege, G.: 1912, 'Anmerkungen zu: Philip E. B. Jourdain, The development of the
theories of mathematical logic and the principles of mathematics', The Quarterly
Journal of Pure and Applied Mathematics 43,237-269; repr, in KS, pp. 334-341.
Frege, G.: 1952, Translations from the Philosophical Writings of Gottlob Frege, ed. by P.
Geach and M. Black, Basil Blackwell, Oxford.
Frege, G.: 1964, Begriffsschrift und andere Aufsiitze, ed. by I. Angelelli, Georg Olms,
Frege, G.: 1964, The Basic Laws of Arithmetic, Exposition of the System, trans!. and ed.
by M. Furth, University of California Press, Berkeley and Los Angeles.
Frege, G.: 1967, Kleine Schriften, ed. by I. Angelelli, Wissenschaftliche Buchgesellschaft,
Darmstadt, and Georg Olms, Hildesheim. (Referred to as KS.)
Frege, G.: 1969, Nachgelassene Schriften, ed. by H. Hermes, F. Kambartel, and F.
Kaulbach, Felix Meiner Verlag, Hamburg. (Referred to as NS.)
Frege, G.: 1972, Conceptual Notation and Related Articles, trans!' and ed. by T. W.
Bynum, Clarendon Press, Oxford.
Frege, G.: 1979, Posthumous Writings, trans!. by P. Long and R White, Basil Blackwell,
Angelelli, I.: 1967, Studies On Gottlob Frege and Traditional Philosophy, D. Reidel,
Aristotle: 1928, Analytica Posteriora, in The Works of Aristotle, Vol. I, ed. by W. D.
Ross, Oxford University Press, London.
Aristotle: 1928, Metaphysica, in The Works of Aristotle, Vol. VIII, ed. by W. D. Ross,
The Clarendon Press, Oxford.
Bell, D.: 1979, Frege's Theory of Judgement, The Clarendon Press, Oxford.
Dancy, R: 1975, Sense and Contradiction: A Study in Aristotle, D. Reidel, Dordrecht.
Dancy, R: 1983, 'Aristotle and Existence', Synthese 54,409-442.
Dummett, M.: 1981, Frege: Philosophy of Language, 2nd ed., Duckworth, London.
(First published in 1973.)
Frede, M.: 1967, Priidikation und Existenzaussage, Hypomnemata 18, Vandenhoeck &
Ruprecht, Gottingen.
Goldfarb, W. D.: 1979, 'Logic in the Twenties: the Nature of the Quantifier', The
Journal of Symbolic Logic 44, 351-368.
Haaparanta, L.: 1985a, Frege's Doctrine of Being, Acta Philosophica Fennica 39.
Haaparanta, L.: 1985b, 'On Frege's Concept of Being', in Knuuttila and Hintikka
van Heijenoort, J.: 1967, 'Logic as Calculus and Logic as Language', Synthese 17,
van Heijenoort, J.: 1985, 'Frege and Vagueness', in this volume.
Hermann, G.: 1801, De emendanda ratione graecae grammaticae, Gerhard Fleischer,
Hintikka, J.: 1973, 'Aristotle and the Ambiguity of Ambiguity', in J. Hintikka, Time and
Necessity: Studies in Aristotle's Theory of Modality, The Clarendon Press, Oxford,
Hintikka, J.: 1979a, 'Frege's Hidden Semantics', Revue Internationale de Philosophie
Hintikka, J.: 1979b, '''Is''.. Semantical Games, and Semantical Ralativity', Journal of
Philosophical Logic 8, 433-468.
Hintikka, J.: 1981a, 'Kant on Existence, Predication, and the Ontological Argument',
Dialectica 35,127-146.
Hintikka, J.: 1981b, 'Semantics: A Revolt Against Frege', in G. Fl0istad (ed.), Contem-
porary Philosophy, Vol. 1, Martinus Nijhoff, The Hague, pp. 57-82.
Hintikka, J.: 1981c, 'Wittgenstein's Semantical Kantianism', in E. Morscher and R
Stranzinger (eds.), Ethics, Proceedings of the Fifth International Wittgenstein Sym-
posium, HOlder-Pichler-Tempsky, Vienna, pp. 375-390.
Hintikka, J.: 1983, 'Semantical Games, the Alleged Ambiguity of "Is", and Aristotelian
Categories', Synthese 54, 443-468.
Hintikka, J.: 1985, 'The Varieties of Being in Aristotle', in Knuuttila and Hintikka
Kahn, {:.: 1973a, 'On the Theory of the Verb "To Be"', in M. K. Munitz (ed.), Logic and
Ontology, New York University Press, New York, pp. 1-20.
Kahn, c.: 1973b, The Verb 'Be' in Ancient Greek, D. Reidel, Dordrecht.
Kahn, C.: 1985, 'Retrospect on the Verb 'to be' and the Concept of Being', in Knuuttila
and Hintikka (1985).
Knuuttila, S.: 1985, 'Being qua Being in Thomas Aquinas and John Duns Scotus', in
Knuuttila and Hintikka (1985).
Knuuttila, S. and Hintikka, J. (eds.): 1985, The Logic of Being: Historical and Critical
Studies, D. Reidel, Dordrecht.
Leibniz, G. W.: 1969, Philosophical Papers and Letters, 2nd ed., ed. by L. E. Loemker,
D. Reidel, Dordrecht.
Mates, B.: 1979, 'Identity and Predication in Plato', Phronesis 24, 211-229.
Owen, G. E. L.: 1965, 'Aristotle on the Snares of Ontology', in R Bambrough (ed.), New
Essays on Plato and Aristotle, Routledge and Kegan Paul, London, pp. 69-95.
Schirn, M. (ed.): 1976, Studien zu FregelStudies on Frege I-III, Frommann-Holzboog,
Stuttgart - Bad Cannstatt.
Stuhlmann-Laeisz, R: 1975, 'Freges Auseinandersetzung mit der Auffassung von
'Existenz' als ein Priidikat der ersten Stufe und Kants Argumentation gegen den
ontologischen Gottesbeweis', in C. Thiel (ed.), Frege und die moderne Grundlagen-
forschung, Anton Hain, Meisenheim am Glan, pp. 119-133.
Williams, C. J. F.: 1981, What is Existence?, Clarendon Press, Oxford.
Wittgenstein, L.: 1961, Tractatus Logico-Philosophicus, The German text of Ludwig
Wittgenstein's Logisch-philosophische Abhandlung with a new Translation by D. F.
Pears and B. F. McGuinness, Routledge & Kegan Paul, London. (Referred to as
Woods, M. J.: 1975, 'Substance and Essence in Aristotle', Proceedings of the Aristotel-
ian Society 75,167-180.
A logically perfect language (Begriffsschrift) should satisfy the conditions, that every
expression grammatically well constructed as a proper name out of signs already
introduced shall in fact designate an object, and that no new sign shall be introduced as
a proper name without being secured a reference. The logic books contain warnings
against logical mistakes arising from the ambiguity of expressions. I regard as no less
pertinent a warning against apparent proper names having no reference. The history of
mathematics supplies errors which have arisen in this way. ([7), p. 41; trns., [10], p. 70)
Here again we likewise see that the laws of logic presuppose concepts with sharp
boundaries, and therefore also complete definitions for names of functions ... In Vol. 1
we expressed this as follows: every function-name must have a reference. ([4), p. 78;
trans., [10), p. 170)
After reaching the end in this way, one may reread the Exposition of the Begriffsschrift
as a connected whole, keeping in mind that the stipulations that are not made use of
later, and hence seem superfluous, serve to carry out the basic principle that every
correctly-formed name is to denote something, a principle that is essential for full rigor.
([3), p. 12; trans., [9), p. 9)
No methodological principle was more important to Frege than the one
at stake in these passages: in a properly constructed scientific language
every name (including function-names as well as object-names) must
have a reference. In his eyes the repeated failures of his fellow mathe-
maticians to be certain of satisfying this tenet was one of their most
grievous errors. Thus it was entirely in keeping with this that he proved
that every name in his own system has a reference. The foundations for
the proof appear in Sections 28-30 of the Grundgesetze (13]) with the
proof itself taking place in Section 31. Evidently, Frege thought that he
need only show that his primitive names were referential since the
referentiality of names composed from them would follow by an easy
induction.! Accordingly, the proof focused almost exclusively on the
"basis" case.
Frege's proof anticipates contemporary proofs that a given valuation
for a language assigns a unique value to each sentence of the language.
Since many authors regard those proofs 'as too obvious to state, it might
be hard to imagine any difficulty with Frege's proof. There is agreement
among Frege scholars that serious problems exist with his proof but that
L. Haaparanta and J. Hintikka (eds.), Frege Synthesized, 177-195.
1986 by D. Reidel Publishing Company.
is as far as the agreement goes. Some have claimed that there is no way
to assign references uniquely to all sentences of Frege's language,
because doing so would establish the consistency of his system.
have claimed that the proof is circular; while most recently it has been
claimed that the proof is perfectly valid but rests upon a contradictory
assumption ([11], pp. 157-158). I scarcely exaggerate in saying that
each person interested in the proof has thought that it involves one big
mistake, and the dispute concerns the location of that mistake. I will give
here still another account of the proof and show that the proof rests
upon not one, but several, big mistakes whose significance has not been
previously appreciated.
Some of the problems with Frege's proof involve simple oversights;
others result from deficiencies in the semantical formalism Frege used;
still others are due to conflating substitutional and referential inter-
pretations of logical operators. Finally, one is due to Frege's use of the
semantical version of Axiom V as a premise. As it turns out, semantical
formalisms adequate for Frege's proof were not available until several
generations later. They also took forms which would not have been
acceptable to Frege.
These formal considerations bear upon the interpretation of Frege's
philosophy. Several scholars have wondered whether Frege's proof and
the sections of the Grundgesetze related to it reflect the signs of much
broader semantical and ontological doctrines - such as a contextual
theory of abstract entities. ([2], pp. 139-141; [16].) Certainly, those
sections demand interpretation in the light of Frege's wider doctrines.
But, as will emerge below, the nature of Frege's mistakes and his
inability to give a fully satisfactory proof attenuate conclusions about
his wider doctrines drawn from the proof itself or its attendant seman-
tical formalism. For Frege's proof may have reflected no ontological or
semantical doctrines at all; it may simply have been the best solution he
found for the mathematical problem of proving the referentiality of his
I am not fully prepared to take such a conservative stand. I, too,
suspect that deep issues may have been at stake in the proof. For, as we
will see, there is a simpler and logically more satisfactory version of
\"loth Frege's semantics and his proof which it is hard to believe that he
could have simply overlooked. In my opinion, Frege rejected this alter-
native because he thought that it would not properly represent classes as
dependent upon concepts.
1. Formation rules
The first step towards proving that all well-formed names are referential
is to delineate the class of well-formed names. Frege does this with
formation rules which are totally foreign to contemporary logicians.
Instead of constructing "(x) (y) (x = y)" by first forming "x = y" and
applying the quantifiers "(x)" and "(y)" to it, Frege starts with the first
level function name "; = and an object name A. He then forms
the first level function name "A = ;" which he combines with the
second level function name "(y)o(y)" to yield "(y) (A = y)". Deleting
the "A" from this, he obtains the function name "(y) (; = y)", which
is, in turn, combined with "(x)o(x)" to yield "(x) (y) (x = y)". Frege
calls the method by which two names are combined to yield a new one
the first way of forming names. The method which forms a new name by
deleting a well-formed part of a name he calls the second way. The
closest Frege comes to an inductive definition of the class of well-
formed names is to follow his presentation of the two ways of forming
names with the statement "all correctly-formed names are formed in this
manner".3 There is no problem in giving such an inductive definition,
however ([14J, pp. 76-78).
Although Frege never discusses alternative formation rules, it seems
clear that the rules he does give are chosen to reflect the way in which
objects and functions of the various levels may be combined. Forming
"(x) (y) (x = y)" from "(x)" and "(y) (x = y)" by juxtaposition would
belie the incomplete nature of quantification qua second level function.
Frege often warned against misrepresenting the unsaturated nature of
functions by using names or variables for them which did not, them-
selves, exhibit unsaturatedness.
He would hardly have violated these
strictures for metamathematical convenience. This passage makes it
quite plain:
For the same reason many of Peano's designations in which a function-letter occurs
without an argument are to be rejected. They contradict the very essence of a function,
... Such designations, which belie the real situation, may indeed seem at first sight
convenient, but in the end they always lead into a morass .... But the designations that
promise best to survive are those which adapt themselves most readily to diverse
requirements and can be applied over the most extensive domain, and this because they
fit the subject-matter best. We shall not be able to discover such designations if we are
merely concerned to cope with the case in hand and do not pursue our reflections
further, but only if we attain the deepest possible insight into the nature of the subject-
matter ([6], p. 156).
2. Semantical rules
Just as Frege's ontological views preclude the use of concatenation-style
formation rules, they also exclude the Tarski approach to semantics.
According to Frege, the denotation of a compound name is yielded by
application of the functions denoted by some of its components to the
arguments denoted by its other components. The universal quantifier
symbol, for instance, denotes a second level function, and Frege's
["(x)F(x)"] is to denote the True if for every argument the value of the function is
the True, and otherwise is to denote the False" ([3], p. 12, trans. [9], p. 42).
tells us which function this is. A Tarski-style rule, such as (x)Sx is true if
Sx is true for every assignment of values to x; and is false otherwise,
(where a domain, interpretation and assignment to the free variables in
S is presumed fixed), can be taken as defining a function denoted by the
universal quantifier symbol. But, and this is the important point, it can
also be taken as merely giving conditions for a universal quantification
to be true. Furthermore, the Tarski approach extends to sentences
containing free variables in a way that a Fregean approach should not.
For Frege did not regard open sentences as names and reduced their
content to that of their universal closures ([3], Sections 17 and 32).
Thus the denotation of the universal quantification, "(x) (x = x)", which
is a name, cannot be construed as a function of the denotation of the
"x = x"; for the latter has a denotation only in the derivative sense of
being stipulated to be coreferential with the former. As we shall see,
Frege's non-Tarskian approach precludes the use of a simple inductive
proof to show that if his simple names denote then so do all names
which are correctly formed from them; although that very proof wouid
seem altogether obvious to us today. (Frege's semantical conditions are
like Tarski's in being recursive. Yet Frege's entire conception is unlike
Tarski's in not relativizing interpretations to domains or to assignments
to nonlogical symbols. The idea that a sentence could be true for one
interpretation and false for an other, while familiar to Frege, was totally
inimical to him. For him logic deals with one domain of each logical
type - the universal domain of that type.)5
Frege explains the truth-function and quantifier symbols by directly
specifying the functions which they denote. However, the explanation of
the abstraction symbol does not follow suit. Instead, it is based on the
stipulation "that the combination of signs 'iF(x) = yG(y)' has the same
denotation as '(x) (F(x) = G(x))'" ([3], p. 16; trans. [9], p. 45). This does
not assign denotations to abstracts; it only interprets identities between
them, and leaves open questions concerning identities between abstracts
and other terms. This odd feature of Frege's method, which has puzzled
and provoked most Frege scholars, is responsible for several of the
difficulties with his proof.
It is quite possible that Frege's unusual semantical stipulation was
largely responsible for his undertaking his proof in the first place. For in
Section 10 he attempts to deal with referential indeterminancies engen-
dered by the stipUlation and he refers to this in Section 31. It is also
possible that he regarded his stipUlation and the proof as justifying the
introduction of classes. One can hardly fail to believe that some deep
motive grounded his procedure, since Frege almost certainly considered
the alternative of interpreting the abstraction operator by stipulating
that it designates a second level function which maps each first level
function onto its course-of-values. In fact his words,
If I say generally that
"iF( x)" denotes the course-of-values of the function F( ;),
this requires a supplementation ... ([3], p. 15, trans., [9], p. 44),
virtually amount to such a stipulation. (Be that as it may, only the first
explanation figures in either Frege's semantical discussions or his proof.)
I will postpone further speculation on Frege's motives until later in this
3. The Criteria of Referentiality
Frege Qpens Section 32 of the Grundgesetze with these words:
In this way it is shown that our eight primitive names have denotation, and thereby that
the same holds good for all names correctly compounded out of these. However, not
only a denotation, but also a sense, appertains to all names correctly formed from our
signs. Every such name of a truth-value expresses a sense, a thought. Namely, by our
stipulations it is determined under what conditions the name denotes the True. The
sense of this name - the thought - is the thought that these conditions are fulfilled ([3],
p. 50-51; trans. [9], pp. 89-90).
(This passage is the source of the famous truth-condition theory of
meaning attributed to Frege by Wittgenstein and others.) As the passage
shows, Frege thought that his proof of referentiality not only established
that each well-formed name has a unique reference but also determined
the identity of each name's reference. (This is not to say that their
references were effectively determined or that the proof was construc-
tive.) It is quite clear that Frege thought that the proof showed not only
that "(x) (y) (x = y)", for example, denotes a truth-value but also that it
denotes the True just in case every object x is identical with every object
y. Had Frege not believed this he would have hardly claimed that his
referential stipulations also sufficed to determine truth conditions for
the sentences of his system.
The proof thus had two tasks: 1) to establish that each name has a
(unique) reference and 2) to fix the identity of that reference. Unfortu-
nately, Frege's semantical formalism was not equal to the second task.
Frege needed a semantical formalism which specifies the reference of
each compound name in terms of the reference of its component names.
Only in this way can the reference of each compound name be deter-
mined inductively from the reference of simple names. Now one might
say, "What could be easier? '(x) 0 (x)' refers to the universal quantifier
function; '- .;' refers to the horizontal function; so '(x) (- x)' refers to
the True in case every object is the True, and refers to the False other-
wise." Yes, there is no problem here; but consider instead the case of
"(y) (x) (x = y)". This can be formed from the universal quantifier
functor and the first level function-name "(x) (x = ';)". Hence we
should be able to derive its reference from those of these components.
But then we must already have determined a reference for "(x)
(x = .;)". However, "(x) (x = .;)" must be formed using the "second
way", that is, by forming the name "(x) (x = A)", where A is an object
name, and then dropping the occurence of A. But there is no analogous
method for obtaining the reference of the name "(x) (x) (x = .;)". We
cannot start with the object (x) (x = A) and "knock out" the object A in
analogy to Frege's second way of forming names. For the same object
can be the value of many functions for many different arguments; and,
thus, there is no inverse operation to map a pair of objects into a unique
function. Instead the reference for "(x) (x = .;)" must be obtained by
applying the universal quantifier function to the identity relation. This
would either require universal quantifiers which can apply to two place
functions, ones which can apply to three place ones, etc, or it would
require an operation which applies to functions of any number n of
places and carries them to functions of n-l places in accordance with
the behavior of the universal quantifier. The latter alternative would
violate Frege's theory of types and the former would require having
infinitely many primitives. The abstraction operator poses similar prob-
lems. (To the modern logician the quickest solution would be to add
functional abstraction to Frege's metalanguage. Yet Frege could not
accept this; for him functional abstracts do not designate functions, they
designate objects!)
In Grundgesetze, Section 29, Frege "solved" this problem by intro-
ducing criteria of referentiality which give conditions for a name to be
referential but which fail to determine what the name's reference is. We
read, for example, that "a name of a first-level function of one argument
has a denotation (denotes something, succeeds in denoting) if the proper
name that results from this function-name by its argument places being
filled by a proper name always has a denotation if the name substituted
denotes something" ([3], p. 46; trans. [9], p. 84). As far as these criteria
go, "has a denotation" and "denotes something" can be one place
predicates which ascribe to a name no more connection to objects,
functions, and truth-values than does the predicate "is a quantifier".
This hints that all there is to an expression having a reference is that
it always yields a referential expression when properly combined with
referential expressions. This in turn is akin to the view that an expres-
sion is referential so long as its occurence in a sentence does not
disqualify the sentence itself from having a truth-value. So, perhaps, a
contextual theory of reference is implicit in the Grundgesetze. Despite
the suggestiveness of both these criteria and the very title of Section 29
("When does a name denote something?") and the clear need to account
for Frege's procedure, we should resist attributing a contextual view of
reference to Section 29 of the Grundgesetze. Such an interpretation is
virtually scotched by the opening lines of section 30, where we read:
The foregoing provisions are not to be regarded as definitions of the phrases "have a
denotation" or "denotes something", because their application always presupposes that
we have already recognized some names as denoting. They can serve only in the
extension step by step of the sphere of such names. From them it follows that every
name formed out of denoting names does denote something ([3], p. 46; trans. [9], p. 85).
First, as the passage clearly states, the criteria do not give the entire
account of reference; they neither define what it is for an expression to
refer nor do they suffice for establishing the referentiality of each
expression of Frege's system. They are to be used to show that names
formed from names already known to denote also denote. Plainly, on
pain of an infinite regress, the referentiality of some names must be
established by other means. The remainder of Section 30 discusses the
two ways in which new names may be formed from the names already at
hand. Section 31 then opens with the remark that given what has been
said before, one need only show that the simple names of the system
denote in order to show all the names of the system do, and then turns
to the proof that the simple names do denote. Thus it is clear that the
primary purpose - and probably the sole purpose - of the criteria of
Section 29 is to enable Frege to carry out his proof. To clinch the case,
remember that simple names, with the notable exception of the abstrac-
tion function name, are directly assigned references.
One might object that all this is compatible with a quasicontextual
theory of reference. According to such a theory (some) simple names
are referential because they refer to specific functions or objects, while
other simple names and all compound ones are referential only in the
sense of meeting the criteria of Section 29. But the idea that there might
be such a distinction between names of the same syntactic category
seems utterly unFregean. This is, perhaps, too strong; there is a passage
in which Frege discusses, though in the most tentative terms, an idea
similar to the contextual account. The text occurs in the appendix to the
second volume of the Grundgesetze as part of his reaction to the Russell
Paradox ([4], pp. 225-256; trans. [9], pp. 129-130). Frege says that if
we are unwilling to recognize classes as "proper objects", that is, "as
admissible arguments for every first level function" then we must regard
class names as syncategorematic and without denotation. He continues,
'They would in this case have to be regarded as parts to signs that had
denotations only as wholes." Now this sentence supports my contention
that Frege would mark no distinction between the ways in which names
of the same syntactic category can refer; names that do not refer
standardly are not really names at all. (Frege proceeds to reject the
syncategorematic approach, because it excludes classes and thereby
numbers, from the range of object quantifiers).
There is a curious footnote to the sentence last quoted that I find
difficult to reconcile with my interpretation. The footnote refers the
reader to Vol. 1, Section 29, where the criteria of referentiality are
given. Since the footnote is appended to the sentence stating that class
names refer only in the sense of occurring in wholes which have
denotation, it seems to imply that the criteria of referentiality in Section
29 do not give sufficient conditions for genuine reference. In a sense,
this does support my view, since it says that contextual reference is not
reference at all. It also demolishes my view that Frege thought that the
criteria of referentiality afforded a means for proving that certain names
have reference in the full sense. There is a way out. The sentence prior
to the one footnoted says that a proper object must be admissible as an
argument for every first-level function. But a corollary of this is that a
proper name refers to an object if and only if substituting it into the
argument place of any first level function name which refers to a
function yields a name which denotes the value of that function for the
object as argument. This, of course, resembles the criterion of referen-
tiality of proper names. Thus it is quite possible that the otherwise
puzzling footnote was intended for the sentence prior to the one to
which it was in fact affixed. If we" accept this explanation then there are
no problems with my reading of the text in question. We can safely
conclude that the criteria of Section 29 were intended neither as a
contextual nor as a quasicontextual theory of reference.
It is worth recalling at this point that there are other explanations for
the form taken by the criteria of Section 29, which have nothing to do
with contextual views of reference. The first is that Frege may have
known no other way to carry out the inductive part of his proof. The
second is that he may have realized that approaches of the sort we
considered above would conflict with his fundamental principles. I
accept the first explanation. As noted earlier, Frege was apparently
unaware of the failure of his proof to determine the identity of the
reference of each name. Furthermore, another problem with Frege's
proof - his use of what I will call the substitutional-referential approach
to quantification and abstraction - indicates that he may have been
unaware of the subtle differences between a name being referential in
the sense of his criteria and referring to some particular entity. Finally,
the semantical formalisms (the predicate functor logics and algebraic
semantics) which measure up to a Fregean proof of referentiality have
only been recently developed.
Do the criteria of referentiality succeed in at least establishing that
every name is referential given that the simple names are? Yes, but they
establish too much. For they state, in essence, that a name of a given
type is referential provided it yields a referential name whenever
properly combined with other referential names.
By appealing to them
one can introduce names into Frege's system which the criteria will
mark as referential but which fail to refer according to Frege's more
familar canons of referentiality. For instance, let us suppose that all the
names in the system refer in the usual sense of designating an entity of
the type appropriate to their syntactic category. Now let us introduce a
new first-level function name, "S", with the stipulation that "S(E)"
refers to the reference of E if E has one. Here E may be any proper
name, including one which has not yet been introduced. Then the
criteria of Section 29 will deem" S" as a referential function name,
because whenever it is combined with a referential proper name the
resulting compound name has a denotation, namely, that of the proper
name. But plainly "s (';)" does not denote a totally defined function.
Consequently, our introduction of it violates Frege's principle of com-
pleteness ([4], sects. 56-65). Yet once again I cannot believe that
Frege was aware of this consequence of his criteria or the difficulties
it raises for his proof.
4. The Proofin Section 31
In this section Frege tries to prove that his simple names have refer-
ences, assuming that once this basis case is established the referentiality
of his compound names will follow automatically from the criteria of
Section 29. He takes the referentiality of the truth-functors to be imme-
diate and pauses briefly over the quantifiers. Here he uses the substitu-
tional-referential approach. He shows that "(X)0(X)" is referential by
showing that "(x)F(x)" is, for every referential function name, F(';). In
other words, he shows that every referential substitution for "0" yields a
referential compound instead of showing "(x) 0 (x)" denotes a function
which has a value for every first level function as argument. He does
that in tum by noting that "F(A)" denotes the True for every referen-
tial object name A or it denotes something else (again, rather than by
arguing that the function denoted by "F(';)" has a value for each
argument). As we have already seen, this form of reasoning would lead
Frege to other consequences which would be totally unacceptable to
him. The major concern in Section 31, however, is with the abstraction
Due to its unusual semantical stipulation, Frege approaches the
abstraction sign differently. He argues that he can restrict himself to
occurences of abstracts within identity contexts, and notes that by using
the criteria of referentiality he need only show that if F is referential
then so is the function name "iF(x) = ';". To do so he need only show
that "iF(x) = A" is referential for each referential object name A. But in
Section 10 he had identified the truth-values with two value-ranges.
Therefore, he restricts himself to those A which are referential abstracts
of the form "yG(y)". Thus everything reduces to showing that
"iF(x) = yG(y)" is referential. And by the semantical stipulation for
abstracts this is if "(x) (F(x) = G(x" is. Of course, it is referential,
given that F and G are.
There are several problems with this argument. First, Frege has failed
to provide for all the contexts in which an abstract may occur. One of
these contexts is "(g) (g(';)", and the proof becomes circular with
respect to it. The context is a referential first level function name.
Hence, following the model of the rest of Frege's proof, the way to show
that an abstract A is a referential abstract is to show that "(g) (g(A)" is
referential (in addition to showing that A = .; is). However that is
shown by showing that F(A) is referential for every referential first-level
function name F(';), and "(g) (g(';)" is one of these; so we have gone in
a circle.
(Note that we do not have this problem with showing that "(g)
(g( ';)" is referential; for we can restrict ourselves to function names
already known to be referential and use the semantical condition for the
universal quantifier.)
Bartlett, who was the first to see that Frege's proof was faulty,9
observed that it is circular in another sense. He noted that the proof
shows only that if F(';) is referential then so is "iF(x)", and that in
admitting the abstraction operator we can form first-level function
names in which the abstraction operator itself occurs. Thus to apply the
italicized principle non-circularly we must order the abstracts appropri-
ately, and the presence of impredicative abstracts is incompatible with
the ordering required. Parsons follows Bartlett in making a similar
Of course, as I have just remarked, even the proof of the
italicized principle fails.
On the other hand, Martin sees no problem at all with the features of
the proof considered so far, and thinks that its only flaw is that it rests
upon a contradictory assumption - the semantical embodiment of
Axiom V. I find his assessment dubious. The Bartlett-Parsons criticism
points out that Frege tried to run an inductive proof while confusing the
basis and inductive cases. An inductive proof should furnish a scheme
for establishing the theorem's assertion with respect to any finite initial
sequence of its instances, and that is exactly what Frege's proof fails to
do. Martin apparently thinks that there is another procedure, which is at
least implicit in Frege's proof, and that it can handle with ease the
impredicative abstracts, such as "y(j) (x/(x) = y)".
Here is how the Martin method works. We can prove that
(1) yif) (Xf(x) = y)
is referential by first applying the method of Section 31 (the basis case)
to reduce the referentiality of (1) to that of
(2) yif) (x/(x) = y) = xF(x),
where F( s) is a referential function-name. The referentiality of (2) then
reduces to that of
(3) (z) [if) (x/(x) = z) = F(z)].
Then we apply the inductive step, taking the universal quantifier to be
referential, and reduce the referentiality of (3) to that of
(4) if) (x/(x) = B) = F(B),
where B is any referential object-name. Since F(s) and B are referential
and the identity symbol is already known to be referential, the referen-
tiality of (4) reduces to that of
(5) if) (x/(x) = B),
and since the universal quantifier symbols are already known to be
referential we can reduce this case to
(6) xG(x) = B,
where G( s) is referential function-name. Then returning to the method
of the basis case we replace B by an abstract formed from a referential
function-name, say, xF(x), and reduce the question of the referentiality
of (6) to that of
(7) (x) (G(x) = F(x,
which is referential since F and G are. Undoubtedly, this method can be
applied to any wellformed abstract of Frege's system.
Thus if we assume 1) the criteria of referentiality, 2) the semantical
version of Axiom Y, and 3) the referentiality of the truth-function and
quantifier symbols, then we can argue validly that each wellformed
name is referential. For in a finite number of steps Martin's procedure
will reduce the question of the referentiality of any name to that of a
name composed of the quantifier and truth-function symbols. Thus it
could be argued that Frege had another "proof' of referentiality in
mind; to wit, one which first establishes that all names constructed from
the truth-functions and quantifiers are referential and then uses this to
show that the abstraction operator is referential.
Is this what Frege had in mind? I think not. For this account fails to
respect Frege's division of his proof into a basis case (all simple names
are referential) and an inductive case (if N is a name correctly formed
from referential names then N is referential too). Martin's method
"proves" that the abstraction functor (a simple function name) is refer-
ential by using the results of the inductive case. In addition, Frege's
remark that the abstraction operator greatly complicates matters sug-
gests that he did not think that a separate proof of referentiality for the
truth-function and quantifier symbols could be carried over intact once
the abstraction symbol was added.
I must confess some uneasiness with this criticism, since it is possible
that Frege attempted to give an induction within the basis case, and
confused this with an argument like Martin's. Whether this was Frege's
procedure or not, it is not generally sound. Consider a language con-
taining a supply of predicates and the description operator. Suppose
that no vacuous descriptions are constructable in this language. Then
suppose that we add a new predicate F, which is referential considered
in itself and together with the other predicates of the language. It is still
entirely possible that the definite description, ''the F", will fail to refer.
Then what should we the blame for the failure of reference? Surely, the
problem lies with the whole system of names and is not a fault of the
predicates or of the description operator taken singly. The proof
considered above suffers from the analogous mistake of assuming that if
the truth-function and quantifier symbols are referential then they auto-
matically remain so in the presence of the abstraction symbol. This
assumption, in tum, can be traced to the criteria of referentiality, whose
other faults we have already noted.
A common opinion is that no sound proof of referentiiility is possible
for Frege's system because it would amount to a proof of its consistency.
Here is how Sluga has recently put it:
Since he treats sentences as names of truth-values, the proof that every name has exactly
one reference amounts to a proof that no sentence refers to both the True and the False.
If Frege's argument were successful it would in effect give us a consistency proof for the
system of the Grundgesetze ([15), p. 167).11
But without some qualification that is clearly mistaken. The set of
sentences of a formal language and the set of its theorems are not
necessarily identical. It is quite easy to give examples of interpreted
formal languages with sentences having unique truth-values and incon-
sistent sets of theorems. Thus referentiality does not imply consistency.
Furthermore, it is also easy to construct consistent formal languages
which have nonreferring terms, and are thus not referential.
The independence of referentiality from consistency can be demon-
strated for the symbolism of the Grundgesetze too. By using a standard
interpretation (or Frege's) for the quantifier and truth-function symbols
and by assigning each abstract the True as its denotation, one can secure
referentiality while falsifying Frege's Axiom V. This together with
Russell's contradiction establishes referentiality and inconsistency. On
the other hand, by using a partial valuation for Frege's language we can
verify a subset of his axioms. Then by taking these as the full set we
obtain a consistent but not fully referential system (this is analogous to
Quine's mixing of virtual and real classes).
But we cannot give a sound interpretation of Frege's system in which
abstracts denote value-ranges construed as objects that satisfy Axiom V.
The Russell paradox shows that there are no such objects. This means
that no sound repair of Frege's original proof is possible (perhaps, this is
all that Sluga and others have meant). Despite this there is no reason to
think that we cannot interpret abstracts as denoting objects like value-
ranges, so long as we lay down sufficiently weak axioms governing them.
There is also an indirect connection between consistency proofs and
Frege's particular proof. That proof could be seen as an attempt to use
Axiom V, or its semantical correlate, to eliminate abstracts in favour of
formulas of second order logic. If all occurrences of abstracts were
eliminable in favor of second order logic, we would have a consistency
proof of Frege's system relative to second order logic. Thus Martin is
right; there would be no way to eliminate abstracts entirely from Frege's
system even if the referentiality proof were correct. I find little reason to
believe that Frege had the elimination of abstracts in mind when he gave
his proof. When Russell proposed the no-class theory to Frege, he used
a definition quite similar to Axiom V. But Frege was only concerned
with Russell's failure to represent functions as incomplete, and, though
he did compare Russell's abstracts with his, he did not show much
interest or enthusiasm for the no-class idea itself ([51, pp. 160-161).
We have just seen that referentiality and consistency are independent.
Does referentiality bear on other major methodological issues? At the
beginning of this paper I emphasized that there was a strong connection
for Frege between referentiality and good methodology. Some of the
passages I cited were portions of Frege's criticisms of the faulty defini-
tions, such as contextual definitions, implicit definitions, conditional
definitions or creative definitions, which many of his contemporaries
used. It is thus ironic that a proof of referentiality for the primitives of a
system in no way insures against improper definitions. Frege seems to
have seen this too, and laid down rules of definition which can be shown
to satisfy Lesniewski's criteria of eliminability and noncreativity.12
Whether or not the primitives of a system are referential, definitions in it
which satisfy these criteria will be formally satisfactory in the sense of
giving rise to a conservative extension of the system. If the primitives are
referential besides, then such definitions will guarantee that the defined
expressions are too.
However, there is no methodological impropriety in presenting,
developing and studying uninterpreted formal systems. Hence Frege's
requirement of referentiality is not a requirement for rigorous work with
formal systems - at least not as such work is done today. Furthermore,
even if rigor did demand that formal systems be interpreted, referen-
tiality need not be a requirement of rigor. For some rigorously inter-
preted systems might contain nondenoting terms. Referentiality is not an
uncontroversial requirement of rigor, despite the emphatic protests by
Frege with which I opened this paper.
Granted, Frege believed that a formal system of logic cannot be
applied unless it is fully referential; thus he portrayed formalism as a
grievous methodological error. Many formalists criticized by Frege did
employ faulty definitions and committed other methodological sins;
however, formalism itself can be given rigorous foundations. (Indeed,
Frege saw how to do it. (Cf. [13], pp. 54-62. Formalism is at worst a
philosophical error and not a methodological one. Frege would like us
to think that referentiality is a requirement of rigor; and it is, given his
views on logic and inference. But we must recognize that those have
never been uncontroversial views. From the point of view of rigorous
mathematical practice, both of today and of Frege's time, referentiality
proofs are merely embellishments.
I have argued that Frege's proof contains a number of mistakes and
confusions, that it would not prove what he wanted even if it were
correct, and, finally, that the demands of rigor do not even require it.
Yet we still have no explanation for Frege's giving the proof in the first
place. As I noted earlier, several possibilities come to mind. Frege did
interpret his language referentially, and, thus, the proof was entirely
appropriate for a rigorous presentation of his language. Perhaps that
was the sole motive for Frege's proof. Another possibility, suggested by
the criteria of referentiality, is that Frege used the proof to buttress a
contextual theory of reference. I argued against that interpretation
earlier. Possibly Frege felt that both the demands of rigor and the
peculiarity of his semantics for abstraction necessitated giving the proof.
This explanation seems the most plausible to me.
But this just raises another question. Why did Frege use the strange
semantical stipulation for abstraction? Was it in order to guarantee that
Axiom V would be true by virtue of the meanings of its terms? Suppose
that he had introduced value-ranges in his metalanguage (as he in fact
did), with the stipulation that he would use the words "the function
has the same value-range as the function as having the same
denotation as the phrase "the functions and have identical
values for identical arguments" (Cf. [3], section 3). But then suppose that
he went on (as he did not), to stipulate that the abstraction operator is a
second-level function which maps first-level functions onto their value-
ranges. His proof would have been greatly simplified and Axiom V,
being a formalization of his meaning condition, would remain true in
virtue of the meaning of its terms.
However, there is a serious flaw in Frege's metalanguage, and it
would persist even if Frege made the changes just mentioned. Frege
introduced the term ''value-range'' by means of a contextual stipulation.
It is dangerously similar to definitions which he strongly castigatedY
Value-ranges themselves are not necessarily at hand prior to Frege's
stipulation nor are they introduced by an explicit existential postulate.
Frege could have avoided this situation by insisting instead that value-
ranges exist and that it must be recognized as a fundamental logical law
that there is a one to one association between them and first-level
Then why didn't he do that? One is again tempted to see Frege as
resisting the recognition of value-ranges in their own right, to see them
as epiphenomenal, or along the lines of Russell's no-class theory or the
results of a contextual view. But, as emerges in his discussion of the
Russell paradox, Frege clearly thought of value-ranges as full-fledged
objects and would reject any resolution of the paradoxes which treated
them as less than SUCh.
The explanation for Frege's treatment of
value-ranges seems to me to be connected with his insistence that the
concept of class is a derivative concept of logic and that classes must be
conceived of as extensions of concepts - the more fundamental logical
If we introduced value ranges, and classes, by means of an
axiom that read,
With each function there is associated in a one to one fashion an
object, its value-range,
we would fail to emphasize the "derivative" status of classes. Indeed,
Frege might have supposed that since his readers might identify value-
ranges with one of their own pet entities, e.g., with classes in Schroeder's
sense, the logical self-evidence of the axiom introducing them would not
be assured. I think that this explanation is as good an interpretation of
the very puzzling treatment of value-ranges in the Grundgesetze as we
currently have.
My account of Frege's attempt to prove his system referential por-
trays him as confused and groping. I will neither moderate nor apologize
for it. But Frege's greatness demands that it be put into perspective.
Prior to Frege no one had clearly grasped the notion of a formal system,
no one had even formulated the distinction between a system and its
interpretation;nor had anyone tried to state or prove theorems relating
them. Frege was the first to do all of these. Moreover, in assessing
Frege's failure we must recall that no satisfactory general semantical
theories for formal languages existed until Tarski's. If, in trying to rank
Frege against other great figures of early mathematical logic, we
compare his attempt with Hilbert's early effort (1904) to prove the
consistency of arithmetic, the result is an embarrassment to Hilbertp6
The University of North Carolina
Chapel Hill
1 The criteria of referentiality given in Section 29 state, roughly, that a name is
referential in case it yields referential names whenever combined properly with other
referential names. The opening paragraph of Section 31 contains the sentence: "By what
has been said, it is necessary to this end only to demonstrate of our primitive names that
they denote something".
2 This view, which has been advanced by Bartlett and Sluga, is related to Frege's own
remark to Russell (in the letter of 22.6.1902) that his "explanations to Sect. 31 do not
suffice". But note that this remark is made with reference to the specific semantical
stipulations for abstracts which he used in Section 31. See Note 9 below. It does not
preclude proofs based upon seman tical stipulations other than those used by Frege. See
[1], p. 75; [15J, p. 167.
3 [3], p. 48; trans., [9], p. 86.
4 For two examples see: [8], pp. 663-665; trans., [10], pp. 113-115; [4], p. 148, Note
2; trans., [10], p. 180.
5 See the Frege-Hilbert and Frege-Peano exchanges.
6 Martin was, to my knowledge, the first to remark on this problem with Frege's proof
([11], p. 163).
7 In fact Frege fails to lay down criteria which cover each way of combining names in
his system.
S This is not the circle discussed by Bartlett or by Parsons, but it has been noted by
Thiel ([17], p. 82).
9 In the letter to Russell giving his first reactions to the contradiction Frege wrote:
my Law Y (Sect. 20, p. 36) is false, and ... my explanations in Sect. 31 do not
suffice to secure a meaning for my combinations of signs in all cases ([5], p. 132).
It is thus clear that Frege realized that something was wrong with his proof, but given
that he connected the problem to Axiom Y, it is most likely that he saw his proof was
based upon an incorrect semantical stipulation for abstraction and was unaware of the
other problems with the proof.
10 See [1], pp. 72-75; [12], p. 190.
II See also [I],p. 75.
12 Lesniewski's notion of creative definitions is not the same as Frege's. To Frege a
creative definitiol'l amounts to mislabeling, as a definition, an existence postulate or
theorem. Frege discusses "creative definitions" in [4], Sect. 143; his principles of
definition are given in [3], Section. 33.
13 This did not escape Frege's notice. In [4], pp. 148-149 Frege argues that since he
introduces value-ranges by means of an axiom rather than a definition his procedure is
not open to the criticisms he had directed at his colleagues. My claim, however, is that
his metalanguage uses such a definition.
14 The question of why Frege introduced value-ranges in his metalanguage by means of
a' contextual stipulation is related to, but distinct from, the question of why he was
willing to settle by stipulation the question of whether, say, Julius Caesar is a value-
15 This is especially clear in [5], pp. 191-192. See [13], pp. 206-207 for further
16 I would like to thank Catherine Elign, Susan Hale, Philip Kitcher and Andrew Rein
for their help with this paper.
[1] Bartlett, J.: 1961, Funktion und Gegenstand, Dissertation, Ludwig-Maximilians-
Universitiit, M. Weiss, Munich.
[2] Dummett, M.: 1981, The Interpretation of Frege's Philosophy, Harvard University
Press, Cambridge.
[3] Frege, G.: 1893, Grundgesetze der Arithmetik, Vol. I, H. Pohle, Jena.
[4] Frege, G.: 1903, Grundgesetze der Arithmetik, Vol. II, H. Pohle, Jena.
[5] Frege, G.: 1980, Philosophical and Mathematical Correspondence, edited by H.
Hermes, F. Kambartel, C. Thiel and A. Veraat, abridged from the German by B.
McGuinness and translated by H. Kaal, University of Chicago Press, Chicago.
[6] Frege, G.: 1979, Posthumous Writings, edited by H. Hermes, F. Kambartel and F.
Kaulbach, translated by P. Long and R. White, University of Chicago Press,
[7] Frege, G.: 1892, 'Uber Sinn und Bedeutung', ZeitschriJt fUr Philosophie und
philosophische Kritik 100, 25-50.
[8] Frege, G.: 1904, 'Was ist eine FunktionT, in Festschrift fUr Ludwig Boltzmann
gewidmet zum sechzigsten Geburtstage, 20. Februar 1904, A. Barth, Leipzig.
[9] Furth, M.: 1964, The Basic Laws of Arithmetic: Exposition of the System, a partial
translation of [3] , University of California Press, Los Angeles.
[10] Geach, P. and Black, M.: 1952, Translations from the Philosophical Writings of
Gottlob Frege, Blackwell, Oxford.
[11] Martin, E.: 1982, 'Referentiality in Frege's Grundgesetze', History and Philosophy
of Logic 3, 151-164.
[12] Parsons, c.: 1965, 'Frege's Theory of Number', in Philosophy in America, edited
by M. Black, Cornell University Press, Ithaca.
[13] Resnik, M.: 1980, Frege and the Philosophy of Mathematics, Cornell University
Press, Ithaca.
[14] Resnik, M.: 1963, Frege's Methodology, Dissertation, Harvard University.
[15] Sluga, H.: 1980, Frege, Routledge and Kegan Paul, London.
[16] Snapper, J.: 1974, 'Contextual Definition: What Frege Might Have Meant but
Probably Didn't', NoUsS, 259-272.
[17] Thiel, c.: 1965, Sinn und Bedeutung in der Logik Gottlob Freges, Verlag Anton
Hain, Meisenheim am Glan.
Logicism by the end of the 19th century was a philosophical doctrine
whose time had come, and it is Gottlob Frege to whom we owe its
arrival. "Often," Frege once wrote, "it is only after immense intellectual
effort, which may have continued over centuries, that humanity at last
succeeds in achieving knowledge of a concept in its pure form, in
stripping off the irrelevant accretions which veil it from the eyes of the
mind" ([Fd] , p. xix). Prior to Frege, logicism was just such a concept
whose pure form was obscured by irrelevant accretions; and in his life's
work it was Frege who first presented this concept to humanity in its
pure form and developed it as a doctrine of the first rank.
That form, unfortunately, has become obscured once again. For
today, as we approach the end of the 20th century, logicism, as a
philosophical doctrine, is said to be dead, and even worse, to be
impossible. Frege's logicism, or the specific presentation he gave of it in
[Gg], fell to Russell's paradox, and, we are told, it cannot be resurrected.
Russell's own subsequent form of logicism presented in [PM], moreover,
in effect gives up the doctrine; for in overcoming his paradox Russell
was unable to reduce classical mathematics to logic without making at
least two assumptions which are not logically true; namely, his assump
tion of the axiom of reducibility and his assumption of an axiom of
infinity regarding the existence of infinitely many concrete or non
abstract individuals.
Contrary to popular opinion, however, logicism is not dead beyond
redemption; that is, if logicism is dead, then it can be easily resurrected.
This is not to say that as philosophical doctrines go logicism is true, but
only that it can be logically reconstructed and defended or advocated in
essentially the same philosophical context in which it was originally
formulated. This is true especially of Frege's form of logicism, as we
shall see, and in fact, by turning to his correspondence with Russell and
his discussion of Russell's paradox, we are able to fonnulate not only
one but two alternative reconstructions of his form of logicism, both of
which are consistent (relative to weak Zermelo set theory).
In regard to Russell's form of logicism, on the other hand, our
L. Haaparanta and J. Hintikka (eds.), Frege Synthesized, 197-252.
1986 by D. Reidel Publishing Company.
resurrection will not apply directly to the form he adopted in [PM] but
rather to the form he was implicity advocating in his correspondence
with Frege shortly after the completion of [POM]. In this regard, though
we shall have occasion to refer to certain features of his later form of
logicism, especially in our concluding section where a counterpart to the
axiom of reducibility comes into the picture, it is Russell's early form of
logicism which we shall reconstruct and be concerned with here.
Though Frege's and Russell's early form of logicism are not the same,
incidentally, they are closely related; and one of our goals will be to
reconstruct or resurrect these forms with their similarity in mind. In
particular, it is our contention that both are to be reconstructed as
second order predicate logics in which nominalized predicates are
allowed to occur as abstract singular terms. Their important differences,
as we shall see, will then consist in the sort of object each takes
nominalized predicates to denote and in whether the theory of predica
tion upon which the laws of logic are to be based is to be extensional or
The doctrine of logicism can be succinctly stated in the following two
fold claim: (1) that all of the concepts of classical mathematics are
explicitly definable in terms of purely logical concepts; and (2) that all of
the theorems of classical mathematics can be derived from the laws of
logic through purely logical deductions (cf. [Camap]). This is not a
doctrine about the reducibility of mathematics to set theory, it should be
noted, but about the reducibility of classical mathematics to the con
cepts and laws of logic. In other words, it is not a doctrine about the
reducibility of mathematics to a theory of membership (in a class or set)
but about the reducibility of mathematics to a theory of predication, and
in particular to a theory about the concepts which predicates stand for
in their role as predicates. For this reason both Frege and Russell
maintain that the logistic framework within which classical mathematics
is to be represented must consist at least of a second order predicate
logic where quantification is not only with respect to the role of singular
terms but to that of predicates as well. Indeed, it was Frege himself who
first formulated and developed standard second order predicate logic,
and he did so precisely as a framework within which classical mathe
matics was to be reduced to logic.
The distinction between predicates and singulars terms, it must be
emphasized, is fundamental to both Frege's and Russell's forms of
logicism; and to fail to attend to this distinction is to fail to understand
the nature of predication in either framework. In particular, the attempt
to characterize predication in terms of a first order theory of exemplifi
cation (or a first order theory of membership without an axiom of
extensionality) is a nonstarter as far as either form of logicism is
concerned. Predicates are not singular terms or what Frege called
"proper names" in his extended sense, and the role of concepts in
predication is not that of objects or individuals to which other objects or
individuals stand in a relation of exemplification.
Now this does not mean that predicates cannot be nominalized and
transformed into singular terms the way that 'human', 'triangular', 'wise',
etc. can be transformed into 'humanity', 'triangularity', 'wisdom', etc.; or
to use Frege's example, the way 'horse' can be transformed into 'the
concept horse'. Rather, the point is that any account of nominalized
predicates as abstract singular terms presupposes an account of their
role as predicates; and in particular any relational predicate such as
'exemplifies' or 'falls under', as in 'Socrates exemplifies humanity' and
'Bucephalus falls under the concept horse', is to be viewed as derived
from an account of predication in which predicates do not have such
nominalized forms. Indeed, the priority of the role of a predicate as a
predicate over the corresponding role of its nominalization as an
abstract singular term is in fact one of the ways we are to understand
Frege's famous context principle (cf. [Fd], p. xxii). For it is only in the
context of a sentence that a predicate can occur as a predicate, and it is
only throught a correlation with such occurrences that we are to under
stand the role of a nominalized predicate as an abstract singular term.
This sort of correlation is fundalnental not only to Frege's logicism but
also to the reconstruction we shall propose.
What distinguishes a predicate from its nominalization, according to
Frege, is that a predicate has an unsaturated nature which is essential to
its role in predication. That is, whereas a nominalized predicate is a
complete, saturated expression in its own right, the predicate itself is in
need of supplementation, and it is this unsaturatedness or need of
supplemenatation which is the basis of its predicative nature. Moreover,
because this unsaturatedness is essential to its role in predication, the
concept which a predicate stands for (bedeutet) is said by Frege to have
a corresponding unsaturated nature as well. Indeed, it is precisely the
corresponding unsaturatedness of a concept which Frege identifies as
its predicative nature (cf. [G & Bj, p. 47, [PW], p. 177).
Now just as the predicative nature of a predicate excludes its being a
singular term, so too, according to Frege, the predicative or unsaturated
nature of a concept excludes it from being the dena tatum (Bedeutung)
of a singular term; and therefore what a nominalized predicate denotes,
according to Frege, is not the concept which the predicate otherwise
stands for in its role as a predicate. Instead, according to Frege, the
concept "must first be converted into an object, or, speaking more
precisely, represented by an object" ([G & Bj, p. 46), and it will be this
object, or concept-correlate as Frege also calls it, which is the real
denotatum of the nominalized predicate. Thus in particular, what 'the
concept horse' denotes, according to Frege, is not a concept but an
object, that is, a concept-correlate.
This is not the conclusion Russell comes to in [POMj and thereafter,
it should be noted; and in fact, as already indicated, Russell's different
view in this regard is one of the principal differences between his and
Frege's form of logicism. Russell did hold something rather similar to
Frege's view in an unpublished manuscript of 1898, it might be noted;
but in [pOM], it is clear, he explicitly rejects the earlier view. Thus
(keeping in mind that terms and predicates for Russell are individuals
and concepts, respectively, and not expressions) in 1898 he wrote that
"the peculiarity of predicates is that they are meanings," and "although it
is impossible to speak of meanings without making them subjects ... ,
yet meanings as such are the antithesis of subjects, are destitute of being,
and incapable of plurality. When I say 'Socrates is human,' human as
used in this judgment does not have being, and is not a logical subject. I
am, in a word, not asserting a relation between two subjects. As soon as
I make human a term, ... I have added something, namely being, one-
ness, and diversity of being from other terms which human as predicate
did not not possess" ([AMRj, book I, p. 10).
Russell briefly restates this view in [pOMj, observing that "it might be
thought that a distinction ought to be made between a concept as such
and a concept as used as a term" (p. 45); but he quickly rejects it,
claiming that "inextricable difficulties will envelop us if we allow such a
view" (ibid.). The difficulties in question, it turns out, are those con-
nected with such claims as that the concept horse is not a concept,
which Russell thinks is false and leads to a contradiction. His conclusion
is that "terms which are concepts differ from those which are not, not in
respect of self-subsistence, but in virtue of the fact that, in certain true
or false propositions, they occur in a manner which is different in an
indefinable way from the manner in which subjects or terms of relations
occur" (ibid., p. 46). In other words, concepts, including relations, are
individuals (since 'term' for Russell is synonymous with 'individual' - cf.
[POM], p. 43), and they are denoted as such by nominalized predicates
occurring as abstract singular terms. Yet, predicate expressions are not
singular terms and the concepts which predicate expressions stand for in
their role as predicates do not occur in propositions the way that
individuals denoted by singular terms do. That is, even though concepts,
according to Russell, do not have an unsaturated nature, nevertheless
they do have a predicative nature in virtue of which they can occur in
propositions in "a manner which is different in an indefinable way" from
that in which individuals denoted by singular terms occur.
Now the "indefinable way" in which a concept can occur in a
proposition as a concept and not as a term, according to Russell, is also
a feature of what he calls a propositional function; and in fact in [pOM]
a propositional function can occur in a proposition only in this "indefin-
able way". That is, unlike concepts, propositional functions in [POM]
are not individuals. To be sure, to each concept there corresponds a
unique propositional function in the propositional values of which the
concept occurs as a concept; but upon discovering his paradox, Russell
was led in [POMI to doubt that every propositional function is either
itself a concept or that it has a concept corresponding to it. Of course,
"apart from the contradiction in question," Russell observes, "this point
might appear to be merely verbal: 'being an x such that fjJx,' it might be
said, may always be taken as a predicate [concept]. But in view of our
contradiction," Russell continues, "all remarks on this subject must be
viewed with caution" (ibid., p. 88). In particular, the contradiction is
avoided for Russell in [POM) "by the recognition that the functional
part of a propositional function is not an independent entity" (ibid.); and
in fact it was precisely for this reason that Russell was led in his
commentary on Frege (in Appendix A of [POM]) to claim that "the
word Begrijf is used by Frege to mean nearly the same thing as
propositional function" (p. 507).
In his later form of logicism in [PM], Russell was able to avoid his
paradox while also claiming that propositional functions are single
entities after all; i.e., that propositional functions are individuals, albeit
of a higher order/type than concrete individuals, a fact which is often
missed or ignored by philosophers and logicians alike since, contrary to
his earlier practice, Russell chose to use 'individual' in [PM] only to refer
to concrete individuals. There is no point in distinguishing concepts
from propositional functions in such a framework, needless to say, and
in fact we shall not assume any such distinction even in our resurrection
of the form of logicism which is implicit in Russell's correspondence
with Frege shortly after [POM] was written. That is, having noted the
difference between concepts and propositional functions which Russell
thought he was committed to in [POM] as a result of his paradox, we
shall nevertheless assume that concepts, including relations, are none
other than propositional functions, since that in fact was what Russell
originally thought and returned to in his later form of logicism. We
return in this way to the original context of Russell's paradox as it
applies to Russell's early form of logicism no less so than as it applies to
Frege's form of logicism. In that regard, in other words, we may assume
that the only important difference between Frege's and Russell's early
form of logicism which need concern us at this point is that whereas
according to Frege nominalized predicates denote concept-correlates,
i.e., objects somehow correlated with concepts, for Russell nominalized
predicates denote the same concepts or propositional functions which
the predicates in question otherwise stand for in their role as predicates.
What is common to both forms of logicism, on the other hand, is that
the concepts assumed in each form have a predicative nature, and that it
is this predicative nature which is the basis of the laws of logic.
Standard second order predicate logic with identity, we have already
noted, was first formulated by Frege as a framework in which to carry
out the reduction of classical mathematics to logic. The reduction is not
forthcoming, to be sure, without a logistic treatment of nominalized
predicates or some such equivalent device, but still it is a framework in
which the laws of logic have their basic form prior to any assumption
about how nominalized predicates are to be interpreted. In this regard it
is a system of the laws of logic which are common to both Frege's and
Russell's early form of logicism; and for that reason we shall tum to its
formulation first.
Accordingly, let us use 'x', 'y', and 'z', with or without numerical
subscripts, to refer (in the metalanguage) to individual variables, and
similarly let us use 'pn" 'Gn' and 'Rn' to refer to n-place predicate
variables. (JVe shall usually delete the superscript when the context
makes clear the number of subject or argument positions that go with a
predicate variable or constant; and we shall only use 'Rn' when n > 1.)
As primitive logical constants, let us take -> (the material conditional
sign), - (the negation sign), = (the identity sign), and 'if (the universal
quantifier sign). As usual, we understand the juxtaposition of signs to
represent their concatenation. We shall also use parentheses and
brackets as auxiliary signs.
Ignoring the introduction of predicate and individual constants for
the special applications of logic, the basic or atomic formulas are all of
the expressions of the forms x = y and pn(XI> ... , x
), where n is a
natural number. Well-formed formulas, or wjfs, are then defined as the
members of the smallest set K containing the atomic formulas and such
that - <jJ, (<jJ -> 'IjJ), ('if x)<jJ, ('if P)<jJ are in K whenever <jJ, 'IjJ are in K and x
and P are an individual and an n-place predicate variable, respectively,
for all natural numbers n. For convenience, we shall use '<jJ' and ''IjJ' to
refer (in the metalanguage) to wffs (with and without predicate and
individual constants as well). We assume the usual notions of bondage
and freedom of variables, of one variable or constant being free for
another in a given wff, and of the proper substitution of a wff for an n
place predicate variable (relative to n individual variables occurring free
in that wff as subject-position indicators).
No function symbols other than predicate variables (and constants)
have been introduced into our present logical syntax, it should be noted;
and in this regard, it might be said, our formalism is more in line with
Russell's form of logicism than with Frege's. For whereas Frege
explained predication in terms of the mathematical notion of function-
ality (cf. [G & B], p. 47), Russell explained mathematical functionality in
terms of predication, i.e., in terms of the predicative nature of proposi-
tional functions. That is, according to Russell, "the sort of function
which is fundamental in logic is the propositional function; and the
functions customary in mathematics are defined by means of this" (EA],
p. 261). Thus "if f(x) is not a propositional function, its value for a given
value of x ... is the term y satisfying, for the given value of x, some
relational proposition; this relational proposition is involved in the
definition of f(x), and some such propositional function is required in
the definition of any function which is not propositional" ([POM], p.
All functions other than propositional functions, in other words, can
be identified with many-one relations (cf. [POM], p. 83); and this, from
either a philosophical or a logical point of view, is not an unimportant
observation. Stated in this way, however, the observation in no way runs
counter to Frege's form of logicism. For just as Frege analyzed all truth-
functions in terms of those for negation and the material conditional,
so too he could have analyzed all functions in terms of those which he
calls concepts and relations, i.e., in terms of functions from objects
to truth-values. (Frege never spoke of relations as concepts; i.e.,
concepts were always unary functions from objects to truth-values for
Frege. For convenience, however, we shall speak here of relations as
relational concepts; and whether a concept is unary or relational, we
shall in either case refer to the saturated object corresponding to that
concept as a concept-correlate.)
There is something in Russell's way of making the above observation,
incidentally, which should not be overlooked; namely, that functionality
presupposes predication in that it depends essentially on the unity of a
proposition. Curiously, though Frege himself did not explain function-
ality in terms of predication, nevertheless precisely this sort of con-
sideration, whether applied to the unity of a sentence or to the thought
(Gedanke) expressed by that sentence, seems to be fundamental to his
notion of unsaturatedness, which of course is the basis of his notion of
functionality. Thus in regard to the unsaturated nature of a predicate as
the predicative component of a sentence, Frege argued that "this
unsaturatedness ... is necessary, since otherwise the parts [of the
sentence) do not hold together" ([PW), p. 177); and, similarly, "not
all the parts of a thought can be complete; at least one must be
'unsaturated', or predicative; otherwise they would not hold together"
([G & B), p. 54). In other words, though Frege explained predication in
terms of his mathematical notion of functionality, ultimately his only
argument for the unsaturatedness of functions is his argument for the
unsaturatedness of the functions involved in predication, whether that
predication be syntactical or otherwise. In this regard our present for-
malism neither distorts nor runs counter to Frege's form of logicism;
and indeed, if anything, it rather emphasizes what is really fundamental
in Frege's view of logicism.
The laws of logic, according to Frege, must be universal, but only in the
sense that they must be applicable to any objects whatsoever. Frege
does not mean to deny, in other words, that there are any existential
posits among the laws of logic. To be sure, none of Frege's axioms in his
Begriffsschrift are other than universal in form, but this does not mean
that no existential posits are provable on the basis of these axioms. In
particular, the (impredicative) comprehension principle,
(CP) (3P) (Vx) . .. (Vxn) [F(x), . .. ,x
) ... j,
where is a wff (pure or applied) in which P does not occur free and
XI' .. , xn are pairwise distinct individual variables occurring free in ,
is easily seen to be provable on the basis of Frege's basic law (lIb).
Besides being provable, however, (CP) can be taken as an axiom
schema and Frege's basic law (lIb) derived instead. That is, together
with the remaining axioms and rules of standard second order predicate
logic with identity, Frege's basic law (lIb), which we can formulate as
(UI2) (VP)'IjJ -> 'IjJ[/F(xI"" ,x
is provable on the basis of (CP) (cf. [Henkin]). Now although there is
something to be said in favor of having the basic laws of logic all be
universal in form, there is also something to be said for putting one's
existential posits up front. In addition, there is something appropriate
about avoiding the notion of proper substitution in stating the basic laws
of logic, especially such a complex notion as the substitution of a wff for
a predicate variable (relative to certain individual variables free in that
wff as subject-position indicators).
Such an observation applies on the first order level as well, inciden-
tally, especially if we are to adhere to Frege's view that "correctly-formed
names must always denote something" ([Gg], vol. 1, 28); for of
course that view amounts to the assumption that logic is not free of
existential presuppositions regarding singular terms. That is, where a is a
singular term in which X does not occur (free), the fact that a denotes
something (as a value of the bound individual variables), i.e., (3x) (a =
x), is provable on the basis of Frege's basic law (IIa):
) ('V x)'IjJ -> 'IjJ( a/ x).
But, conversely, given the remaining axioms and rules, (UI
) is also
provable on the basis of (:Ix) (a = x) (cf. [K & MD.
A substitution free axiom set for standard second order predicate
logic with identity can be described, accordingly, as follows. As axioms
(or basic laws of logic) we need take only all wffs (pure or applied)
which are either tautologous or of one of the following forms:
(AI) (Vu) [ -+ 1/J]
-+ [(Vu) -+ (Vu)1/J],
(A2) -+ (Vu),
(A3) (3x) (a = x),
(LL) (a= b) -+ ( - 1/J),
(CP) (3P) (V Xl) ... (V xn)
[F(XI , ... , xn) - j,
where u is an individual or a pre-
dicate variable,
where u is an individual or pre-
dicate variable not occurring free
in '
where a is a singular term in
which x does not occur free,
where a, b are singular terms and
1/J comes from by replacing
one or more free occurrences of
b by free occurrences of a,
where Fn does not occur free in
, and Xl' ... , Xn are among the
distinct individual variables oc-
curing free in .
As inference rules we need take only modus ponens and universal
(MP) if f- and f- ( -+ 1/J), then f- ;
(UG) if f- , and u is an individual or a predicate variable, then f-
Unlike Frege, incidentally, Russell does not take (Ul
) and (UI
) as basic
laws, but instead he takes their contrapositives, i.e., their corresponding
form as existential generalizations (cf. [PM], *9.1):
) 1/J(a/x) -+ (3x)1/J,
(EG2) 1/J[/F(XI"'" xn)] -+ (3P)1/J.
Of course, Russell does not distinguish these two laws and expresses
) in the form of (EG
); but that is because his "individual" variables
are really metalinguistic variables subject to systematic ambiguity. The
substitution of wffs for predicate variables which is involved in (EG
) is
only implicit in Russell, moreover, and (EG
) is really effected through
his use of the cap-notation, (Xj, ... , xn)' for the representation of
propositional functions. Such a use of the cap-notation, needless to say,
amounts to applying a syntactical operation for the generation of com-
plex predicates. An alternative notation which we shall adopt here is
Alonzo Church's A-operator for functional abstraction. That is, instead
of Russell's notation ... , xn) we shall use [AXI ... for the
expression of a complex predicate. Russell's "primitive proposition"
) can then be restated as follows:
1Jl([A..Xi .. P) (3P)1Jl.
According to Russell, incidentally, "the above primitive proposition
gives the only method of proving 'existence-theorems'" ([PM], p. 131).
That is, "in order to prove such theorems, it is necessary (and sufficient)
to find some instance in which an object possesses the property in
question. If we were to assume 'existence-axioms', i.e., axioms stating
(:lz) . fjJz for some particular fjJ, these axioms would give other methods
of proving existence. Instances of such axioms are the multiplicative
axiom (*88) and the axiom of infinity .... But we have not assumed any
such axioms in the present work" (ibid.).
This is somewhat misleading, needless to say, since it suggests that in
logic "existence-theorems" are only conditional theorems, and this we
know from our discussion of (CP) is false. Thus, in particular, where
ai' ... , an are singular terms which are free, respectively, for Xl' ... , Xn
in fjJ, the basic law regarding complex predicates corresponding to
Russell's of his cap-notation is the principle of A-conversion:
Generalized, this principle can be stated as follows:
(V/A-Conv) (Vx
) ... (VXn)([AXI ... XnfjJ] (Xl" .. , Xn) .... fjJ);
and of course from this, (CP) follows by (EG
It is historically noteworthy, incidentally, that the first explicit use of
(CP) as a basic law of logic occurs in [Tarski] where it is credited to
Lesniewski (cf. [Henkin], p. 203). Lesniewski apparently, referred to
special cases of (CP), each of which is really an instance of (V / A-Conv),
as "pseudo-definitions", which of course is quite appropriate since (CP)
is the logical basis of all explicit definitions of predicate constants. That
is, it is on the basis of (CP) that any such definition can be shown to be
noncreative and that the predicate constant so defined is eliminable in
principle. It is in terms of these "pseudo-definitions", in other words,
that the reduction of classical mathematics to logic is to be effected.
No reduction of mathematics is forthcoming, however, without a logistic
treatment of nominalized predicates, or some such equivalent syntactic
device such as Frege's notation for value-ranges (Wertverliiufe). Value-
ranges, it should be noted, are not sets or classes in the sense of being
composed of their members; rather, they are the saturated logical
objects which Frege also informally called concept-correlates. That is,
value ranges are for Frege the denotata of norninalized predicates - or
at least that is what we claim in our reconstruction and shall attempt to
show in what follows. If we are right in this claim, then we may in effect
identify Frege's implicit logic of nominalized predicates, that is, the logic
which is implicit in his informal remarks, with his explicit logic of value-
ranges. Indeed, this way of viewing Frege's theory of value-ranges, we
claim, will not only provide the essential rationale for his basic Law V
regarding value-ranges but it will also explain why his theory of value-
ranges is not really a second-order set theory.
In justifying our basic claim that value-ranges are concept-correlates,
accordingly, let us turn first to 'On Concept and Object', where Frege
explicitly states that an expression of the form 'the concept F' denotes
not a concept but an object which is somehow correlated with that
concept. In particular, in response to Benno Kerry's objection that
concepts are objects, and, moreover, objects other than their extensions,
Frege explains that "in my way of speaking expressions like 'the concept
F' designate not concepts but objects" ([G & Bj, p. 48). That is, "if he
[Kerry] thinks that I have identified concept and extension of concept,
he is mistaken; I merely expressed my view that in the expression 'the
number that applies to the concept F is the extension of the concept
like-numbered to the concept F' the words 'extension of the concept'
could be replaced by 'concept'. Notice carefully that here the word
'concept' is combined with the definite article" (ibid.). In other words,
according to Frege, 'the extension of the concept like-numbered to the
concept F' denotes the same object as that which is denoted by 'the
concept like-numbered to the concept F'. In a footnote of his original
draft of this passage, Frege writes that "the question whether one should
simply put 'the concept' for 'the extension of the concept' is in my view
one of expediency" ([PW], p. 106). Of course, by the extension of a
concept Frege means none other than a value-range (ct. [Gg], vol 1, 3).
The above explanatory remark is about as explicit as Frege gets in
identifying concept-correlates with value-ranges. More indirect, but also
more important, evidence for this identification can be found in the
connection Frege implicitly makes between second level concepts and
concept-correlates on the one hand and that which he explicitly makes
between second level concepts and value-ranges on the other. A second
level concept, we should explain, is essentially what a variable binding
operator on wffs to wffs (such as a quantifier) stands for, or, equiva-
lently, it is a concept corresponding to an open wff whi<:h may be used
in a third order comprehension principle to specify such a variable
binding operator, such as the wff (Vx) [F(x) -+ G(x)j which specifies the
second level relational concept of the subordination of one first level
concept to another (cf. [NBC-I]). A first level concept, i.e., one which a
predicate stands for, is said by Frege to fall within a second level
concept in a way analogous to (but still not the same as) the way that an
object is said to fall under a first level concept.
Now the connection between second level concepts and concept-
correlates which is implicit in "On Concept and Object" is the thesis that
corresponding to each second level concept there is a special first level
concept such that a first level concept, say, G, falls within that second
level concept if, and only if, the object correlated with G, i.e., the
concept G, falls under the corresponding special first level concept; or in
symbols (in the monadic case), where subject-position occurrences of a
predicate variable are nominalized occurrences of that variable:
(VQx) (3F) (VG) [(Qx)G(x) ++ F(G)j.
Thus, for example, corresponding to the second level concept of
(objectual) existence, i.e., the second level concept that first order
existential quantifier phrases stand for, there is the special first level
concept of being realized; and the correspondence, according to Frege,
is so tight that even the same thought is expressed by 'there is a square
root of 4' and 'the concept square root of 4 is realized' (cf. [G & Bj, p.
49f). A concept-correlate is realized, in other words, if, and only if,
there exists an object which falls under the concept in question.
Needless to say, but if value-ranges really are concept-correlates,
then Frege's thesis connecting second level concepts with concept-
correlates should connect second level concepts with value-ranges; and
indeed Frege explicitly states this to be the case in [Ggj, vol. 1, 25. That
is, according to Frege, "second level functions can be represented in a
certain manner by first level functions, whereby the functions that
appear as arguments of the former are represented by their value-
ranges" (op. cit.); or in symbols (in the case of unary concepts):
(VQx) (3F) (VG) [(Qx)G(x) - F(tG(E].
The singular term tG(E) is of course Frege's notation for the extension
(value-range) of the concept G; and the above thesis, it should be noted,
amounts in effect to a restatement of Frege's context principle regarding
such expressions. That is, just as it is only in the context of a sentence
(or of a wff in general) that a predicate can occur as a predicate, it is
only through a correlation with such occurrences of a predicate that we
are to understand the role of the name for a value-range; and of course
it is precisely the same thesis, and therefore the same restatement of his
context principle, which Frege implicitly gives for nominalized predi-
cates, i.e., for abstract singular terms of the form 'the concept G'. We
may read 'tG(E)', in other words, either as 'the concept G' or as 'the
extension of the concept G', which, as already noted, is exactly what
Frege said in the footnote referred to above ([PW] , p. 106).
For convenience, we shall hereafter refer to the above thesis (in
either form) as Frege's double correlation thesis. This is because Frege
assumes as part of his thesis both a one-to-one correlation between
second level concepts and certain special first level concepts on the one
hand, and a one-to-one correlation between first level concepts and
certain special objects called concept-correlates or value-ranges on the
other. Our two alternative reconstructions or resurrections of Frege's
logicism, as we shall see, will turn precisely on a minor modification of
one or the other of the correlations involved in this thesis.
There is at least one other place in Frege's wntmgs which clearly
indicates that value-ranges are concept-correlates, and which, given
Frege's extensional view of concepts (as functions from objects to truth-
values) provides the essential rationale for his Basic Law V. Both the
identification and the rationale, it should be noted, are again based on
Frege's double correlation thesis, only now applied to second level
relational concepts, and in particular to the second level relation of
mutual subordination.
Indeed, Frege's Basic Law V, viz.,
tF(E) = tG(E) ++ (Vx) [F(x) ++ G(x)j,
should be viewed precisely as a special instance of his double correla-
tion thesis. For what is indicated on the right hand side of this law is
none other than the second level relation of material equivalence or
mutual subordination of two first level concepts; and on Frege's exten-
sional view of concepts (as functions from objects to truth-values) such
an equivalence amounts in effect to their "identity". That is, Frege's
Basic Law V amounts to correlating the first level relation of identity
with his second level relation of mutual subordination. Such a correla-
tion is needed, Frege observes, since "to construe mutual subordination
simply as equality is forbidden by the basic difference between first and
second level relations. Concepts cannot stand in a first level relation.
That wouldn't be false, it would be nonsense. Only in the case of objects
can there by any question of equality (identity). And so the said transfor
mation [from mutual subordination to identity] can only occur by
concepts being correlated with the same object. It is all, so to speak,
moved down a level" ([PW], p. 182, italics added). Of course, if concept-
correlates are identical when the concepts in question are mutually
subordinate, then concept-correlates are none other than the extensions
of the correlated concepts; i.e., then concept-correlates are value-ranges.
The identification of value-ranges as concept-correlates also explains,
it should be noted, why Frege's theory of value-ranges is really not a
second order set theory. For as a concept-correlate, Frege observes, a
value-range "simply has its being in the concept, not in the objects which
belong to it" (ibid., p. 183). That is, unlike sets whose existence or being
is constituted by their members, concept-correlates, and therefore
value-ranges, are "logical objects" whose sole determination is given by
Frege's double correlation thesis, which, as already noted, is really a
restatement of his context principle applied specifically to nominalized
predicates. It is this thesis, in other words, which explains how "by
means of our logical faculties we lay hold upon the extension of a
concept, by starting out from the concept" (ibid., p. 181); for as also
already noted, a value-range, and therefore a concept-correlate, is none
other than the extension of a concept. In this regard, accordingly, the
confusion sometimes made of Frege's theory of value-ranges with a
second order set theory might best be obviated by directly describing
his theory of value-ranges as a theory of concept-correlates; that is, by
describing his logicism as a second order predicate logic with nominal-
ized predicates.
Unlike Frege for whom the extension of a concept "has its being in the
concept, not in the objects which belong to it," Russell originally took
the extension of a concept to be a class, or rather he took it to be what
he called a class as many as opposed to a class as one; and a class as
many, according to Russell, is essentially many, i.e., it is essentially
composed of its members (cf. [POM), chap. VI). Being many, however, a
class as many is not a single object, according to Russell, and therefore
it could not occur in a proposition as a term, which in effect defeated
the whole point of Russell's original form of logicism.
Of course, prior to the discovery of his paradox, Russell assumed that
a class as one always existed corresponding to a class as many (cf.
[pOM), p. 104); and as so determined, a class as one was also composed
of its members. But since a class as one is an individual, according to
Russell, then it, unlike its corresponding class as many, can occur in a
proposition as a term; and therefore the fact that a class as many, that is,
the extension of a concept, was not a single object was without any real
effect in Russell's original form of logicism.
With the discovery of his paradox, however, Russell gave up the
assumption that a class as one always existed corresponding to a class as
many. That is, Russell came to believe that it was this assumption which
was "the source of the contradiction" (ibid.). In particular, in regard to
his paradox of the class of classes that are not members of themselves,
Russell found that he could only conclude that "the classes which as
ones are not members of themselves as many do not form a class - or
rather, that they do not form a class as one, for the argument cannot
show that they do not form a class as many" (ibid., p. 102). Having given
up this assumption, however, Russell in effect was forced to give up his
original form of logicism.
Shortly after completing [POM) , Russell gave up not only the
assumption that a class as one always existed corresponding to a class as
many but he gave up assuming that there are any classes at all - i.e.,
classes in the sense of objects that are essentially composed of their
members. Thus, in his May 24, 1903 letter to Frege we find Russell
writing that "I believe I have discovered that classes are entirely super-
fluous. Your designation e(E) can be used for itself, and x n e(E)
for (x)" ([PMq, p. 158). (Frege's notation 'x n ecp(E)' in effect
amounts to 'x E ecp(e),' as Russell observed in [POMI, p. 512.) The
important suggestion here, it should be noted, is that propositional func-
tions are individuals after all, and therefore they can occur in proposi-
tions as terms instead of the classes as ones Russell originally assumed
to correspond to their extensions. That is, instead of assuming that
there are any classes, whether as "ones" or as "manys", Russell is now
proposing what later he called his "no classes" theory; i.e., the theory
that propositional functions are single entities (individuals) after all, and
that all talk of classes is to be reduced to talk of propositional functions.
Frege's response to Russell's letter is of course predictable. "I cannot
regard your attempt to make classes entirely dispensable as successful,
the reason being that you use function letters in isolation" ([PMq, p.
160). In other words, according to Frege, "to use a function sign in
isolation is to contradict the nature of a function, which consists in its
unsaturatedness" (ibid.). Russell's reply in turn is of course also predict-
able by now, for he writes that "it is not dear to rile that it is never
permissible to use to function letter in isolation" (ibid., 166). That is,
Russell simply refuses to accept the unsaturated nature of concepts.
Despite the stalemate on this point, however, the exchange is instruc-
tive for our present purposes. For to do what Russell suggests, Frege
(ibid., p. 161) observes that "we would first have to transform all [unary]
function names in such a way that there was only one argument place
and that was on the right-hand side. Thus, we would have to transform,
e.g., 'x
= l' into '(e
= l)x', and 'x(x - 1) (x + 1) = 0' into '[(e - 1)
(e + 1) = Olx', so that we could write
= 1)x::? [e(e-l) (e+ 1) = O]x
and for this, according to your [Russell's] definition,
= 1) c [e(e- 1) (e+ 1) = OJ.''
Now what should be especially noted here is that Frege uses the sipritus
asper or rough breathing operator to generate a complex predicate as
opposed to his use of the spiritus lenis or smooth breathing operator to
generate a complex singular term (for a value-range). Stated in terms of
our own notation, and using the A-operator to generate complex predi-
cates, Frege's point is that in transforming a wff into a complex
predicate, [AX(X)], we must keep in mind that [A.x(x)] really has a pair
of parentheses (and commas as well in the case of a relational predicate)
accompanying it. Thus whereas
(Vy) ([AX(X)] (y) [AX1jJ(X)] (y
is a well-formed formula in which two A-abstracts occur as complex
predicates, the expression
R([AX(X)], [AX1jJ(X)]),
where R is a 2-place predicate constant, cannot be well-formed or
meaningful as far as Frege is concerned, since as a complex predicate a
A-abstract "would be defined only in connection with an argument sign
following it, and it would nevertheless be used without one; it would be
defined as a function sign and used as a proper name, which will not do"
In his reply on this point, Russell is undaunted, insisting, as already
noted, that an expression for a propositional function can at least in
some cases be used as a singular term. This will not do as it stands, of
course, but we can reconstruct Russell's position here by distinguishing
between two transformations of a wff (x) where Frege has acknowl-
edged only one. That is, we can first transform (x) into the complex
predicate [AX(X)] () which does have an accompanying argument or
subject position (indicated by the last pair of parentheses); and then we
can transform this complex predicate into the singular term [A.x(x)]
which does not have an accompanying argument or subject position. A
confusion might arise here if we allow ourselves to speak of [AX(X)] as
both a predicate and a singular term; but so long as we understand that
when used as a predicate it must be accompanied by a pair of paren-
theses (and commas in the case of a relational predicate), no confusion
should arise if these parentheses (and commas) are informally dropped
for abbreviatory purposes.
Frege, incidentally, is not unaware of this second transformation of a
predicate into a singular term; for in applying Russell's suggested nota-
tion in just this way, he notes that "we would have 'i( e
= 1) = i[( e - 1)
(e + 1) = 0]" which does not differ essentially from my 'f(e
= 1) =
e[e e - 1) (e + 1) = 0]'" (ibid., p. 162). But in that case, Frege observes,
Russell's suggested notation for nominalized predicates "would lead to
the same difficulties as my value-range notation" (ibid., p. 161). That is,
other than assuming that nominalized complex predicates denote con-
cepts or propositional functions as single entities, Russell's suggested
use of such singular terms in 1903 was no less immune from his
paradox than was Frege's use of the same singular terms to denote
In turning to Russell's paradox and its resolution in our reconstruction
of Frege's and Russell's early form of logicism, i.e., the form of logicism
Russell was implicitly advocating in his correspondence with Frege, let
us first give a more explicit formulation of both the logical grammar and
the logical principles involved in the above exchange. As indicated,
instead of using both the spiritus asper or rough breathing operator for
the generation of complex predicates from wffs and the spiritus lenis or
smooth breathing operator for the generation of complex singular terms
from wffs, we shall follow the reconstructed Russellian strategy sug-
gested above and use only the A-operator for both purposes. Also, for
convenience of expression we shall informally drop parentheses and
commas when referring to predicates. It should be noted in this regard,
however, that using one operator and adopting an informal convention
of dropping parentheses and commas when referring to predicate
expressions in no way prejudges the case in favor of Russell's view that
concepts have a saturated or individual nature. Aside from simplicity
and economy of notation, in other words, the convention allows us to
formulate a logical grammar which is common to both forms of logic-
ism. (We will later introduce an intensional operator for Russell's form
of logicism as well, but this will not affect the grammar which is
common to both forms of logicism.)
In describing our logical grammar we shall for convenience of exposi-
tion identify the diffferent types of meaningful expressions by associat-
ing them with different natural numbers, where 0 is understood to
represent the type of all singular terms, 1 the type of all wffs or proposi-
tional forms, and n + 1, for n > 1, the type of all n-place predicate
expressions. Individual variables, accordingly, are of type 0, proposi-
tional variables are of type 1, and n-place predicate variables are of type
n + 1. We continue to ignore the introduction of special individual and
predicate constants, and, for nEro, we recursively define the meaning
ful expressions of type n, in symbols, MEn> as follows:
every individual variable (or constant) is in MEa, and every
n-place predicate variable (or constant) is in both MEn+ I and
if a, b E ME
, then (a= b) E MEl;
if :n E MEn+I' and aI' ... , an E ME
, then :n(a
, ... , an) E
if E MEl, and Xl' ... , xn are pairwise distinct individual
variables, then [AXI ... E ME
if E MEl, then - E MEl;
if and ljJ and in MEl> then ..... ljJ) E MEl;
if E MEl' and a is an individual or a predicate variable,
then ('V E MEl;
if E MEl, then E MEo; and
if n > 1, then MEn MEo.
Singular terms, which we shall also refer to simply as terms, are now
understood to be all the members of MEo; and for n > 0, we under-
stand the members of ME
to be n-place predicate expressions. We
are in general to think of each n-place predicate expression as having n
argument or subject positions associated with it, and, as in clause (3)
above, these are all understood to occur within parentheses and to be
separated from one another by commas. Wffs or propositional forms
are of course all the members of MEl. Note that whereas by clause (9)
every predicate expression is a term, not every wff is a term. We differ
in this regard from what Frege would allow; but our difference is
negligible since by clause (4), where n = 0, is a wff if is a wff,
and by clause (8) is a term. In other words, besides O-place
predicate variables (and constants), wffs are terms only when prefixed
by the A-operator. We shall in general read as 'that f when it
occurs in a wff as a term, i.e., when it occurs in one of the arguments or
subject positions of a predicate expression.
It is clear of course that predicate expression occurring in the argu-
ment or subject positions of other predicates, or of themselves as well,
are intended to represent the nominalized predicates that occur in natu-
rallanguage. For this reason, we shall refer to such occurrences of a
predicate as nominalized occurrences of that predicate, acknowledging
thereby that it has been transformed into an abstract singular term. Note
that adding such suffixes as '-ity', '-ness', or '-hood' to nominalized
occurrences of predicates would be completely superfluous here since
such occurrences are already formally identified as subject position
occurrences. The same observation applies, needless to say, to the use of
such related phrases as 'the concept P or 'being an P. Such phrases and
suffixes are important in transformational grammar. no doubt, since
they serve to mark derived nominal expressions in the surface grammar
of English; and in that regard we shall ourselves use such expressions
when translating or verbally stating certain theses in English. Neverthe-
less, it is sheer sophistry to insist that such surface grammatical features
of English either must or should occur in our "deep structure" logical
forms, as though a logical error were being committed otherwise.
As logical principles regarding this grammar and with respect to
which Russell's paradox is to be derived, we assume exactly the same
axioms and inference rules of standard second order predicate logic
with identity already described in 3, but understood now to apply to
wffs containing nominalized predicates as well. We assume in this
regard (but avoid going into the details here) the obvious definitions of
bondage and freedom of terms and predicate expressions in wffs and
A-abstracts, and also when one such expression can be properly substi-
tuted for another of the same type. In general we shall use a *-label in
referring to axioms and other theses so as to remind ourselves that we
are now dealing with wffs which may contain nominalized predicates as
singular terms. Thus, e.g., Axiom (A3), which is now referred to as
(A3*), has not only
(3y) (Fn-= y)
but also
(3y) ([AXI ... x
] = y)
as an instance. Similarly, Leibniz' Law, (LL), which is now referred to as
(LL *), has not only
but also
pn = Gn -. (V'YI) ... (V'Yn) [F(YI' ... ,Yn) ... G(YI'' Yn)]
[AXl .. xn] = [AXI .. xntp] -. (V'Yl) ... (V'Yn) ([AXl
xn] (Yt> ... ,Yn) - [Axl .. xntp] (YI' ... ,Yn
as an instance. Together with (A-Conv*), moreover, this last instance of
(LL *) has the following generalized form of Frege's basic law (Vb) as an
[AXj . .. = [AXj ... Xn1fJ] -+ (V'Xj) . .. (V'xn) ++ 1fJ).
Note incidentally that the comprehension principle (CP), now referred
to as (CP*), does not have the following standard formulation of Rus-
sell's paradox as an instance:
(3F) (VG) [F(G) ++ - G(G)].
This is because concepts are posited by (CP*) only by means of condi-
tions that apply to all individuals or object, whether those individuals be
abstract or concrete. It is for exactly the same reason, moreover, that
one cannot define in this context a predicate, say, 'Impredicable', as fol-
(V' G) [Impredicable( G) ++ - G( G)].
For, as already indicated, an explicit definition of a predicate constant
must be based upon the comprehension principle (CP*) in the sense of
being one of its existential instantiations, and the above fails in this
regard for the same reason that the preceding wff is not an instance of
Russell does give another version of his paradox, however, in terms
of ''what seems like a complex relation, namely the combination of non-
predicability with identity" ([POMj, p. 97); and this version is an
instance of (CP*):
(3F) (V'x) (F(x) ++ (3G) [x= G& - G(x)]).
A contradiction is derivable from this instance of (CP*), it should be
noted, only because
(UIf) -+
is derivable from (A3*) and (LL*). (By (LL*),
F= x -+ -+
and therefore by (UG), (A1 *), (A2*) and tautologous transformations,
(3x) (F= x) -+ -+
(UIf) then follows, . accordingly, by (A3*) and modus ponens.) Indeed,
by weakening either (A3*) or (LL*) in certain obvious ways it can be
shown that the resulting second order logic of nominalized predicates is
consistent, not inconsistent. The weakened version of (LL *) is appro-
priate, it should be noted, only when identity is defined (or replaced) by
indiscernibility. (Cf. [NBC-4], 4.6 for a consistency proof when (LL *) is
weakened to its version for indiscernibility; and also 4.1 0 for a
consistency proof when (A3*) is weakened instead.) We do not con-
template rejecting or modifying (LL *) here, however, and although
(A3*) will be weakened in our second alternative reconstruction of
Frege's logicism, we shall retain it in our first and more fundamental
reconstruction, since, together with (LL *), it implies that part of Frege's
double correlation thesis, namely that every concept has a unique
saturated concept-correlate, which will remain intact in our first recon-
struction of Frege's logicism. Of course, since this correlation for
Russell is really an identity, then (A3*) cannot be weakened in any
reconstruction of Russell's early form of logicism.
Now in his correspondence with Frege, it should be noted, Russell
does suggest weakening one other principle, namely, the comprehension
principle (CP*). Thus, in his December 12, 1904 letter to Frege, we find
Russell writing that "I believe the contradiction does not arise from the
nature of a class, but from the fact that certain expressions of the form
() . F(x, x, )
... do not represent [propositional] functions of x. That is, we have
f- ::(:3F):: -'(3!):.(x):fx= (). F(x,x,).
This is easy to prove ill the case of
x = (). :::> - x .
For this proposition denies fj(f)} for any f" ([PMq, p. 167). Re-
stated in our own present notation, Russell's particular example
amounts to
(VG) [x = [AyG(y)] -> - G(x)],
and given the identity
G= [AyG(y)]
as a law of the logic of norninalized predicates, Russell's particular
instance of (CP*) which he wants to reject is
(3F) (VX) (F(x) - (VG) [X= G .... - G(X)]),
which in the present context is also easily seen to lead to a contradic-
tion. The editors of [PMC] erroneously claim there is no contradiction
here (cp. p. 168), incidentally, and their explanation suggests that they
are ignoring Russell's original proposal that f= But even without
this particular identity, it is clear that a contradiction is derivable on the
basis of (LL*), or what amounts to Frege's basic law (Vb), and the
following instance of (CP*) which is clearly intended by Russell in the
letter in question:
(3F) (V x) (F(x) - (VG) [x = [AyG(y)] .... - G(x)]).
Now it is clear of course that Russell is not proposing that we are to
reject all instances of (CP*) in his letter to Frege, but only those of a
certain form. What is not clear, on the other hand, is the precise
delimitation of the excluded forms in question; that is, it is not clear
what restricted form (CP*) is to have according to Russell. What we
shall suggest in our reconstruction of this early form of Russell's
logicism is that (CP*) is to be restricted in accordance with the theory of
simple types Russell described in Appendix B of [pOM), though applied
now to propositional functions as. individuals rather than to classes as
many as the extensions of propositional functions. Such a theory, in
other words, seems to be what Russell had in mind in his correspond-
ence with Frege shortly after the completion of [POM).
Before turning to the specifics of Russell's proposal, however, let us
note that independently of its contradictory instances (CP*) is a conse-
quence of a still simpler form of comprehension principle, namely,
(Cpn (3P) ([AXI ... = F),
where P does not occur free in For by (LL *) and (V / A-Conv*),
= F .... (Vx
) (Vxn) [F(x
, , x
) -
from which (CP*) follows by (UG), (Al*), tautologous transformations
and (Cpn. A restricted form of (Cpn, needless to say, will imply only a
restricted form of (CP*); and for this reason we shall apply our recon-
struction of Russell's proposal to (Cpn instead.
Four months prior to his completion of [POM] in Dec. 1902, Russell
wrote Frege suggesting that ''the contradiction could be resolved with
the help of the assumption that ranges of values are not objects of the
ordinary kind; i.e., that needs to be completed (except in special
circumstances) either by an object or by a range of values of objects or
by a range of values of ranges of values, etc. ''This theory," Russell
observed, "is analogous to your theory about functions of the first,
second, etc. levels. In x ("\ u it would be necessary that u was the range
of values of objects of the same degree as x: x ("\ x would therefore be
nonsense. This view would also be useful in the theory of relations"
([PMC] , p. 144). This suggestion was subsequently described as the
simple theory of types which appears in Appendix B of [POM].
In his reply to Russell and the suggestion "that we are to conceive of
ranges of values and hence also of classes as a special kind of objects
whose names cannot appear in all argument places of the first kind"
(ibid., p. 145), Frege noted that "a class would not then be an object in
the full sense of the word, but - so to speak - an improper object for
which the law of excluded middle did not hold because there would be
predicates that could be neither truly affirmed nor truly denied of it.
Numbers would then be improper objects" (ibid., italics added). In his
discussion of Russell's paradox in the appendix to Volume 2 of [Ggj,
Frege reiterated this objection, nothing that if "classes were proper
objects, the law of excluded middle would have to hold for them"
([G & Bj, p. 235).
Now it is noteworthy that in 1895 Frege had already considered and
rejected the restriction on the laws of logic which would be necessary
for the kind of hierarchy of classes (and later of propositional functions
as individuals) Russell was suggesting as a way of avoiding his paradox.
The circumstances of this rejection had nothing to do with Russell's
paradox, needless to say, since Frege was at that time unaware of the
paradox; but rather they had to do with E. Schroder's conceptual
difficulties with the empty set as an extension consisting of nothing, on
the one hand, and with his notion of a singleton as an extension which
was identical with its only member, on the other (cf. [G & Bj, pp.
86-106). Be that as it may, Schroder, as Alonzo Church has observed,
anticipated the theory of types when he took ''the universal class 1
which appears in his algebra, not as an absolute universal class, but as
composed of all the elements of a certain domain fixed in advance"
([Church), p. 150). Once such a universal class or "manifold" was given,
moreover, a second may be obtained (to which the algebra is to be
applied in turn) "by taking the subsets of the first to be the individuals of
the second" (ibid.); and by continuing in this way a "hierarchy of reine
Mannigfaltigkeiten may be extended to infinity" (ibid.). The crucial
restriction Schroder imposed on this hierarchy was that no subset of the
domain of "individuals" considered at any stage of the hierarchy was to
be among the individuals of that stage, and that consequently the laws of
logic, as the laws of his algebra, were to be restricted in any given
application only to the individuals of the stage in question.
It was precisely this sort of restriction of the "field for our logical
activities" ([G & B), p. 92), in other words, which Frege criticized and
rejected in his 1895 review of Schroder's book. For ''whereas elsewhere
logic may claim to have laws of unrestricted validity, we are here
required to begin by delimiting a manifold with careful tests, and it is
only then that we can move around inside it" (ibid.).
Russell was on the mark, it should be noted, when he observed that
such a restriction on the laws of logic was already imposed by Frege on
his functions or concepts of first, second, etc. levels. But then for
Russell, it must be remembered, these functions or concepts do not have
an unsaturated nature; and as far as Russell is concerned if the laws of
logic can be restricted when applied to concepts of different levels, then
they can also be restricted when applied to the extensions of these
concepts as well. Indeed, once classes as the extensions of concepts are
eliminated from logic altogether, then as far as Russell is concerned the
restrictions in question are essentially the restrictions already imposed
by Frege on the concepts of different levels.
Now the crucial point ill Russell's view of Frege's hierarchy, it must
be emphasized, is that concepts do not have an unsaturated nature; i.e.,
that despite their predicative nature (as functions from individuals to
propositions), concepts for Russell are abstract individuals. This is
essential to Russell's interpretation, in other words, since if concepts
really do have an unsaturated nature as Frege claims, then they cannot
be construed as objects or abstract (higher order) individuals; and as a
theory of different types of individuals, the theory of types, whether
simple or ramified, would then not be applicable to Frege's concepts of
different levels. Indeed, it is precisely because concepts, and functions in
general, have an unsaturated nature according to Frege that Church
rejects "the claim sometimes made on behalf of Frege that his Stufen ...
constitute an anticipation of the simple theory of types" ([Churchl, p.
151). And in his rejection of classes as "improper objects" Frege himself
points out that "there is nothing 'unsaturated' or predicative about
classes that would characterize them as functions, concepts, or rela-
tions" ([G & BI, p. 235). Numbers in particular are objects, according to
Frege, not concepts or "improper objects" as they would have to be on
Russell's proposal.
There is a way, it turns out, of reconstructing Russell's proposal while
still agreeing with Frege that if nominalized predicates denote objects as
individuals "in the full sense," then the laws of logic, and the law of
excluded middle in particular, must not be restricted when applied to
such objects. We shall do so, moreover, by applying the notion of a
simple logical type not to a description of Frege's Stufen or levels of
concepts as "improper objects" the way Russell suggests but rather as a
description of the conditions determined by Frege's double correlation
thesis for positing first level concepts and their corresponding concept-
Now the point of our reconstruction is that if second level concepts
can be correlated with certain first level concepts, then third level
concepts can be similarly correlated with second level concepts, and
therefore, by the product of these correlations, third level concepts can
in effect also be correlated with first level concepts. Similarly, fourth
level concepts can be correlated with third level concepts and therefore
with first level concepts as well. In general, in other words, all concepts
of whatever level can in effect be correlated with first level concepts,
and these in turn can be correlated with certain (saturated) objects
called concept-correlates. Thus instead of speaking of a higher level
concept Q falling within a concept M of one level higher we shall
instead generalize Frege's double correlation thesis and speak of the
concept-correlate of the first level concept corresponding to Q falling
under the first level concept corresponding to M. Doing so, however,
requires that the conditions for specifying the first level concepts and
concept-correlates in question must be stratified in a way corresponding
to the stratification of the higher level concepts to which these first level
concepts and their concept-correlates correspond. The comprehension
principle (CPt), in other words, must be restricted in a way that corre-
sponds to the stratification of the unsaturated concepts of Frege's
We shall actually need a more stringent form of stratification than
Frege allows, it turns out for reasons shown below, and specifically one
which assumes that there are no unequal higher level relations (or at
least none that are involved in Frege's double correlation thesis); that is,
we shall be required to assume, on pain otherwise of generating
Russell's paradox after all, that higher level relations are only homo-
geneously stratified. This in fact will be the only modification of Frege's
original form of logicism which we shall assume in our first reconstruc-
tion of Frege's logic.
Returning to the logical grammar of 7, accordingly, we shall say that
a formula or A-abstract of that grammar is homogeneously stratified if,
and only if, there is an assignment t of natural numbers to the set of
terms occurring in (including itself if is a A-abstract) such that (1)
for all terms a and b, if (a = b) occurs in then tea) = t(b); (2) for all n
1, all n-place predicate expressions n and all terms ai' ... , an> if
, ... , an) is a wff occurring in then (i) tea;) = teak)' for 1 j, k
n, and (ii) ten) = teal) + 1; and (3) for all mEw, all individual
variables x, ... , X
, and all wffs tfJ, if [AXI ... xmtfJ] occurs in then (iii)
t(x;) = t(x
), for 1 j, k m, and (iv) t([AXI ... XmtfJ]) = t(Xl) + 1. If
clauses (i) and (iii) are dropped and clauses (ii) and (iv) are replaced by
the weaker requirement that ten) = 1 + max [t(a
), ... , t(a
)] and t([AXI
... xmtfJ]) = 1 + max [t(x
), ... , t(x
)], then we shall say that is hetero
geneously stratified, or simply stratified; and if clause (1) as well as
clauses (i) and (iii) are dropped and clauses (ii) and (iv) are replaced by
the still weaker requirement that max [t(a
), ... , t(a
)] < ten) and
max[t(xI)' ... , t(Xm)] < t([AXI ... XmtfJ]), then we shall say they is
cumulatively stratified.
We include the idea of cumulative stratification here because it is a
natural generalization of the idea of a stratified hierarchy of concepts.
Involved in a such a generalization, in other words, is L1.e suggestion that
we should restrict (CPt) as little as possible in our representation of
Frege's double correlation thesis; that is, that we should replace (CPt)
by the cumulatively stratified comprehension principle, (CSCPt), which
is exactly like (CPI) except for the added constraint that the A-abstract
in question must be cumulatively stratified. Unfortunately, however, this
is too much of a generalization in our present context, since the
A-abstract [AX (3 G) (x = G & - G(x)] involved in Russell's paradox is
easily seen to be cumulatively stratified. That is, (CSCpn is an insuffi-
cient restriction of (Cpn since it still leads to Russell's paradox.
Now although [Ax(3G) (x = G & - G(x] is cumulatively stratified,
it is not heterogeneously stratified, and therefore it will not fulfill the
conditions for the (heterogeneously) stratified comprehej1sion principle;
(SCPI), which is exactly like (CPt) except for the constraint that the
A-abstract in question must be (heterogeneously) stratified. This princi-
ple, it should be observed, is in accordance both with the simple theory
of types briefly described in Appendix B of [POM] and with Frege's
hierarchy of unsaturated concepts. That is, although unsaturated
concepts cannot be cumulatively stratified, higher level relational con-
cepts can be inhomogeneously stratified. Such higher level relations are
referred to by Frege as unequal leveled relations. Thus, for example, the
second level relation of an object to a concept under which that object
falls is said by Frege to be an uneq1lal second level relation because it
has as arguments both a saturated object and an unsaturated first level
concept ([Gg], vol 1, 22). Needless to say, but (SCPt) posits a first level
relation corresponding to tJ:>Js second level unequal relation of sub-
sumption; that is,
) ([Axy(3G) (x= G& G(y] = R)
is an instance of (SCpn; and in that regard (SCPt) is easily seen to be in
full accordance with Frege's double correlation thesis.
Unfortunately, however, precisely because predication stands for a
relation according to (SCPt), then being impredicable with respect to
this relation is also specifiable in terms of (SCPt). That is, since
[Axy(3G) (x = G & G(y] is heterogeneously stratified, then so is
[AZ- [Axy(3G) (x = G & G(y)] (z, z)]; and therefore despite its com-
plexity of expression in terms of A-abstracts Russell's paradox of the
concept which is predicable of itself if, and only if, it is not predicable of
itself is derivable on the basis of (SCPt) after all.! It follows, accordingly,
that if Frege's double correlation thesis is to apply to all higher level
unsaturated relations, then we must assume that no such relations are
inhomogeneously stratified, i.e., that there are no unequal higher level
relations. In particular, we must not assume that there can be such an
unequal second level relation as the sUbsumption of an object under a
concept; for it is fundamental to Russell's paradox that predication
cannot stand for a relation between an object and a concept-correlate.
If unsaturated higher level concepts, including relations, are not to be
inhomogeneously stratified, then the appropriate restriction of (CPt),
needless to say, is the homogeneously stratified comprehension principle,
(HSCP!), which is exactly like (CPt) except for the added constraint that
the A-abstract in question must be homogeneously stratified. Indeed, in
the second order logic of nominalized predicates in question, which
hereafter we shall refer to as AHST*, the only A-abstracts recognized as
well-formed are those which are homogeneously stratified. This does
not mean, it should be noted, that every wff of AHST* must be homo-
geneously stratified, but only that the A-abstracts occurring in such wffs
are. That is, in general, where n is a natural number, a meaningful
expression of type n (as defined in 7) is a meaningful expression of
type n in AHST* if, and only if, every A-abstract occurring in that
expression is homogeneously stratified.
It follows, accordingly, that if F is a I-place predicate variable (or
constant), then F(F) and - F(F) are both wffs of AHST*; and similarly if
[Ax] is homogeneously stratified, then both [AX] ([AX]) and its nega-
tion are well-formed formulas of AHST*. In other words, the laws of
logic, and the principle of excluded middle in particular, apply in AHST*
to concept-correlates as the denotata of nominalized predicates no less
so than they apply to objects or individuals in general; and in that
regard, therefore, concept-correlates may be said to be objects or
individuals "in the full sense" as far as AHST* is concerned.
The basic laws of logic according to AHST*, it should be emphasized,
are exactly those of standard second order predicate logic with identity
already described in 3, but extended now to include homogeneously
stratified A-abstracts and nominalized predicates. In other words, by an
axiom of AHST* we understand any wff of AHST* which is either
tautologous or of one of the following forms:
(Vu) [ --+ 'IJI] --+ [(Vu)
--+ (V u) 'IJI] ,
--+ (Vu),
where u is an individual or
predicate variable,
where u is an individual or
predicate variable not occur-
ring free in '
(A3*) (3x)(a= x),
(LL*) (a= b) -+ ... 'I/J],
(A-Conv*) [Ax
, , an)
.... .. , ajx
where a is a singular term of
,1.HST* in which x does not
occur free,
where a, b are singular terms
of AHST* and 'I/J comes from
by replacing one or more
free occurrences of b by free
occurrences of a,
where a
, , an are singular
terms of AHST* and each a
free for Xi in
(IdA) [Ax
, . .. , xn)] = P, where P is an n-place predi-
(HSCPA) (3P) ([Ax1 ... = P),
cate variable or constant,
where Fn does not occur free
in and [Ax1 ... is
homogeneously stratified.
Modus ponens and universal generalization (of an individual or pre-
dicate variable) are still the only inference rules, and theoremhood and
derivability from premises are defined in the usual way.2
Now it is important to note that if is a A-free wff containing no
nominalized occurrences of predicates, then [Ax1 ... is homo-
geneously stratified, and therefore by (HSCPA), the comprehension
principle (CP) of standard second order predicate logic with identity is
easily seen to be provable in AHST*. In other words, any A-free wff
containing no nominalized occurrences of predicates which is a theorem
of standard second order predicate logic with identity is a theorem of
AHST* as well. In this regard, AHST* goes beyond the laws of standard
second order predicate logic only in its recognition of homogeneously
stratified A-abstracts as complex predicates and in its logistic treatment
of nominalized predicates, whether simple or complex, as singular
terms. Both of these features are incorporated in the comprehension
principle (HSCPI), which, as already indicated, is in full accordance
with Frege's double correlation thesis, so long as we assume that
unsaturated higher level concepts are only homogeneously stratified.
In regard to the question of the consistency of AHST*, let us note first
that the full AHST* system can be shown to be consistent relative to
monadic AHST* (i.e., AHST* restricted to monadic predicates only) in
essentially the same way that the simple theory of types can be shown to
be consistent relative to the simple monadic theory of types. That is, it
can be shown that if monadic AHST* is consistent, then the full AHST*
system is also consistent (cf. [NBC-4], 4.8). Secondly, by interpreting
monadic predication as membership, monadic AHST* can be readily
shown to be consistent relative to R. Jensen's system NFU ("New
Foundations with Urelements"); and therefore if NFU is consistent, then
so is the full AHST* system. In [Jensen], however, Jensen has shown that
NFU is consistent relative to weak Zermelo set theory;3 and therefore,
by putting these results together, we can make the following consistency
claim regarding the full AHST* system (cf. [NBC-4], 4.9).
THEOREM: If weak Zermelo set theory is consistent, then so is AHST*.
We observed in 7 that (LL*) and (A-Conv*) together yield the follow-
ing generalized form of Frege's basic law (Vb):
.. = [AXI ... xntP] -+ ('Vx
) .. ('Vxn) ++ tP)
It was this law, it will be remembered, which together with (A3*) and
the unrestricted comprehension principle (CPt) led to Russell's paradox.
Given the restriction of (CPt) to (HSCPt), however, Russell's paradox is
no longer derivable (if weak Zermelo set theory is consistent), and we
are still able to maintain Frege's basic law (Vb) as a law of logic.
Of course, this is only one direction of Frege's Basic Law V. The
other direction, i.e., Frege's basic law (Va), is the following principle of
(Ext*) ('Vx
) ('Vxn) ++ tP) -+ [Ax
.. = [AX
xn tP]
This principle, needless to say, is not provable in AHST*, and in that
regard, it might be claimed, it is not a "law of logic". For Frege,
however, it is a law of logic because concepts and relations are
(unsaturated) functions from objects to truth-values. That is, according
to Frege, ''what two concept-words mean [bedeuten] is the same if and
only if the extensions of the corresponding concepts coincide" ([PW], p.
122). In making this claim, Frege is aware that he has made "an impor-
tant concession to the extensionalist logicians. They are right," he
claims, ''when they show by their preference for the extension, as against
the intension, of a concept that they regard the meaning [Bedeutung]
and not the sense of words as the essential thing for logic. The inten-
sionalist logicians are only too happy not to go beyond the sense; for
what they call the intension, if it is not an idea, is nothing other than the
sense. They forget that logic is not concerned with how thoughts,
regardless of truth-value, follow from thoughts, that the step from
thought to truth-value - more generally, the step from sense to meaning
has to be taken in. They forget that the laws of logic are first and
foremost laws in the realm of meanings [Bedeutungen] and only relate
indirectly to sense" (ibid.). In other words, as far as the laws of logic are
concerned, "concepts differ only so far as their extensions are different"
(ibid., p. 118); and "therefore just as proper names can replace one
another salva veritate, so too can concept-words, if their extension is the
same" (ibid.).
Regardless of his commitment to the principle of extensionality as a
law of logic, however, it must not be overlooked here that Frege still
maintains "that the concept is logically prior to its extension" ([G & B],
p. 106), and that he regards "as futile the attempt to take the extension
of a concept as class, and make it rest, not on the concept, but on single
things" (ibid.). In other words, despite his commitment to the principle
of extensionality, Frege's second order logic of value-ranges as concept-
correlates is not a second order set theory in the sense in which sets are
essentially constituted or composed of their members. That this is so is
especially brought out by the reconstruction and identification of
Frege's form of logicism with the system AHST* + (Ext*) as a second
order logic of nominalized predicates.
Our first reconstruction of Frege's form of logicism, accordingly, is its
reconstruction as the system AHST* + (Ext*). The question whether this
system is consistent, and, if consistent, whether it suffices for the reduc-
tion of classical mathematics is answered, it turns out, by its relation to
Jensen's modification of Quine's well-known "set" theory NF; i.e., by its
relation to NFU. In particular, the concept-correlates of monadic
AHST* + (Ext*), it turns out, are none other than the "sets" of NFU;
and therefore classical mathematics is reducible to AHST* + (Ext*) at
least to the same extent that it is reducible to NFU. Moreover, by
representing concepts by their extensions, monadic AHST* + (Ext*) can
be easily seen to be equiconsistent with NFU; and therefore since
monadic AHST* + (Ext*) is already equiconsistent with the full AHST*
+ (Ext*) system, then AHST* + (Ext*) is equiconsistent with NFU.
In other words, by defining membership as follows,
XE y=df(3F) [y=F&F(x)],
Jensen's "set" theory NFU can be shown to be contained within
monadic AHST* + (Ext*); and, similarly, by interpreting monadic
predication as membership in NFU, monadic AHST* + (Ext*) can be
translated into NFU so that the translation of a theorem of monadic
AHST* + (Ext*) is a theorem of NFU (cf. [NBC-4], 4.9).4 But monadic
AHST* + (Ext*), for reasons already indicated, is equiconsistent with
the full AHST* + (Ext*) system (cf. [NBC-4], 4.8); and therefore we are
able to establish the following result.
THEOREM: AHST* + (Ext*) is consistent if, and only if, Jensen's "set"
theory NFU is consistent; and therefore AHST* + (Ext*) is consistent if
weak Zermelo set theory is consistent.
This theorem is perhaps not surprising, it might be said, since both
AHST* + (Ext*) and NFU are constructed as type-free counterparts of
the theory of simple types (with an axiom of extensionality for each type
greater than 0). This is somewhat misleading, however, since AHST* +
(Ext*) is really a reconstruction of standard second order predicate
logic with nominalized predicates and not, strictly speaking, a recon-
struction of the theory of simple types as a theory of "improper objects".
The guiding principle of our reconstruction is indeed in accordance with
a homogeneously stratified hierarchy of unsaturated concepts since that
principle is none other than a generalized form of Frege's double
correlation thesis (minus unequal leveled relations); but that is not quite
the same as its being a reconstruction of the theory of simple types (with
axioms of extensionality for each type greater than 0) as a theory of
"improper objects". This is especially so, moreover, insofar as the latter
is intended as a model of all of the finite "stages" at which classes are
generated from concrete individuals (as objects of type 0) in accord-
ance with the power set axiom of set theory. For as a theory of classes
which are essentially constituted or composed of their members, the
number of classes of any given type in the theory of simple types will be
less than the number of classes of the next succeeding type; and this,
needless to say, runs directly counter to Frege's double correlation
thesis. Thus, in particular, if second level concepts are to be correlated
with certain first level concepts and first level concepts are to be
correlated in turn with their saturated concept-correlates, then there can
be neither more second level concepts than first level concepts nor
more first level concepts than objects. In this regard, since AHST* +
(Ext*) was constructed in accordance with Frege's double correlation
thesis (minus unequal leveled relations), it is misleading to describe it as
a type-free reconstruction of the theory of simple types.
Now Quine's "set" theory NF, on the other hand, was originally
constructed precisely as a first order counterpart of the theory of simple
types as a theory of classes which are essentially constituted or com-
posed of their members (d. [Quine], 40-42). Its failure to be what it
purports, however, is precisely what makes NF so controversial as a
theory of sets. For as a first order theory NF is committed to construing
all individuals as "sets", since by the axiom of extensionality every
individual in NF either has a member or is identical with the empty
"set". That is, unlike the situation in AHST* + (Ext*), NF cannot
"recognize" the concrete individuals of the theory of simple types
without first transforming them into "sets" (e.g., by identifying them with
their singletons). In this regard, NF fails to capture an important feature
of the theory of simple types as a theory of classes which are composed
of their me.mbers, since, as such, classes are ultimately founded on
concrete individuals or urelements which are not themselves classes.
NFU, needless to say, was designed to overcome precisely this failure of
In both NF and NFU, however, there is a universal "set", a fact which
runs counter to the way classes are generated by "stages" which never
end in the theory of simple types, thoagh not of course counter to the
Fregean view that the universal concept of self-identity should have an
extension as its concept-correlate. Furthermore, in both NF and NFU
every "set" has an absolute complement, which also runs counter to the
strictly relative complements a class has at each "stage" in the theory of
simple types, but not counter to the absolute complements which
concept-correlates have as the extensions of unsaturated concepts. There
is a symmetry between the small and the large in both NF and NFU, in
other words, which runs counter to the limitation of size doctrine which
is incorporated in the theory of simple types as a theory of classes that
are essentially constituted or composed of their members. The same
symmetry obtains, of course, between the extensions of concepts in
2HST* + (Ext*), since as concept-correlates these extensions have their
being in the concepts whose extensions they are, and not in the objects
which are their members. That is, there can be no asymmetry between
the small and the large among the extensions of concepts in 2HST* +
(Ext*), since by the laws of logic regarding concepts the extensions of
concepts must satisfy the conditions of a Boolean algebra; and this is so,
moreover, precisely because "the extension of a concept simply has its
being in the concept, not in the objects which belong to it" ([PW), p.
183). In this regard, it makes more sense to identify NFU-"sets" with the
concept-correlates of monadic 2HST* + (Ext*) than to attempt to
construe them as sets which are essentially constituted or composed of
their members.
Now there is a result of Ernst Specker's which shows that NF is
equiconsistent with the theory of simple types as a theory of classes if
we add to the latter the assumption that all of the classes of anyone type
can be correlated one-to-one with the classes of the next succeeding
type and that all of the classes of urelements can be correlated one-to-
one with these urelements, an assumption which Specker calls "com-
plete typical ambiguity" (cf. [TA], p. 118). Such an assumption, needless
to say, runs directly counter to the idea of classes as composed of their
members, and therefore it fails to explain in what sense NF is to be
viewed as a theory of sets. It does not run counter to the idea of
NF-"sets" as concept-correlates, on the other hand, and indeed, given
the assumption that every object is a concept-correlate it conforms
perfectly to our generalized form of Frege's double correlation thesis for
unsaturated (unary) concepts. This assumption, formalized as (\Ix)
(3pl)(X = F), was in fact briefly considered by Frege himself in a
footnote to [Gg) , vol. 1, 1O; and, indeed, it is easily seen that NF is
contained in mST* + (Ext*) + (\Ix) (3P) (x = P) in precisely the same
way that NFU is contained is 2HST* + (Ext*). In this regard, we
maintain, it is more appropriate to identify NF-"sets" with the concept-
correlates of monadic AHST* + (Ext*) + (\Ix) (3P) (x = F) as a recon-
struction of Frege's form of logicism than to construe either NF or
2HST* + (Ext*) + (\fX)(3Pl)(X = F), as type-free reconstructions of
the theory of simple types, even when Specker's axiom of "complete
typical ambiguity" is added to the latter. A similar observation applies,
needless to say, to the identification of NFU-"sets" with the concept-
correlates of monadic HST* + (Ext*). In other words, despite the
original motivation or purpose for its construction, NFU makes more
sense as a partial description of the concept-correlates of AHST* +
(Ext*) as a reconstruction of Frege's form of logicism than it does as a
reconstruction of the theory of simple types as a theory of classes which
are composed of their members.
The appropriate definition of the natural numbers in NFU, inciden-
tally, is precisely the well-known Frege-Russell definition and not either
von Neumann's or Zermelo's purely set-theoretic definitions. Given the
Frege-Russell definition of the natural numbers, moreover, we can go
on to construct all of the integers, rational numbers and real numbers
using only the resources already available in NFU, and therefore in
AHST* + (Ext*) as well. In other words, except for the assumption of an
axiom of infinity the reduction of classical mathematics to logic, and
especially of arithmetic to logic, as originally conceived by Frege is fully
realized in our reconstruction of Frege's form of logicism as AHST* +
Finally, in regard to the only "existence-axioms" described as such by
Russell, viz., the axioms of choice and infinity, we should take note of
Specker's result in [AC] that the axiom of choice is disprovable in NF;
and therefore, since the axiom of choice is provable for finite sets, the
existence of an infinite "set" is provable in NF. But since NF is
contained in AHST* + (Ext*) + ('Ix) (3P) (x = F), then Specker's
results apply here as well. That is, in AHST* + (Ext*) + ('Ix) (3F) (x =
F), the axiom of choice is disprovable and the axiom of infinity is
provable. Of course, this does not mean (as it did in Russell's later form
of logicism) that there are infinitely many concrete individuals, but only
that the total number of objects, whether abstract or concrete, is infinite.
However, by another argument of Specker's regarding NF, we can prove
in AHST* + (Ext*) + ('Ix) (3F) (x = F), that there are infinitely many
natural numbers in the sense in which these are defined in the Frege-
Russell manner (cf. [Quine], p. 299); and that, needless to say, is exactly
what Frege thought held in his form of logicism. Specker's proofs do not
apply to NFU, however, and indeed Jensen (op. cit.) has shown that the
axiom of infinity is not provable in NFU. The same observation applies,
needless to say, to AHST* + (Ext*).
The only fundamental difference between Frege's and Russell's early
form of logicism so far emphasized is that for Russell concepts are their
own concept-correlates. That is, Russell refuses to accept the unsaturated
nature of concepts, and he assumes accordingly that nominalized pre-
dicates denote the same concepts which these predicates otherwise
stand for in their role as predicates. The prdicative nature of a concept
for Russell, in other words, does not consist in its being an unsaturated
function from objects to truth-values, but rather only in its being a
function whose values are propositions. That is, concepts for Russell are
none other than propositional functions, and propositional functions are
themselves individuals.
As functions whose values are propositions rather than truth-values,
however, propositional functions are intensional and not extensional
entities; and this in fact is another fundamental difference between
Frege's and Russell's early form of logicism. For when applied to
AHST*, this difference is reflected in the acceptance of (Ext*) by Frege
as a law of logic as opposed to its rejection as such by Russell. That is,
besides assuming that concepts have an individual nature, Russell also
assumes that in general they are only intensionally individuated.
What exactly Russell means by the intensionality of a proposition,
and thereby of a propositional function as well, he never says. Neverthe-
less, in our reconstruction of his early form of logicism, we shall assume
that propositions can be represented by (or rather correlated with)
functions from possible words to truth-values. This is not what a
proposition really is according to Russell, needless to say, especially
since possible worlds would themselves be constructed in Russell's early
framework in terms of propositions about merely possible as well as
actual individuals;5 but it does serve as an intuitive guide regarding the
intensional individuation of propositional functions. Thus, instead of
Frege's principle of extensionality, (Ext*), being a law of logic, Russell
(or so we shall assume) would have as a law of logic the corresponding
principle of intensionality.
(DExt*) o (V'Xj) ... (V'xn) (rp ... 'I/J) - [.hj ... xnrp] = [AXj ... Xn'I/J].
This means that we shall need to take 0 as a new primitive logical
constant of our logical grammar, and that we shall need to add the
following clause in the definition of a meariingful expression given in 7:
(10) ifrp E MEl' then 0 rp E MEl'
We shall retain all of the axioms and inference rules of AHST*, needless
to say, except that now these axioms and rules are understood to apply
to wffs containing occurrences of 0 as well. In addition, we shall also
assume the axiom schemas of the S5 modal propositional logic and the
rule of modal generalization; i.e., the rule that if is provable, then so is
O. For convenience, we shall refer to the resulting system as OAHST*.
Our initial proposal, accordingly, is that Russell's early form of logicism
is to be reconstructed as OAHST* + (OExt*).6
Unfortunately, however, OAHST* + (OExt*) does not suffice for
Russell's view of classes as analyzable in terms of concepts. In particu-
lar, assuming that classes are to be represented in terms of concepts (as
individuals), we shall need an account of how this representation is to be
given; and assuming the adequacy of that account we shall need to
establish the following thesis as one of its consequences:
(itF)(3G)(Cls(G) & (Yx) [F(x) +> G(x)]).
Now without going into the details here, we shall only say that Russell's
own later contextual analysis in [PM] of 'CIs', or of expressions for
particular classes as "incomplete symbols", will not suffice in our
present context, since as reconstructed here, Russell's contextual analy-
sis would have the inappropriate consequence that all concepts are
"classes"; i.e., that concepts are extensionally, not intensionally, indi-
viduated after all. Instead of Russell's analysis, however, we can begin
with the notion of a rigid (n-ary) propositional function as a concept
which has the same extension in every possible world. A class, on this
analysis, will then simply be a rigid concept. Indeed, if concepts are
themselves individuals, as Russell claimed, there would be little or no
point in distinguishing a rigid concept from the class which is the
extension of that concept, at least not if classes are really "superfluous"
in Russell's form of logicism? Accordingly, where definitions' of predi-
cate constants are given in terms of homogeneously stratified A-
abstracts, we define the notion of rigidityn as follows:
=df [Ax(3P) (x = P & (ity!) ... (itYn) [OF(Yl' ... , Yn)
V 0 - F(y), ... ,Yn)])]'
Thus, an n-ary relation-in-extension, on this analysis, is simply a rigid
relation (in intension), and a class is simply a rigid
Cls=df [Ax(3Fl) (x= F & Rigidl(F].
Now the fundamental new assumption which we need as a "law of logic"
in Russell's intensional form of logicism is not just the above thesis that
every concept is extensionally equivalent to a rigid concept (i.e., that the
extension of a concept exists), but rather the following more general
principle of rigidity:
(PR) (V p)(3Gn)(Rigid
(G) & (Vx/) . .. (Vxn)[F(x/, . .. ,x
- G( xl> . . . , x
Our proposal, accordingly, is that Russell's early form of logicism is to
be reconstructed as the system DAHST* + (DExt*) + (PR).
It is clear, of course, that by interpreting D as double negation,
DAHST* + (DExt*) + (PR) collapses to just AHST* + (Ext*). That is,
on our reconstruction, Russell's (early) intensional form of logicism is
equiconsistent with Frege's extensional form; and therefore by the
equiconsistency of the latter with Jensen's "set" theory NFU, we have
the following result (cf. [NBC-4], 6.3).
THEOREM: DAHST* + (DExt*) + (PR) is equiconsistent with both
AHST* + (Ext*) and Jensen's "set" theory NFU.
Classical mathematics, it is clear, is reducible to our reconstruction of
Russell's intensional form of logicism no less so than it is to our recon-
struction of Frege's extensional form. The identification of NFU-"sets"
with Frege's value-ranges as concept-correlates is perhaps more plau-
sible than their identification with rigid concepts; but then their identi-
fication with concepts to begin with should also more readily obviate
their confusion with sets as essentially constituted or'composed of their
In reconstructing Frege's logicism, it should be emphasized, we have
made only one relatively minor change in his overall view. We have
assumed, in particular, that all higher level unsaturated concepts are
only homogeneously stratified; that is, that there are no unequal higher
level relations. Only on the basis of this assumption, in other words, can
we maintain that (CPt) is to be restricted to (HSCPt) in accordance with
Frege's double correlation thesis.
There is an alternative to this assumption, however, which allows us
to retain both (CPt) and to avoid Russell's paradox. This alternative, is
implicit in Frege's discussion of Russell's paradox and is contained in his
suggestion that we might "suppose there are cases where an unexcep-
tional concept has no class answering to it as its extension" ([G & B], p.
235). In other words, instead of maintaining for each singular term the
dubious existential presupposition that that singular term actually
denotes (a value of the bound individual variables), we are to allow on
this suggestion that some singular terms, and in particular some nomi-
nalized predicates, are denotationless.
Given the standard second order predicate logic with nominalized
predicates described in 7 - that is, the logistic context in which
Russell's paradox was originally formulated - the proposal in question,
accordingly, amounts to replacing the dubious axiom (A3*), which
explicitly expresses all such existential presuppositions, by the following
weaker, but also unexceptionable, law of logic:
(A3**) (\Ix) (:Jy) (x= y).
A little more is actually needed on this proposal, however, since without
(A3*) the identity wff (a = a), also an unexceptionable law of logic, is
no longer derivable. In replacing (A3*) by (A3**), in other words, we
shall also need to add (a = a) as an axiom, where a is any singular term.
Similarly, in dropping all existential presuppositions regarding singular
terms, we shall need to replace (A-Conv*) by the presupposition free
form of A-conversion:
(:J/A-Conv*) [AX
] (a
, ... ,an) ++ (3x
) . .. (:JXn)
= Xl & ... & an = Xn & ),
where no Xi is free in any a
, for all i, j such that 1 i, j n.
Now with the replacement of (A3*) by (A3**) and (a = a), and of
(A-Conv*) by (:J/A-Conv*), the principle of universal instantiation (VIi)
regarding singular terms is no longer provable, it turns out, except in the
following qualified form:
(:J/VIi) (:Jy) (a= y) -+ [(\lx) -- (a/x)],
where a is any singular term which is free for X in and in which y has
no free occurrences. With this qualification, however, what follows from
Russell's paradox as described in 7 is not a contradiction but only that
the complex predicate [AX(:JG) (x = G & - G(x] is denotationless in
its occurrences as a singular term: i.e., instead of a contradiction,
- (3y) ([AxC:lG) (x== G& - G(x) = y)
is provable. In other words, even though (CP!) posits the existence of an
unsaturated concept corresponding to [Ax(3G) (x = G & - G(x] as a
predicate, nevertheless by Russell's argument it is provable that there
can be no saturated object corresponding to [A.x(3G) (x = G & - G(x]
as a singular term; or in Frege's words here is a case ''where an
unexceptional concept has no class answering to it as its exten-
Unfortunately, however, mathematics, and arithmetic in particular, is
not reducible to the system resulting from the above changes; for unlike
the situation in AHST* where (A3*) remains in force, but where only
A-abstracts which are homogeneously stratified are recognized as well-
formed (and where (CPt), accordingly, is replaced by (HSCPl, we can
no longer prove that there are any objects denoted by nominalized
predicates, or, equivalently, that there are any concept-correlates at all.
(The logical grammar of the system in question, it should be empha-
sized, is just the full unrestricted grammar described in 7, and the
axioms are (Al*), (A2*), (A3**), (a = a), (LL*), (Idi), (3/A-Conv*) and
That there might be no concept-correlates at all, incidentally, was not
a possibility Frege refused to consider in his discussion of Russell's
paradox. For in commenting on the proposal in question, Frege seemed
to interpret this alternative as one in which we can only "regard class
names as sham proper names, which would thus not really have any
reference" (ibid. p. 236). We agree, it should be noted, that for Frege
this apparently meant that "class" names "would have to be regarded as
part of signs that had reference only as wholes" (ibid.), as, for example,
names of natural numbers might occur as parts of quantifier expres-
sions, but in our present framework this can also be formulated as the
Abelardian thesis that no concept is a thing (cf. [NBC-4], 4.1):
('VF') - (3x)(F'=x).
In other words, instead of maintaining only that some nominalized
predicates must be denotationless, Frege generalized this alternative
into the Abelardian view that all nominalized occurrences of predicates
are denotationless. Abelard, incidentally, apparently believed that the
same universal can be shared by different individuals, but he refused to
grant that what individuals have in common is a ''thing''; that is, he
denied that universals are individuals. In this regard, he might be said to
have anticipated Frege's view of concepts as unsaturated functions. Be
that as it may, nevertheless it was Frege's view throughout most of his
career that nominalized predicates, and abstract singular terms in
general, denoted objects even though these objects could not themselves
be unsaturated concepts. For Abelard, on the other hand, at least on our
interpretation of his view here, nominalized predicates were singular
terms which simply failed to refer to any individual at all. Such a view,
needless to say, is obviously "safe" from Russell's paradox, and indeed
the result of adding the Abelardian thesis to the system in question can
be shown to be consistent (cf. [NBC-4], 4.10).
Though consistent, however, such a view does not lead to a recon-
struction of Frege's form of logicism; and in fact, it leads directly away
from it, a direction which Frege was not disinclined to take in 1924/5 at
the end of his long and brilliant career in defense of logicism. Thus, in
speaking of "the formation of a proper name after the pattern of 'the
extension of the concept a'" ([PW), p. 269), Frege not only noted that
"because of the definite article, this expression appears to designate an
object" (ibid.), but he went on to suggest that "there is no object for
which this phrase could be a linguistically appropriate designation"
(ibid.). Weare misled by language here, he suggests, and "from this has
arisen the paradoxes of set theory which have dealt the death blow to
set theory itself' (ibid.).
These final thoughts of a great logician are unfortunate, however, for
the paradoxes, and Russell's paradox in particular, do not affect set
theory as a theory of classes which are composed of their members, but
rather affect only a theory of classes as the extensions of concepts; and
whereas set theory has continued to flourish throughout the 20th
century, it is the theory of classes as the extensions of concepts, and
thereby logicism itself, which has fallen into disrepute. Such disrepute is
not deserved, however, since logicism, and Frege's form of logicism in
particular, can be easily reconstructed in such a way as to avoid the
paradoxes. In this regard, we have not only the reconstruction of Frege's
logicism already given, but the following alternative reconstruction as
The proof that some nominalized predicates must be denotationless in
the system presently under consideration, it should be noted, does not
show that all must be denotationless; i.e., the Abelardian thesis is not
provable in our modified second order logic with nominalized predi-
cates as described above in 14. We may assume, in other words, that
some concepts do have concept-correlates after all.
Such an assumption, needless to say, can be given in a number of
different ways, such as those corresponding to the existence conditions
for sets in different set theories. The latter, however, will result only in a
second order theory of sets in the sense in which sets are essentially
constituted or composed of their members; and dropping the axiom of
extensionality so as to avoid calling such objects sets is really pointless,
since their existence conditions are in accordance with the limitation of
size doctrine, which in tum is based on the notion of sets as composed
of their members. Such an assumption, in other words, will not result in
a coherent form of logicism since as objects which are essentially consti-
tuted or composed of their members sets are mathematical and not
logical objects. The relevant assumption, in this regard, must be based
on Frege's double correlation thesis and the way it pertains to the
positing of concept-correlates as logical objects and not on conditions
that pertain to the existence of sets as composed of their members.
Now in returning to Frege's double correlation thesis in our present
framework where A-abstracts need not be homogeneously stratified, we
should note that there is a difference in applying the thesis in the
positing of first level concepts from applying it in the positing of con-
cept-correlates. In partiCUlar, since the full unrestricted comprehension
principle, (CPr), is to remain as an axiom schema, we in effect retain the
unrestricted form of Frege's double correlation thesis as it applies to the
positing of first level concepts; i.e., the form in which it posits first level
concepts corresponding to unequal leveled or inhomogeneous higher
level relations no less so than to homogeneous or equal leveled
relations. Thus, corresponding to Frege's unequal second level relation
of subsumption or predication, there is a first level relation posited by
(CPt); that is,
(3R2) ([Axy(3G) (x= G& G(y] = R)
is provable in the system in question. Of course, by Russell's argument,
this same predicate when nominalized simply fails to denote (a value of
the bound individual variables). In other words, as applied to concept-
correlates, Frege's double correlation thesis must be restricted; and in
particular it is not to apply in general to inhomogeneous higher level
relations. The assumption we shall make here is that it is to apply at
least to all higher level unsaturated concepts which are homogeneously
stratified; that is, to all of the concepts which have concept-correlates in
In order to formulate our assumption as an axiom schema, we shall
say that a meaningful expression (as defined in terms of the logical
grammar of 7) is bound to individuals if, and only if, for all natural
numbers n, all n-place predicate variables P, and all , if (V P) is a wff
occurring in then for some individual variable x and some wff 1/J, is
the wff [(3x) (P = x) -> 1/J]. In other words, to be bound to individuals,
every predicate quantifier occurring in must refer only to those
concepts posited by (CPr) which have corresponding concept-correlates.
Our assumption may now be stated as the following axiom schema:
(3/HSCPr) (3y) (a
= y) & ... & (3y) (a
= y)
-> (3y) ([AXI ... x
] = y),
where [AXI ... x
] is homogeneously stratified, is bound to individ-
uals, y is an individual variable not occuring in ' and aI' ... , ak are all
ofthe variables or non-logical constants occurring free in [AXI ... xn]'
The axiom schemas of our present system, accordingly, are (Al*) ,
(A2*), (A3**), (a = a), where a is any singular term, (LL*), (Idr),
(3/A-Conv*), (CPr) and now (3/HSCPI) as well.
(Modus ponens and
universal generalization are its only inference rules.) Because of its
relation to our earlier system, we shall refer to this system hereafter as
We must again emphasize that unlike the situation in AHST*, A-
abstracts are not required to be homogeneously stratified in HSTi. That
is, the meaningful expressions of HSTi are just those described in the
logical grammar of 7. Also, unlike AHST*, the system HSTi is free of
existential presuppositions regarding singular terms, including of course
nominalized predicates as abstract singular terms. It follows, accord-
ingly, that HSTi is not a conservative extension of AHST*, since
whereas, by (A3*), (V P) (3x) (F = x) is provable in AHST*, this same
wff is actually disprovable in HST!. Nevertheless, since every wff of
AHST* is provably equivalent, again by (A3*), to a wff which is bound
to individuals, HSTi may be said to contain AHST* in the sense of the
following lemma (which is easily seen to hold).
LEMMA: If is a wff of AHST* which is bound to individuals, y is an
individual variable not occurring in , and aI' ... , ak are all the
variables or non-logical constants occurring free in rp, then IAHST* rp only
(3y) (a
= y) & ... & (3y) (a
= y) -> rp.
Restricting ourselves to pure wffs, i.e., wffs in which no predicate or
individual constants occur, if follows by the above lemma that every
sentence (or wff with no free variables) of AHST* which is bound to
individuals and provable in AHST* is therefore provable in HST;.,* as
well. In addition, by (3IHSCPI), every object which is a concept-
correlate in AHST* is also a concept-correlate in HSTf, and therefore,
for the same reason, every object which is a concept-correlate in AHST*
+ (Ext*) is also a concept-correlate in HST! + (Ext*). In other words,
since Jensen's "set" theory NFU is contained in AHST* + (Ext*), then in
the sense of the above lemma NFU is also contained in HST! + (Ext*);
and in that regard classical mathematics, and arithmetic in particular, is
reducible to HST! + (Ext*) at least to the same extent that it is reduci-
ble to NFV. (Actually, HST! + (Ext*) is an improvement over AHST* +
(Ext*), and therefore over NFU as well, for the same reason that
Quine's "set" theory ML is an improvement over NF; in particular, in
both HST! + (Ext*) and ML, mathematical induction is provable in an
unrestricted form (cf. [Quine], 42).) Our proposal that HST! + (Ext*)
be taken as an alternative, and perhaps a preferred, reconstruction of
Frege's form of logicism is in that case quite in order.
Finally, it should be noted that just as Hao Wang was able to prove
the consistency of Quine's set theory ML relative to NF, we are able to
prove the consistency of HST! + (Ext*) relative to AHST* + (Ext*), and
therefore relative to NFU as well (cf. [NBC-4], 6.4).
THEOREM: If AHST* + (Ext*) is consistent, then so is HST! + (Ext*).
By the intensional counterpart of HST! we shall understand the system
HST!o which is developed from HST! in the same way that DAHST*
was developed from AHST*. That is, the axiom schemas of HST!o are
just those of HST!, extended now to apply to wffs containing D as well,
plus those of the SS modal propositional logic. The inference rules are
of course the same as those for DAHST*, viz., modus ponens, universal
generalization and modal generalization.
Now it is perhaps noteworthy that insofar as (DExt*) is to be
assumed as well, then we need not take 0 as a primitive logical constant
since it can in that case be contextually defined as follows:
=df = [A(\fX) (x= x)].
That is, since the biconditional,
.... = [A (\fx)(x = x)]
is provable in HSTio + (DExt*), then we need not take 0 as a primitive
logical constant of this system since it can be introduced by means of
the above contextual definition. (Some, but not all, of the modal axioms
will then become redundant.) This definition is not available in DAHST*
+ (DExt*), our reconstruction of Russell's early from of logicism, since
it cannot be used in that context to explain occurrences of 0 in wffs
which are not homogeneously stratified. That is, the above definiens, as
a wff of AHST*, would restrict the application of 0 to wffs which are
homogeneously stratified, since only these wffs can occur within the A-
abstracts of AHST*. But since there are wffs of AHST*, including tau-
tologous wffs, which are not homogeneously stratified, defining 0 as
above would in that case fail to validate the rule of modal generalization
when applied to those axioms of AHST* which are not homogeneously
stratified. It is only in HSTi, in other words, where A-abstracts are not
required to be homogeneously stratified that the above definition of 0
will suffice.
In addition to the principle of intensionality, (DExt*), we shall also
need to assume the principle of rigidity, (PR), if we are to represent
classes in this alternative the way they are represented in DAHST* +
(!;JExt*), + (fR\.our reconstruction of Russell'searlvJorm_ofJolicisltL.
Of course, with the addition of (PR) we can bring about the same
reduction of classical mathematics to logic as is already achieved in
HSTi + (Ext*), which contains all NFU-"sets" as concept-correlates.
Moreover, we can prove the relative consistency of HSTio + (DExt*) +
(fR) in essentially the same way as we proved the relative consistency of
HSTi + (Ext*).
THEOREM: If DAHST* + (DExt*) + (PR) is consistent, then so is
HSTio + (DExt*) + (PR).
Despite its relative consistency, however, HSTl'o + (Ext*) + (PR)
cannot be construed as a reconstruction of Russell's early form of
logicism. For as an abstract singular term a nominalized predicate,
according to Russell, denotes the same concept as an individual which
the predicate in question otherwise stands for in its role as a predicate;
and therefore whatever is the value of a bound predicate variable,
according to early Russell, is also a value of the bound individual
variables. That is, in Russell's early form of logicism no predicate can
stand for a concept in its role as a predicate and yet fail to denote an
individual when transformed into an abstract singular term. But since
there are predicate expressions satisfying just these conditions in HSTAo,
and therefore in HSTAo + (DExt*) + (PR) as well, then neither HSTl'o
nor any of its extensions can be the basis of a reconstruction of Russell's
early form of logicism. In other words, unlike Frege's form of logicism,
we have no way in Russell's framework by which to explain why some
concepts have individuals as concept-correlates while others do not,
since on Russell's account concepts are their own concept-correlates.
We might adopt a mixed strategy here, however, whereby concepts
are unsaturated functions in Frege's sense but still have propositions as
values in Russell's sense. Concepts are then not individuals after all, and
there is nothing odd or contradictory in the idea of some concepts
having a concept-correlate while others do not. Of course, if in addition
concept-correlates are none other than the extensions which these
concepts have in different possible worlds, then the resulting framework
may be described more as an intensionalized form of Frege's logicism
than as a counterpart to Russell's. In this regard, we would not adopt
(DExt*) but (Ext*) instead, since even intensional concepts which are
materially equivalent will have the same extension, and therefore con-
cepts which are materially equivalent in a given possible world will have
the same concept-correlate in that world. In such a framework, needless
to say, the principle of rigidity, (PR), would be completely superfluous.
There is a problem here with HSTAo as the basis for such an
intensionalized form of Frege's logicism, however, since by (LL *),
(VP) ('iGn) (F= G -> DF= G)
is provable in HSTl'o, and therefore all concepts, even when intensional-
ized, will always have the same extension. We can of course modify
(LL *) so that it remains valid only for extensional contexts, that is, only
for wffs in which D does not occur; for in that case concepts having the
same extension in one possible world need not necessarily have the
same extension in every possible world. That is, in that case,
(3P) (3Gn) (F= G& <> F G)
will be consistent and not inconsistent (cf. [NBC-4], 6.5). Thus, if
is the system resulting from HSTTo by weakening (LL *) in this
way, but also by adding the commutative law,
(V - 0 (V
as an axiom schema (since concepts themselves remain the "same" from
world to world), then we can identify this intensionalized form of
Frege's logicism with + (Ext*). The extensional system HST! +
(Ext*) described earlier, needless to say, is a subsystem of +
(Ext*), and by interpreting 0 as double negation the two systems
collapse into one, thereby showing the consistency of + (Ext*)
relative to HST! + (Ext*).
The mixed strategy described above favors Frege's form of logicism
over Russell's, it should be noted, since unsaturated concepts, even
when intensionalized, were assigned their extensions as their concept-
correlates in the different possible worlds. There is another mixed
strategy, however, which is more in line with Russell's form of logicism
than with Frege's, and in fact which may be taken as a counterpart even
to Russell's later form oflogicism in (PM].
On this mixed strategy, the predicative nature of a concept consists in
its having both an intensional and an unsaturated nature, though the
latter is understood in even a more radical sense than Frege would have
allowed. In particular, on this account all predicable concepts are
cognitive capacities to identify (in a classificatory sense), characterize or
relate objects in various ways, and the purely dispositional or non-
occurrent nature of such a capacity is none other than its unsaturated
nature. That is, on this account, concepts are intersubjectively realizable
cognitive abilities which may be exercised by the same person at
different times as well as by different persons at the same time; and in
that regard concepts are neither ideas nor mental images in the sense of
particular mental occurrences. The exercise (or saturation) of a concept
on this account does indeed result in a mental event; i.e., a mental act,
and, if overtly expressed, a speech act as well; but the concept is not
itself the mental or speech act as an event but rather what accounts for
that act's predicable nature. Second level concepts are really referential
concepts on this account, incidentally, and it is their joint exercise or
mutual saturation with predicable concepts (in a kind of mental chem
istry) which is the basis of our particular thoughts and acts of communi
cation (cf. [NBC-4], chap. 2, and [NBC-3], 11-15).
Predicable concepts, on this version of conceptualism, are not
independently real properties or relations, it must be emphasized; and,
unlike referential concepts, their primary role in thought and communi
cation is not referential but predicative. Yet, by a curious development
of the interplay between language and thought, predicable concepts as
cognitive capacities can be transformed into secondary or derived
abilities which enable us to apply predicable concepts in a denotative
manner corresponding to the use of nominalized predicates in natural
language. It is by means of such a secondary or derived application,
moreover, that we purport to refer to independently real platonic forms
as the denotata of nominalized predicates. Thus, for example, not only
do we predicate of a shape that it is triangular, and of a person that he is
wise by applying a predicate concept in each case, but, in addition,
we also purport to denote the properties of triangularity and wisdom,
respectively, by applying these same concepts denotatively.
Purporting to denote and actually denoting are of course not the
same, and in fact, despite all our purportings, there may be no inde
pendently real properties or relations at all which are actually denoted
by any nominalized predicates; or at least that is a view which is
compatible with our present account of concepts as unsaturated cogni
tive capacities. Adopting the Abelardian thesis here, however, will not
result in a counterpart to Russell's form of logicism. In other words, we
shall consider here only the alternative platonic view according to which
most, even if not all, of the predicable concepts which we can form and
articulate actually do denote an independently real property or relation
when applied denotatively. That some predicable concepts which we
can form and articulate must fail to denote such a property or relation
will, on such a view; be an interesting but hardly problematic fact. What
would be problematic, of course, is the different platonic view that
concepts are themselves the independently real properties and relations
they purport to denote when applied denotatively while admitting that
only some of them are individuals while others are not.
Now it might objected that if concepts are cognitive capacities, then
none of them can be impredicatively formed in the Russellian sense in
which they can be specified only be means of a wff which contains a
quantifier ranging over all concepts. In other words, as based on the
capacities of the human intellect, the laws of compositionality regarding
concept-formation in the sense intended here exclude the possibility of
validating the full impredicative comprehension principle (CPt). In this
regard, only a conceptualism which is "constructive" in a sense corre-
sponding to the theory of ramified types will be permitted.
While such a constructive conceptualism can be formalized as a logic
of nominalized predicates without resorting to Russell's treatment of
concepts as "improper objects" for which the law of excluded middle is
to be restricted, we shall, for reasons of space, forego such a presenta-
tion here; for our concern is with the validation, not the invalidation, of
(CPt). In this regard, it should be noted, there is a form of concep-
tualism, which we shall call holistic conceptualism, which agrees that the
initial, and perhaps most important, stages in concept-formation are
"constructive" in the above sense, but which nevertheless maintains that
there is also a stage of concept-formation (usually occuring only in post-
adolescence) at which so-called impredicative concept-formation
becomes possible. Such a stage is realized, moreover, only through our
capacity for language, and in particular through our capacity to use
language for the expression of "constructible" concepts. Impredicative
concept-formation, in other words, is a mediated process, and language
and the linguistic ability to use predicate expressions is the means used
to master and direct such a process. Holistic conceptUalism, in this
regard, presupposes constructive conceptualism as an antecedent stage
of conceptual development, and in particular as a stage which is subse-
quently reconstructed through a certain reiterable pattern of reflective
abstraction which proceeds through the ramified hierarchy of predicable
concepts and which is finally completed or closed only by an idealized
transition to a limit at which impredicative concept-formation becomes
possible (cf. [NBC-4], chap. 2).
According to holistic conceptualism, in other words, the impredica-
tive comprehension principle (CP) of standard second order predicate
logic is valid as a description of the laws of compositionality for
concept-formation, as long as the predicate quantifiers occurring therein
are taken as ranging over all predicable concepts, including those that
are only impredicatively specifiable. A holistic conceptualism which also
adopts the platonic assumption described above regarding the denota
tive application of predicable concepts will in that case also validate
(CP1) as well. In other words, HSTio can serve as the basis for a
formulation of Conceptual Platonism as described above. Moreover, if
thoughts are assumed to be at least in principle always overtly express
ible in language, and 0 is interpreted as ranging over all possible
contexts of use of language, that is, in the sense of pragmatics, then the
principle of intensionality, (DExt*), as a principle regarding the condi
tions under which predicable concepts are understood to denote the
same property or relation when applied denotatively, would seem also
to be valid in Conceptual Platonism.
HST!o + (DExt*) does not itself suffice for the reduction of classical
mathematics to logic, however; and in particular the axiom of infinity is
not provable in HSTio + (DExt*) just as it is not provable in either
HST! + (Ext*) or HST]j + (Ext*). That there are infinitely many
properties and relations as platonic forms is no doubt plausible, if we
are assuming that there are platonic forms as individuals to begin with;
but it is not in any case a consequence of Conceptual Platonism in the
sense in which the latter assumes only that predicable concepts denote
properties or relations when applied denotatively. In this regard,
Conceptual Platonism is very much like Russell's later form of logicism
in [PM], since an axiom of infinity is required in that framework as well.
In [PM], however, the axiom of infinity requires that there be infinitely
many concrete individuals, and that assumption is less plausible than
one regarding properties and relations as abstract individuals. In
particular, an axiom of infinity in the form:
(Inf*) (VE"') (T/Gk)(F 'f G), where n 'f k,
is especially natural, since it only assumes that different properties or
relations are denoted by predicable concepts of different degrees
The situation is more problematic for an account of classes as the
extensions of concepts in HST!o + (DExt*), however, though such an
account is forthcoming if we also assume the principle of rigidity, (PR).
But (PR), unfortunately, cannot be validated in Conceptual Platonism,
since as a principle about concepts as cognitive capacities it amounts to
an unwarranted (and in fact contraindicated) reducibility axiom to the
effect that every predicable concept is materially equivalent to one
having the same extension (or rigid property or relation when applied
denotatively) in every possible context of use. (PR), in other words, can
be validated only by the full platonic view which identifies concepts as
being themselves the independently existing properties and relations
they otherwise purport to respresent, which is to say that (PR) can be
validated only in Russell's early form of logicism reconstructed as
DAHST* + (DExt*) + (PR).
Russell's own reducibility axiom in his theory of ramified types is also
said to be unwarranted (and contraindicated) in his later form of
logicism, incidentally. If it is, then this can only be- so because the
concepts or propositional functions assumed in the theory of ramified
types are not indepedently real properties and relations, as Russell in
fact described them to be, but are rather really concepts as cognitive
capacities in the sense intended here. In that case, however, they cannot
also be the logical objects in terms of which classical mathematics is to
be explained, since in fact they are not objects at all. Russell's later
form of logicisim, in other words, is better understood as a form of
Conceptual Platonism, though apparently one based only on construc-
tive rather than holistic conceptualism as well.
The upshot is that although HSTto + (DExt*) + (Inf*) , but not
HSTto + (DExt*) + (Inf*) + (PR) , can be taken as a counterpart to
Russell's later form of logicism, nevertheless neither HSTto + (DExt*)
+ (Inf*) nor Russell's theory of ramified types suffices as an adequate
framework for logicism. This still leaves our reconstruction of Russell's
early form of logicism intact, however, and it in no way affects either of
our reconstructions of Frege's form of logicism. In other words, as
philosophical doctrines go, logicism, whether in Frege's or Russell's
early form, is alive and well.
Indiana University
I I am indebted to Edmund Gettier for bringing this fact about (SCpn to my attention.
2 The rewrite law
[AXI ... = [AYI ... .. , Yn1xn)],
where no Yi occurs in is derivable in AHST* on the basis of the principle of exten-
sionality, (Ext*), i.e., Frege's basic law (Va), or its intensional counterpart, (OExt*),
described in 11 and 13 below. If neither of these principles are assumed, then the
rewrite law must be taken as an additional axiom schema of AHST*.
3 Weak Zermelo set theory is the restriction of Zermelo set theory to those instances of
the Aussonderungsaxiom in which all quantifiers in the comprehension clause are
limited or restricted; that is, in which all quantifiers have the form (''Ix E y) or (3x E
4 Strictly speaking the translation proceeds through an intermediate system NFU'
which is obtained from NFU by adding A as an individual constant together with the
axiom - (3x) (x E 1\) and the modification of NF's comprehension principle for NFU
to the following:
(CP-NF')(3y)((y= A V (3x)(x E y)] & (lfx) [x E y .... IPI)
where IP is a stratified wff of NFU' in which y does not occur free. Similarly, instead of
AHST* + (Ext*) we translate the A-free wffs of its A-free counterpart HST* + (Ext*) into
the first order wffs of NFU'. (AHST* + (Ext*) is a conservative extension of HST* +
(Ext*) since every wffs of AHST* is provable equivalent in AHST* + (Ext*) to a A-free
wff.) The important clause in the translation, besides that translating monadic predica-
tion into membership, is the clause for predicate quantifiers:
trslfFI)IP) = (ifF) [F=A V (3X)(XE p) - trs(IP)].
where '-' (bar) is a one-to-one mapping of predicate and individual variables onto the
individual variables.
5 Cf. [NBC-2] for a description of Russell's early framework as a form of possibilism.
6 Note that by (LL*),
(a= b) - D (a= b)
is provable in DAHST* + (DExt*), and therefore so are
... xn] = F),
where x does not occur free in a, and Fn does not occur free in [Axl ... x
], and [Axl
... xn] is homogeneously stratified. From these principles, the following commutative
laws follow:
(lfx)D91 .... D(lfx)91
(lfP)D91 .... D(lfP)91.
As applied to concrete individuals, the first of these commutative laws makes sense only
in Russell's early possibilist framework (cf. [NBC-2]).
7 This approach towards the representation of classes and relations-in-extension was
first given by Montague in [FP], p. 132. It also occurs in [Gallin], p. 77, where the
principle of extensional comprehension is the type-theoretical counterpart of the prin-
ciple of rigidity described below.
8 As indicated for AHST*, the rewrite law,
[Axl ... x
91] = [AYI .. Yn91(Y/xl , , Yn!xn)],
where no Yi occurs in 91, is provable in this system on the basis of either (Ext*) or
(DExt*). If neither of thege principles are to be assumed, however, then the rewrite law
must be added as an axiom schema.
Frege, Gottlob:
[Fd] The Foundations of Arithmetic, trans, by J. L. Austin, Harper & Bros.,
N.Y. 1960.
[G & B] Translations from the Philosophical Writings of Gottlob Frege, eds. P.
Geach and M. Black, Blackwell, Oxford, 1952.
Die Grundgesetze der Arithmetik, vols 1 and 2, Hildesheim, 1962.
Posthumous Writings, eds. H. Hermes, F. Kambartel, F. Kaulbach; trans.
by P. Long and R. White, Blackwell, Oxford 1979.
[PMC] Philosophical and Mathematical Correspondence, eds. G. Gabriel, H.
Hermes, F. Kambartel, C. Thiel, A. Veraart; trans. by H. Kaal, University
of Chicago Press, Chicago, 1980.
Russell, Bertrand
[AMR) An Analysis of Mathematical Reasoning, unpublished ms. of 1898, in the
Bertrand Russell Archives, McMaster University, Hamilton, Ontario,
[POM] The Principles of Mathematics, 2nd ed., W. W. Norton & Co., N. Y., 1937.
(PM] Principia Mathematica, co-author, A. N. Whitehead, Cambridge Univer-
sity Press, 1913.
(EA] Essays in Analysis, ed.D. Lackey, Braziller, N. Y., 1973.
Carnap, Rudolf
[Carnap] 'The Logicist Foundations of Mathematics', Erkenntnis 2 (1931); trans. by
E. Putnam and G. J. Massey and reprinted in Philosophy of Mathematics,
eds. H. Putnam and P. Bennacerraf, Prentice Hall, 1964.
Church, Alonzo
[Church] 'Schroder's Anticipation of the Simple Theory of Types', The Journal of
Unified Science (Erkenntnis). 9 (1939): 149-152.
Cocchiarella, Nino
[NBC-I] 'A Second Order Logic of Variable-Binding Operators', Reports on
Mathematical Logic 5 (1975): 3-18.
[NBC-2] 'Meinong Reconstructed Versus Early Russell Reconstructed', Journal of
Philosophical Logic 11 (1982): 183-214.
[NBC-3] 'Philosophical Perspectives on Quantification in Tense and Mopal Logic',
in Handbook of Philosophical Logic, vol. 2, eds, D. Gabbay and F.
Guenthner, D. Reidel, Dordrecht, 1984, pp. 309-353.
[NBC-4] Logical Investigations of Predication Theory and the Problem of Univer-
sals, forthcoming as vol. 2 of Indices, Bibliopolis Press, Naples, 1986.
Gallin, Daniel
[Gallin] Intensional and Higher-Order Modal Logic, No. Holland, Amsterdam,
Heqkin, Leon
[Henkin] 'Banishing the Rule of Substitution for Functional Variables', The Journal
of Symbolic Logic, vol. IS (1953): 201-208.
Jensen, Ronald
[Jensen] 'On the Consistency of a Slight(?) Modification of Quine's New Founda-
tions', Synthese 19 (1968): 250-263.
Kalish, Donald and Montague, Richard
[K & M] 'On Tarski's Formalization of Predicate Logic with Identity,' Arch. for
Math., Logik und Grundl. 7 (1965): 61-79.
Montague, Richard
[FP] Formal Philosophy, ed. R. H. Thomason, Yale University Press, New
Haven, 1974.
Quine, Willard Van Orman
[Quine] Set Theory and its Logic, Harvard University Press, Cambridge, 1963.
Specker, Ernst
[TAJ Typical Ambiguity', in Logic, Methodology and Philosophy of Science,
eds. E. Nagel et al., Stanford University Press, 1962: 116-124.
[AC] The Axiom of Choice in Quine's New Foundations for Mathematical
Logic', "Froc. of the Nat. Acad. of Sciences 39 (1953): 972-975.
Tarski, Alfred
[TarskiJ 'Der Wahrheitsbegriff in Formalisierten Sprachen', Studia Philosophica 1
(1936): 261-405.
Today we find ourselves at the outset of a golden age in the interpreta-
tion of Frege's philosophical writings. Judged by the number of articles,
books, and seminars addressing his thought, interest in Frege is at an
all-time high. More importantly, as Frege has come out of the shadow of
Russell and Wittgenstein into the fun light of critical attention, the
degree of sophistication of discussion has achieved a quantum improve-
ment. Many factors conspired to bring about this result, but two events
may be singled out as having madt;: contributions both to the
resurgence of interest in and to our greater understanding of Frege's
First is the publication, more than sixty years after his death, of that
part of his Nachgelassene Schriften which survived the vicissitudes of
the intervening years. These papers appeared in German in 1969 and in
English in 1979.
Some of the contents are rough in form, though not
without value. We are offered, for example, tables of contents and
partial drafts of a textbook on logic and its philosophy which Frege
made starts on at various crucial periods of his life. Even draft frag-
ments of this sort permit important inferences from the order of
presentation and different emphases given various topics to conclusions
about the explanatory priorities Frege associated with his central
technical concepts. But not all of the selections represent rough cuts or
abandoned projects. Included are some fully polished articles, dealing
with Frege's most central technical concepts - fine examples of his
concise, sometimes lapidary mathematician's prose - which he had tried
unsuccessfully to publish. In a number of cases, these additional texts
permit the resolution of exegetical disputes occasioned by what can now
be seen to be accidental lacunae and merely apparent emphases in the
canonical published corpus.
The other landmark event is the publication in 1973 of Dummett's
monumental and long-awaited full-length treatment of Frege's philoso-
phy of language.
It would be difficult to overestimate the significance of
L. Haaparanta and J. Hintikka (eds.), Frege Synthesized, 253-295.
1986 by D. Reidel Publishing Company.
this classic work. Anyone interested in the interpretation of Frege must
give it the same close attention owed to the primary texts. Its clarity of
thought, patient rehearsal of considerations, and exercise of the best
critical judgement in final appraisal will not be soon equalled. This essay
will not offer a systematic account of Dummett's views, since the most
important of these are so intimately tied up with the development of
powerful novel approaches to contemporary philosophy of language as
to defy brief characterization, even by their author. The original volume
has now been supplemented with another containing many valuable
amplifications and clarifications.
The result is a 1300 page corpus
which, Dummett's complaints
to the contrary notwithstanding, by now
deserves to be considered as setting out the canonical reading of Frege.
It is so considered by the authors discussed below, and forms the
background against which their own accounts are set out.
Two examples will serve to indicate the sort of interpretive advance
signalled by these events. First, it was widely believed in the 1950's and
1960's that Frege did not intend the distinction between sense and
reference to apply to functional expressions such as predicates, but only
to complete expressions such as terms and sentences.
Although the
famous essay on sense and reference does not discuss such an applica-
tion of that distinction, the Nachlass makes clear that this is only
because that discussion was reserved for a further article which is quite
explicit in its endorsement of that application, but which was repeatedly
rejected for publication until Frege abandoned the attempt. Several
other passages reprinted in [19] decisively refute the interpretation
which would restrict the distinction to complete expressions. A some-
what less important mistake may also be mentioned as indicative, which
was done in as much by Dummett's arguments as by the unearthing of
further evidence. In 'On Sense and Reference' Frege says "One might
also say that judgements are distinctions of pans within truth-values,"
and that "the reference of the word is part of the reference of the
sentence."6 These remarks have sparked the attribution of a variety of
bizarre ontological views to Frege, centering on the notion of the True
as representing the whole world, sometimes conceived as a Tractarian
world of facts, sometimes as comp'osed of objects (and what about the
False?). The remarks stem from a hasty assimilation, soon explicitly
rejected, of the relation between the argument of a function and the
value it determines to the relation of part and whole. For although the
function 'capital of .. .' takes the value Stockholm when Sweden is taken
as argument, Sweden is not part of Stockholm. Dummett's discussion of
this issue has permanently disposed of the temptation to take these
remarks seriously as interpretive constraints. We shall see below, how-
ever, that there remain genuine controversies which are not so easily
disposed of (concerning the senses and referents of functional expres-
sions) which may be regarded as successors to these two mistaken lines
of thought.
Dummett has shown that Frege should be treated as a modern
thinker in the sense that one can think about contemporary philosophi-
cal issues of considerable significance by thinking about his concepts
and their explanatory deployment, and that one cannot think about
those concepts and their principles of deployment without thinking
about such contemporary issues. In what follows those concepts are
approached from three different directions. First, an attempt to interpret
and develop Frege's technical scheme in light of contemporary discus-
sions of the issues he was addressing is considered. Then attention is
turned to an argument to the effect that ignoring the historical context in
which Frege developed his theories, treating him we might say merely as
a contemporary, leads to substantive misinterpretation of those theories.
,Finally, following one strand of the account of the path by which Frege
developed and defended some of his central concepts, leads to a novel
diagnosis of the status of those concepts.
One important recent offering is David Bell's book Frege's Theory of
This is a clear and well-written work. The issues it raises and
the form in which they are addressed merit the attention of anyone
interested in the significance for current inquiry of Frege's strategic
deployment of a battery of technical concepts to explain various aspects
of linguistic practice. Its title is worthy of some consideration. It is a
measure of the degree of sophistication of contemporary Frege commen-
tary that a controversy exists even over how one should describe the
topic which his philosophical work addresses. Of course no one disputes
his concern with the foundations of mathematical reasoning and knowl-
edge, expressed above all in his three books, the Begriffsschrift, the
Grundlagen der Arithmetik, and the Grundgesetze der Arithmetik But
the more general conceptual framework he found it necessary to
elaborate in order to express clearly and precisely his claims about the
nature of mathematics and its objects cannot easily be characterized
without prejudging substantial issues of interpretation. It may seem
obvious that Frege was pursuing a project in the philosophy of lan-
But such a description is misleading in the context of Frege's
own insistence on the priority of thoughts (though not of thinkings) to
their linguistic expression. For he was interested in natural languages
only insofar as they permitted rough formulation of objective and
language-independent thoughts, and he crafted artificial languages only
as more adequate means for their expression. It would be inappropriate
to build into the description of the subject-matter at the outset a post-
Wittgensteinian conviction of the wrong-headedness of such an
approach, by assimilating his concerns to contemporary investigations
under the rubric "philosophy of language". One of the major theses of
Sluga'S book, discussed below, is that such Whiggish presuppositions of
continuity of concern have consistently led Frege's readers to overlook
important strands of his thought. Dummett has also suggested "theory of
meaning" as a general characterization, but this seems to apply better to
his own enterprise than to Frege's. For 'meaning' is correlative to
'understanding', and Frege's concern lay at least equally with reference,
which is not in general grasped when one understands a claim, as with
the sense which must be grasped in that case.
In his discussion of the book,9 Dummett objects that Bell has
misdescribed his topic, in that Frege's treatment of the act of asserting is
the topic of only one chapter, while the rest of the book talks about the
notions of sense and reference. This seems unfair, for the heading
"theory of judgement" ought to entitle Bell to offer an account of the
contents which are judged as well as of the acts which are the judgings of
those contents. It has the advantage of placing Frege's concerns in
appropriate historical and philosophical context. Bell's denomination of
Frege's topic as judgement displays his recognition of the importance
Frege, in company with Kant and Wittgenstein, placed on inverting the
traditional order of explanation which took concepts as primary and
sought to account for judgments in terms of them. At least until 1891,
Frege clearly regarded the claim that concepts can only be understood
as the products of analysis of judgements as one of his most central
insights. Although Bell does not say so, it is equally clear in the
Begriffsschrift that Frege completes the inversion of the classical priority
of concepts to judgements and judgements to syllogisms by taking the
contents of sentences (judgement in the sense of what is judged rather
than the judging of it) to be defined in terms of the inferences they are
involved in.1O Concepts are to be abstracted from such judgements by
considering invariance of inferential role (which pertain only to judg-
ments) under various substitutions for discriminable (possibly non-
judgemental) components of the judgement. Both in the introduction to
BGS and in his essay on "Boole's logical Calculus and the BGS",ll the
virtue of the purely formal perspicuous language of inference in non-
formal contexts is described as its permitting for the first time the
scientific formation and expression of concepts. Although it is for this
reason that Frege called his first work a "concept script", he later came
to believe this phrase misleading precisely because it obscured his
doctrine of the primacy of judgements. On the other hand, it would be
equally misleading to describe Frege simply as a theorist of inference, in
spite of the explanatory priority he accorded to it. For his primary
theoretical focus always lay on the sentential and thence sub-sentential
contents attributable to different expressions in virtue of the roles they
played in inference, as revealed by their behavior under substitution. So
"judgement", which is (a translation of) an expression Frege himself
used pretheoretically to describe the object of his theorizing, seems a
good choice to delimit his subject matter.
Like any other choice, however, it does prejudge some controversial
issues of interpretation, for instance that concerning the persistence in
Frege's thought of the so-called "context principle". It is often unclear
exactly what this principle means, but the canonical statement of it is the
Grundlagen claim that "only in the context of a sentence does a word
have any significance". (I use 'significance' here for Frege's 'Bedeutung'
because in 1884 he had not yet distinguished Sinn from Bedeutung, and
the undifferentiated term should be marked.) It is often claimed,12 even
by those such as Dummett who take the putative change in view to be a
serious mistake, that when Frege achieved his mature views in 1891
with the formulation of that crucial distinction he discarded the context
principle. If that is so, then Bell's choice of "theory of judgement" to
describe the topic of the mature semantic views he discusses would be
misleading or simply incorrect. As we shall see below, Sluga argues that
Frege never relinguishes the context principle. Bell does not argue this,
however, nor does he even claim it. He is simply silent on this issue, as
on others concerning detailed questions about the attribution of various
views to Frege based on textual evidence.
Bell's enterprise lies in a different direction entirely. He is concerned
to look closely at the explanatory roles played by Frege's various
concepts and at the ways in which Frege takes them to be related, in
order to refine and reconstruct a broadly Fregean account of the nature
of judgement. In keeping with this aim, he is not engaged in the exegesis
of Fregean texts, and freely discards from his reconstruction a number
of doctrines which Frege clearly held, in favor of incompatible princi-
ples (for instance, in Bell's reconstruction functional expressions are
assigned senses but not referents). His project is to salvage from Frege's
account those insights which can be put together to form a workable
theory of judgement. The result is broadly Fregean in endorsing the
following "major strands" of Frege's theory:13
I. There is the methodological principle that 'we can distinguish parts
in the thought corresponding to the parts of a sentence, so that the
structure of the sentence serves as a model of the structure of the
II. A thought is (a) objective, (b) the sense of an indicative sentence ....
III. A thought must have at least one 'unsaturated' or functional element,
otherwise its elements would fail to coalesce and would remain merely
disparate atoms.
IV. In a thought the complete elements refer (if at all) to objects.
The nature of this enterprise makes it hard to evaluate its success. There
are many issues one would think to be central to any attempt to offer a
theory of judgement which Bell nevertheless does not address. For
instance, although he argues that it would be wrong to require an
account of judgement to restrict itself to the form of an account of the
propositional attitude constructions used to attribute judgements to
others, he does not justify the books failure to present any such account
as a proper part of such a theory. Again, althought it has been suggested
above that Bell was not obliged to restrict his attention to the notion of
assertoric force (the analysis of the act of judging), one would certainly
like a fuller and more satisfactory account of that notion than the
cursory sketch we are offered.
The book does its work in a sort of
methodological no-man's land between textual exegesis and theory
construction owing allegiance only to the phenomena it seeks to
theorize about.
This is not to say that the analysis is not enlightening, however. Bell is
at his best when dissecting the explanatory role assigned by Frege to his
technical concepts. When he succeeds we learn both about Frege and
about the phenomena. Consider for instance the notion of Bedeutung.
Bell tells us that:
... Frege had not one, but two notions of reference. These notions hang together so well
in the case of singular terms that they are hard to distinguish in this context. In the case
of predicates, however, they are not only distinguishable, they are different to reconcile.
One notion is this: the reference 01 an expression is that extra-linguistic entity with
which the expression has been correlated or which it picks out. The other notion of
reference is that it is a property which an expression must possess if that expression is to
be truth-valuable (to coin a phrase). By truth-valuable I mean such that it either
possesses a truth-value, or is capable of being used (and not just mentioned) in a
sentence which possesses a truth-value.
Bell claims that although in the case of singular terms one notion can
play both of these roles, since for them to be truth-valuable just is to be
correlated with an object, in the case of sentence and functions the two
notions diverge. All that Frege ever offers in the way of evidence for the
application of the notion of reference to expressions in these categories
is considerations showing them to be truth-valuable. Since he does not
distinguish the two different notions of reference which he has in play,
he feels entitled to conclude that they possess reference in the first sense
as well. But this is a non sequitur, or at any rate a transition which must
be justified, and not simply assumed on the basis of the conflation of the
two different senses of Bedeutung. Thus Bell rejects the notion of truth-
values as objects, and of functions as the references of functional
expressions, as excess conceptual baggage mistakenly mixed in with the
second notion of reference, which is the only one doing any explanatory
work for these categories.
This analysis is clear-headed and valuable but can be faulted on two
grounds, each of which amounts to a request for further analysis. First,
as Dummett points out,16 the characterization of the second notion of
reference does not seem right. For as Bell has described it, reference is a
property which an expression either has or lacks, depending upon
whether sentences containing it can have or always lack truth values.
But Frege's notion is that in addition to having or lacking reference,
expressions which have reference can have different references, accord-
ingly as they make different contributions to the truth-values of sentences
containing them. The test is always substitutional - two expressions
which have reference have different references iff in some context the
substitution of one for the other changes a true sentence into one
which is not true. Others who have noticed the distinction Bell is after
have put things better. For instance, Tugendhat
(who seems to have
introduced this line of thought) calls this non-relational sense of refer-
ence "truth-value potential" and in effect identifies the truth-value
potential of a sub-sentential expression with the equivalence class of
expressions intersubstitutable salva veritate.
The sharpening of Bell's distinction (which makes it similar to that
between 'referent' and 'reference' which Dummett uses throughout [8])
does not affect his criticism of the inference from possession of refer-
ence in this non-relational sense to possession of reference in the
relational sense, of course. But it does affect a further use he wants to
make of the distinction, to argue that it is incorrect to think of predicate
expressions as having a reference at all, even in the non-relational sense.
For here Bell argues that Frege incorrectly takes as a necessary and
sufficient condition for the truth-valuability (in Bell's sense) of predi-
cates that they have sharp boundaries. He accordingly takes it that the
assignment of reference to predicates is motivated only by this require-
ment, and so that showing the untenability of such a requirement is
sufficient to show the inappropriateness of assiging reference to pre-
dicate expression at all. This line of argument is undercut by seeing that
there is more to the second notion of reference than truth-valuability.
Since the denial of the cogency of the application of the notion of
reference to predicates (or function expressions generally) is one of the
main innovations of Bell's analysis, his failure adequately to characterize
that part of Frege's notion of reference which remains when one takes
away correlation with an extra-linguistic object has serious conse-
quences for the subsequent course of his argument.
Dummett, however, rejects not only Bell's characterization of the
second notion of reference, but also the claim that there are two notions
of reference. He claims that the relational and the nonrelational senses
represent "two ingredients of one notion". The second "tells us what
Frege wanted the notion of reference for, and the oher tells us how he
thought that it applied to the various categories of expression".18 It may
be granted that the explanatory work Frege wanted the notion of
reference for is its truth-value potential or contribution to the truth-
conditions of sentences, and that he thought that the intersubstitutability
equivalence class of equipollent expressions was determined by the
correlation of all and only its members with the same extra-linguistic
entity. But it would still remain to be asked, for instance, whether the
identity of the correlated object and the nature of the correlation can be
inferred from the semantic equivalence class of expressions they deter-
mine, as Frege's arguments concerning the reference of sentences and
functional expressions would seem to require. Such a question is in no
way made less urgent or easier to answer by rephrasing it in terms of
two ingredients of one notion rather than in terms of the relations of two
notions. In the final section of this paper it will be argued that this
difficulty is one instance of a quite general definitional failure of Frege's
part, one which in another context he tried unsuccessfully to resolve in a
purely technical way.
Putting the issue in these terms raises the second source of dissatis-
faction with Bell's argument. For the sort of question just raised seems
no less important or difficult for the paradigmatic case of singular terms
than for the parts of speech Bell finds problematic. The basic substitu-
tional/inferential methodology which yields the sense of
reference as an equivalence class of expressions vastly underdetermines
the correlated objects and mode of correlation invoked by the relational
sense even for proper names. Tugendhat, having formulated the non-
relational notion of reference, takes it to be the notion of reference,
discarding correlation with an object as a realistic confusion best
extruded from Frege's thought. Sluga follows Tugendhat in this regard.
The reason in each case is that all that Frege's analysis of the use of
expressions seems to require is the sorting of expressions according to
the non-relational sense of substitutional role. The semantic analysis he
developed is a method for the perspicuous codification of inferences.
Truth is what is preserved by good inferences, and sub sentential expres-
sions can be grouped into co-reference classes accordingly as inter-
substitution within the classes preserves such good-iriference potentials.
Such an approach can give rise to specification of the conditions under
which two expressions have the same reference, but how can it warrant
a claim that the shared reference is to be identified with some object
(among all those which in one way or another could be taken to deter-
mine the same coreference classes) specified otherwise than as the
reference of an expression? The answer seems to be that Frege's
arguments for this identification are straightforwardly substitutional
ones, in particular that for any singular term t we can always substitute
(saving the inferential potentials) the. term the object referred to by
the singular term 't'. The expressions which license intersubstitution
of expressions are identity locutions (as Frege had argued in the
Grundlagen), and so we are correct to say that the object referred to by
the singular term 'Julius Caesar' is Julius Caesar. Whether this fact has
the significance Frege thought it had is another matter. I
One of the most important discoveries of the early 1970's, both from
the point of view of the interpretation of Frege and of the philosophy of
language generally (for once, made independently of Dummett) con-
cerns the need to distinguish two different explanatory roles which are
conflated in Frege's technical concepts of sense. Kripke and Putnam
independently argued 20 that the cognitive notion of the sense of an
expression, what one who has mastered the use of that expression may
thereby be taken to understand, and the semantic notion of the sense of
that expression, what determines the reference of the expression, cannot
in general be taken to coincide. In particular, in the case of proper
names no knowledge or practical capacity which can plausibly be
attributed to an ordinary competent user of the name will suffice to
determine the object of which it is a proper name. A similar point can
be made about the use of natural kind sortals. Since Frege had required
that his notion of the sense of an expression play both the cognitive and
the semantic role, and since for an essential range of expressions no
single notion can do so, it is apparent that his concept must be refined
by dividing it into two distinct sense-concepts, whose interrelations it
then becomes urgent to investigate.
A further distinction within the semantic notion of sense has been
urged by a number of writers, on the basis of the consideration of the
behavior of indexical or token-reflextive expression.
In Kaplan's
idiom, we must distinguish for such expressions between their character,
which is associated with the expression type, and the content associated
with each contextually situated token(ing) of that type. The distinction in
question is evident in the following dialogue:
A: I am anxious to get started.
B: No, it is possible that you are eager, but I am the anxious one.
We are concerned with the semantic notion of the sense of an expres-
sion, that is, with the way in which its reference is determined. In one
sense both tokens of "I" have their reference determined in the same
way, for in each case it is the speaker responsible for the tokening who
is referred to. These expressions share a character. But in another sense
A's token of "I" and B's token of "you" have their reference determined
in the different ways (e.g. for the purpose of tracking the referent through
the other possible worlds which must be considered to evaluate the
model qualifications in B's remark). The referents of these tokenings will
coincide in every possible world relevant to the evaluation of these
utterances, in virtue of the identity of their contents. The characters of
these expressions, together with the context in which they are uttered,
determine a content which in tum determine a referent in every possible
world. It is this latter task with which the semantic notion of sense is
charged for nonindexical expressions. Such expressions may accordingly
be thought of as those whose character determines a content without
needing to be supplemented by a context. The point is that as we ask
about what would be true in other worlds of the individual picked out
by B's indexical utterance there is a double relativity to possible worlds,
accordingly as those worlds can be relevant to the two different stages in
the determination of a referent. First, since B's remark could have been
addressed to someone other than A, we must consult the world-context
in order to determine what content is fixed by the character of the
exression when uttered in that context. The individual concept so
determined as a content can then be tracked through various possible
worlds and assigned referents in each, so that model claims can be
Without referring to either of these antecedents, Bell distinguishes
two notions of expression sense in a way which partakes of some of the
features of each of the other distinctions. He calls his two notions "input
sense" and "output sense", and introduces them by reference to two
Fregean principles:
PSI: The sense of a sentence is determined by the senses of its com-
ponent parts,
PRI: The truth-value of a sentence is determined by its sense. (And, of
course, how things stand.)
His claim is that although the ''two principles depend for their plausi-
bility and usefulness on there being a sense of 'sense' which remains
constant throughout", in fact they demand different ones. Input sense is
that notion of which principle PS I holds, and output sense is that notion
of sense of which PRI holds. Input sense is that which is preserved by
correct translations and that for which synonymy claims assert identities
of sense. Sub-sentential expressions have input senses (''meanings''), and
these combine to determine the input senses of sentences containing
them. Output senses are defined as what is common to claims such as
''Today I ate plum pudding," and "Yesterday you ate plum pudding".
The input senses of sentences together with a context of utterance
determine such output senses. The output senses of sentences are what
can meaningfully be described as true or false, as per principle PR2.
As described so far, Bell's distinction amounts to the claim that the
cognitive/semantic and character/content partitions of the notion of
sense ought to be seen as coinciding. For the compositionality of sense
is a postulate required for the explanation of the possibility of under-
standing complex expressions, so that it must be input senses which are
in the first instance grasped cognitively. Semantic senses, determining
truth values of sentences, are in turn identified with output senses. But
since the latter are determined by the former together with a context of
utterance and the distinction is enforced by attention to indexical
expressions, the character/content distinction is likewise subsumed by
the difference between input and output senses.
Such an identification is clearly subject to a number of objections, as
consideration of the quite different motives and functions of the con-
flated distinctions indicates. But these difficulties may not be insur-
mountable. Perhaps a useful view could be elaborated based on the
assimilation of the sense in which the referent of a proper name token is
determined not by what its utterer understands by it, but only by this
together with a causal, historical, and social context in which the token
is embedded, on the one hand, and the sense in which the reference-
determining sense of a token of "yesterday" is given not just by what one
can understand as the meaning associated with the expression type, but
only this together with a concrete context of use. But Bell does not
attempt to develop such an account. In part this is because he has
nothing whatever to say about what "contexts" are, or how these
together with input senses determine output senses. And it is just here
that all the detailed work is involved in making out either half of such
an assimilation, and hence in justifying their conflation. But Bell is
precluded from addressing such a task by other, less defensible features
of his view.
For Bell denies that sub-sentential expressions have output senses at
all, claiming that "output sense is essentially sentential",23 No argument
or even motivation for this position is presented. It is suggested that for
sentences the distinction between input senses and output senses corre-
sponds to that between sentences and the statements they can be used to
make, and that it is better to think of the former not as possessing truth-
values which change, by contrast to statements whose truth-values do
not, but rather to think of the former as not the kind of thing which can
have truth values at all. But no reason is given for not extending this
distinction to sub-sentential expressions. The distinction hetween the
two varieties of sense is introduced, as indicated above, in terms of two
Fregean principles. PR2, the 'sense determines reference' principle, is
quoted at this portion of the argument as restricted to sentences and
truth values. But of course the principle Frege uses is not so restricted.
Indeed, when Bell first introduces it some sixty pages earlier it is in
unrestricted form. He has just been discussing the principle he calls
PRI, that the reference of complex expressions is determined by the
references of their components (which Bell discards because as we have
seen he does not attribute reference of any kind to functions). He says:24
Elsewhere in his writings, however, he seems to invoke a quite different principle which
we can call PR2. It is this: (a) the reference of any expression is determined by its sense,
(b) the sense of a complex expression is determined by the senses of its component
Two features of this definition deserve comment. First, part (b) of
principle PR2 as here stated is what he later calls PS I and is concerned
precisely to distinguish from PR2. Second, part (a) of this original state-
ment differs from the later version in not being restricted to sentences.
Neither of these substantial changes in the significance of his expression
"PR2" is announced, acknowledged, or motivated in the intervening
text. Such carelessness in specifying a central interpretive principle
which one has taken the trouble to name for clarity of reference is bad
enough under any circumstances. It is unforgivable when essential
features of one's own claims and their justifications depend precisely on
the matters obscured by the sloppiness. As things stand, the reader is
left with no idea why in using the two principles PR2 and PSI (=
PR2(b) in earlier statement) to distinguish two notions of sense one
should employ the later version of PR2 rather than PR2( a) from the
earlier version, which is the principle Frege endorsed. Apart from the
invocation of PR2, output senses are specified as what is common to the
two "plum-pudding" sentences quoted above. As our sketch of the
character/content distinction show, it is not at all obvious why this
characterization should not extend to what is common to 'today' and
'yesterday', on the one hand, and'!' and 'you' on the other.
Bell does, however, employ the restriction of output senses to
sentences to argue for a further point. For he claims that the "context
principle" of the Grundlagen may be understood in terms of the fact
that terms only have input senses, which together with the input senses
of other expressions determine sentential input senses, which in context
determine a truth-value. Since the reference of terms matters only in
determining truth-values, it is "only in the context of a sentence that a
term have a reference". Clearly nothing can be made of this line of
thought in the absence of a rationale for its basic premises.
These difficulties with the distinction between input senses and out-
put senses also make it difficult to evaluate another novel interpretive
suggestion which Bell offers. He concludes his discussion of the senses
of proper names with the claim
"The sense of a proper name, then, is
that it purports to refer to a determinate object of a given sort with
which it has been conventionally correlated." The sense of a proper
name is here taken as "that which one understands when one is able to
use it correctly".26 As indicated above in the discussion of the relation of
the cognitive notion of sense to Bell's notions, this must be the input
sense, for sub sentential expressions aren't supposed to have output
senses. It is accordingly obscure what the connection is supposed to be
between the senses Bell is offering a theory of here and the determina-
tion of referents for the proper names they are senses of. What then are
the criteria of adequacy for an account of what a name user must be
taken to understand? Bell examines the conditions under which we
would want to deny that someone had mastered the use of a name, and
concludes that in addition to using it as a singular term one must at least
know some sortal under which the referent is taken to fall in order to be
judged a competent user. This is useful as a necessary condition, but
much less plausible as a sufficient condition to be taken to be using an
expression as a proper name. For a sufficient condition would seem to
require that one be appropriately connected to a community of users of
the name, perhaps an historically extended one, whose joint use does
determine a referent, though no individual's use need do so. It is not
obvious that merely believing that some conventional correlation has
been established with an object of the right sort is sufficient to be
appropriately connected with the community of users of that name. In
any case, to argue for such a principle would require looking at how
input senses and various specific sorts of context can together determine
output senses and eventually referents for the names in question, and
this Bell does not undertake.
Bell wants his notion of proper name sense in order to develop an
appropriate account of the senses of functional expressions. This latter
task is made especially urgent by the confrontation between his denial
that the referents of functions have any explanatory value, on the one
hand, with the undeniable importance in Frege's scheme of functions
and concepts understood as functions, on the other. Bell's reconstruc
tion reconciles these ideas by interpreting concepts and functions as the
senses rather than the references of functional expressions. A concept,
accordingly, is to be understood as a function which can take as
arguments proper name senses of the sort he has described, and yield
thoughts, the senses of sentences. While this identification of concepts
must be seen as a revision rather than an interpretation of Frege's
thought, it might seem that, setting that identification aside, at least the
account of the senses of functional expressions as functions from the
senses of argument expressions to the senses of value expressions ought
to be uncontroversial. It is not, and it is instructive to see why not.
As Bell has pointed out in his discussion of senses generally, the
concept of sense is required to play two distinguishable roles. First, the
sense of a component of a complex expression must contribute to the
determination of the sense of that complex. But also, the sense of the
component must determine a reference for that component. This gives
us two different ways to think about the senses of functional expressions
such as predicates. On the one hand they must combine with the senses
of terms to yield the senses of sentences. On the other hand they must
be the way in which a function from objects to truth values is deter
mined or given. It is not obvious that these two jobs can be done by one
notion. In particular, Dummett has argued
that "once the proper name
has specified the way in which the object is given, then it has made its
contribution to the sense of the sentence; if it had not, then it would be
impossible to see how its sense could both contribute to the sense of the
sentence and consist in the way in which the object is given." That is,
maintaining the coincidence of the two roles of sense in the case of
proper names (presumably where our grasp is firmest) commits us not
only to their divergence for functional expressions, but also to which
half we give up, namely the identification of their senses with sense
functions. Geach has objected to this doctrine of Dummett's,28 and it is
instructive to examine Dummett's response.
It is not disputed that once a sense has been assigned to a predicate, a
function from the senses of proper names to thoughts is determined. For
according to Dummett the predicate sense is the way in which a
function from objects to truth values is given. Hence, when that function
is supplemented by an object, it determines a way in which a truth-value
is given, that is, a thought. But since a term sense will determine such a
supplementing object (according to the second role of senses mentioned
above), the predicate sense will induce indirectly a function from term
senses to sentences senses. As Dummett says, "the question is whether
the sense of the predicate just is that function."
To argue that it is not, Dummett appeals to a further thesis of Frege's
about senses, namely that the senses of component expressions are parts
of the senses of the complex expressions in which they occur. We have
seen that it is a mistake to think of functions or their arguments as parts
of the values they generate, as Frege's retraction of his careless claim
that objects are parts of truth values shows. But since Frege did hold
that predicate senses are parts of thoughts, we would be committing
precisely this howler if we identified those senses with functions taking
term senses into thoughts. This is an ingenious counter-argument, but it
cannot be considered decisive. For while it would be a howler to treat
functions and their arguments generally as parts of the values they
determine (as in the combination of Sweden and the function the capital
of ... to yield Stockholm), this consideration does not show that
particular functions and kinds of function cannot have values which
contain the functions or their arguments as parts. Stockholm is part of
the value of the function the country of which . . . is the capital. And
mathematical examples of function-values which contain functions as
parts in the set-theoretic sense are easy to come by. (One thinks of the
story of the oracle who offered to answer a single question, and upon
being asked "What is the ordered pair whose first element is the best
question I could ask you, and whose second element is its answer?"
replied (falsely, I suppose): "the ordered pair whose first element is your
question and whose second element is this answer.")
Insulated from this dispute about sense functions by his distinction
between input senses and output senses, Bell backs up his commitment
to treating the senses of functional expressions as functions by citing a
number of passages, both published and from the posthumous works, in
which Frege unequivocally describes such senses as "unsaturated",
"incomplete", and "in need of supplementation", going so far in fact as
to say that "The words 'unsaturated' and 'predicative' seem more suited
to the sense than to the reference".3o To motivate his identification of
concepts with sense functions, Bell argues as follows.
The only reason
Frege had for believing in concepts as predicate referents was the need
to deal with a situation in which predicates have a sense and so
determine a thought, but lack a reference, and so determine a thought
which has no truth-value. The only case where this can happen which
does not reduce to the failure of a term to have a reference is where the
predicate is not defined for the sort of argument to which it is applied.
But this sort of case can be much more plausibly excluded by considera-
tions concerning predicate senses. For such cross-categorial predications
(such as "Julius Casesar is the sum of two prime numbers") ought
properly to be seen as not s\:lcceeding in expressing thoughts at all. Bell's
solution accordingly is to see predicates as having sortal restrictions
associated with their argument places, which together with the 'sortal
physiognomy' he has already assigned to proper name senses yields the
result he desires. One of the benefits which might be derived from such
a radical reconstruction should be made manifest by the discussion to
be given below to the difficulties ensuing from Frege's insistence that
functions be defined for all arguments whatsoever. However, as before,
the evaluation of this thesis about senses must await some resolution of
the general questions Bell has left open concerning his distinction
between input and output senses.
Hans Sluga's new book on Frege in the "Arguments of the Philoso-
phers" series
represents an approach complementary to Bell's in
almost every regard. It's central aim is to reread Frege's work in the light
of that of his precursors and contemporaries, rather than by reference to
his successors in the analytic tradition, as has been traditional. Although
Frege's unprecedented innovations in symbolic logic have made it
natural to think of him exclusively in the role of the founder of a
tradition - as a man without a past - Sluga argues that we ignore at our
peril his intellectual climate and .the influences which conditioned
various aspects of his technical concepts and of the explanatory tasks he
set for them. Sluga's task is not purely historical, however. For he is also
concerned to set out and justify novel readings of some of Frege's purely
philosophical doctrines, readings which are suggested and motivated by
the historical recontextualization he recommends. The result is a stimu
lating new picture of Frege's thought which will be of interest even to
those who are not in the end persuaded in detail by it. Furthermore,
since the narrative strategy employed is to trace the development of
Frege's ideas chronologically (starting, as it were, before he was born)
and surveying all of his important writings seriatim, this book is excel
lently constructed to serve as an introduction to these ideas (as Bell's or
Dummett's books, for instance, could not) as well as to challenge
The book's historical orientation, then, is not adopted only for its
own sake, but also in order to guard against blinding ourselves to inter
pretively significant features of Frege's work by the importation of
anachronistic prejudices. Accordingly, it is primarily in terms of the
philosophical illumination they provide for our appreciation of Frege's
concepts and claims that we must evaluate the success of Sluga's various
invocations of historical influence. The claimed influences may be
considered under four headings. First, a view is presented about who
Frege took to be his philosophical opponents. Next, Leibniz is identified
as a precursor. Third, claims are made about the influence of two
logicians of the generation preceding Frege's, Lotze and Trendelenburg.
Finally and most significantly, it is claimed that overlooking the intellec
tual debt which Frege owes to Kant has most seriously distorted our
understanding. We will consider these claims in this order.
In his first chapter, Sluga is concerned to refute the claim that "In a
history of philosophy Frege would have to be classified as a member of
the realist revolt against Hegelian idealism, a revolt which occurred
some three decades earlier in Germany than in Britain."33 In this aim he
succeeds unequivocally. Hegelianism had ceased to be dominant or even
popular in German philosophical circles some years before Frege was
born. The view against which Frege was reacting is the scientific natural
ism which Sluga claims was held by the physiologists turned philoso
phers Vogt, Moleschott, Buchner, and Czolbe, popularized during
Frege's lifetime by Haeckel, and shared with some reservations by
Gruppe. Ontologically this view is a reductive materialism, and epistem
ologically it is an empiricist psychologism. Sensations are viewed as
material processes of the brain. Concepts, and hence the thoughts
constructed from them are taken to be reflections of such sensations.
Logic is seen as the study of the laws of thought, that is, as an empirical
investigation seeking to establish the natural laws governing the associa-
tion of concepts in judgment and of judgments in inference. It is this
psychologism which Frege so vigorously opposed, and on those rela-
tively few occasions when he describes his opponents as 'idealists' it is
clearly this school which he has in mind.
This is a point of no small moment, especially in the context of an
evaluation of Frege's role as progenitor of the analytic tradition. For his
over-arching objection to the naturalists is their failure appropriately to
distinguish between the normative and ideal order of correct inference
and justification on the one hand, and the descriptive and actual order
of causation and empirical processes on the other. Their concommitant
confusion of features of cognitive acts with features of the contents of
those acts is merely the expression of this original sin. And in his
insistence on the centrality of this basic distinction Frege is at one with
Kant and the post-Kantian idealists, and at odds with the primarily
physicalist and empiricist tradition in Anglo-American philosophy
which he fathered, and in the context of which it has been natural for us
to read him.
Throughout his book Sluga talks about Leibniz' influence on Frege,
but when he specifies the details of this influence his claims turn out to
be quite weak. Like Leibniz (and Kant), "Frege is interested in the study
of logic and the foundations of mathematics because they allow one to
ask in a p'recise form what can be known through reason alone."35 Aside
from this general rationalist commitment to the possibility of a priori
formal knowledge, the only Leibnizian doctrine which is attributed to
Frege is the endorsement of the project of the universal characteristic.
Frege explicitly describes the motivation for his Begrijfsschrift in this
way. That at this level of generality Frege owes a debt to Leibniz is
hardly a novel or surprising claim, however. Sluga also discusses the
influence of Trendelenburg, but in the end the claims seem to come to
little more than that he was the conduit through which Frege became
familiar with Leibniz' ideas.
It is otherwise with the connection discerned between Frege and the
logician Lotze. The suggestion of influence here has specifically been
denied as "a remarkable piece of misapplied history".36 Yet in this case
Sluga shows sufficiently striking similiarities to make the hypothesis of
influence persuasive. It is known that Frege read Lotze. Indeed it has
been argued that the theory of judgement in opposition to which he
presents his innovation in the Begriffsschrift just is Lotze's formulation.
The essay immediately preceding 'The Thought' in the journal in which
it was originally published, which Sluga takes to have been intended by
the editors as an introduction to Frege's essay, mentions Frege in the
context of an exposition of Lotze which highlights several Fregean
From Sluga's account of Lotze's views (as presented in the
Logik of 1874 and an earlier work of 1843) one can extract eight points
of similarity with Frege.
First, Lotze inveighs against psychologism and indeed is the figure
Frege's contemporaries would probably have identified as leading the
battle against the dominant naturalism of the day and in favor of a more
Kantian position. Next, Lotze was a logicist about mathematics,
although there is no hint in his works that he took the detailed working
out of such a reduction to logic as part of what would required to justify
this view. Third, Lotze insists, against empiricistic sensationalism, upon
the distinction between the objects of our knowledge and our recogni-
tion of such objects, in much the same terms that Frege did. Fourth,
Lotze emphasized and developed the Kantian strategy of explaining
concepts as functions (though of course he does not have the notion of
functions as unsaturated which Frege derived from his own substitu-
tional method of assigning contents to sub-sentential expressions). Fifth,
Lotze attacks the empiricists with a distinction between the causal
conditions of the acquisition of concepts and the capacity to use such
concepts in correct reasoning which mastery of the concepts consists in
(see note 34 above). Next, Lotze offers a theory of identity statements
according to which the two terms share a content, but differ in form.
This is the Begriffsschrift view, and the language survives into the
opening paragraphs of 'Uber Sinn und Bedeutung.' Seventh, Lotze
endorses the Kantian principle of the priority in the order of explana-
tion of judgements to concepts which Frege endorses in the Grundlagen.
Lotze does not succeed in being entirely consistent on this point, since
he also is committed to atomistic principles which are not obviously
compatible with the view on the priority of judgements. Although Sluga
does not say so, those who take Frege not to have discarded the context
principle in the post-1890 writings must find a similar tension in some
of the procedures of the Grundgesetze. Finally, Lotze is committed to
the objectivity of sentential contents, and treats them as neither mental
nor physical just as Frege does. Lotze, however, specifically denies that
this objectivity is grounded in the correlation of sentences with objects
such as Frege's thoughts appear to be, taking a more Kantian position.
Sluga, as we shall see below, argues that despite apparent statements to
the contrary we should understand this to be Frege's view as well.
This is a suggestive set of similarities to find in a prominent near-
contemporary logician with whose work Frege was familiar. Recogniz-
ing them as important need not commit one to minimizing the signifi-
cant, perhaps dominant, differences in outlook which remain between
Lotze's revived Kantianism and Frege's philosophical e1aboration of his
semantic methodology (although Sluga does on occasion succumb to the
temptation to treat Frege's agreement with Lotze on one point as
evidence that he probably agreed with him on others). Only according
to the crudest notion of what philosophical originality consists in is
there any incompatibility between finding enlightenment in the demon-
stration that these general principles were in the air and so came
complete with a history and a tradition, on the one hand, and the
appreciation of the genius shown in the use such adopted and adapted
raw materials were put to in service of quite a different explanatory
project on the other.
Sluga's most important and sustained argument, however, concerns
the influence of Kant on Frege. He claims that Frege should not be
thought of as a dogmatic realist about physical objects nor as a Platonist
about abstract objects, as he almost universally has been thought of. He
should be seen rather as a Kantian whose realistic remarks are to be
interpreted as expressing that merely empirical realism which is one
feature of transcendental idealism. This is certainly a radical reinterpre-
tation. What evidence can be adduced for it? Sluga's considerations may
be assembled as five distinct arguments.
First it is pointed 'out that Frege joined a philosophical society whose
manifesto is explicitly idealist and Kantian, and that he published in
their journal. By itself, this shows little, for Frege had so much trouble
getting his work into print and finding others willing to discuss it that we
cannot be sure how much he would have put up with to secure such
opportunities. The rationale Sluga suggests
is that ''what tied him to
the idealists was primarily his opposition to the various forms of
naturalism". Specifically, Frege and the idealists (a) were anti-psycho-
logistic, (b) endorsed an objectivist epistemology (taking the contents of
judgements to be independent of their entertainment by thinkers), and
(c) endorsed a rationalistic a priorism about mathematics. These points
are well taken, but the views involved are all consistent with Platonism
and realism generally as well as with transcendental idealism. Indeed
Sluga admits that "one can read much of Frege and not raise the
question of transcendentalism". So we must look elsewhere for a
warrant for such an attribution.
The second argument concerns Frege's attitude towards the truths of
geometry.40 It is remarked to begin with that in his Habilitationsschrift
Frege held a Kantian view on this topic, saying that geometry rests "on
axioms that derive their validity from the nature of our capacity for
intuition (Anschauungsvermogen)". Furthermore, throughout his career
Frege describes geometrical knowledge as synthetic a priori, and on this
basis rejects non-Euclidean geometry as false. From this fact Sluga
concludes: "Frege held a Kantian view of space and hence a transcen-
dentally sUbjective view of the objects that occupy it." The only
elucidation offered of this crucial 'hence' is the later statement that
"Frege's view must be close to Kant's: Empirical objects are in space
and time, but space and time are a priori forms of sensibility. That
seems to imply that for Frege empirical objects can only be empirically
real, but must be transcendentally ideal." That Kant believed the two
views to be linked in this way falls far short of showing that Frege did
so. Certainly such an argument cannot be taken to undermine an inter-
pretation which takes Frege's realistic remarks about physical objects at
face value, and admits that his views are inconsistent to the extent that
be never confronted these latter with his views about geometry with an
eye to reconciling them. On the other hand some interpretive cost is
clearly associated with attributing such an inconsistency to Frege.
The next two arguments must be judged less satisfactory.41 First,
Sluga argues that in the context of Kantian transcendentalism (as just
discussed) Platonic realism looks like dogmatic metaphysics. So Frege
should have been expected to argue that views (a) through (c) above, on
which he argees with the idealists, cannot in fact be warranted tran-
scendentally. But Frege nowhere argues this. The trouble with this
argument is that there is no evidence that Frege did not, as most of his
contemporaries did, read Kant's transcendentalism as a form of psy-
chologism. If he had done so, he would have dismissed it and so not felt
the force of the demand in question. Sluga next argues that every claim
of Frege's that can be taken as evidence of Frege's realism can be
matched by a passage in Lotze, who had a Kantian idealistic theory of
validity. This argument seems to do no more than restate the point that
(a) through (c) are consistent with either position. For it is a criterion of
adequacy of anyone's transcendentally idealistic position that it have
room for all of the claims the realist wants to make, suitably reinter-
preted. Further, Frege does insist that thoughts are independent, not just
of this thinker or that, but of the very existence or even possibility of
thinkers at all. This seems to contradict Lotze's account of objectivity as
rule-governed intersubjectivity.
Sluga'S final argument is weightier and involves more interpretive
work, both in construction and evaluation. The basic claim is that "there
are strewn through Frege's writings statements that appear irreconcil-
able with Platonic realism. In particular the central role of the Fregean
belief in the primacy of judgements over concepts would seem to be
explicable only in the context of a Kantian point of view."42 Arguing in
this way obviously commits Sluga to showing that Frege does not
discard the context principle when he arrives at the distinction between
sense and reference. We will see below that he contributes significant
new considerations to that debate in furtherance of this aim. But the
incompatibility of realism with the recognition of the primacy of judge-
ments must also be shown. The latter view is 'Kantian', but it does not
obviously entail transcendental idealism, which is the view in question.
Sluga takes the principle of the primacy of judgements to serve the
purpose for Kant
of refuting any atomistic attempt to construct
concepts and judgements out of simple components, and in particular to
resist the empiricist sensationalist atomism of Hume. Such a view is
indeed incompatible with the reism of Kotarbinski (to which Tarski's
recursive semantics owes so much), which sees the world as an arrange-
ment of objects out of which concepts and judgements must be con-
structed set-theoretically.44 But the Kantian principle need not be taken
to be incompatible with Platonic realism about abstract entities such as
thoughts which are the contents of judgements. Given that the context
principle does not show that Frege was a transcendental idealist about
thoughts, it seems also open to him to hold some form of realism about
other objects, provided thoughts retain an appropriate priinacy (as,
given the very special status of truth in the late works, even those who
see the context principle as discarded are committed to granting) even if
he has not discarded that principle. So if the case for the persistence of
the context principle can be made out, it should be taken as showing
Frege was a Kantian in the sense of holding the context principle,
not in the sense of being a transcendental idealist.
Still, this point is worth establishing for its own sake. Sluga correctly
sees the Begriffsschrift as the confluence of three lines of thought: (1)
that judgements, as involved in inference, are the original bearers of
semantic significance, so that it is only by analyzing such judgements
according to the procedure of "nothing invariance under substitution"
that such significance can be attributed to sub-sentential expressions
('the primacy of judgements'), (2) the Leibnizian idea of a perfect
language, and (3) the idea of reducing mathematics of logic. Assuming
the context principle was thus "anchored deeply in Frege's thought, it is
implausible to conclude with Dummett that in his later years Frege
simply let it slip from his mind."45 Sluga advances five arguments for the
persistence of the principle, and along the way addresses two commit-
ments of Frege which have been taken to be incompatible with such
First, Sluga offers an important consideration which has not pre-
viously been put forward in the extensive literature discussing this
question. The first of the 1891-2 essays which Frege wrote is a seldom
read review of L. Lange's The Historical Development of the Concept of
Motion and its Foreseeable End Result called 'The Principle of Inertia'.
In it Frege argues at some length that the concepts of a theory are not
given prior to and independent of that theory. Rather those concepts
can be arrived at only by analyzing the contents which the judgements
constituing the theory are given by the inferences concerning them
which that theory endorses. This is a significant new piece of evidence
supporting Sluga'S view. The only question which might be raised about
it is that since this semi-popular piece does not deploy the full-blown
apparatus of sense and reference it may be wondered whether the views
there expressed were confronted by Frege with that apparatus, or
whether the essay might not be seen as merely the latest of his early
works. But to take such a line would be to concede a lot, and future
claims that the context principle was discarded will have to confront this
argument of Sluga's in detail.
Next Sluga offers a novel reading of the essay on the distinction
between sense and reference which denies that, as has often been
claimed, that distinction as there presented applies primarily to singular
terms and their relations to the objects which are their referents, and
hence commits Frege to an assimilation of sentences to terms which is
incompatible with the context principle. The strategy here is, in effect, to
deny that 'Bedeutung' as Frege uses it ever has the relational sense
which indicates correlation with an object. Relying on the Tugendhat
essay mentioned above in connection with Bell, Sluga understands
'Bedeutung' as a nonrelational semantic potential defined paradigmati-
cally for sentences, in virtue of their role in inference. The introduction
of this notion in the context of the consideration of identities involving
singular terms is seen as a rhetorical device of presentational signifi-
cance only. In the final theory subsentential expressions are taken to
inherit indirect inferential significances in virtue of their substitutional
behavior in sentences, which alone are directly inferentially and hence
semantically significant. Thus 'Bedeutung' is paradigmatically a senten-
tial notion.
To this analysis is conjoined an account of 'Sinn' as a cognitive
notion, as what matters for knowledge. But again, the units of knowledge
are judgements, and sub sentential expressions can become relevant only
insofar as they can be put together to form sentences which can express
judgements. So sense also should be seen as primarily a sentential
notion, which applies to sub sentential expressions only in a derivative
way. This line of thOUght concerning senses is then combined with that
concerning reference in a subtle and sensitive account of the puzzling
relations between the Lotzean rendering of identity locutions offered in
the Begriffsschrift and its successor in 'Uber Sinn und Bedeutung'.
The previous discussion of Bell's interpretation suggests that these
readings leave something to be desired. Sluga does not acknowledge the
existence of any passage or considerations indicating that Frege does
have a relational notion of reference in play. Yet such passages and
considerations do exist, and merely elaborating the nonrelational ver-
sion of Frege's concept, as Sluga does, does not obviate the necessity of
investigating the relations between the two notions and the possibilities
for reconciling them. Similarly, Sluga pushes his discussion of the notion
of sense no farther than the discrimination of the cognitive role played
by that concept. He has nothing to say about the semantic notion of
sense, or accordingly about how senses are to be understood as deter-
mining references, even nonrelational references. On these points
Sluga's analytic net does not have as fine a mesh as Bell's. As a result,
his ingenious interpretation of sense and reference will require further
filling-in before its eventual promise can be assessed.
The overall interpretation which results from all of these arguments,
however, is challenging and powerful. The primary objections . to the
persistence of the context principle are that Frege nowhere explicitly
endorses that principle after the 1884 Grundlagen formulation, and that
the principle is incompatible with two central doctrines of the 1891-92
essays, namely the semantic assimilation of sentences to terms and the
account of concepts as functions from objects to truth-values. Sluga
claims that his readings of the "Inertia" essay and of USB meet these
objections. He does not say in detail how the doctrine about functions is
to be reconciled with the context principle, but does argue that the
"Inertia" essay justifies us in attributing Bedeutung to any expression
which makes an appropriate contribution to the possession of truth
values by sentences containing it. Thus function-expressions may be
assigned (nonrelational) reference on this account. Using intersubsti-
tution equivalence classes to move from Tugendhat's nonrelational
sentential semantic significances to those of sub-sentential expressions
does indeed justify such an attribution. But in the "Inertia" essay, Frege
seems to be using "concept" in the ordinary sense rather than his
technical one, that is, to refer to the senses of predicate expressions
rather than their references. This being the case, it is not clear how the
envisaged reconciliation of the context principle with the view of
concepts as functions from objects to truth values is to be achieved.
Besides the evidence of the essay on inertia, Sluga offers two further
reasons to deny that the later Frege is silent on the topic of the context
principle. First, he mentions in several places the posthumously
published 'Notes for Ludwig Darmstaedter' (of 1919) as showing that
Frege continued to endorse the principle. He does not say what pas-
sages he has in mind, but he presumably intends the following:
is distinctive about my conception of logic is that I begin by giving pride
of place to the content of the word 'true', and then immediately go on to
introduce a thought as that to which the question 'Is it true?' is in
principle applicable. So I do not begin with concepts and put them
together to form a thought or judgement; I come by the parts of a
thought by analyzing the thought." Such a passage does show that
sentences play a special explanatory role for the late Frege, but that
much is not in question. At most such claims would show that a version
of the context principle held for senses, confirming Sluga's claim that the
cognitive origins of the concept of sense require that priority be given to
sentences. No version of the context principle for referential signifi-
cances follows from these claims. Unfortunately, Sluga never says what
exactly he takes the context principle to be, whether a doctrine about
senses, references, or both. Frege's original formulation, of course,
preceded his making this distinction. So perhaps the best conclusion is
that Sluga takes the principle to persist as applying to senses, that is that
it is only in the context of a thought that a term or other sub-sentential
expression expresses a sense. This seems to be something Frege indeed
did not surrender. Such a reading has the additional advantage that the
doctrine that concepts are functions from the references of singular
terms to truth-values is not incompatible with it.
The final argument fares less well. It is claimed that Frege's late
treatment of real numbers shows that his practice is still in accord with
the context principle.
Here the point seems to be that the real numbers
are given contextual definitions. Such an argument would be relevant to
a context principle applying to reference rather than senses, since Frege
does not pretend to specify the senses of numerical expressions in his
formal definitions. But the definition of real numbers he offers is of just
the same form as the Grundgesetze definition of natural numbers. If this
style of definition does exhibit commitment to a form of the context
principle, that case should be argued for the more central and important
case of natural numbers. It is not clear how such an argument would go.
One of the themes around which Sluga usefully arranges his presenta-
tion of Frege's development is that of the pursuit of the definition of
purely logical objects. The reason offered for the somewhat misleading
order, of presentation pursued in DSB, which seems to give pride of
place to singular terms rather than sentences, is that the road from the
Grundlagen account of numbers to that of the Grundgesetze needed to
pass through a more thorough understanding of identity claims. Sluga is
quite clear that for Frege, beginning with the Grundlagen, the only
concept we have of an object is as that which determines the semantic
significance of a singular term. For an expression to play the semantic
role of a singular term is for it to make a certain contribution to the
inferential potential of sentences containing it, a contribution which is
constituted by the appropriate (truth preserving) substitutions which can
be made for that expression. The substitution inference potential of a
singular term is in tum codified in the endorsed identity claims involving
that term. That what we mean by 'object' is according to Frege
exhausted by our conception of that the recognition of which is
expressed in identity claims in virtue of their licensing of intersubstitu-
tion is one genuinely transcendental element in his thought about which
Dummett, Sluga, and Bell agree.
In the Grundlagen Frege argued that according to this criterion,
number-words are singular terms, so that if statements about them are
ever objectively true or false they must be so in virtue of properties of
the objects which are identified and individuated in assertions of
numerical identities. The logicist thesis that the truths of mathematics
are derivable from the truths of logic by logical means alone accordingly
entails that numbers are purely logical objects, in the sense that the
identities which express the recognition and individuation of these
objects are themselves logical truths. Sluga's ingenious suggestion is that
Frege's concern in DSB with the nature of synthetic or potentially
knowledge-extending identities specifying ordinary objects should be
understood as a stage in the working out of his mature account of
analytic (logically true) identities required for the adequate specification
of the logical objects treated in the Grundgesetze. The specific interpre-
tive use to which Sluga puts this general insight is hard to warrant,
For he claims that the difference between these two sorts of identities
resides in the fact that the identities by which logical objects are
identified and individuated express coincidence not just of reference,
but also of sense.
It is not clear what reasons there are to accept this
reading, nor what interpretive advantages would accrue from doing so.
For Frege explicitly affirms on a number of occasions that the two
expressions '2
' and '2 + 2' express different senses. And he seems
committed to this view by structural principles of his approach, in
particular by the compositionality principle as it applies to senses.
Different function-expressions appear in these two complex designa-
tions, and the senses of components are parts of the senses of com-
plexes containing them. Nor does the fact that such identities are to be
logically true entail that they express identities of sense rather than
merely of reference. Identity of sense would of course be sufficient for
identity of reference. But we are often told that logic need be concerned
only with truth and reference, and Frege's view seems to be that it can
be logically true that two different senses determine the same reference.
This mistake aside, Sluga's tracing of the development of Frege's
attempts to define abstract objects of the sort instantiated by logical
objects is a valuable contribution, and raises issues of the first impor-
tance for our understanding of the constraints on interpretations of
Frege's technical concepts. The story begins with the second definition
of number which Frege tries out in the Grundlagen. It states that two
concepts have the same number associated with them iff the objects
those concepts are true of can be correlated one-to-one.
He rejects
such a definition as inadequate to specify numbers as objects, on the
grounds that it will not determine whether, for example, Julius Caesar or
England are identical to any number. Such a definition settles the truth-
values of identities (and hence the appropriateness of substitutions) only
for terms which are the values for some argument expression of the
function-expression "the number of the concept ... ". This procedure
would be legitimate only if we had independently defined the concept
(function from terms to truth values) number signified by this function-
expression. But it is not possible simultaneously to specify that function
and the objects for which it yields the value True. If objects had been
specified by this definition, then there would be a fact of the matter as to
whether Julius Caesar was one of them. But the definition does not
settle this issue either way. On the basis of this objection, Frege moti-
vates his third and final definition of numbers, considered below.
Sluga traces through the later works Frege's efforts to clarify the
specification of numbers in such a way that it will not be subject to this
objection, culminating in the Grundgesetze account of courses of values.
Given the centrality to Frege's project of producing an adequate defini-
tion of number this progress is of interest for its own sake. But the task
of responding to the objection to the second GL definition of number is
made especially urgent for interpreters of Frege by a consideration
which Sluga does not mention. For the specifications of the abstract
objects in terms of which Frege's semantic analysis proceeds (e.g. sense,
reference, thought, truth-value) are of the same objectionable form as
the second GL definition of number. Nothing we are ever told about the
senses of singular terms or sentences, for instance, settles the question
of whether Julius Caesar can be such a sense. Though this may seem like
a question of no interest, some interesting questions do take this form.
For in interpreting the notion of sense one is concerned both With
subdividing the explanatory functional role played by the concept (as
exhibited in the discussion of Bell) and with the possibility of identifying
senses with things otherwise described - for example, the uses of
expressions, sets of possible worlds, mental representations. Frege him
self addresses such issues when he denies that the senses of sentences
are to be identified with ideas in people's minds. How is the identity he
wishes to deny given a sense?
All that is given is a criterion determining when the senses associated
with two expressions are the same (namely if they are intersubstitutable
without change of cognitive value - Erkenntniswerte). If something is
not specified as the sense associated with an expression (compare:
number associated with a concept) its identity or nonidentity with
anything which has been so given is entirely undetermined. Frege's
procedure for introducing his technical concepts such as sense is
invariably to attempt to specify simultaneously a realm of abstract
semantic interpretants and a function which assigns a member of this
realm to each expression.
We are, for instance, to associate truth-values with sentences. But we
are told only that the truth-value associated with p is the same as the
truth-value associated with q just in case for no occurrence of p (either
as a free-standing sentence or as a component in a more complex
sentence) can a good inference be turned into a bad one by substituting
q for that occurrence of p (reading the principle that good inferences
never take true premises into conclusions that are not true as defining
truth-values in terms of the goodness of inferences). Even conjoining
such a specification with the stipulation that the truth-value associated
with the sentence '2 + 2 = 4' is to be called 'the True' does not settle the
question of whether Julius Caesar is that truth-value. He had better not
be, for if the logicist program of GG is to be successful, truth-values
must be definable as purely logical objects. The current question is how
the identity which is denied here is given a sense so that something
could count as justifying that denial. The functions which associate the
various kinds of semantic significances with expressions are always of
the form: f(x) = fey) iff R(x, y), where x and y range over expressions,
and R is some relation defined in terms of the inferential potentials of
those expressions. These are exactly the kind of definition Frege found
wanting in GL.
Seeing how Frege believes he can overcome the objectionable in-
determinateness of concepts such as that determined by the second GL
definition of number is thus a matter of considerable importance for the
appraisal of his success in specifying his own technical concepts, as well
as for the narrower project of introducing numbers as logical objects.
The third and final definition of number which Frege offers in GL is:
"the Number which belongs to the concept F is the extension of the
concept 'equal (Gleichzahlig) to the concept F' "50 The number three is
thus identified with the extension of the concept, for example, "can be
correlated one-to-one with the dimensions of Newtonian space". This
definition does not have the form Frege had objected to. However, it
essentially involves a new concept, 'extension', which has not previously
appeared in GL, nor indeed anywhere else in Frege's writings. In a
footnote to the definition, Frege says simply "I assume that it is known
what the extension of a concept is." Sluga points out that this defini-
tionally unsatisfactory situation is not remedied in the remainder of the
book. The result is scarcely up to the standards of definition to which
Frege held others and himself. The project of GL could not be counted
a success until and unless it could be supplemented with an account of
the extensions of concepts.
Six years later, in 'Funktion und Begriff', Frege offers such an
account. The general notion of a function is explicated, and concepts are
defined as functions from objects to truth-values. The extension of a
concept is defined as the 'course of values' (Wertheverlauf) of that
function. This is the first appearance of the concept of a course of
values. Since extensions are reduced to them, the residual definitional
burden bequeathed by GL is put off onto this new concept. What Frege
tells us here is just that the course of values associated with function F is
the same as the course of values associated with function G just in case
for every argument the value assigned to that argument by F is the same
as the value assigned to it by G. The trouble with such a stipulation, as
Sluga says, is that it has exactly the objectionably indeterminate form of
the second GL definition of number which it is invoked to correct.
Frege wants to associate with each function a new kind of object, a
course of values. This domain of objects and the function which assigns
one to each function are introduced simultaneously. The result is that it
has not been determined whether Julius Caesar is the course of values
of any function. A given course of values has only been individuated
with respect to other objects specified as the courses of values asso-
ciated with various functions. In sum, the courses of values in terms of
which the extensions of concepts are defined suffer from exactly the
defect of definition which extensions of concepts were introduced to
rectify or avoid.
In the Grundgesetze when courses of values are introduced this
difficulty is explicitly acknowledged and described in the same terms
used to raise the original objection in GL (though without reference to
the earlier work). Frege introduces the same principle for determining
when the courses of values of two functions are identical, and then
points out that such a principle cannot be taken to determine any
objects until criteria of identity and individuation have been supplied
with respect to objects which are not given as courses of values. He
proposes to supplement his definition so as to satisfy this demand. His
proposal is that for each object not given as a course of values it be
stipulated to be identical to an arbitrary course of values, subject only to
the condition that distinct objects be identified with distinct courses of
Frege expresses the function which assigns to each function an object
which is its course of values by means of an abstraction operator
binding a Greek variable. The course of value of a function F is written
as ' 0 (F 0). Axiom V of the Grundgesetze tells us that:
(a) 'O(FO)='a(Ga)iff('Vx) [Fx(=) Gxl.
Frege recognizes that this principle alone does not suffice to determine
the identity of objects which are courses of values. To show this he
points out that if X is a function which yields distinct values if and only
if it is applied to distinct arguments (what we may call an "individuation
preserving" function), then:
(a') X('O(FO=X('a(Ga iff ('Vx) [Fx(=) Gxl
without it's having been settled for instance whether
(a") X('O(FO = 'a(Ga)
for any F and G (including the case in which F = G). The by now
familiar point is that (a) only determines the truth-values of homo
geneous identities, those both terms of which are of the form 'O(FO).
And (a') only determines the truth-values of identities which are
homogeneous in that both terms have the form X('O(Fo. But (a") asks
about heterogeneous identities, whose terms are of different forms.
Another identity which is heterogenous and whose truth-value is ac-
cordingly not settled by principle (a) is Julius Caesar = 'O(Fo).
To fix up this indeterminateness, which would result from taking
Axiom Valone as the definition of courses of values, Frege proposes to
supplement it by stipulating the truth-values of the heterogenous iden-
tities. Actually, he is required to specify the inferential behavior of
course of value expressions in all contexts in which they can appear. In
Frege's terminology such contexts are functions, so this requirement is
equivalent to the demand that it be determined for every single-argu-
ment function-expression what value is designated by the substitution of
any course of values expression in its argument place. Doing so will then
determine all of the properties of the objects designated by expression
of the form for those properties just are concepts, that is,
functions whose values are truth-values. Among those properties are
individuative properties, the facts corresponding to identity contexts
involving course of values expressions. Thus the Grundlagen require-
ment that to introduce a new set of objects one must settle all identities
involving them is in the Grundgesetze motivated by the omnicontextual
condition. (It is worth noticing, as Sluga points out, that there is an
endorsement of a strong context principle in Frege's claim that what it is
to have introduced expressions of the form as the names of
definite objects is for the truth-values of all sentential contexts in which
those expressions can be substituted to have been settled.) In fact, in the
spare environment of GG it turns out that it is not only necessary to
settle the truth-values of all identities involving course of value expres-
sions in order to satisfy the omnicontextual requirement, but sufficient
as well.
Indeed, in the system of the Grundgesetze at the time courses of
values are introduced the only objects already defined are the two truth-
values, and so the only heterogenous identities Frege explicitly
addresses are those involving a course of values and a truth-value. But
he must justify the general procedure of stipulating truth-values for
heterogenous identities, and not just his application of it. For if he does
not, then the GG definition of number will still be open to the objection
to the second GL account of number (that it has not been settled
whether Julius Casesar is one) which drove him to define the extensions
of concepts and hence courses of values to begin with. Indeed, the
concept "logical object" will not have been defined if it has not been
settled whether Julius Caesar is one. Further, as we have seen, Frege's
own definitions of his technical terms in general suffice only to deter-
mine the truth-values of homogeneous identities, for example, identities
of two truth-values, or two senses, or two references, but not the
heterogeneous identities which would be required to make the claim
that Julius Caesar = the Bedeutung of the expression 'Julius Caesar', or
that a certain linguistic role is the sense of some expression.
In particular, Frege's substitutional-inferential methodology deter-
mines only the nonrelational sense of 'Bedeutung', according to which
expressions are sorted into substitutional equivalence classes as having
the same Bedeutung. For Frege to add to this determination of homo-
geneous identities (both of whose terms are of the form "the Bedeutung
of the expression t") the relational sense of reference in which these
Bedeutungen are identified with objects suitably related to all and only
the members of the nonrelational substitutional equivalence class of
expressions is precisely to stipulate the truth-values of the heterogene
ous identities. The question of whether such a procedure can be justified
on Frege's own terms is thus exactly the question of whether the two
notions of Bedeutung can be made into "two aspects of one notion" as
Dummett claims and Frege is committed to, or whether they are simply
conflated without warrant, as Bell claims. Following Sluga's develop-
ment of Frege's attempted definition of terms which refer to logical
objects thus leads to the argument which must justify the identification
of the things playing the two explanatory roles which Bell has shown
must be distinguished under the heading "Bedeutung".
In Section 10 of GG Frege offers his justification of the procedure of
stipulating the heterogeneous identities, in an argument which Currie
has called ''brilliantly imaginative".SJ The argument is a difficult one, and
we shall have to examine it with some care. What is to be shown is that
it is legitimate to stipulate (a) above, determining the homogeneous
identities involving courses of values, together with the following stipu-
lation for heterogenous identities:
(b) 'T(LT) = tl and 'a(Ma) = 0.,
where t} 0. and (3x) (Lx Mx). L and M are to be arbitrary
functions, and t} and 0. are terms which are not of the form 'a(Pa). For
the purposes of the GG argument, the terms in question are "the True"
and "the False". In the context of Sluga's point that Frege's defense of
his own view against his objection to the second attempted definition of
number in GL must be traced through the account of extension to the
account of courses of values, it will be worth keeping in mind that for
this purpose the argument must apply equally to the case in which t} is
"Julius Caesar" and 0. is "England". To emphasize this requirement, the
exposition of Frege's argument which follows will use those values for t}
and 0. rather than the truth-values which Frege employed. In any case
the poinI is that distinct objects which are not given as courses of values
are stipulated to be identical to the courses of values a like number of
arbitrary distinct functions. The task is to show that such a stipulation is
The strategy of the argument is to construct a domain of objects of
which (a) and (b) can be proven to hold. To start, suppose it has been
stipulated that:
(c) -lJ(FlJ)= -y(Gy)iff('1x) [Fx(=) Gxl,
that is, we stipulate the homogeneous identities for terms of the form
- lJ(FlJ), where the function which associates objects so denominated
with functions F is unknown except that principle (c) holds. As was
pointed out above by means of (a') and (a"), the fact that both (a) and
(c) hold does not in any way settle the heterogenous identities one of
whose terms is a course of values and the other of which is of the form
-lJ(FlJ). The next step is to use the arbitrary distinct functions Land M
of (b) to construct an individuation preserving function X as above. The
function Xis defined by five clauses:
(1) X(Julius Caesar) = -lJ(LlJ)
(2) X( - lJ(LlJ = Julius Caesar
(3) X(England) = - y(My)
(4) X( - y(My) = England
(5) For all other y, X(y) = y.
The function X is constant except when it is applied to either the two
objects which are not specified as the result of applying - -abstraction
to some function (Julius Caesar and England, or the True and the False)
or to the result of applying - -abstraction to the arbitrarily chosen
functions L and M. In these special cases, the function X simply
permutes the distinguished values.
X is constructed to be individuation preserving, so that a correlation
is preserved between distinctness of its arguments and distinctness of its
values. It follows then that:
(d) X( - lJ(FlJ = X( - y( Gy iff (V x) [Fx (=) Gxl.
In these terms we could now define the course of values notation (which
has not previously appeared in this argument) by agreeing to let:
(e) 'a(Fa) =df X( -lJ(FlJ for all functions F.
Given the definition (e) and the truth of (d), principle (a) for courses of
values follows immediately. The truth of (d), as we have seen, follows
from (c), together with Clauses (1)-(5) defining the function X But
Clauses (2) and (4) of that definition, together with (e), entail principle
(b) concerning courses of values (with the substitution of Julius Caesar
for t1 and England for Thus given only the homogeneous identities in
(c) we have constructed courses of values in such a way that their
homogeneous identities in (a) can be shown to hold and in such a way
that heterogeneous identities can be proven for two of them, since
'a(La) = Julius Caesar (= X( - 1J(L1J) and 'DeMo) = England
(= X( - y(My). The legitimacy of stipulating heterogeneous identities
in the context of a principle determining homogeneous ones has been
shown by reducing the questionable stipulation to the composition of
two obviously acceptable forms of stipulation: the specification of the
values which the function X is to take for various arguments (in
particular in Clauses (2) and (4, and the introduction of the expression
"'a(Fa)" (previously without a use) as a notational abbreviation of
"X( -1J(F1J".
This imaginative argument is Frege's ultimate response defending his
account of number and of logical objects generally against the objec-
tions he had raised but not answered in the Grundlagen. Seen in that
context, the argument is fallacious. The problem concerns the extremal
Clause (5) of the definition of the individuation preserving function X If
that clause is expanded to make explicit what is contained in the condi-
tion ''for all other y" it becomes:
(5') (lty) [(y Julius Caesar & y - 'fj(L'fj) & Y England
& y - y (My = > X(y) = yJ.
It may then be asked whether it is appropriate at this point in the
argument to make use of a condition such as y - y(My). If the term
substituted for 'y' is also represented as the product of applying
- -abstraction to some function, then clause (c) will settle the truth-
value of the resulting identity. For it settles just such homogeneous
identities. But what of the case in which the identity is heterogeneous?
All that has been fixed concerning - -abstraction is principle (c), which
says nothing about such identities. Indeed, the whole strategy of the
argument depends upon starting from a specification of purely homo-
geneous identities with one sorf of abstractor (-) and using the
function X to construct an abstractor (') for which the heterogenous
identities are specified. Nothing which has been said, or, given the
strategy just indicated, could be said, settles a truth-value for hetero-
geneous identities such as
(t) Julius Caesar = - y(My)
(g) England = - 'fj(L'fj).
For all that principle (c) concerning - -abstraction and the distinctness
of the functions L and M settle, (f) could be true and (g) false. Given the
truth of (f), substituting in Clause (4) would yield that x(Julius Caesar)
= England, and so by clause (1) that England = - rJ(L-fJ), that is, that
(g) is true. So the definition of X presupposes valuations for hetero-
geneous identities which it is in no way entitled to.
Matters are just as bad if we consider some other object, say the
direction of the Earth's axis (also discussed in GL). It has nowhere been
determined whether it is identical to - rJ(LrJ) and so falls under Clause
(2), or identical to - y(My) and so falls under Clause (4), or to neither
and hence falls under Clause (5). The definition of X, in terms of which
it is to be shown acceptable to stipUlate heterogeneous identities for
- -abstraction, is well-formed only if the heterogeneous identities
involving - -abstraction have already been settled. They have not been
settled. Further, to add to the argument the assumption that truth-values
for such heterogeneous identities involving expressions of the form
- o(Fo) have been settled is to assume exactly what the argument as a
whole is supposed to show, namely that such matters are open for
stipulation in the first place (so long as suitable care is taken to match
distinct objects with the result of abstracting distinct functions). If more
is supposed about - -abstraction than principle (c) fixing homogeneous
identities, the question will be begged. And without some supposition
about heterogeneous identities the argument does not go through.
The intent of the offending extremal clause is to deal with all objects
which can be represented by expressions of the form - o(Fo), where F
L and F M. Distinct objects not so representable are each to be
dealt with by a pair of clauses, letting the function X permute them with
the result of abstracting from corresponding arbitrarily chosen distinct
functions. There is nothing in general wrong with such a definitional
strategy. It may not be used in the context of this argument, however.
The distinction between the cases which are to be dealt with by paired
specific stipulations and those which remain to be dealt with by the
extremal stipulation cannot be made precise without begging the ques-
tion. For that distinction corresponds to the distinction between hetero-
geneous identities and homogeneous ones, in the sense of stipulations for
objects not representable by expressions of the form - o(Fo) and those
which are so representable. This distinction is not one which a principle
like (c) specifying the purely homogeneous identities permits us to
make, and we are entitled to presuppose no more than such a principle.
Put otherwise, the form of definition essentially requires that there be a
pair of specific clauses dealing with every object whose individuation
with respect to the results of applying - -abstraction to functions has
not been settled by principle (c). But this class of objects cannot be
described or specified in the terms permitted for the definition if it is to
play its appointed role in the larger argument.
The only way in which this situation might be remedied would be if
there were some property available which could be independently
appealed to in order to distinguish the two kinds of cases. Thus if to (C)
were added:
(c') (\fy)[P(y)(=) (3F)(y= -o(Fo]
then the extremal clause in the definition of X could be amended to
(5") -y(My(=)X(y)=y]
In the context of (c'), (5") will have the desired effect of applying only
to objects which can be designated by expressions of the form - o(Fo),
where F L and F M. More important, (c') would ensure that the
identities in (5") are homogeneous with respect to - -abstraction, and
hence have had their truth-values settled by (c). It was the failure to
ensure the homogeneity that was responsible for the inadequacy of the
original definition of X
The trouble with this way out is that no such independently specifi-
able property is available. Already in the Grundlagen Frege had pointed
out that the account of when the numbers associated with two concepts
were identical (settling identities homogeneous with respect to the form:
the number of the concept F) could be defended against his objection if
the concept " .. is a number" were available. For then the truth-values
of the heterogeneous identities (such as those involving Julius Caesar)
could be settled by specifying that for any t, if t is not a number, then it
is not identical to the number of any concept. But the problem the
desired definition was to solve was precisely that of specifying the
concept " ... is a number", as the current task is to specify the concept
" ... is a course of values". It would be circular to use for the property P
" ... is an x such that there is an F such that x = - o(Fo)" For that
would precisely presuppose that the heterogeneous identities have some-
how already been settled, rather than independently settling them. Nor
could some property such as " ... is not in the causal order" be used, for
there are other logical objects (such as the True and the False) whose
individuation with respect to objects specified by - -abstraction has not
been determined. Nor in any case would such a property be available to
a logicist.
Frege's argument does not work, then, and it cannot be made to
work. If the Grundgesetze is meant to offer an account of number which
will meet the demands set by the Grundlagen, then it is a failure by
Frege's own standards. Further this failure is not a matter of the incon-
sistency of the later system. Although Axiom V is the CUlprit in both
cases, it is different features of that principle which are found objection-
able in the two cases. The current complaint is that settling the truth-
values of the homogeneous identities alone, as that principle does, is
definitionally too weak to meet the requirements imposed by the
discussion of GL. Those demand the justification of the stipulative
extension of the definition to heterogeneous identities. That it leads to
inconsistency, on the other hand, shows that that Axiom is inferentially
too strong. Putting aside the question of inconsistency which makes the
claim counterfactual, even if the account of courses of values in GG
were technically adequate, it would not be philosophically adequate as a
specification of its objects and concepts. For it has not settled whether
Julius Caesar is the number three, nor shown that stipulating an answer
in the case of logical objects such as the truth-values is a legitimate
procedure. Nor can this be shown with the materials at hand.
I take it that this definitional inadequacy has not been remarked upon
for two connected reasons. In the purely technical context of the
Grundgesetze the stipulation of the two heterogeneous identities con-
cerning the truth-values and arbitrary distinct courses of values is in fact
perfectly acceptable. Further, provided that it is stipulated that neither
of the truth-values is identical to the result of applying - -abstraction to
any function, Frege's argument shows that his procedure is in order. It is
only in the larger philosophical context provided by Sluga's historical
tracing of the stages in Frege's development of an answer to his own
objections to the second attempted definition of number in GL, from
the invocation of the extension of a concept in the third and final GL
definition, via the reduction of concepts to a special kind of function
and of extensions to courses of values in 'Funktion und Begriff, to the
{inal attempt to define courses of values adequately in the early sections
of the Grundgesetze that it can be seen that satisfying the purely techni-
cal constraints will not suffice to render the definition of courses of
values (and hence of logical objects generally) adequate by the philoso-
phical standards Frege has insisted upon.
But the result is significant not only for the appraisal of the success in
its own terms of Frege's account of the logical objects which were his
explicit subject matter in GG. For as we have seen, the technical
philosophical concepts Frege developed to use in that discussion, such
as reference, and sense, and truth-value, are all given the same form of
definition as courses of values are, which individuates them only homo-
geneously. Thus " ... we cannot say what the sense of an expression is.
The closest we may approach to this is to say that the sense of a given
expression E1 is the same as the sense of another expression, E
."52 It
follows that so far as interpretation (rather than further development) of
Frege's concept of sense is concerned, one can only subdivide the
explanatory roles played by his concept, but cannot identify anything as
playing those roles. Thus it is legitimate and valuable to distinguish the
cognitive role from the semantic role played by senses, or sense as
content from sense as character, or input and output senses as Bell does.
But to entertain hypotheses about whether thoughts are mental pictures
(as Frege did by denying this) or sets of possible worlds, or denizens of
some extra-causal realm is to consider claims which have been given no
sense by Frege's purely homogeneous specification of the entities in
question. Truth values are similarly immune from heterogeneous iden-
tification, from identification in any other form than as the truth-value
associated with some expression.
Probably most important is the case of singular term reference. Here
Frege tried explicitly to supplement the purely homogeneous sorting
into semantic equivalence classes of the reference associated with various
expressions (the nonrelational sense of 'Bedeutung') with the stipulation
of heterogeneous identities involving the references of expressions and
ordinary objects. In accord with his inferential/substitutional method-
ology, these stipulations are grounded in the intersubstitutability for all
terms t of the term itself and the expression 'the Bedeutung of t'. Bell
has shown how much of Frege's conceptual scheme depends upon the
assumption that such heterogeneous identities are determined (and
hence a relational sense of reference applies) for other parts of speech,
given only the determination of the homogeneous identities (settling a
non-relational sense of reference) which is all that is available for expres-
sions of these other categories. Pursuing further a line of thought Sluga
initiated has shown that this assumption is indeed unwarranted, and that
even Frege's attempted stipulation of coincidence of relational and
nonrelational senses of 'reference' in the case of singular terms has not
been justified by Frege's own standards. Thus extending Sluga's argu-
ment permits better understanding of the philosophical status of Frege's
technical concepts in general, and in particular of the two sides of the
concept of reference which Bell, following Dummett, has so usefully
University of Pittsburgh
I See [19].
2 [8].
3 Dummetl's [10].
4 [10], pp. xii-xvi.
5 The most influential proponents of the view were Grossmann in [16), and Marshall, in
in [24] and [25).
6 p. 65 in [14].
7 [3). Bell is Lecturer in Philosophy at the University of Sheffield.
8 As in the title of Dummett's book, [8).
9 [10), pp. 476-95.
10 In Section 3 of BGS, reprinted in [14], Frege says that the begriffliche Inhalt of two
judgements is the same just in case "all inferences which can be drawn from the first
judgement when combined with certain other ones can always also be drawn from the
second when combined with the same other judgements."
11 [19], pp. 9-46.
12 For instance by Resnik in [28] and [29], and Angelelli in [1].
13 [3], pp. 139-40.
14 I have my say in [4].
IS [3], p. 42.
16 [10], pp. 47S-9.
17 in [32].
18 [10], p. 479.
19 I have argued that a purely intralinguistic anaphoric account of such facts can be
offered by construing 'refers' and its cognates as complex pronoun-forming operators, in
20 Kripke in [23], and Putnam in [27].
21 Most prominently, Perry in [26], and Kaplan in [20] and [21].
22 [3], p. 112.
23 [3], p. 115.
24 [3], p. 5l.
25 [3], p. 64.
26 [3], p. 65.
27 First in [S], pp. 293-4, then at greater length as chapter 13 of [10]. Citations here
from p. 251 of the latter.
28 in [15].
29 [10],pp.251-2.
30 [19], p. 119. Bell's other citations are in [3], p. 72.
31 [3], pp. 74-8.
32 See [31]. Sluga is Professor of Philosophy at the University of California at Berkeley.
33 Dummett in [7], p. 225, quoted by Sluga in [31], p. 8.
34 In Dummett's defense it should be said that in the final chapter of [8] the main
historical significance of Frege's work is taken to be precisely his anti-empiricist and
anti-psychologist shifting of concern from the acquisition of concepts to what such
mastery consists in - from how the cognitive trick is performed [e.g., by material beings
of our sort] to what counts are performing it. The injudicious invocation of a dominant
Hegelianism as the psychologistic culprit is explicitly made subsidiary to this central
35 [31], p. 59.
36. Dummett, in [9].
37 Bierich in [2].
38 [31], pp. 53, 192.
39 [31], pp. 59-60.
40 [31], pp. 44-5, and 106.
41 [31], p. 60.
42 [31], p. 60. pp. 43)[31], p. 91.
44 cf. [31], p. 181.
45 [31], p. 95.
46 [19], p. 253.
47 [31], p. 134 and Note 21 to chapter 4.
48 e.g. in [31],p. 156.
49 Sections 62,63 of [12].
50 Section 68 of [12]. Equality of concepts in the sense invoked here has been defined
as obtaining iff the objects of which the concepts are true can be put into one-to-one
51 in [6], p. 69.
52 Bell in [3], p. 55 [emphasis in original]. He-goes on to point out an analogy with the
Fregean concept of concept reference: "concept words refer, but we cannot stipulate
what it is they refer to". But in this case the reasons are purely substitutional, since
expressions like "the concept horse" will never be intersubstitutable with predicative
function-expressions. This shows that all heterogeneous identities involving function-
expressions on one side and singular terms on the other must be false.
[I] Angelelli, I.: 1967, Studies on Gottlob Frege and Traditional Philosophy, Reidel,
[2] Bierich, M.: 1951, Freges Lehre von dem Sinn und der Bedeutung der Urteile und
Russells Kritik an dieser Lehre, Dissertation, Hamburg.
[3] Bell, D.: 1979, Frege's Theory of Judgement, Oxford University Press, Oxford.
[4] Brandom, R: 1983, 'Asserting', NoUs 1, 637-650.
[5] Brandom, R: 1984, 'Reference Explained Away', Journal of Philosophy 84,
[6] Currie, G.: 1982, Frege: An Introduction to His Philosophy, Barnes and Noble,
Totowa, New Jersey.
[7] Durnmett, M.: 'Frege, Gottlob', in [11], Vol. 4, p. 225.
[8] Dummett, M.: 1973, Frege: Philosophy of Language, Harper and Row, New York.
[9] Durnmett, M.: 1976, 'Frege as Realist', Inquiry 19, 476-485.
[10] Dummett, M.: 1981, The Interpretation of Frege's Philosophy, Harvard University
Press, Cambridge.
[11] Edwards, P. (ed.): 1967, Encyclopedia of Philosophy, The Macmillan Company &
The Free Press, New York/Collier - Macmillian Ltd., London.
[12] Frege, G.: 1967, Grundlagen der Arithmetik, transl. by J. L. Austin, Northwestern
University Press, Evanston, Illinois.
[13] Frege, G.: 1967, Grundgesetze der Arithmetik, transl. by M. Furth, University of
California Press, Berkeley, California.
[14] Geach, P. and Black, M. (eds.): 1970, Philosophical Writings of Gottlob Frege,
Blackwell, Oxford.
[15] Geach, P.: 1975, 'Review of Dummett's Frege: Philosophy of of Language', Mind
[16] Grossmann, R.: 1961, 'Frege's Ontology', Philosophical Review 70, 23-40.
[17] Gunderson, K. (ed.): 1976, Language, Mind, and Knowledge, Minnesota Studies in
the Philosophy of Science, Vol. 7, University of Minnesota Press.
[18] Harman, G. and Davidson, D. (eds.): 1972, Semantics of Natural Language, Reidel,
[19] Hermes, H., Kambartel, F., and Kaulbach, F. (eds.): 1981, Gottlob Frege: Post
humous Writings, transl, by P. Lond and R. White, University of Chicago Press,
[20] Kaplan, D.: 'The Logic of Demonstratives', in [33].
[21] Kaplan, D.: Demonstratives, 1980 John Locke Lectures, Oxford University Press,
Oxford, forthcoming.
[22] Klemke, E.: 1968, Essays on Frege, University of Illinois Press, Urbana.
[23] Kripke, S.: 1972, 'Naming and Necessity', in [18).
[24] Marshall, W.: 1953, 'Frege's Theory of Functions and Objects', Philosophical
Review 62, 347-390.
[25] Marshall, W.: 1968, 'Sense and Reference: A Reply', in [22].
[26] Perry, J.: 1977, 'Frege on Demonstratives', Philosophical Review 86, 474-497.
[27] Putnam, H.: 1972, 'The Meaning of Meaning', in [18].
[28] Resnik, M.: 1967, 'The Context Principle in Frege's Philosophy', Philosophy and
Phenomenological Research 27,356-365.
[29] Resnik, M.: 1967, 'Frege's Context Principle Revisited', included in [30] pp.
[30] Schirn, M. (ed.): 1967, Studien zu Frege, Vol. III, Frommann-Holzboog, Stuttgart
and Bad Cannstatt.
(31) Sluga, H.: 1980, Gottlob Frege, Routledge and Kegan Paul, London.
[32] Tugendhat, E.: 1970, 'The Meaning of "Bedeutung" in Frege', Analysis 30,
[33) Uehling, T., Wettstein, H., and French, P. (eds.): 1978, Contemporary Perspectives
on Philosophy of Language, Midwest Studies in Philosophy, University of
Minnesota Press, Minneapolis, Minnesota.
In the 1880s, two men who had both been trained as rpathematicians
wrote short books defending the idea that arithmetic has an intimate
relation with logic. Neither work was exactly a commercial or intellectual
success. Frege's Grundlagen (Frege 1884) went virtually unnoticed, and
Frege recorded his disappointment and frustration in the introduction
to his Grundgesetze (Frege, 1893, p. xi; Furth, 1967, p. 8). Dedekind's
monograph, Was Sind und was Sollen die Zahlen? (Dedekind, 1888),
fared only a little better. Before Dedekind published it, he had been
encouraged by the interest of other mathematicians in his project. In
1878, for example, Heinrich Weber urged him not to postpone his
planned study on the number concept (Dugac, 1976, p. 273). However,
when the book appeared, it made comparatively little stir: certainly,
Dedekind's discussion of the natural numbers aroused nothing like the
interest excited by his study of the real numbers (Dedekind, 1872).
Although many of Dedekind's contemporaries viewed him as an impor
tant mathematician, they did not rank Was Sind und was Sollen die
Zahlen ? among his major achievements.
Today, Frege's Grundlagen is widely appreciated as a philosophical
masterpiece. In retrospect, the mathematicians who ignored it appear as
men who failed to recognize a pioneering work. Yet, in the intervening
century, Frege's study of the foundations of arithmetic has been applied
in a number of different philosophical enterprises. In the early decades
of the century, for example, Frege's defense of logicism was used to
support a program with quite different motivations from his own?
Faced with the problem of accounting for the status of arithmetic, a
traditional bastion of rationalist epistemology, the logical positivists and
their empiricist successors, sought comfort in Frege's claim that arith
metic is disguised logic. More recently, philosophers of mathematics
have approached Frege's work not as an arsenal that can be raided in
the defense of empiricism but as a treasure trove of philosophical
insights about mathematics. Despite the fact that Frege's own solutions
L. Haaparanta and J. Hintikka (eds.), Frege Synthesized, 299-343.
1986 by D. Reidel Publishing Company.
to the problems he posed are not widely accepted, his conception of the
central questions of the philosophy of mathematics is frequently viewed
as definitive. If the history of philosophy in general is a series of
footnotes to Plato, then the history of the philosophy of mathematics in
the last thirty years appears as a series of footnotes to Frege.
Because our current understanding of what the philosophy of mathe-
matics ought to be is so dominated by Frege's view of the field, it is
important to discuss the way in which Frege's enterprise was born. My
aim will be to show how Frege's own philosophical preconceptions led
him to identify, and to bequeath to his successors, a misguided picture
of the central problems of the philosophy of mathematics. To achieve
this, I shall use Dedekind as a foil. Nearly a century after it was
published, Dedekind's monograph on the concept of natural number
persists in its state of benign neglect. Dedekind appears to us as a lesser
Frege, a man who groped toward some Fregean insights but who only
saw dimly what Frege saw clearly. I shall attempt to show that
Dedekind's work does not deserve to be ignored, that it responds to a
different set of philosophical problems than those identified by Frege,
and that a bedekindian view of mathematics might provoke new and
profitable philosophical inquiries.
Frege's logicism was the product of his philosophical reflections on the
state of late nineteenth century mathematics, reflections that were by no
means the innocent philosophical thoughts of a philosophically untu-
tored mathematician. In his earliest writings we can already discern a
philosophical view, inherited from Kant. Frege's inaugural dissertation,
presented in 1873, is devoted to the problem of giving a geometrical
representation of imaginary points and figures. The young mathemati-
cian who regaros this problem as an important question for his research
is already concerned with the general project of showing that new
developments within mathematics are compatible with the epistemo-
logical ideals of the discipline. The first sentence provides a clear view
of Frege's intentions:
When one considers that the whole of geometry ultimately rests upon axioms, which
receive their validity from the nature of our faculty of intuition, then the question of the
sense of imaginary figures appears to be well-justified, since we often attribute to such
figures properties which tontradict our intuition. (Frege/ Angelelli, 1967, p. 1)3
These are not the words of a philosophical naif. The linking of
geometrical axioms to our faculty of intuition reveals the influence of
Kant, and it is the background of Kantian epistemology that gives
significance to Frege's mathematical enterprise.
Yet, as every reader of Frege knows, he was to break explicitly with
Kant over the status of arithmetic. Grundlagen is devoted to defending a
non-Kantian account of arithmetic, and it contains a respectful critique
of Kant's own sketchy ideas about the subject. The divergence from
Kant's specific proposal for arithmetic was already clear by 1874. In the
opening section of the dissertation he presented on his arrival at Jena,
Frege writes.
It is thus clear that for so comprehensive and abstract a concept as that of the concept of
quantity there can be no intuition. Because of this there is a significant difference
between geometry and arithmetic in the way in which their fundamental propositions
are grounded. Because the object of arithmetic is not intuitable (keine Anschaulichkeit
hat), the fundamental propositions of the subject cannot stem from intuition. (Fregel
Angelelli, 1967, p. 50)4
Significantly, while rejecting one part of Kantian epistemology, Frege
makes explicit his commitment to other parts. Within Kant's epistemo-
logy for mathematics we can distinguish two main doctrines. Many
prominent sections of the Critique reflect a particular type of apriorism
about mathematics, the idea that mathematics is a corpus of a priori
knowledge. Kant is committed to what I shall call an apriorist program.
He believes that it is possible to give an epistemological reconstruction
of mathematics which will make it clear that the truths asserted by
mathematicians can be (and perhaps are) known a priori. This general
idea is articulated in the second main doctrine, a specific proposal about
the way in which the apriorist program will go. On Kant's view, all
mathematical truths can be known by means of inferences from funda-
mental propositions which are themselves knowable a priori through
pure intuition. Frege rejects this specific proposal, while honoring the
more general commitment. Moreover, his dissent about the details is
oply partial: throughout his career, Frege continues to espouse Kant's
explanation of the a priori knowability of geometry.
Let me make this interpretative claim more precise by defining the
notion of an apriorist program. I shall assume that a necessary condition
for a person to know that p is for that person's belief that p to have been
produced in an appropriate way. This very general assumption about
what knowledge is has not only been ably defended in the recent
epistemological literature; it is also an assumption that both Kant and
Frege would have accepted.
With respect to any item of knowledge,
there must be a tree-like structure which represents the causal process
through which the belief in question was produced. The ultimate causal
nodes in this structure are basic beliefs, beliefs which can be arrived at
without inference from other beliefs (Kornblith, 1980). To reconstruct a
body of knowledge is to reveal the causal networks underlying the
statements that are known, showing what items of knowledge are basic,
and what chains of inference lead from these pieces of basic knowledge
to the rest of the corpus. To reconstruct a body of a priori knowledge
one must demonstrate how that knowledge results from statements that
are knowable a priori without inference from other statements by means
of chains of inference along which apriority is transmitted. Someone
who believes that mathematics is a body of a priori knowledge is
committed to the possibility of a reconstruction of this kind. Such a
person must hold that there is a feasible apriorist program for mathe-
matics, a program which is to be carried out by completing the following
tasks: (a) identifying a set of basic a priori statements (statements that
can be known a priori without inference from other statements); (b)
explaining the nature of the processes through which a priori knowledge
of the basic a priori statements can be gained; (c) identifying a set of
apriority-preserving rules of inference, rules which license inferential
transitions that always yield a priori knowledge of the conclusion given a
priori knowledge of the premises; (d) showing that all accepted mathe-
matical statements can be obtained from basic a priori statements
through a sequence of inferences each of which is licensed by an
apriority-preserving rule of inference.
Kant's influential contributions to the philosophy of mathematics
consist in his general account of what a priori knowledge is and his
suggestions about how such knowledge can be obtained. As I have
indicated elsewhere, (Kitcher, 1980a), the Kantian idea of a priori
knowledge as knowledge which is independent of experience can be
developed in a coherent way. Given this development, we are able to
define the notions of a basic a priori statement and of an apriority-
preserving rule of inference, thus making clear the concept of an
apriorist program and identifying the claims to which someone wllO
believes in the apriority of mathematical knowledge is committed. But
Kant not only offered this very general characterization of a priori
knowledge. He also proposed that there are two ultimate sources of a
priori knowledge; conceptual analysis and pure intuition. Kant's own
philosophy of mathematics proceeds from the identification of what an
apriorist program is to the task of explaining the nature of the processes
- processes of pure intuition - through which he takes basic a priori
mathematical knowledge to be obtained.(Thus Kant focusses on task (b)
above.) Kant does not try to specify the statements which are basic a
priori statements or the rules of inference that preserve apriority. His
attitude seems to be that that is the business of the mathematician,
perhaps that mathematical proofs are in order as they are.
The similarities and differences between Kant and Frege can now be
described more precisely. Like Kant, Frege is committed to apriorism
and his work reflects acceptance of the Kantian idea of an apriorist
program. Moreover, Frege endorses Kant's conception of the possible
sources of a priori knowledge. The most obvious disagreement concerns
the way in which areas of mathematics are taken to be related to basic
sources of a priori knowledge. Frege contends, of course, that Kant was
wrong to hold that arithmetical knowledge is based' ultimately on pure
intuition. Instead, Frege believed that our knowledge of arithmetic is to
be traced to the source of our knowledge of logical truths. This shift in
the specifics of the apriorist program causes Frege to emphasize differ-
ent parts of the enterprise. Thinking that our source of a priori knowl-
edge of logic can do more than Kant had given it credit for, Frege sees
the importance of identifying the basic a priori statements - basic laws
of logic - on which arithmetical knowledge depends, of specifying the
apriority-preserving rules of inference, and of showing that the state-
ments and rules so identified suffice for the reconstruction of arithmetic.
Thus tasks (a), (c), and (d) receive a great deal of attention from Frege,
while task (b) is ignored. From Frege's perspective, the basic epistemo-
logical issue, the question of explaining the ultimate sources of a priori
knowledge, had already been settled. But for Frege (unlike Kant)
mathematics itself needs reform. The subject will not stand forth as an a
priori science until the ordinary proofs of mathematicians are improved.
Frege's inaugural dissertation addresses one of the obvious problems
that arise for Kant's favored apriorist program. Late nineteenth century
geometry had advanced beyond the discipline whose epistemological
foundations Kant had considered. Frege's project is to show how geo-
metrical ideas that initially appear beyond the scope of pure geometrical
intuition can be accommodated within the Kantian program. But
Frege's early reflections on arithmetic must have raised for him a more
serious difficulty. Having arrived at the judgment that it is impossible to
trace arithmetical knowledge to propositions known through pure
intuition, Frege set for himself a new apriorist program, one that would
require a revision of the usual ways of presenting proofs within arith-
metic. Arithmetic must be constructed from the fundamental laws of
logic. So Begriffsschrift was born. Having undertaken to provide arith-
metic ''with the most secure foundation," Frege formulated the problem
as follows:
.. , I first had to ascertain how far one could proceed in arithmetic by means of
inferences alone, with the sole support of those laws of thOUght that transcend all
particulars. My initial step was to attempt to reduce the concept of ordering in a
sequence to that of logical consequence, so as to proceed from there to the concept of
number. To prevent anything intuitive (Anschauliches) from penetrating here unnoticed,
I had to bend every effort to keep the chain of inferences free of gaps. In attempting to
comply with this requirement in the strictest possible way I found the inadequacy of
language to be an obstacle; no matter how unwieldy the expressions I was ready to
accept, I was less and less able, as the relations became more and more complex, to
attain the precision that my purpose required. This deficiency led me to the idea of the
present ideography. (Frege/van Heijenoort, 1967, pp. 5-6)
Formalization entered the philosophy of mathematics in the service of
Frege's epistemological ideals.
When we reconstruct Frege's route to logicism, one question stands
out. Why did he believe that the logical principles he identified were a
priori and the rules of inference he set forth apriority-preserving? I shall
have more to say about this in what follows, but, for the moment, let me
use my comparison of Frege with Kant to offer a quick explanation of
why Frege is so silent about fundamental epistemological issues. At the
one place in his published writings where Frege confronts the question
of ''why and with what right" we acknowledge a law of logic to be true,
Frege announces that, as a logician, the issue is none of his business
(Frege, 1893, p. xvii; Furth, 1967, p. 15). I suggest that Frege saw his
own work as the counterpoise to Kant's. Kant had characterized the
notion of a priori knowledge, thereby making clear the idea of an
apriorist program. Kant had also identified the sources of a priori
knowledge, one that would generate knowledge of analytic truths and
one that would yield knowledge of synthetic truths. Kant's philosophy of
mathematics was incomplete in its failure to show in detail how the
whole of mathematics could be traced to these sources. Where it was in
error was in its assignment of arithmetic to the wrong source, to pure
intuition instead of to the source of a priori knowledge of analytic
truths. Frege's enterprise is directed at remedying the incompleteness
and correcting the error, but, in this enterprise, certain basic features of
the Kantian framework are taken for granted.
Frege's work is intermediate between the projects of general epis-
temologists (like Kant) and those of practicing mathematicians (like
Frege's contemporaries who saw no point in what he was doing). There
is a pervasive philosophical myth, to which Frege himself contributed, to
the effect that his probing of the foundations of arithmetic is an out-
growth of the nineteenth century tradition of making analysis rigorous.1
A careful look at the history of research in the foundations of nineteenth
century analysis explodes this myth. The work of the great mathemati-
cians - men like Cauchy, Weierstrass, and Dedekind - who struggled
with the foundations of analysis was motivated by a quite different set of
interests from those which provoked Frege's metamorphosis from
mathematician to philosopher.
Consider the case of Cauchy. Cauchy was not moved to offer new
definitions of 'limit', 'continuity', 'convergence', and 'derivative' because
he yearned to show that analysis was genuinely a body of a priori
knowledge. His aims were far more pragmatic. Certain important
research problems of nineteenth century analysis could not be resolved
using the techniques that Cauchy and his contemporaries had inherited.
The mix of algebra and geometry constructed by Euler, d'Alembert and
Lagrange does not make it possible to determine whether the sum of an
infinite series of continuous functions is always continuous. A decision
on this issue was needed if the merits of various methods for solving
partial differential equations were to be clearly understood. Cauchy's
new analysis was born from a desire to tackle the urgent problems of
mathematical research, and Cauchy was quite happy to rely on the
older ideas of the eighteenth centJry when he thought he could get away
with it. When we read Cauchy (or Abel, or Dirichlet) through Fregean
spectacles, their texts are extremely puzzling. There are constant lapses
from the ideal of giving full algebraic derivations, periodic invocations
of geometrical evidence, and references to eighteenth century notions
(like that of infinitesimal) which we might have thought Cauchy and his
contemporaries were in the business of eliminating. Removing the
Fregean spectacles improves our vision. Cauchy et al. developed mathe-
matical concepts and methods for the purpose of resolving technical
questions, and they used those concepts and methods where they found
them useful (or even indispensable). They had no general interest in
showing the apriority of analysis, or in eschewing old-fashioned tech-
niques that might more readily be applied in some contexts.
Ironically, Cauchy'S own reform of analysis showed for the first time
how certain old-fashioned techniques, specifically the method of infini
tesimals, might lead to false conclusions. Of course, from the seven-
teenth century on, mathematicians had been aware that they did not
fully understand why infinitesimalist reasoning works. But no critic -
not even Berkeley - suggested that some conclusions generated by such
reasoning were false. Although Cauchy's work was rightly admired for
its ability to yield recognizably correct answers to certain traditional
questions, his employment of some traditionally sanctioned shortcuts
generated anomalies. The late nineteenth century rigorization of analysis
- whose full flowering is the work of Weierstrass and Dedekind -
was a response to mathematical difficulties which Cauchy'S work had
unearthed. Because some of Cauchy'S techniques gave demonstrably
false results and because Weierstrass needed reliable methods to tackle
recherche questions in elliptic function theory , Weierstrass was forced
to develop the austere style of analysis which has dominated the subject
ever since. Similarly, Dedekind's work on continuity was motivated by
the desire to establish the existence of limits in cases where Cauchy had
had to appeal to geometric analogies, analogies which were, by 1850,
recognizably faulty. The program of rigorization of nineteenth century
analysis was kept in motion by the mathematical difficulties left over by
the latest achievement.
Frege arrived on the scene at a time when the search for rigor had
temporarily come to a halt and when mathematicians were eager to put
to work the tools that they had acquired from Weierstrass. (Frege was to
appreciate that this was the attitude of his mathematical contemporaries
- see [Frege, 1893, p. xii; Furth, 1967, pp. 9-10; also Frege/Long et
al., 1979, pp. 165-66]- although he continued to attack it with wither-
ing scorn.) His campaign for attention to the foundations of arithmetic
fell on deaf ears, precisely because it was felt that the available concepts
and methods of mathematics were adequate to the important research
problems and that little would come of foundational ventures in an area
which had yielded no anomalies. If Frege's contemporaries did indeed
say of his work "Metaphysica sunt, non leguntur." then they were only
remaining true to the motivations of Cauchy and Weierstrass.
Frege's enterprise thus falls between the very general project of the
epistemologist, whose interest in mathematics is one of integrating the
subject within a global account of human knowledge, and the activity of
mathematicians, whose concern with foundational issues is fueled by
their recognition of current inability to solve interesting technical ques
tions. If successful, an enterprise like Frege's could point in either (or
both) of two directions: it could show the mathematicians that certain
types of inquiry need to be reformulated as well as reveal to philoso
phers that certain epistemological goals mayor may not be attainable.
But an enterprise that borrows so heavily from general epistemology is
vulnerable to the possibility that, if some of the epistemology that is
taken over is misguided then the project will lack the epistemological
significance that is attributed to it. As I shall suggest below, despite the
rich heritage of Frege's work in logic, his philosophy of mathematics is
flawed in precisely this way.
Like Frege, Dedekind was led to investigate the foundations of arith
metic for philosophical reasons; however, his motivation was interest
ingly different from Frege's. The reconstruction of arithmetic offered in
Was Sind und was Sollen die Zahlen? grew out of Dedekind's earlier
work on continuity and irrational numbers. As we shall see, all of
Dedekind's foundational work is pervaded by a particular conception of
mathematics, which differs from Frege's not only in its characterizations
but in the questions that it seeks to resolve.
Yet it is easy to assimilate Dedekind to the Fregean program, to see
him as a pygmy working beside a giant. When we read Dedekind after
reading Frege, we find cryptic phrases that remind us of Frege's goals.
Consider Dedekind's account of how he was led to his investigation of
continuity. He reports his early experience of trying to teach differential
In discussing the notion of the approach of a variable magnitUde to a fixed limiting
value, and especially in proving the theorem that every magnitude which grows
continually, but not beyond all limits, must certainly approach a limiting value, I had
recourse to geometric evidences. Even now such recourse to geometric intuition in a
first presentation of the calculus, I regard as extremely useful from the didactic
standpoint, and indeed indispensable, if one does not wish to lose too much time. But
that this form of introduction into the differential calculus can make no claim to being
scientific, no one will deny. (Dedekind/Beman, 1901, p. 1)
Here we might discern the Fregean project of carrying out an apriorist
program that would correct the Kantian mistake of tracing our knowl-
edge of arithmetic and analysis to pure intuition. However, I think that
Dedekind has something different in mind. Although geometrical dia-
grams may sometimes help us to see that results about the continuity of
functions or the convergence of sequences are true (indeed they may be
the best means of convincing the beginner that these results are true)
they have two deficiencies. First, the transition between the algebraic
notion of continuity and the geometrical notion of continuity is not as
simple as analysts like Cauchy had believed; the uncritical use of that
transition (and others like it) can generate mistakes. Second, even if the
transition were to produce correct results, it would be a detour. We
venture into geometry because we lack any algebraic way of formulating
that continuity of the real numbers which is so easily represented
geometrically, and because, as a result of this lack, we are unable to
establish on a purely algebraic basis theorems about the existence of
limits. 10
Why should we adopt this construal of Dedekind's intent rather than
that which assimilates him to the Fregean enterprise? My answer is
drawn from the structure of the argument advanced in Dedekind's
memoir on continuity. Dedekind sets himself the task of providing an
algebraic characterization of the continuity of the real numbers, and
showing that this characterization suffices for the derivation of standard
theorems about the existence of limits. Far from proposing that the
principle of continuity which he introduces is evident a priori, Dedekind
recommends it by explaining how it can be used to demonstrate results
about limits which had baffled his predecessors. As I interpret him,
there is a body of prior mathematical work which stands in need of
algebraic systematization - systematization which is intended not to
provide us with a priori knowledge but to improve our mathematical
understanding and our ability to solve mathematical problems - and his
proposal to define continuity should earn our acceptance because of its
ability to answer to that need. (For reasons to be given below, I suggest
that this style of argumentation is an important pattern through which
mathematical knowledge develops, and that our understanding of mathe-
matical knowledge will be advanced by taking such patterns seriously.)
Nevertheless, even though Dedekind's work on continuity is free
from commitment to the kind of apriorist views which motivate Frege, it
does contain philosophical views about mathematics, ideas which, as we
shall see, motivate his subsequent studies of arithematic. Quite early in
his memoir, Dedekind adopts what appears to be a primitive construc-
tivist approach to what mathematics is about: the human mind "creates"
the numbers, and thereby fashions "an exceedingly useful instrument"
for itself. This constructivist attitude erupts ih the substantive mathe-
matics. As Russell pointed out, Dedekind does not identify the real
numbers with sets of rationals. Rather, he "creates" real numbers to
correspond to certain partitions of the rationals (Dedekind cuts),
thereby provoking Russell's gibe that he has obtained "all the advan-
tages of theft over honest toil." However, this taunt is unfair to
Dedekind's intentions. Dedekind does not interpret mathematics as
describing some pre-existent realm of abstract objects (sets) among
which the real numbers are to be found. Rather, he suggests that the
numbers are products of acts of construction. His project is to charac-
terize the way in which the real numbers are constructed, by giving a
new d