RUDOLF CARNAP, LOGICAL EMPIRICIST

SYNTHESE LIBRAR Y

MONOGRAPHS ON EPISTEMOLOGY,

LOGIC, METHODOLOGY, PHILOSOPHY OF SCIENCE,

SOCIOLOGY OF SCIENCE AND OF KNOWLEDGE,

AND ON THE MATHEMATHICAL METHODS OF

SOCIAL AND BEHA VIORAL SCIENCES

M anaging Editor:

J AAKKO HINTIKKA, Academy of Finland and Stanford University

Editors:

ROBERT S. COHEN, Boston University

DONALD DAVIDSON, Rockefeller University and Princeton University

GABRIEL NUCHELMANS, University of Ley,den

WESLEY C. SALMON, University of Arizona

VOLUME 73
Photograph by Adya, 1962
 RUDOLF CARNAP,
LOGICAL EMPIRICIST
M aterials and Perspectives

Edited by

JAAKKO HINTIKKA

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

 Library of Congress Cataloging in Publication Data

Main entry under title:

Rudolf Carnap, logical empiricist.

(Synthese library ; v. 73)

Inc1udes bibliographies.
CONTENTS: Homage to Rudolf Camap.-Hempel, C. G. Rudolf
Camap, logical empiricist.-Wedberg, A. How Carnap
built the world in 1928.-Eberle, R. A construction of
quality c1asses improved upon the Aufbau. [etc.]
1. Carnap, Rudolf, 1891-1970--Addresses, essays,
lectures. 1. Hintikka, Kaarlo Jaakko Juhani, 1929-
B945.CI64R8 193 75-12709

ISBN 978-94-010-1809-8 ISBN 978-94-010-1807-4 (eBook)

DOI 10.1007/978-94-010-1807-4

All Rights Reserved

Copyright © 1975 by Springer Science+Business Media Oordrecht
Originally published by D. Reidel Publishing Company, Dordrecht, Holland
Softcover reprint ofthe hardcover Ist edition 1975
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,
inc1uding photocopying, recording or by any informational storage and
retrieval system, without written permission from the copyright owner
 TABLE OF CONTENTS

PREFACE IX

ACKNOWLEDGMENTS XI

HOMAGE TO RUDOLF CARNAP XIII

CARL G. HEMPEL / Rudolf Camap, Logical Empiricist

ANDERS WEDBERG / How Camap Built the World in 1928 15
ROLF A. EBERLE / A Construction of Quality Classes Improved
upon the Aufbau 55
RUDOLF CARNAP / Observation Language and Theoretical
Language 75
DAVID KAPLAN / Significance and Analyticity: A Comment on
Some Recent Proposals of Camap 87
RYSZARD W6JCICKI / The Factual Content of Empirical
Theories 95
P. M. WILLIAMS /On the Conservative Extensions of Semantical
Systems: A Contribution to the Problem of Analyticity 123
JOHN A. WINNIE / Theoretical Analyticity 143
ANDERS WEDBERG / Decision and Belief in Science. Comments
on Rudolf Camap's Views in 'Empiricism, Semantics, and On-
tology' 161
HERBERT G. BOHNERT / Camap's Logicism 183
J AAKKO HINTIKKA / Camap's Heritage in Logical Semantics 217
BARBARA HALL PAR TEE / The Semantics of Belief-Sentences 243
VIII TABLE OF CONTENTS

ASA KASHER / Pragmatic Representations and Language-Games:

Beyond lntensions and Extensions 271
RUDOLF CARNAP / Notes on Probability and lnduction 293
RICHARD C. JEFFREY / Carnap's Inductive Logic 325
RISTO HILPINEN / Carnap's New System of lnductive Logic 333
THEO A. F. KUIPERS / A Generalization of Carnap's lnductive
Logic 361
WILHELM K. ESSLER / Hintikka versus Carnap 365
JAAKKO HINTIKKA / Carnap and Essler versus lnductive Gener-
alization 371
ABNER SHIMONY / Carnap on Entropy: lntroduction to 'Two
Essays on Entropy' by Rudolf Carnap 381
INDEX OF NAMES 397
 PREFACE

There already exists an impressive survey and evahiation of Camap's

thought in the form of the Library of Living Philosophers volume ded-
icated to him. This volume appeared in 1963, but most of the contribu-
tions to it were actually written in the early or middle fifties. So much
has happened since in several fields close to Camap's interests - and so
much happened in his own mind before his death - that it seems amply
motivated to attempt to re-evaluate his philosophy and relate his thought
to subsequent developments. The editor's belief in the continued rele-
vance of Camap's ideas has led him to emphasize strongly the latter aim
of this volume. This conviction will therefore explain a feature of the
present collection which might otherwise seem strange. Several of the
papers below deal less with Camapian exegesis than with the later de-
velopments which Camap's work was instrumental in starting. It is my
strong belief that these new theories and conceptualizations are the best
proofs of the lasting importance of Camap's thought, and also that this
recent work helps us to understand and to put into perspective Camap's
philosophy even when it does not deal with it in so many words. This
serves to explain the inclusion of such largely non-exegetical papers as
those by Eberle, W6jcicki, Williams, Winnie, Hall Partee, and Kasher.
Another consequence of my belief in the lasting relevance of Camap's
thought is that 1 have not felt any obligation to approach it in terms
different from those in which one would approach contemporary con-
tributions to philosophical discussion. Hence 1 have not tried to avoid
even sharp criticism of Camap in those papers that deal directly with
his ideas. How Camap reacted to such serious criticisms as came to his
attention is beautifully shown by David Kaplan's reminiscences of his
early encounter with Camap (reprinted below in 'Homage to Camap').
I do not think Camap would have felt differently about posthumous
discussions and criticisms.
In fact, the very criticisms of Camap that Kaplan mentions are those
presented in his contribution to the present volume. This short paper
x PREFACE

has enjoyed a wide informal circulation and bas been discussed in the
litera ture but has never before appeared in print. It is complemented
here by the able papers by Wedberg, Bohnert, Jeffrey, and HiIpinen.
An aspect of Carnap's work which has untiI now remained almost
completely unknown is the work he did in the early fifties on entropy
and on certain related notions. The bulk of this work is now being pub-
lished by the University of California Press. Professor Abner Shimony's
contribution below places Carnap's work in this area firmly into an in-
teresting wider historical and systematic perspective.
I feeI especially fortunate in being able to include in this volume Pro-
fessor Hempel's paper 'Rudolf Carnap, Logical Empiricist'. I do not
think a better overall characterization of Carnap's thought is possible.
It seems to me eminently appropriate to use its title as the heading for
this whole volume.
Having known Carnap personally, it seemed to me monstrous to edit
a book on him without trying to convey some idea of him as a person.
For this purpose, I have included in this volume the personal recollec-
tions and statements that were published under the title 'Homage to
Carnap' shortly after his death. I hope that they do justice to this heading
here, too.
JAAKKO HINTIKKA
 ACKNOWLEDGMENTS

The papers by Hempel, Wedberg ('How Carnap Built the World in

1928'), Williams, Hintikka ('Carnap's Heritage'), Jeffrey, Hilpinen, and
Kuipers first appeared in the Carnap number of Synthese 25 (1972-73),
No. 3--4. Professor Wedberg has revised his paper substantialIy; Hintikka
has changed the title of his contribution, which originalIy had the title
'Carnap's Semantics in Retrospect'.
Carnap's 'Notes on Probability and Induction' likewise appeared fust
in the same number of Synthese. They were edited by Professor Arthur
Benson, with the help of Professor Risto Hilpinen. Professor Benson has
provided additional corrections for the present version. The 'Notes' re-
appear here with the permission of Mrs Hanneliese Carnap Thost, holder
of alI the copyrights to this material.
Carnap's 'Observation Language and Theoretical Language' is a
translation of 'Beobachtungssprache und theoretische Sprache', Dia-
lectica 12 (1958), 236-248, by Professor Herbert Bohnert. It appears here
with the kind permission of the Editor of Dialectica, Professor Henri
Lauener, and of the Estate of Rudolf Carnap. I am grateful to Professor
Bohnert for suggesting the inclusion of this useful but relatively in acces-
sible paper in this volume.
The contributions by Eberle, Kaplan, W 6jcicki, Bohnert, and Kasher
appear here for the fust time. Professor Bohnert's paper has grown out
of his paper at the 1971 International Congress of Logic, Methodology,
and Philosophy of Science in Bucharest. A summary of that earlier paper
appeared in the Abstracts of the Bucharest Congress.
The notes by Essler and Hintikka ('Carnap and Essler versus Inductive
Generalization') first appeared in Erkenntnis 9 (1975). They appear here
with the permission of the Editor of Erkenntnis, Dr. Wilhelm Essler.
The 'Homage to Carnap' and the paper by John A. Winnie, 'Theo-
retical Analyticity' first appeared in Roger C. Buck and Robert S. Cohen
(eds.), PSA 1970 in Honor of Rudolf Carnap: Proceedings of the 1970
Biennial M eeting of the Philosophy of Science Association (Boston Studies
XII ACKNOWLEDGMENTS

in the Philosophy ofScience), D. Reidel Publ. Co., Dordrecht and Boston,

1971, pp. xiii-Ixviii and pp. 289-305. They appear here with the pennis-
sion of the editors of Boston Studies.
Anders Wedberg, 'Decision and Belief in Science' appeared in the
Danish Yearbook of Philosophy 1 (1964) 139-158. It appears here with
the pennission of Professor Wedberg - and with a number of corrections
and changes he has made in it for the present printing.
Barbara Hall Partee, 'The Semantics of Belief-Sentences' appeared in
Hintikka, Moravcsik, and Suppes (eds.), Approaches to Natural Language
(Synthese Library), D. Reidel Publishing Company, Dordrecht and
Boston, 1973, pp. 309-336. It appears here with the pennission of the
author and of the editors.
Abner Shimony's paper reprinted here is his Introduction to Two Es-
says on Entropy by Rudolf Carnap, edited by himself, the University of
California Press, Berkeley and Los Angeles, 1976.
All the pennissions are gratefully acknowledged.
 HOMAGE TO RUDOLF CARNAP

At the Boston meeting, October 23rd, 1970, Adolf Grtinbaum, President

of the Philosophy of Science Association, introduced the Carnap Memo-
rial meeting.
We are assembled here to pay tribute to Rudolf Carnap because his
work made an immense contribution to 20th century philosophy of
science. But we also honor his memory because his life as a human
being enriched the lives of those who hll;d personal contact with him
professionally or otherwise. A few ofthese colleagues, friends or former
students of his will address you now. Herbert Feigl, Richard Jeffrey,
Willard Quine and Abner Shimony will do so in person themselves.
Cari Hempel is still recovering from eye surgery and has asked me to
read his tribute to you on his behalf.

Our great master and good friend, Rudolf Carnap, pas sed away on
September 14, 1970. His death carne rather swiftly (at the age of 79), after
a brief but severe illness. The world has lost one of its truly great thinkers.
After Bertrand Russell who, perhaps along with GottIob Frege, was his
prime 'Făther figure', Carnap was in the opinion (not only) ofhis disciples
one of the greatest 10gicians and philosophers of science of our century.
Since it is impossible to summarize in a few minutes his tremendously
important contributions, 1 shall take the liberty of reminiscing a bit about
the forty-four years of his friendship 1 was privileged to enjoy.
There is a great deal about Carnap's life and work that can be learned
from the volume The Philosophy of Rudolf Carnap, that magnificent book
in the Library of Living Philosophers (Voi. 11, Open Court, 1963) edited by
Paul Arthur Schilpp. There is a highly interesting autobiographical ac-
count of Carnap's intellectual development, and there are also his many
thoroughgoing replies, remarks and rebuttals in response to the tweny-
six articles by disciples as well as critics. Carnap devoted a great deal of

J. Hintikka (ed.) , Rudo!! Carnap, Logica! Empiricist, XIII-LXVIII. AII rights reserved.
Copyright © 1975 by D. Reide! Publishing Company, Dordrecht-Holland.
XIV HOMAGE TO RUDOLF CARNAP

thought and energy to these parts written by him for the Schilpp volume.
Even before Carnap carne for his first visit to the Vienna Cirele (1925),
some of us - especially Schlick, Hahn, Waismann and r, had read some
ofhis early work (Der Raum, Die Aufgabe der Physik, and the voluminous
typescript which was later to become - in somewhat abridged form - his
first major contribution to epistemology, published under the somewhat
romantic titIe (dreamed up by Schlick) Der Logische Aufbau der Welt.
r was tremendously impressed by this exact logical reconstruction of the
concepts of empirical knowledge. Once r carne to know Carnap better
personally, i.e., after he had moved to Vienna, 1 kept pestering him with
my objections to his phenomenalism (i.e., the more exact and much more
fully worked out develoPlllent of the epistemologies of Mach, Avenarius
and the early B. Russell). Although r pleaded for a critically realist view
(essentially along the lines of, e.g. O. Kiilpe and the early Schlick),
Carnap was firm in what he took to be a 'metaphysically neutral' stand-
rejecting as cognitively meaningless both subjective idealism and critical
realism. But in contradistinction to the positivism of Mach and Ostwald
he always emphasized his 'empirical realism' in regard to the 'external
world', the entities of microphysics, and of scientific theories in general.
Carnap retained, though in a semantically refined and elaborated form,
his metaphysical neutrality. This was explicitly stated in his by now justly
famous and 'elassical' artiele, 'Empiricism, Semantics, and Ontology'
(1951). Ontologically oriented metaphysicians have often been surprised
over Carnap's indifference to what they consider matters of profound
truth or falsity. Given Carnap's distinction between questions concerning
the categorial frame of knowledge, and questions pertaining to issues
within a given frame, one can understand (even if one does not fully
agree) that he regarded 'ontological' questions as issues about the choice
of a language.
r remember vividly Carnap's first lecture (1925) to the Vienna Cirele.
He presented his Space-Time topology (the 'C-T System', cf. Introduc/ion
to Symbolic Logic, Chapter G, p. 197 ff.) in the manner an engineer might
explain the structure of a machine he had just invented. To the non-
logicians Carnap indeed seemed to be no philosopher at alI. Some of this
sort of misunderstanding was Carnap's fate throughout most of his long
life. But this did not seem to disturb him at aII.
As r recall it, Moritz Schlick and Hans Hahn were to decide on an
 HOMAGE TO RUDOLF CARNAP xv

appointment of a Privatdozent in Philosophy at the University of Vienna

(in 1925-26). They found it difficult to choose between Carnap and
Hans Reichenbach who by then were good friends of each other, as well
as of Schlick's. 1 am sure it was Hans Hahn, the great Vienna mathe-
matician and admirer of Bertrand Russell who finally swayed the com-
mittee in favor of Carnap. Hahn felt that Carnap was in the process of
fulfilling the program that Russell had only informally sketched in his
Our Knowledge ofthe External World. This, in fact, was Carnap's achieve-
ment in the Logische Aufbau. (Impressive developments, such as Nelson
Goodman's The Structure of Appearance were to follow.)
Beginning with a (second) reading and discussion of Wittgenstein's
Tractatus, Carnap elearly became the leading spirit of the Vienna Cirele.
Ris task-oriented, unemotional (hardly ever ego-involved), penetrating
and thorough ways of work, thought and expression impressed most of us
much more than the intuitive (really basically artistic, though ingenious)
manner of philosophizing of the elusive and hermitic Wittgenstein. (Only
Schlick and Waismann succumbed to the almost hypnotic spell of
Wittgenstein.)
Carnap characterized hi!nself as 'schizoid' (rather than 'cyelothymic'),
as 'introvert' and 'cerebrotonic'.
1 remember him always as extremely rational, calm, well-balanced,
eager to absorb new information, never reluctant to change his mind even
in vitally important matters. While his great sense of humor was not
immediately obvious even to his elose friends, it did manifest itself in his
occasional irony (and even in 'pulling someone's leg'). The chapter (12)
on 'The Magic View of Language' in his Philosophical Foundations of
Physics (Basic Books, New York, London, 1966) wiU surprise readers
who know Carnap only from reading his austerely 'exact' books.
Carnap, originally very much like Schlick, was reluctant to propa-
gandize and proselytize in behalf ofthe Vienna Cirele's Logical Positivism
(or Logical Empiricism). His was a quiet, sober way of teaching and dis-
cussing - with colleagues as well as with students. But the enthusiasm and
the enormous energy of his friend Otto Neurath proved infectious. Thus
he collaborated wholeheartedly in the composition of our 'deelaration of
independence' (from traditional philosophy) in the famous pamphlet:
Wissenschaftliche Weltauffassung der Wiener Kreis (1929). For the ten
following years he was an active co-editor, with Hans Reichenbach. ofthe
XVI HOMAGE TO RUDOLF CAR NAP

periodical Erkenntnis. Many articles ofthe greatest significance for logical

theory, foundations ofmathematics, epistemology, and the philosophy of
the empirical sciences, etc., appeared in that memorable publication. There
were important contributions by Carnap, Reichenbach, M. Schlick,
F. Waismann, O. Neurath, P. Frank, E. Schrodinger, P. Jordan,
C. G. Hempel, W. Kohler, P. Hertz, M. Strauss, K. Ajdukiewicz, K.
Grelling, and P. Oppenheim, A. J. Ayer, E. Nagel, and many others.
Carnap was also a key figure in the Unity of Science Congresses (e.g.,
Prague 1929, Konigsberg 1930, Paris 1935, etc.). I also managed to persua-
de him to speak in a special session (with long discussion, eminent discus-
sants) of the International Congress of Philosophy in Mexico City, 1963.
On several occasions I was privileged to watch (and even in a very
minor way, to catalyze) some of the major forward steps of Carnap's
thought. 1 remember a long walk and discussion in the beautiful Tiirken-
schanzpark of Vienna. It was then (perhaps in 1928) that Carnap devel-
oped a first sketch of the idea of a logical syntax of language. Quick with
labels for new conceptions (rather than carrying them out in exact detail)
1 told Carnap that the syntax (he then still called it 'semantics', a term
which he applied - following Tarski after 1934 or 1935 in the later well-
established sense) he was thinking of as formulated in a metalanguage,
amounted to a 'Hilbertization' of Principia Mathematica. He accepted,
smilingly, my designation as essentially correct.
1 believe Carnap was the first to understand and to endorse Kurt Godel's
epochmaking discovery (made, 1 think, in 1930 - first published in 1931)
regarding the essential incompleteness of a large class of mathematical
systems. Godel, in turn, had utilized Carnap's (metalinguistic, i.e., syn-
tactical techniques) in his famous proof.
Throughout the many years of our intimate friendship 1 visited Carnap
in Chicago, at Harvard (Ernest Nagel was with us there in 1936), a sum-
mer in Maine, Princeton, Sante Fe and Los Angeles, usually once a year,
sometimes for several weeks, sometimes just for a few days. (In the sum-
mers of 1930,1931,1932, and 1935 we spent wonderful vacations together
in the Austrian Alps.) It was during the summer vacation of 1932 (in
Burgstein, Tyrol) that 1 introduced Karl Popper to Carnap. Several weeks
of exciting discussions in that beautiful mountain spot ensued. Popper
reported briefly about that episode in his contribution to the Carnap-
Schilpp Volume. As is well known, Carnap and Popper diverged sharply
 HOMAGE TO RUDOLF CARNAP XVII

later, especially in matters of inductive logic. Other, to me unforgettable,

occasions were our reunion (for the autumn quarter) of 1940 at Harvard-
with a vertitable 'galaxy' of stars: B. Russell, A. Tarski, R. von Mises,
E. G. Boring, S. S. Stevens, W. V. O. Quine, I. A. Richards, and others;
and the small conference in St. Louis, 1946 with intensive discussions
between Carnap, H. Reichenbach, Norbert Wiener el al. on issues of
probability and induction. Carnap and 1 conducted a seminar together
with Paul Feyerabend and F. von Hayek at Alpbach, Tyrol in 1964; and
colloquia at the University of Hawaii, as well as a most pleasant vacation
in the Hawaiian Islands in 1967. We exchanged letters frequently, and had
occasional long distance telephone conversations between Los Angeles
and Minneapolis during the last few years. Physicialism, Philosophy of
Physics, Unity of Science, Semantics, Psychology and Logical Behavior-
ism, and Inductive Logic were the major topics of our many conversations.
Carnap was always a patient and attentive listener - and a most construc-
tive critic of my (often deviant) ideas.
In 1938 we stayed on for two or three days after the meeting of the
American Philosophical Association at Urbana, Illinois and I urged
Carnap to apply his enormous analytic powers to the problems of induc-
tion and probability (this had been one of my hobbies ever since 1922-
but Carnap had paid only scant attention to these matters, leaving them
largely to R. von Mises and Reichenbach, later also to Ernest Nagel).
Carnap immediately began sketching in many hours of intensive discus-
sion ofwhat later became his great and influential work in Inductive Legic.
Even ifhis (stiII highly controversial) contributions to this field ofthought
wiII eventually be displaced by different approaches (some already in
existence!), there can be no question as to their great importance. If
Carnap failed in this ambitious, (and 1 don't say he did) enterprise (of
more than 30 years) he failed as nobly as did (if I may mention a very
distant analogue) Einstein with his Unified Field Theory. Carnap's work,
as well as his socialist-pacifist-world federation ideals (here he always
acknowledged the inci si ve influence of his late friend Otto Neurath) will
live on. His work williong continue to be influential, and his great person-
ality will be remembered and cherished by his friends as long as they live!

Minnesola Center for the Philosophy of Science, HERBERT FEIGL

Universily of Minnesota
XVIII HOMAGE TO RUDOLF CARNAP

REFERENCES

.For more detailed accounts of Carnap's role in the Vienna Cirele and in recent philoso-
phy generally, see his autobiography in the Schilpp Volume; also V. Kraft, The Vienna
Cirele (A. Pap, transl.) Philosophical Library, New York, 1953; J. Jorgensen's The
Development of Logical Empiricism in International Encyelopedia of Unified Science,
VoI. II, No. 9, 1951; A. J. Ayer (ed.) Logical Positivism, The Frec Press, New York,
1959; and H. Feigl, 'The Wiener Kreis in America' in D. Fleming and B. Bailyn (eds.)
The Intellectual Migration. Europe and America 1930-1960, Harvard University Press,
1969.

At this gathering in memory of Rudolf Carnap, it may be fit for me to

record a few of the impressions of Carnap the man and Carnap the
philosopher that 1 formed as his student and later also as his friend.
1 first met Carnap in 1929, at the Prague conference on the epistemology
of the exact sciences. I had gone there from Berlin to meet the author of
two books, published the year before, which had struck me as singularly
powerful and illuminating: Der Logische AuJbau der Welt and Schein-
probleme in der Philosophie. Living in a philosophical climate where loose
speculative thinking and convoluted writing were often thought marks of
profundity, I was captivated by the extraordinary lucidity, rigor, and
solidity of Carnap's analyses and reconstructions, as well as by the bold-
ness and ingenuity ofthe logical imagination from which they had sprung.
With his neat distinction between cognitively meaningful and meaning-
less components in philosophical problems and doctrines, Carnap ap-
peared to me to hold the key to the solution or dissolution of any philoso-
phical perplexity; his mode of analysis seemed to put all philosophical
issues into a hard, clear light, leaving nothing in a mysterious penumbra,
nothing in a frustratiilg fag.
an meeting him in Prague, I rai sed the question of coming to Vienna
for the following se mester to study with him, and Carnap readily agreed to
let me attend his seminar in addition to his lecture course. In subsequent
years, I visited him for shorter periods in Vienna and then in Prague;
and it was Carnap who made it possible for me in 1937 to come to this
country when he secured a year's assistantship for me at the University of
Chicago, where he was then teaching.
Carnap was a dedicated teacher. Ris lectures were carefully planned and
 HOMAGE TO RUDOLF CARNAP XIX

organized and were delivered in a calm and sober manner, without any
trace ofhistrionics. His style oflecturing and of discussing bespoke a deep
and inspiring concern with the exploration of philosophical issues and an
abiding dedication, so evident also in his writings, to the highest standards
of clarity, precision, and cogency of reasoning.
His students must have appreciated his open-minded readiness to
listen to their questions and to given serious attention to their objections -
an attitude that stood in marked contrast to the traditional aloofness of
most professors at continental universities. He was fair and even-tem-
pered; and he never put down a student by ridicule or by display of
professorial superiority. Pretentiousness and self-importance were totalIy
and naturalIy alien to him.
His openness to objections, whether in class discussion or in profes-
sional exchanges, was genuine: he treated a debate not as a duel aimed at
scoring points against an opponent, but as a joint effort to advance
knowledge and understanding; and he was prepared, therefore, to change
his views in response to telling arguments.
As is well known, he did change his views on various subjects: he
broadened the earlier testability requirement for empiricalIy significant
sentences into a requirement of confirmability, and later liberalized it
further in an effort to give an adequate account of the logical status of
theoretical terms. He similarly liberalized, step by step, the idea that
scientific terms should be definable by means of an observational vocabul-
ary; he broadened his conception of philosophy as the logical syntax
of the language of science by introducing semantic and pragmatic con-
siderations as welI. In these and other cases, Carnap characteristicalIy did
not just abandon an earlier theoretical conception which had shown itself
to be inadequate: he replaced it by a carefulIy elaborated new, more
flexible and inclusive conceptual system.
Carnap's readiness to make such changes has sometimes been referred
to an ironical overtones, almost as though it reflected a weakness of moral
fibre, akin to a betrayal of one's professed moral principles in times of
trial. But surely, readiness to change one's beliefs in response to pertinent
reasons is a prerequisite for alI objective theoretical inquiry; and in ad-
hering to this standard, Carnap was implementing his dedication to the
conception of philosophical exploration and discussion as objective
theoretical pursuits.
 xx HOMAGE TO RUDOLF CARNAP

Carnap had a wide-ranging and intense intellectual curiosity that ex-

tended far beyond the borders ofhis philosophical work; he had advanced
training in mathematics and physics; he read and thought a good deal
about psychological, social, and political issues; and he wondered about
future developments in these fields. Re was intrigued by science fiction;
and in a conversation about suspended animation, Carnap once remarked
that if it were feasible, he might exchange whatever years remained in his
present life for a change of being put into that quiescent state and resus-
citated at intervals, to witness different stages of the future.
Re kept steadily at work, even when a painful back obliged him to
remain stretched out on his couch; and in the waking hours during the
days of his final illness, he spoke of his plans for further work.
To observers who did not know him closely, Carnap may well have
appeared as intellect and rationality incarnate, as a powerful thinker
moving in a theoretical realm far distant from the domain of human
passions and hopes and fears; but this picture does not do him justice.
Carnap formed very strong human bonds; he was a warm-hearted and
loyal friend, a most interested and sympathetic listener, and a broad-
minded observer of the human scene who sought earnestly and with
imagination to understand - sometimes in psychoanalytic terms - person-
alities of a case quite different from his own.
Ris deep social and political commitments and his human compassion
speak from a memorandum that he must have written not long before his
death. It is a report, sent to a number of friends and colleagues, about a
visit he paid in January 1970, when he was nearly 79, to two Mexican
philosophers in a prison in Mexico City, where they had been detained as
political prisoners since September 1968. Carnap describes how, in the
presence of police guards, he talked with the two men, de Gortari and
Molina, about their interests, their work, and their prospects for release;
and he ends with a passage from which I would like to quote:

Mrs. de Gortari (who had accompanied Carnap) handed me two blank cards and
asked me to write a few words for each of the two philosophers .... 1 wrote for each of
them some words of admirat ion for their fortitude, tenacity, and stoic equanimity
with which they bear their hard fate, devoting their time to positive, fruitful work; and
1 also expressed the hope that the day of liberation would not be too much delayed .. , .
Both men read the cards and were visibly moved; they said that they would keep and
cherish the cards forever.
Suddenly trumpets and drums gave the signal for the end of the visiting period. 1
 HOMAGE TO RUDOLF CARNAP XXI

took a very cordial farewell with an abrazo from de Gortari and then from Molina.
They expressed their thanks very warmly; they said that this had been their best day
since September 1968. Then 1 walked with Mrs. de Gortari the long way between the
rows of barracks. Several times 1 looked back; 1 saw the two men standing in the door-
way and 1 waved my hand to them.

The words we are saying tonight are words that we would have like to
write on little cards for Carnap: alas, it is too late to give them to him.

Princeton University CARL G. HEMPEL

While we mourn Carnap's death, let us not omit to celebrate his life'
I knew him during his last twenty-five years - the period which he
devoted primarily to inductive logic - so I will talk about that, and say a
bit about how he seemed, through my eyes, at various stages.
I first heard of Carnap in a metaphysics seminar at Boston University,
when I was seventeen. Professor Brightman had mentioned that the
Logical Positivists (who were atheists) held metaphysics to be meaning-
less, and that seemed right. I elected to write a paper on them, and became
captivated by a littIe book - a 'Psyche Miniature' - by Rudolf Carnap:
Philosophy and Logical Syntax. I never looked back. As soon as I could,
after the war, I went to Chicago and took ali of Carnap's courses. (During
the war, stationed for a while in Chicago, I had managed to visit one ofhis
classes: I sat at the back of the room, full of reverence and misery and
longing, and never spoke to him.)
After five years, having exhausted my government subsidy, I took my
M. A. and went out into the Great World, which I found wanting. So
eventually I returned to Academia - to Princeton - and wrote a doctoral
dissertation with Hempel and Putnam on Carnapian inductive logic.
Following an implausible suggestion of Hempel's, I sent a copy to Carnap,
and he read it! (1 did not know, then, that he read everything people sent
him, and took notes, and wrote long, careful replies. Of course, he had
read my M. A. thesis in Chicago, but that was his job!) In this reply,
he pointed out that unless he had missed something, there was a serious
error in the last chapter: A construction which I had assumed would yield
probability measures would not do so. John Kemeny had proved as much,
and his proof was reproduced in some dittographed lecture notes of
XXII HOMAGE TO RUDOLF CARNAP

Carnap's, which he sent me. (The notes also reported work by my friend
and fellow-student at Chicago, Abner Shimony.) 1 consoled myself with
the reflection that it was too late for Princeton to withdraw my doctorate;
that the rest of the dissertation survived; and that the final section of
Carnap's monograph, The Continuum of lnductive Methods, was tarred
with the same brush.
That was in 1957. 1 was in England at the time. Carnap suggested that
when 1 returned to the United States, 1 visit him and talk about probability
and induction. Re wrote to me as to a kindred spirit whose opinion he
valued. I was thrilled. I went.
It turned out that his system of inductive logic was a work in progress,
and that much had changed since Chicago - since the massive tome,
Logical Foundations of Probability (I950). I had attended his lectures on
inductive logic, in the late forties, but had heard them rather passively.
1 had had some further thoughts about the matter at Princeton where, in
Rempel's seminar, I had turned away from Carnap's conception of prob-
ability, toward the subjectivistic or personalistic view of Ramsey and
DeFinetti and Savage; but it had never occurred to me that Carnap
himself had changed. But in 1958 I learned that he had never stopped
changing.
The Logical Foundations of Probability had been projected as the first
of two volumes under the general title, 'Probability and Induction'.
The Continuum oflnductive M ethods (I952) was to have been a preliminary
study for the second volume. But The Continuum was no sooner published
than vitiated, in Carnap's eyes: None ofthe c-functions in the continuum
were adequately responsive to analogy influence. In 1952-3, at the In-
stitute for Advanced Study, Carnap and Kemeny developed a generaliza-
tion of the continuum, within which they hoped that an adequate induc-
tive method might be found. (Carnap had already worked out a more
limited generalization, which was eventually published in 1959, as Anhang
B, Section VIII of Carnap and Stegmliller, lnduktive Logik und Wahr-
schein/ichkeit. )
My wife and I spent the summer of 1959 in Los Angeles, where 1
worked with Carnap nearly every day. (Re worked every day, especially
Sundays, on grounds of atheism.) Kemeny carne out for a week or so,
and got an enormous amount done, as did Raim Gaifman, who was then
Carnap's assistant. We would talk for three hours in the morning or the
 HOMAGE TO RUDOLF CARNAP XXIII

afternoon, usually sitting in the garden. At issue might be anything from

a bit of notation to the very core of Carnap's view of induction: Carnap
I

would bring to it the same calm reluctance to abandon debate until all
participants were of one mind, for he had no doubt of the universality of
reason, at least in philosophical matters. Y our reasons for judgement must
have equal authority for me, if they are genuinely reasons; and if they are
not, best suspend judgement. Mere conviction was of no interest to
Carnap.
Memories of Carnap are inseparable from memories of ina, I find. (Ina
was Mrs. Carnap. She invented her name, and insisted on the lower case
'i'.) She was in the background throughout the philosophical talks, and
was a major participant (although not at great length) in the talks about
about everything el se at mealtimes. It was her idea, to replace Volume II
of Probability and lnduction by a nonperiodical journal, to be called
'Studies in Induction and Probability'. Some quick, cheap method of
publication would be used, so that the era of dittographed arcana, dis-
tributed within the circle of the initiates, would end. The volumes would
be generally available; cheap enough for poverty-stricken students to buy;
and fresh enough to prevent fiascos like the one in my dissertation. The
first volume was to contain an article by Carnap whiclr would play the
foundational role of the outdated 1950 book. There would be an article
by me, explaining measure theory and DeFinetti's representation theorem
to readers of Carnap's earlier work. (He had decided to recast his ex-
position in the measure-theoretic idiom current in mathematical statistics
and probability theory.) And then there would be articles by Kemeny and
Gaifman, applying DeFinetti's representation theorem to Carnap's sys-
tem. Again, the system itself would be a generalization of its predecessors.
That was to be the first volume, which the four of us would prepare for
early publication - say, toward the end of 1961. Other volumes were
projected: I have a list of twenty titles of articles which the four of us
proposed to write, and it was hoped that after the first volume, other
collaborators would appear. But Carnap's foundational article grew and
grew. As the years passed, it developed into a sizeable monograph, of
which the sections, dittographed, were circulated among the initiates as
they appeared. And then they were revised in turn, and the revisions
recirculated.
When 1 last saw Carnap, over a year ago, before going off to England
XXIV HOMAGE TO RUDOLF CAR NAP

for a year, he suggested that Volume 1 of the Studies should contain only
the first half of his monograph, since that part was then in final form.
While the book was being printed and bound, he would revise the second
half for Volume 2. 1 agreed to look after the editorial busywork, in Eng-
land. Well, that took longer than we had expected, but now it is done:
Volume l of Studies in Inductive Logic and Probability is now at the
printer's. It will be published in 1971, by the University of California
Press.
After ina died, in 1964, Carnap's daughter, Ranna Thost, carne to
live with him and look after him. She stayed until the end, and after. On
the day folIowing her father's death, she received a packet from his typist.
It was the final revision of the penultimate sections of part 2 of Carnap's
monograph, 'A Basic System of lnductive Logic'. Re must have sent it
off just before going into the hospital. Re had alI but finished it. The rest
is now being revised by two of his associates over the years, Lary Kuhns
and Gordon Matthews, and the whole will appear in Volume 2 of the
Studies, in 1972. Wolfgang StegmlilIer, who heard from him just before
his death, tells me that Carnap intended to write some further material
for the 'Basic System', but no draft ofthis exists. Re was full ofideas and
plans, to the end. It is unlike him, to rest.
But although we shall never hear alI that he had to say, we have not yet
heard the last from our old master, our teacher and friend. Re was agreat,
and good man, and 1 am proud to have known him.

University of Pennsylvania RICHARD C. JEFFREY

Carnap is a towering figure. 1 see him as the dominant figure in philosophy

from the 1930's onward, as Russell had been in the decades before.
Russell's well-earned glory went on mounting afterward, as the evidence
of his historical importance continued to pile up; but the leader of the
continuing developments was Carnap. Some philosophers would assign
this role rather to Wittgenstein; but many see the scene as 1 do.
Russell had talked of deriving the world from experience by logical
construction. Carnap, in his Aufbau, undertook the task in earnest. It was
a grand project, and yet a self-effacing one, when so few philosophers
 HOMAGE TO RUDOLF CARNAP xxv

understood technicallogic. Much ingenuity went into the constructions,

much philosophical imagination, much understanding of psychology and
physics. If the book did not achieve its exalted purpose, it did afford for
the first time an example of what a scientific philosopher might aspire to
in the way of rigor and explicitness. If afforded detailed glimpses also, and
philosophically exciting ones, ofhow our knowledge ofthe extern al world
could in considerable part turn out to be, in Eddington's phrase, a put-up
job. And it provided techniques of construction that continue to be useful.
In his Logical Syntax Carnap again vigorously exploited the resources
of modern logic for philosophical ends. The book is a mine of proof and
opinion on the philosophy of logic and the logic of philosophy. During
a critic al decade it was the main inspiration of young scientific philoso-
phers. It was the definitive work at the center, from which the waves of
tracts and popularizations issued in ever widening cireles. Carnap more
than anyone else was the embodiment of logical positivism, logical
empiricism, the Vienna Cirele.
Ultimately Carnap saw limitations in his thesis of syntax. Thus carne
his third phase: books and papers on semantics, which have given Carnap
a central place in the controversies over modal logic.
Meanwhile Carnap's Logical Foundations of Probability continued to
develop, a monument to his unwavering concern with the logic of science.
Two months ago I had a lively letter from him about some supplementary
work that he was doing on this subject. Aiso he sent me a sheaf of ma-
terial from the new work in progress.
Carnap was my greatest teacher. 1 got to him in Prague 38 years ago,
just a few months after I had finished my formal studies and received my
Ph.D. I was very much his disciple for six years. In later years his views
went on evolving and so did mine, in divergent ways. But even where we
disagreed he was still setting the theme; the line of my thought was largely
determined by problems that 1 felt his position presented.
I first heard about Carnap and his Aufbau from John Cooley in 1931,
when we were graduate students at Harvard. Herbert Feigl was then at
Harvard as an International Rockefeller Fellow. He encouraged me to go
to Vienna and to Carnap the following year if I got a traveling fellowship.
Carnap moved from Vienna to Prague that year, and I followed him.
1 attended his lectures and read his Logische Syntax page by page as it
issued from Ina Carnap's typewriter. Carnap and Ina were a happy pair.
XXVI HOMAGE TO RUDOLF CARNAP

He was 41, she even younger. Along with their intense productivity there
was an almost gay informality. Ifyou combine strong intellectual stimula-
tion, easy laughter, and warm friendliness, you have an unbeatable recipe
for good company; and such were the Carnaps. On a day when Carnap
didn't have to come into the city to lecture, my wife and 1 would ride the
trolley to the end of the line and walk the remaining few blocks to their
littIe house in a suburb calledPod Homolkou. As the name implies, the
place is at the foot of something; and Carnap and Ina would have just
come in, likely as not, from an hour on skis on that very slope. Carnap
and I would discuss logic and philosophy by the hour. My wife and 1
would stay to lunch, or maybe dinner; but, if din ner, that was the end of
philosophy and logic until another meeting. Carnap's habits were already
austere: no science after din ner, on pain of a sleepless night. No alcohol
ever. No coffee.
I was then an unknown young foreigner of 23, with thirteen inconse-
quential pages in print and sixteen at press. It was extraordinary of any-
one, and characteristic of Carnap, to have been so generous of his time
and energy. It was a handsome gift. It was my tirst experience of sustained
intellectual engagement with anyone of an older generation, let alone a
great man. It was my first really considerable experience of being intellec-
tually fired by a living teacher rather than by a dead book. 1 had not been
aware of the lack. One goes on listening respectfully to one's elders,
learning things, hearing things with varying degrees of approval, and ex-
pecting as a matter of course to have to fali back on one's own resources
and those of the library for the main motive power. One recognizes that
his professor has his own work to do, and that the problems and the
approaches that appeal to him need not coincide in any very fruitful way
with those that are exercising oneself. I could see myself in the professor's
place, and I sought nothing different. 1 suppose most of us go through life
with no brighter view than this of the groves of Academe. So might 1
have done, but for the graciousness of Carnap.
At Harvard the following year, 1 lectured on Carnap's philosophy.
Our correspondence was voluminous. He would write in English, prac-
ticing up for a visit to America, and 1 in German; and we would enclose
copies for correction. By Christmas 1935 he was with us in our Cambridge
flat. Four of us drove with him from Cambridge to the Philosophical
Association meeting in Baltimore. The others were David Prall, Mason
 HOMAGE TO RUDOLF CARNAP XXVII

Gross, and Nelson Goodman. We moved with Carnap as henchmen

through the metaphysicians' camp. We beamed with partisan pride when
he countered a diatribe of Arthur Lovejoy's in his characteristically
reasonable way, explaining that if Lovejoy means A then p, and if he
means B then q. I had yet to learn how unsatisfying this way of Carnap's
could sometimes be.
Soon Carnap settled at Chicago. Two years later I took him to task for
flirting with modal logic. His answer was characteristic:
1 do not indulge in this vice generally and thoroughly .... Although we do not like to
apply intensionallanguages, nevertheless 1 think we cannot help analyzing them. What
would you think of an entomologist who refuses to investigate fleas and lice because he
dislikes them?

In 1939 Carnap carne to Harvard as visiting profe5sor. Those were historic

months: Russell, Carnap, and Tarski were here together. Then it was that
Tarski and I argued long with Carnap against his idea of analyticity.
Because of distances our later meetings were regrettably few. In 1949
my new and present wife and I spent some memorable days at the Carnap's
in New Mexico. In 1951 he and I held a symposium on ontology in
Chicago. In 1958 a reunion in California was prevented by an illness of
mine. Finally in 1965. to my delight, I saw him at Popper's colloquium in
London. He looked well and was vigorous and alert. When Popper con-
fronted him on induction his defense was masterly. It carried me back to
his confrontation of Lovejoy thirty years before. It was the same old
Carnap. His tragic death, while still at the height of his powers, marks a
sad date in the history of philosophy.

Harvard University W. V. QUINE

No one can expect to give an adequate impression of Rudolf Carnap to

those whodid not have the privi lege of knowing him. However, we can
hope that reports from people who had diverse relationships to him will
combine to convey a partially accurate conception of him.
Of all the speakers tonight I was probably the farthest from Carnap in
philosophical outlook. This was surely the case when I was actually his
student at the University of Chicago, and it was probably the case there-
 XXVIII HOMAGE TO RUDOLF CARNAP

after, in spite ofsome convergence ofviews. For 1 began to study with him
as an intoxicated metaphysician and in the course of many years became a
sober one. Ris generosity in discussing problems with students, analyzing
their papers, and encouraging them was not in the least contingent upon
their discipleship. Re did not in any way encourage discipleship, which
showed how deeply his opposition to authoritarianism in the political
domain was grounded in his character. Likewise, I think, his famous
Principle of Tolerance in The Logical Syntax of Language was not merely
a device to deal with Scheinprobleme, but an expression of his character.
Carnap was agreat systematic philosopher, who had a coherent ap-
proach to nearly the whole range of classical philosophical problems. The
massiveness and lucidity of his philosophical thought became increasingly
apparent to a student who was challenging him and searching for weak
points in its structure. It was an exercise of great value to Iose in such a
venture. The systematic character of Carnap's work was the expression of
a most remarkable feature ofhis intellect: the complete mastery and recall
of all that he had thought through in the past. If he had once reached a
tentative solution to a problem - even if it had been several decades
previously - one never saw him groping to remember the analysis. The
territory he had once conquered he thoroughly occupied. This was a major
component in the massiveness of this thought.
Ido not want to suggest, however, in pointing to the systematic charac-
ter of Carnap's thought, that he was inflexible. The other speakers tonight
have sketched some of the major changes in his general philosophical
position and in this theory of induction, showing his openness to criticism
and to suggestions for improvements. I want to add that he took particular
delight in technical advances which permitted him to widen the scope of
his investigations without loss of precision. Perhaps the most important
such advance for him was Tarski's work on the concept of truth, which
enabled him to extend his investigations of language from syntax to
semantics. In Testability and Meaning and Foundations of Logic and
Mathematics he himself did the main technical work which permitted the
relaxation and extension of the empiricist criterion of meaningfulness. A
third and more recent example was Kemeny's work on models, which per-
mitted him to define c-functions on languages which could not be readily
analyzed semantically in terms of state-descriptions. The general drift of
his philosophical development towards greater flexibility, openness, and
 HOMAGE TO RUDOLF CARNAP XXIX

richness indicates that his early program for the elimination of metaphysics
was never an attempt to contract the scale of the world or to view it
anthropocentricalIy, as some critics have c1aimed. Rather, it was the result
of his intense desi re to understand things c1early, and my personal im-
pression was that he experienced much joy in finding that c1arity is com-
patible with greater and greater scope.
Finally, one cannot adequately suggest how Carnap appeared to those
who knew him by talking only about his intellectual qualities. To be sure,
his thorough-going fair-mindedness towards others and himself was very
intimately connected with his logicallucidity. But his generosity, sponta-
neous kindness, concern for the oppressed, and courage alI went far
beyond fair-mindedness. The combination of his intellectuai power with
these moral characteristics made him agreat man.

Boston University A. SHIMONY

I met Carnap in person for the first time in the autumn of 1950. Until
then we had only corresponded, the correspondence consisting mostly in
my asking various questions concerning his publications, most of which I
had read many times in full and consulted thousands oftimes for c1arifica-
tion of my ideas (1 must have doubtless spent more time by an order of
magnitude on the study of Carnap's writings than on those of any other
author), and in his patient and detailed replies.
When 1 finalIy arrived in that autumn of 1950 with my family in Chicago
on a felIowship from the Hebrew University, we were lucky in finding
accommodations in walking distance from the university as well as from
Carnap's apartment - we, of course, could not afford a car at that time.
In addition to my attending Carnap's lectures on Analytic Philosophy,
carefulIy prepared and marvelously delivered (1 hope that 1 am not be-
traying my trade as a university teacher when I say that 1 have made in
later years much use of my lecture notes in my own courses on this sub-
ject), he was so kind as to set apart, during a per iod of five months, on
each Tuesday afternoon around to hours for discussions, in which we were
joined, after a few weeks, by Arthur Burks.
Carnap, who kept constantly complaining about his weak memory,
xxx HOMAGE TO RUDOLF CARNAP

always had before him a sheet of paper on which he had noted down the
topics he wanted to discuss with us; we were, of course, always welcome
to rai se questions of our own, an opportunity of which 1 tried to make
maximum use. This was the year in which his Logical Foundations of
Probability had been published, and 1 spent most of my free time in study-
ing this book. Carnap, who had only relatively late gotten interested in
this subject, was by then utterly absorbed in it and working on its planned
sequels. As for myself, to teU the truth, 1 always thought, and stiU con-
tinue to think, of Philosophy of Language as the field to which Carnap
had made the most significant contribution and should have been in a
position to make further decisive contributions, if he only had left himself
sufficient time for this endeavor. My attempts to draw him back to this
field, through leading questions and other ruses, met, though, with only
very partial success. After 1950, he published only two short but highly
important papers in Philosophy of Language, in addition to the Replies
in the Schilpp volume.
1 shall never forget the discussion between Carnap and Quine on the
occasion of the latter's visit to Chicago, in December 1950 or January
1951, in which he gave a masterful presentation of his 'Two Dogmas',
though not in the form published shortly afterwards. There were too not
many of us attending that lecture, and 1 presume that we ali preferred to
watch these two giants join in battle rather than interrupt with our in-
consequential questions and remarks. Needless to say that nobody
'won' - until this very day.
Unti11967, there was never a time when 1 would not make the greatest
efforts to visitCarnap, whenever 1 was in the States. There was never a
meeting from which 1 would not come away with the feeling that 1 had
learned something new and important. TiU today, whenever 1 tackle a
philosophical problem, the first question that comes to my mind is what
did Carnap say on that problem or what would he have said, had it been
put before him. 1 am convinced, and shaU probably remain so to the end
of my life, that this is the best way of approaching these problems.

The Hebrew University of Jerusalem YEHOSHUA BAR-HILLEL

 HOMAGE TO RUDOLF CARNAP XXXI

I. FIRST IMPRESSIONS: THE LOGIC CLASS

I was, candidly, set to encounter aGreat Man as I waited for Symbolic

Logic 320, falI term, 1941, to begin. I was not disappointed. In spite ofthe
inevitable slight shock at finding him no larger than a moderately taU
lifesize, with a small strand of hair standing up in back, there was an
immediate impression of weight. He was not especially overweight physi-
calIy. Rather there was a slowness, even a ponderousness, as of a tank in
preliminary maneuver. The impression of controlled mass was strength-
ened by the slow, cello-like voice. Though German-accented, with the
long syllables dwelt upon, it f10wed with total assurance, without hesita-
tion for words or grammar. The f10w was. not unstoppable, 1 found, on
raising a question; and the Great Man impression grew as the answer carne
with courteous formality, tempered with an almost fatherly gentleness.
Something of this impression was shared, I think, by the other students,
even without benefit of preestablished aura. But some seemed not to find
the f10w as easily stopped. It was not just momentum. There was also a
feeling of careful scheduling of topics for each lecture, a timetable not to
be upset. And there was an impression ofintense concentration underlying
the steadily articulate delivery, a concentration which manifested itself
in his gaze being averted from the c1ass to something c1oser, often his
fingernails, which he would scrutinize as if through a microscope. It con-
tributed to a sense of remoteness which discouraged questions. This remote-
ness, to be sure, did not seem inappropriate to agreat man; or even,
perhaps, to agreat teacher, in the European tradition. lndeed, Carnap
often seemed to exemplify that tradition at its best, within its limitations:
he was kindly but impersonal, vigorously dutiful in supplying outlines,
exercises, and c1ass drills - but ali somewhat unilaterally, with little depend-
ence on feedback. Nothing, for example, could swerve him from spending
the first third of each period in a methodical review of the preced ing period,
or could persuade him that the level of student understanding had gone
beyond that called for by his carefully considered timetable. Replies to ex-
trapolative or speculative questions were consistentIy minimal. On the
other hand, the material he had chosen to convey was testably, indelibly
conveyed.
XXXII HOMAGE TO RUDOLF CARNAP

There was, of course, more to Carnap's method as a teacher than such

a description would indicate, as 1 carne to realize. It may be suggestive to
recall the word-pictures Carnap gives us in his intellectual autobiography
(pp. 4-6 in Schilpp) ofhis own most memorable teachers, Hermann Nohl
and Frege. Nohl is praised for taking "a personal interest in the lives and
thoughts of his students" and because "in his seminars and pri-
vate talks he tried to give us a deeper understanding of philosophers on
the basis of their attitude toward life ('Lebensgefiihl') and their cultural
background." Of Frege he says "Ordinarily we saw only his back, while
he drew the strange diagrams of his symbolism on the blackboard, and
explained them. Never did a student ask a question or make a remark,
whether during the lecture or afterwards. The possibility of discussion
seemed to be out of the question." Without insisting that Carnap made a
conscious choice between being a lovable Nohl and an unapproachable
Frege, it is worth noting that his praise of Nohl is immediately followed
(with apparentIy unintended irony) by the remark that "On the whole, 1
think 1 learned much more in the field of philosophy by reading and by
private conversations than by attending lectures and seminars." The
imprint of Frege, on the other hand, is evident throughout the remainder
of the autobiography.
To be sure, much of Carnap's teaching manner was a reftection of
personality. The natural set of his mind was obviously closer to Frege's
than to Nohl's. Yet within such bounds, there was a conscious choice of
teaching styles. His style was, in fact, quite different in his 'Seminar on the
Principles of Empiricism, 440' which 1 attended concurrently.

II. CONTRASTING IMPRESSIONS: THE SEMINAR

For the seminar, there were no notes. Readings were handled only by a
reference to Ayer's writings as possible background. Exposition proceeded
by a general formulation of a thesis, followed by what, for want of a
better term, 1 shall refer to as pa rabIes or fables. These were not just
related. They were acted out. Many students must still recall the animated
dialogue between the idealist geographer and the realist geographer
concerning the mountain they had just surveyed (and about which
they disagreed only as to its ultimate ontological nature). Or the revelatory
saga of the blind student who arrived, through systematic coaching, defini-
 HOMAGE TO RUDOLF CARNAP XXXIII

tions, and thought, at a theoretical understanding of color words, not only

as applied to physics but as used in connection with perception in others.
In the acting out of these parables, there were amusing imitations of
indignant dispute, with gestures, simulated perplexities, discoveries, and
the like. Yet these interludes did not seem hard to reconcile with the man-
ner of the logic class. It seemed only a deliberate change of pace for a
different1y-textured area of inquiry. Dignity was relaxed but not lost. Nor
was there lost, it may be conceded, a certain essence of European profes-
soro The effect was not unpleasant. An occasional overconscious drollery,
is after all, well-intended. Indeed, the naked implausibility and artificiality
of the plot 'premiss', which would emerge as a parable began, had a
professorial quaintness that won an amiably tolerant audience. And when
the dramatist would earnestly supply circumstantial detail to add 'realism'
to such a fable, the audience, amused and charmed, would visualize as
cooperativelyas possible. The 'concrete' details, however, did sometimes
go to somewhat surprising lengths. Not only did the geographers, between
ontological sallies, compare their findings in terms drawn from the appro-
priate technical disciplines, but, for the blind student case, teachers and
ingenious pedagogical methods were detailed, as were special instruments
of perception, some suggested by current technologies, others more
speculative. But such methods and devices oftcn proved to have their
own epistemological points to make, so that the impression of unnecessary
stage-setting would undergo repeated revisions. Indeed, such revisions
would often continue after the apparent close of a parable. For example,
the methods introduced for the blind student would reappear in new
elaborations in a succeeding fable, this time about a far more restricted
being. The needed e1aborations for this case would turn out to be made
possible by some principle or device mentioned, apparent1y only for
science-fiction effect in the earlier parable.
Typically, such a cumulative linking of fables would build to some
especially significant extremity. The sequence above, for example, con-
tinued on through increasingly grave but iIluminating cases of perceptual
deprivation, to a climax in which a nonhuman superbrain whose only
perceptual ability was that of distinguishing dark from light was led to a
theoretical understanding of the universe, - an understanding as compre-
hensive, cognitively, as that ofthe society ofscientists (by now rather com-
plex) which had educated it.
 XXXIV HOMAGE TO RUDOLF CARNAP

With the stage thus set, Carnap would then be able to sum up, quite
simply, but with focused clarity and indelible impact, the central themes
that had emerged - in this case, about differences between experience and
knowledge, about relations between the communicability of information
and the causal connectedness of the uni verse, about the empirical in-
telligibility of a hypothesis that distinct, causally isolated universes exist. ...
The impact ofthis unusual thesis-parable-summary cycle was, as I think
back, oddly enhanced by the very remoteness that seemed a barrier in the
logic c1ass. While at tirst it was a mere part of the professorial quaintness
that enlisted amused attention as an 'example' (parable) would begin, the
psychological distance would seem to grow - and change in nature - as
unexpected turns of plot appeared. As a story would shift to a future
society, to other-dimensioned creatures, or to interplanetary communica-
tion, the professorial remoteness would, momentarily, become that of a
more genuinely alien, but kindly, mind - with whom communication was
suddenly as vital as if it had just arrived by time-machine. The effect was,
for others, [ think, as well as myself, that the philosophical points, when
perceived, would seem to stand out with a naked independence and uni-
versality, without restriction to era or galaxy.

III. THE LOGIC CLASS RECONSIDERED

The contrasting style of the seminar may dispel an impression of unreflec-

tive rigidity in Carnap's approach to teaching. But it also prompts a
reconsideration of the style of the logic c1ass. Why should it have had to
give such an impression in the tirst place?
To be sure, the two styles revealed certain affinities not apparent at the
outset. The seemingly rambling parables showed, in retrospect, at least as
much planning as the logic course. And their seeming lavishness of image
and detail proved more economically deployed than it tirst appeared. They
did, after aII, maneuver the seminar to some rather special vantage points
- for philosophical panoramas not ordinarily easy to encompass.
But such reflections only underline the contrasts. The unswerving
exposition, unleavened by drama or speculation, must now be regarded as
deliberate, the examples as consciously prosaic. Replies to extrapolative
questions, as already remarked, were consistently minimal. Indeed, the
minimality was so consistent and calculated that one tinally felt it as a
 HOMAGE TO RUDOLF CARNAP xxxv

principle at work, a sort of least-action principle - with respect to commit-

ment in general. But it is important to speli this out a bit.
There was never a suggestion of brusqueness. Every question brought a
kindly and patient response. But if the question suffered from any slight
confusion in the use oftechnical terms, Carnap's reply would not do more
than point that out. Often the tone would be reluctant, sympathetic:
"But in this system one could not say ... because -." There would seem
even to be an apology for the meagerness of the system's resources. But
there would also be agentIe finality to such a response. The lecture would
be resumed. There would seem no impulse to provoke reformulation of
the question, let alone guide it, or even to imagine that there was any
remaining question in the student's mind. If a reformulated question was
nevertheless put, the lecture would again come to a graceful halt and the
reformulation would be patientIy considered as if de novo (no matter how
quickly the reformulation had been interjected.) To be sure, if reformula-
tions persisted, Carnap might, with an air of consciously sympathetic
attentiveness, offer a trial interpretation ofthe student's remarks in a form
beginning :'Would you then mean to say ... ?' But such suggestions would
typically be obvious reductios, made only more frustrating by Carnap's
unchallengeable pretense of earnest misunderstanding. No air of 'con-
scious drollery' ever relieved the problem of dealing with such a ploy.
Indeed, at the time, such misunderstandings would seem discouragingly
genuine. After ali, it seemed not too surprising in a logic teacher, especial-
ly one teaching in a language not his own. 1 began to withhold more ex-
citing questions for discussion at office hours, (there to be carefully re-
formulated in 'his terms'.)
Procedure at office hours, however, was just as in cIass, except for the
greater freedom to pursue questions. With remarkable patience, as 1 now
see it, Carnap would take up each question on my mercilessly long list-
for-the-day. But, as in cIass, when a question was poorly formulated, he
would be of little help beyond pointing out what violation of type-restric-
tions or mismatch of universes of discourse had occurred. Nor would any
sequence of not quite correct reformulations tempt him to guess at, and
re formulate, my question for me. Instead, he would finally suggest, with a
polite sadness, that we pass on to the next question on my list. Even when
a correct formulation was finally arrived at, the answer would still be
disappointingly minimal, always at the least deep-going level appropriate
XXXVI HOMAGE TO RUDOLF CARNAP

to the question - often no more than an indication that we had arrived at

a truism, or a contradiction. And, more often than not, the meat of my
question would seem to have been lost among the reformulations. Either
way was frustrating, since it always seemed to me that r had merely been
trying to follow out consequences of some extraordinarily sweeping
formulation of his. At such times there wouldcome an impression, of
playing a chess-like game with a disguised adversary - beneath the ap-
parent gent1eness and pedantic literalness - whose aim was to prevent any
glimpse of a hidden truth.
Very slowly, however, over many sessions, enlightenments were at-
tained. Most often, they took the form of a colJapsing system of misunder-
standings. Each reformulation, often undertaken with ill-concealed im-
patience, would require smalJ mental readjustments that somehow cleared
the way, so that when the terminal collapse carne, it had the liberating
shock which some ascribe to satori. On rarer occasions, a question would
finally firm up, after shucking off layers of confused formulation, and
Carnap would say something, in a matter-of-fact way, that would sudden-
ly open a door to a whole new realm of concepts, or would turn some large
part of my mental universe upside down.
Yet Carnap never seized seized such moments to drive a point home or
rouse further enthusiasm by opening deeper doors. The response never
went beyond the question at hand. r would have an impression of waiting
layers of interpretations each guarded by a door with a separate key. rf
this was tutelage, it was of a uniquely austere and passive kind. A Zen
master, at least, poses questions to provoke satori. Carnap waited for
questions and then criticized them.
Nevertheless r did come to view this procedure as just such an austere
teaching policy. r feIt r perceived a similar policy ofliteralness and minimal
commitment at work say, in a Supreme Court decision to settle a case on
the basis of a technicality, reserving opinion on deeper issues until they
are more properly formulated. Such a policy has, certainly, its deep
rationale. Both law and logic rest on intricacies of definition and formula-
tion, and one sure way of driving them home is to withhold all response
beyond what is literally called for. That such a mechanical, non-directive
approach can indeed teach is amply attested by the experience of compu-
ter programmers. But it is very demanding, as programmers will also
acknowledge.
 HOMAGE TO RUDOLF CAR NAP XXXVII

The benefits of this austere minimality policy were, however, not

limited to the few who pursued matters at oftice hours. There was some-
thing in the carefully timed classroom exposition alone, in the underplayed
preparation of subtopics that tended to culminate in flashes of insight as
vivid as those provoked by the parables of the seminar.
One of the moments of illumination 1 recall carne in the latter half of the
second term. By that time an undeniably impressive framework of con-
cepts had been assembled and was continuing upward, step by firmly
planted step. What had seemed plodding, now seemed inexorable ad-
vance - purposive and architectonic. After an exploration of the logically
definable properties of relations and their interrelationships, with the help
of many careful blackboard diagrams, Carnap developed the notions of
isomorphism and structure. Isomorphism was not defined for classes
first, then for dyadic relations, and so on, as normal pedagogical might
dictate, but directly, as a schema for the general n-adic case. Conceptuali-
zation capacities were strained and struggling. Carnap went on to define
the structural properties of an n-adic relation as those hereditary with
respect to isomorphism. 1 ventured a bewildered confession - an inability
to visualize very many structural properties for the case n = 1, i.e., for
classes. Indeed, 1 could not see more than one. "And what is that?"
was Carnap's patient question. 1 hopelessly replied that 1 could not really
see how two isomorphic classes could have much in common beyond
the same number of members. There was then, my memory is quite clear,
a dramatic pause. Carnap smiled. Earlier in the course, after defining
cardinals 1, 2, ... Carnap had announced that the concept of cardinal
number in general could only be reached at a later, more advanced point
in the course. The pause allowed the realization - that we had now arrived
at that point - to take its own slow, many-inferenced course. My mental
eye seemed to travel over a synoptic vision of the great structure of ideas
that underlay this hardwon vantage point - and there was time also for
visions of previously unglimpsed realms - realms that were opened up by
coming upon the intuitively familiar concept of cardinal number as a
special case of a new, more general notion of relation number, with its
promise of many strange arithmetics; strange yet already illumined by
that small candle of familiarity. Aiso illumined, then, more dimly, was a
yet vaster world: that of all structural properties.
This vision was quickly followed by another, consolidating one. Carnap
 XXXVIII HOMAGE TO RUDOLF CARNAP

went on to say, after supplying the needed details of definition, that the
importance and power oflogic stern from the fact that aII inferences about
any subject matter depended only on the relational structure, and struc-
tural properties, of that subject matter. Completelycorresponding results
were assured for quite different subject-matters provided only that there
be a structural similarity. And - my thoughts echoed - aII structures and
structural properties are definable by logic alone. We seemed in possession,
then, of aII possible forms of aII possible knowledge!
Such carefully prepared moments seemed to have similar impact on at
least some of the other students. 1 recall attending a talk Carnap gave, for
a student mathematics club, in which several of us from the logic class
were amazed to see Carnap retrace the whole course of our two terms,
right up through the cardinality concept, using only the essential defini-
tions, but with easy cogency. At the end, after a burst of exhiliarated
remarks, we fell silent briefIy. It seemed impossible, both for our great
system, and for our inexorable Carnap, that he should have traversed it
so quickly. One classmate finally wondered aloud:
"Why couldn'the have gone through it Iike that with us?" After a pau se,
in which we refIected upon the long two terms, there were two further
remarks: "Well, the audience didn't get to understand it the way we do."
"With the short treatment, we wouldn't have either."
Ther:! was, 1 am suggesting, a rather special sense of timing and prepara-
tion that went into the understandings achieved during those two long
terms. In part it might be viewed as a sense of the dramatic expressing
itself in a different way. But it went deeper. It seemed to stern from an
insight into the workings of insight. The well-planned thought-explosion
required the careful assembly of combustibles. Each bit ofthe needed
conceptual structures had to be introduced and absorbed on its own terms,
without hint of further exciting implications - even when interconnecting
fuses were set in place. And when the fiare began, carefully estimated time
had to be allowed to ensure that aII the remoter secondary targets were
reached. The fiare, of course, had to be set off by the student, in accord
with his own timing, if it were to have effect. That is, while routine concep-
tual preparation may be set out straight-forwardly, the really difficult
insights, once within range, must be attained by the student on his own.
Insights, like jokes, Iose impact if the point is made explicit.
Such attitudes, 1 am aware, may sound unCarnapian. It was, after aII,
 HOMAGE TO RUDOLF CARNAP XXXIX

Carnap's dissatisfaction with Wittgenstein's remarks on things that could

only be shown, not said, that partly motivated his Logical Syntax. But
the attitude here ascribed to Carnap is not inconsistent. Carnap's belief
in the sayability of anything concerned 'cognitive' language. His with-
holding of literal explicitness is a matter of pragmatics. While his writings
may sometimes suggest that his acknowledgement of a realm of 'prag-
matics' was only to dispose of matters he thought little about, my own
experience has led me to a different estimate.
Perhaps another episode in that experience will illustrate some of this,
including a revealing interplay between the literal-cognitive and the inex-
plicit-pragmatic inCarnap's use of language.
The word 'property' had bothered me throughout the two logic terms.
And I had bothered Carnap about it a bit. He seemed to treat questions
about this apparently central notion with an amiable casualness that
suggested I could not be serious about such a question. A typical reply
might be: "anything that can be said about anything" followed by a few
examples, though examples were hardly my need. In a later more ad-
vanced course, I asked about it a last time, going on at length about
features, traits, tertiary properties, relational properties .... At least that is
the sort of harangue it was. In any case, it apparently convinced
Carnap that I was serious in my confusion and that the time had come
for a move. The genial informality did not appear. Instead Carnap
became totally serious. After a short silence in which the impression of
philosophical majesty grew, he announced "1 shall define 'property',"
and ro se and walked slowly, at his greatest weight, to the board. He wrote
"P(F)=.F=F". 'P' served as a symbolization for 'property'. 'F' was a
monadic predicate varia bIe in the type-theory logic we were using at that
time. 1 stared. It defined F to be a property just on condition that it was
self-identica\. 1 said the obvious: It makes everything a property. "Every-
thing of that type" Carnap amended. A silence began. I was not to be
helped, sink or swim. I was aware of things Carnap had said about
'universal terms' in his Syntax and their borderline closeness to 'meaning-
less' questions. This and other considerations showed well enough, I
thought, the unpalatable point Carnap intended to make: that in this
language 'property' simply corresponds to a grammatical type. No syn-
thetic sentence about the 'nature' of properties can be formulated, even
when the language is fully interpreted. I felt cheated again. My real ques-
XL HOMAGE TO RUDOLF CARNAP

tion seemed to remain at one remove, inaccessible. 1 finally blurted

something of this out. Instead of replying 'cognitively', he formed a few
beginnings of sentences about properties as if groping for something
'significant' to say about them. Each seemed to violate type restrictions
or shift universes of discourse. Then by some quite slight grimace, he
managed to convey a despairing inarticulateness. Somehow the philoso-
phical finality got through to me. My 'real' question, if answered, had to
be answered in a language, and that language would prompt 'real' ques-
tions with respect to ils universal terms, and so ono In the ensuing dis-
cussion, the distinction at stake (between internal and external questions,
as Carnap later put it in a famous article) became more articulate for me,
but it concerned something 1 had already 'seen', something tliat 1 had
been shown, not told about - something shown, moreover, by contempla-
tion of a totally cognitive sentence (an unbendingly literal, hopelessly
uninformative sentence) - and a grimace.
In this episode one can glimpse the dramatist of the seminar, usually
invisible in the logic class, showing himself in conscious command of the
deliberately minimal materials ofthe more difficult genre. There was, one
can see, always a commonality of origin and intent underlying the two
styles, and a careful readiness for interaction.

In the years that followed these early experiences, 1 carne to see the two
contrasting, but subtly interaction, styles - of the logic class and the
seminar - as involving more than pedagogical technique. They were not
confined to the classroom but appeared as basic communicational modes.
1 could recognize and trace them in his writings and in his discussions
with colleagues. (The minimality of the logic class may be more evident in
his writings but the elaborate parable of the seminar wiII remain at least
as well remembered by close philosophical friends and associates. 1 still
cherish a photograph 1 snapped of Carnap in mid-parable at the Institute
of Advanced Studies (circa 1952) at a kind of small 'summit' meeting
arranged by Herbert Peigl for his Minnesota Center for Philosophy of
Science. In the picture, Carnap's eyes are rai sed to his right hand whose
symbolic role has taken it above his head. The index finger points rigidly
downward. Perhaps it is a field vector, an instrumental sighting, or a step
in a rationally reconstructed discovery.)
StiII further, 1 carne to see the two modes of communication as related
 HOMAGE TO RUDOLF CARNAP XLI

to characteristic modes of thought in Carnap, even to features of what

might be called his 'philosophical personality'. They were modes ofthink-
ing and communicating which, whatever the native tendencies, were
developed over years in accord with an increasingly articulated philosophy
of (among other things) thinking and communicating.
To better dissociate them from classroom context, 1 can, for the logic
class mode, speak rather of minimality of commitment, uninstantiated
generality, literalness (often formalized), non-extra-polativeness, or the
like. For the seminar mode 1 must reach a bit further below the pedagogi-
cal surface. Underlying the graphic, acted-out example, as used by Carnap,
1 see his preoccupation with the development of simplified models; e.g.
his treatments of many problems via a 'rational reconstruction' of science
as a language, a store of sentences, or as a sentence-processing system with
protocol-sentence-input. They are often couched in language that seems
to refer to procedures and decisions of working scientists. This small echo
of 'parable' is harmless enough to scientists, who are accustomed both to
idealizations and to informal communication about them, but it has led
some philosophers of science to reject them as unrealistic. Others, at-
tempting to relate Carnap's use of simplified models to their use in, say,
physics, tend to object that the para bie is not spelled out even further: in
sufficient detail so that one could compare the idealized science with
actual science, and ascertain measures of approximation, check predic-
tions from the idealization by observation, and the like. But any student
of a Carnap seminar, who has seen the parable aspects vigorously devel-
oped in varied science fictions is well aware that it is not fact but concept
that is under study. Unlike the physical scientist's approach to complex
fact by simplified model, Carnap's simplified structures are approaches to
areas of conceptual complexity. Scientists, especially mathematicians, are
familiar with this kind of theory (and simplification) too. The mathe-
matical theory of games, for example, and its variety of simplified
subtheories, is not judged by a measured degree of approximation to
actual games. Nor are its suggestive concepts of strategy and payoff to be
viewed as models, in any measurable sense, of preexisting concepts of
strategy or payoff. Similarly, Carnap's 'explications', 'rational recon-
structions', and the like, were never just analyses of given concepts. He
sought to forge systems of concepts to replace the confusion areas.
(Ironically, the fact that Carnap's simplest inductive logic systems are for
XLII HOMAGE TO RUDOLF CARNAP

systems of fixed vocabulary brought complaints that Carnap took no

account of the ro le of conceptual innovation in the development of science,
that he portrayed science as consisting of no more than the confirmation
and disconfirmation of sentences. But, conceptual innovation was
Carnap's business.)
Mention of game theory should not suggest that Carnap's constructions
be judged in terms of theorem interest. Their simplicity was designed to
postpone mathematicalIy interesting complexity where possible. What
theorems there were were simply to show that the concepts defined did
exhibit the proper interrelationships. Similar concerns are shown in
Tarski's famous study of 'The Concept of Truth in Formalized Languages',
in which Tarski proves the law of exc1uded middle - not as a contribution
to mathematics, but to show that his definition of truth for a given system
works. This was the sort of theorem Carnap's structures also proved. It
is almost coincidence that Godel's theorem appeared in time to permit
Tarski to cap his great study with the 'interesting' theorem that no lan-
guage can define its own truth concept.
The final Tarskian theorem does, however, point to a pervasive feature
of formal systems, possibly of alI representational systems, that in a sense
provides an intrinsic rationale for the contrast between Carnap's two
thought modes. The Tarskian theorem, and the Godel incompletability
theorem, with which it is c10sely related, belong to a family of limitative
results which, among other things, seems to put a final bar on reaching
total generality in any definition of such concepts as truth, analyticity,
consequence. Whatever concepts humans can form of these 'ineffables'
seems therefore left to analogy with less comprehensive concepts defined
for specified systems. In other words, the seeker after total generality is
forced to approach such topics by example. In other words, the austere,
uninstantiated generality of the logic c1ass must be supplemented by the
graphic example of the seminar, since what they are examples of can not
always be stated, even in principle. Carnap's explicit struggle to come to
terms with this necessary dualism in approach is writ large throughout
his Logical Syntax of Language (which was strongly influenced by his
conversations with Godel during the period Godel was working on his
famous theorem.) And it was evidenced in many conversations as a
central theme in his thinking. That it was so central is a measure of his
enormous drive toward complete generality, toward finality, solution
 HOMAGE TO RUDOLF CARNAP XLIII

The Syntax was, in a way, an ape x of that drive. The AuJbau program
(construction of the world!), which was the first cIear exemplification of
that drive, paused after its 'first volume', to consolidate and rework, in
the Syntax, its logical underpinnings, borrowed from Principia Mathe-
matica. More positiveiy, it aimed at a maximaIly powerful logic system for
aII future such construction - a project both stimulated by the possibilities
inherent in the generality of G6del's arithmetization idea and, inevitably,
circumscribed by G6del's result. The upshot was Carnap's famous Lan-
guage II, complete with an infinity of 'positions', and 'infinite induction'
rule, and other powerful equipages, which he then, and later, procIaimed
as permanently adequate for empirical science. Yet it was not, and could
not, be the strongest possible, nor could the corresponding logical terms
be totaIly general. It was, therefore, a special case, an example, a paradigm
- material for a parable. Carnap's dissatisfaction is amply evidenced in
the chapters foIIowing those dealing with LII, namely, those of Part IV
on general syntax.
For a philosopher to concentrate very hard on any one thing, even if
that thing be generality, is, for some, a sign of narrowness, the mark ofthe
specialist. A. J. Ayer once divided aII philosophers into pontiffs and
journeymen, with Carnap as chief example of the latter. And Richard
McKeon sorted them into holoscopic and meroscopic types, with Carnap
as the chief example ofthe part-peerer. Ayer's division was more kindly to
Carnap than McKeon's but both missed Carnap's scope. Perhaps it was
because they and similar critics had not my opportunity for wide-ranging
conversations with Carnap over years. But perhaps they had not adequa-
tely reflected on how they would expect a very holoscopic, global mind to
act. Perhaps the only way 1 can convey my counterimpression is by a
parable of my own.
Picture a very holoscopic mind. Suppose it is a very powerful one. After
a survey of the whole scene it would, of course, form plans. The plans
would require deeper study in certain areas. This deeper study would
reveal broader promises and puzzles. Interrelationships would be per-
ceived. Plans and studies, by interaction, would quickly become global.
Science would have to be unified, language systematized, the foundations
of reasoning and experience scrutinized. Many specialized, meroscopic
jobs would have to be done. Some could be done best by the mind itself
- like constructing the needed overall conceptual framework - but time is
XLIV HOMAGE TO RUDOLF CARNAP

limited. Minds must organize. A journal must be started. A manifesto of

the new plan must be issued, congresses scheduled, an encyclopedia
planned. Delays must be expected, of course. Wars. The interaction of
minds is uncertain; the interaction of groups is unguided. In the meantime,
the mind applies itself to those very special but very basic jobs which few
but it can perceive as essential and as promising. This brings the mind fame
as a great specialist.
At some point, of course, the whole plan is seen as unlikely to progress
beyond its most initial phases in the time the mind sees as available to it.
But then it was never unaware of probabilities. As a holoscopic mind, it
understands its predicament perfectly. It stilllikes the plan. 'It proceeds in
its painstaking work as if it had milennia.
My para bie, unlike those of Carnap, has no literal point - only an
impression of a remarkable mind.

Michigan State University HERBERT G. BOHNERT

For most of his life, Carnap was a socialist humanist as wel1 as a scientific
philosopher. In the buoyant decade of socialist Vienna, logical clarification
was thought to be a means ofliberation from simple prejudice and sophis-
ticated mystification, just as empirical science was seen as a tool for
construction and a source of intellectual pleasure. But such a spirit of
enlightenment did not prevail, and darker stronger powers came to
dominate. Carnap lived through the years of fascism, imperialism, wars,
these decades of barbarism, and they entered profoundly into his life.
Ever the man of reason, of sceptical and probabilistic judgment as a guide
for living, he was not a naive Utopian about the chance for humane
relations among men and women in any class-divided or race-divided or
creed-divided society, or in a world of unlimited patriotisms. A socialist
and an internationalist, Carnap nevertheless lived through situations
which demanded defense, retreat, self-criticism, stubborn decency, maxi-
mal intelligence about minimal possibilities - not a life of individual
creativity in a communal society but of individual work and hope within
a context that elicited resistance. In his passionate careful way, he was a
man of the resistance. His first major work, Der logische AuJbau der Welt
 HOMAGE TO RUDOLF CARNAP XLV

of 1928, was written, as he said, in the confident spirit which "acknowl-

edges the bonds that tie men together, but at the same time strives for free
development of the individual." His last effort was an attempt to help
imprisoned Mexican philosophers whom Carnap visited in their jail just
a few months ago before he died. Critical of his own life and work,
Carnap was critical of other men and of their works and institutions; but
always with reason, never with hysteria or fantasy. A Stoic and an
Epicurean, an individualist and a socialist, a happy man and a somber
man - the best of this troubled age.

Boston University ROBERT S. COHEN

When Carnap was invited to come to the Department of Philosophy at

The University of Chicago 1 was its secretary (the nearest thing to a
chairman the department then had). The initiative carne from Morris, but
1 strongly supported it. Since the administration was hostile to the
proposal, this support was doubtless necessary if the project was to
succeed. J well knew how widely Carnap's views and methods diverged
from mine, but then 1 had learned at Harvard and Freiburg in the twenties
that philosophers never agree - well, hardly ever, And 1 thought it would
be exciting to have such arnan. And 50 it was. There was no one else in the
department with whom 1 liked better to talk.
Carnap did not impress me intellectually quite so much as afew
philosophers - Lewis, RusselI, Whitehead, Lorenzen - have done, but
still he was impressive. He was also likable, a very decent human being.
My problem was to convince him that my ideas were clear enough to be
worth discussing. If 1 showed him a manuscript, almost alI the marginal
comments were 'n.e.' for 'not clear'. Thus, when I tried to show him that
'externa!' and 'internal' relations could be clearly distinguished, he said
that this attempt, like G. E. Moore's, was unsueeessful. He did eoneede a
point to the contention that relations between universals and their in-
stanees were external to the universals and intern al to the instances; but he
balked at applying a similar asymmetrical distinetion to temporal relations
of earlier and later, or remembered and remembering experienees. (In
short he stood by what 1 view as Hume's greatest mistake, as Russell,
XLVI HOMAGE TO RUDOLF CAR NAP

Ayer, Von Wright, and many others have done.) I stiU wonder whether
(a) the Carnapian criteria of c1arity are too severe to make it possible to
deal with this and other important philosophical questions, or (b)
whether I simply lacked the necessary skiU and ingenuity. 1 incline to
(a).
1 agreed with Carnap that metaphysical mistakes are lapses from legi-
timate to iUegitimate (confused or inconsistent) uses of words; the dis-
agreement was on Carnap's view that the alternative to metaphysical
mistakes was other such mistakes - or el se no metaphysics. The issue
carne to a focus concerning the relations of truth to time, he holding
with many or most logicians:-- that aU truths obtain timelessly and [
holding that many new truths become true as "the indeterminate future
becomes the irrevocable past" (Peirce). This view reaUy annoyed Carnap!
He had to admit that the Tarski elucidation of 'truth' was 'neutral' to the
issue, and 1 never got any stronger arguments on his side than the con-
tention that common usage and common sense take truths to be timeless
(whereas [ think they are undecided on the question) and the contention
(which he stressed much more) that it is convenient for many purposes in
logic and science to take them to be so, which I grant but regard as
compatible with my position. Here 1 accused Carnap oftrying to get away
with an ontological proof of a metaphysical entity, aUegedly timeless
truth, timeless by a convenient definition. (Of course very abstract truths,
purely logical or metaphysical, are timeless. The dispute concerned
truths about particular events or c1asses of events.)
Charner Perry who (to my relief) followed me as officer of the depart-
ment, used to tease Carnap as foUows. When a highly historical disserta-
tion, and they were mostly that, was the topic of an oral examination,
Perry would solemnly inquire of Carnap if he would like to attend, only
to be met once more with, "I am not much interested in the history of
philosophy.' ,
In departmental seminars Carnap took part loyally and made genuine
efforts to see what speakers were driving at - in glaring contrast to another
philosopher of somewhat similar views (briefly a member of the depart-
ment) who invariably took a single sentence the speaker had uttered and
attacked it, totaUy ignoring the context into which the utterance had
painstakingly been placed. I was amused once when, after a fellow ex-
German temporarily with us had used the expression 'the eternal human',
 HOMAGE TO RUDOLF CARNAP XLVII

1 asked this man, did he mean the expression literally or was it only a
piece of poetry (whatever may be eternal, 1 do not see that humanity can
be so), whereupon Carnap, who had been silent until then, burst out,
"Everything he says is poetry!"
Carnap was mildly interested in, and gave a bit of technical assistance
to, my effort to exhibit a contradiction in the Thomistic idea of omnisci-
ence. His question was, will not a metaphysician always find a way of
wriggling out of any contradictions which his confused employment of
words appears to present? The answer, 1 think, is, yes and no: yes, he will
probably satisfy himself that he has escaped; no, he will (in many cases)
not have escaped without paying a price, the price of shifting from relativ-
ely elear but inconsistent meanings to meanings whose degree of elarity is
so low that the impressiveness of the position is diminished. On this
answer we would, perhaps, almost have agreed.
When Carnap left Chicago 1 felt it as agreat loss.
One can never forget Ina Carnap, her charm, or her devotion to her
husband's work.

The University of Texas af Austin CHARLES HARTSHORNE

10

When Carnap arrived at UCLA in 1954, he was completing work on his

brief but fundamental paper 'The Methodological Character of Theoreti-
cal Concepts'. In a series of previous artieles, beginning in the Vienna
Cirele days, Carnap had presented successive refinements of his basic view
that even for such complex theoretical structures as those characteristic of
modern physics and contemporary theology it must be possible to winnow
the empirically significant from the metaphysically speculative. In his
typically scientific way, Carnap had backed his arguments by offering
precise definitions applicable to formalized theories. 'The Methodological
Character of Theoretical Concepts' contained his most recent, and by
far most sophisticated, such definition.
As a graduate student, in 1958, I studied Church's review of a similar
attempt by Ayer in Language Truth and Logic. Church had shown that
Ayer's definition had unacceptable consequences, and 1 noticed that a
result slightly stronger than Church's could be obtained by an alternative
XLVIII HOMAGE TO RUDOLF CARNAP

proof strategy. Richard Montague found my argument interesting and

suggested that 1 attempt to apply it to the much more complex definition
given in 'The Methodological Character of Theoretical Concepts'. To
my own astonishment, a version of my criticism worked against Carnap
and together with a second, related, criticism seemed to me to throw
serious doubt on the feasibility of the program itself. But my argument
against Carnap's definition was less perfect than Church's argument
against Ayer. Carnap had stated the explicit condition that aII defined
notation be eJiminated before the definition of empirical significance is
applied, and my argument violated this condition.
At Carnap's request 1 prepared a brief written account of my results,
and we made an appointment to discuss the matter at his home later in
the week.
Climbing the garden path to Carnap's secluded home in the Santa
Monica foothills, 1 feIt some trepidation about my forthcoming interview.
1 knew Carnap to be a kind and even gen tIe man, but he was also master-
fuI. In seminars he welcomed questions, and he plainly loved philosophical
discussion. However, he always seemed to be perfectly familiar with
whatever questions or objections were raised. His responses would
typically sharpen the question, place it in a larger context, and then
convincingly dispose of it. One slowly carne to see that Carnap had a very
general and powerful framework which he used to focus imd c1arify
issues preparatory to resolving them. The distinction, for individual
sentences, between those which are empirically significant and those which
are not, like the distinction between those that are analytic and those that
are not, played a very important, if not fundamental, role in that frame-
work. Thus on the issues 1 had raised, 1 simply expected to be set straight.
And in the general way one feels when about to confront agreat power, 1
hoped 1 had not been too audacious in framing my objections.
Carnap waved at me through his study window and rose to greet me
as 1 entered the house. He was tall and slender, probably only a few inches
taller than 1, though he seemed to tower over me. "Ho, Dave" he boomed
in his typical hearty way, which in this instance 1 found somewhat un-
nerving. He seated me in his study and then took my report in his hands
and waved it at me.
"Well", he said, "this is really wonderful." Then, with evident en-
thusiasm, he reflected that he had been quite wrong for about 30 years, and
 HOMAGE TO RUDOLF CARNAP XLIX

his critics who had been arguing that theories must be accepted or rejected
as a whole (he mentioned at least Quine and Hempel) were very likely
correct. He asked a few quick and pointed questions about some varia-
tions on his definition, brushed aside my hesitant mention of the fact
that my argument violated his condition about primitive notation, and
then turned to discussion of how Hilbert's e-operator, applied to an
entire theory, might be used to characterize the empirical content of
theoretical terms. 1 was still somewhat stunned by his whole-hearted
acceptance of my criticism.
At the end of our talk, Carnap congratulated me again on my result
and urged me to communicate it immediately to a number of his co-
workers, especially those who had been critical of his own attempts.
Though we had frequent discussions over the last thirteen years, no sub-
sequent piece of work of mine was ever so critical of him, nor pleased
him so much.
It took me some time to understand Carnap's response. After reading
my report, Carnap had improved his understanding of a subject which he
had studied for many years. It was an advance in Philosophy, and whether
it was initiated by a student's criticism of Carnap or by Carnap's own
work made as little difference to Carnap as it did to Philosophy. His
enthusi~sm for the subject and his drive to understand the phenomena he
studied, completely submerged anyconcern with his own role in the process.
Through the years 1 have observed many other instances of Carnap's
selflessness. But the emotional impact upon a second year graduate student
of seeing Rudo!f Carnap respond to a student's argument with an enthu-
siasm completely unmitigated by his own 30-year investment on the other
side has stayed strongly with me. It was a rare and cherished experience;
Carnap taught much more than logic.
Perhaps only in a person of Carnap's enormous intellectual power and
accomplishment can the love of wisdom be unadulterated by love of self.
But since 1958 1 have tried to be like that.

University of California, Los Angeles DA VID KAPLAN

L HOMAGE TO RUDOLF CARNAP

11

My correspondence with Rudolf Carnap began in 1933. At his invitation

I visited him in Prague for some memorable weeks in the summer of
1934. In 1935 I suggested to the department of philosophy at the U niversi-
ty of Chicago that Carnap be invited as visiting professor for the winter
months of 1936. The invitation was extended to him, and accepted. During
this period I proposed that both Carnap and Reichenbach be invited to
become permanent members of the philosophy department. This double
proposal was made to the Administration, but was successful only with
respect to Carnap. He began his regular teaching in October 1936, and
remained at the University of Chicago until his move to the University of
California at Los Angeles in 1952.
I saw much of Carnap during those years, partly because of our joint
editorship (with Otto Neurath) of the International Encyclopedia of
Unified Science. I visited him only once in California, but our correspond-
ence and our work on this encycJopedia continued until his death in 1970.
I am happy that he saw in print the final bibliography and index mono-
graph ofthis large enterprise, the first monograph ofwhich was published
in 1938.
Rudolf Carnap was one of the finest human beings I have ever known.
His generosity to those in need, the depth and constancy of his personal
friendships, his cJose companionship with his dog, the place of novels and
music in his range of sympathy, the complete absence of pettiness and
petulance, the cJarity and kindness of his smile, the combination of sereni-
ty and concern which he always manifested - these attest the largeness of
the man. I treasure the remembrances of these many years.

University of Chicago and University of Florida CHARLES MORRIS

12

Nobody contests Carnap's greatness as a philosopher and as a man. He

was a pioneer, an untiring worker, and a sower offruitful ideas in modern
scientific philosophy. To me he was also a generous friend and mentor, and
to my husband a cJose associate and collaborator. I should like to convey
some of my feelings and impressions regarding these relationships
 HOMAGE TO RUDOLF CARNAP LI

in the hope to contribute a little to rounding out Carnap's portrait.

Although Carnap's and Hans Reichenbach's friendship dated back to
1923, 1 had not known Carnap before he carne to the University of
California at Los Angeles in 1954 as my late husband's successor in the
Philosophy Department.
1 had heard wonderful things about Carnap from Hans who did not
only admire and esteem him as a philosopher but also loved him as a
person. He had told me how charming, interesting, warm, good, and solid
Carnap was. And so 1 found him to be. 1 remember our first meeting as if
it had been yesterday. It was as if 1 met an old friend. Both he and Ina
were completely informal, offered me the familiar 'Du' immediately (but
explained that 'Carnap' functioned as a first name because they disliked
'Rudolf'). 1 had brought some flowers as a we\coming gift. He acknowl-
edged them by asking "Are they from your garden? 1 hope you did not
spend any money on them!".
1 saw them often and became almost a member of the family; when
relatives or visitors from out oftown carne, 1 would be asked to join them,
or the Carnaps brought them to my house. They were immensely sensitive
and to a large extent 1 owe them my ultimate recovery from the shock 1
had experienced. Occasionally when 1 lost my composure because 1 had
not quite mastered yet the loss of my husband, Carnap would put his arm
around me, offer his shoulder and say "Come here. Cry! Don't have any
moral scruples". And 1 just borrowed a little of his strength.
He was always available, always ready to help selflessly, whether 1
wanted to discuss a philosophical, an editorial, a pedagogical, or a person-
al problem with him; whether 1 asked him to look over a manuscript, or to
write introductory comments to works by Hans that 1 was publishing
posthumously.
We took enumerable walks together in the neighborhood of his or my
house. He loved nature, the mountains, the ocean, the trees, and the
flowers. He was an 'optica\' person and was very excited when he could
distinguish colors again after his cataract operation. He disliked talking
long on the telephone because he wanted to see the other person's face.
Whenever 1 took a trip, he wanted to be shown exact1y on the map where
1 had been and then would reminisce about his own travels and hikes,
or his visits to the archeological excavations in Greece led by his famous
unele Wilhelm Dorpfeld. He always carne to look at the slides 1 had taken.
LII HOMAGE TO RUDOLF CARNAP

Carnap had a tremendous capacity to recall earlier experiences and to

communicate them. He would remember the little streets and paths that
he had traveled in his youth, often on foot, particularly in Austria,
Switzerland and Greece. He had walked over the mountains to Delphi.
He told about his ski trips and the time he had broken a leg and almost
did not make it back bec au se he could hardly walk and was all by himself.
Carnap loved music. Once when I played a Bach record ing for him he
told me that the unaccompanied cello sonatas by Bach were among the
favorite pieces he had played himself on the cello. "Bach is like mathe-
matics" he said.
Carnap's interests ranged widely over areas in addition to philosophy
and the sciences: languages, literature, psychology, education, politics,
and people, and I marveled at his instant recall of so many diverse facts.
He did not subscribe to any daily newspaper but read carefully maga-
zines such as The Nation, Monthly Review, The New York Review of Books,
I.F. Stone's Weekly, and kept abreast by looking at the news on television
in the evening.
He was fabulously organized. He had a file of cards summarizing briefly
in shorthand the content of and his reaction to every book he read, and in
a conversation about a particular work he would frequently pick out the
relevant card and read it to me. He underlined everything he thought
important, even passages on the picture postcards I sent him from abroad.
His days were structured. There was work in the morning, a walk before
lunch, a nap, a snack, more-work, dinner and reading. He kept up this
regime until the end.
During the time Carnap was teaching at UCLA, I attended his various
seminars. I would pick him up in my car from his house in West Los
Angeles. Since I knew that he did not like to talk on the drive to the univer-
sity because he was already concentrating on the material he was going to
present, I kept quiet. His seminars were really graduate lecture courses,
never presentations ofpapers by students. They were attended by many of
his colleagues at UCLA from various departments, by faculty members of
other universities and colleges in the area, and by graduate students. He
always prepared the content most conscientiously and had a folder with
notes in shorthand in front ofhim. But he spoke freely and consulted them
rarely. He encouraged questions and comments. Sometimes Ina carne
along. One semester she taped the whole seminar on 'The Philosophical
 HOMAGE TO RUDOLF CARNAP LIII

Foundations of Physics' which was later edited and published as a book.

an the way home Carnap would always talk animatedly, often discuss
special aspects ofthe respective seminar. After we carne back to his house,
he would rest until dinner while 1 talked to Ina. As often as not the subject
matter of our discussion would be Carnap.
Carnap was frugal. His house was minimally furnished, there were no
pictures on the walls of his rooms or oftice, only book shelves. He did
not drink and did not smoke. 1 knew that he liked marzipan and 1 used to
give him some for his birthday. It was a sight to watch him cutting off
a microscopically thin slice after lunch. He made it last almost until his
next birthday. He seemed to know his own worth, though, bec au se when
1 drew his attention to the fact that he and Bertrand Russell were both
born on May 18, he said with a straight face "ah, so another great man
was born on that date".
Carnap and my husband saw each other on many occasions in Europe
and the United States. Carnap speaks about this relationship in the
Schilpp volume. They carried on a steady correspondence. Much can be
learned from these letters about their concerns. They dealt with mutual
critiques of manuscripts and books, with long discussions of special
philosophical problems, with the possibilities of finding positions for
younger gifted men they had spotted, and, during Hitler's time, with the
urgent necessity of rescuing mutual friends who had to leave Europe.
Many a weB known name is among them, and Carnap and Hans shared their
happiness and relief when they had been successful in placing somebody.
Carnap's letters contain practical advice and Lebensweisheit. For in-
stance, he suggested to send a manuscript by railway express because that
would be the cheapest way. Even though their mutual critiques were very
frank, he advised Hans not to publish too many replies to negative
criticisms of his books and articles by others, but rather to correct eventual
misunderstandings within the framework of other positive contributions.
And 1 know that on the whole Hans took this advice to heart. Another
time Carnap wrote "1 had indeed misunderstood your position. If one has
a different opinion from another, then he easily misunderstands the ex-
planations of the other man. 1 see that in your letter you also misinterpret
my position .... "
In one of his letters Hans offered Carnap a Flint professorship, a
specially endowed chair in the Philosophy Department at UCLA for
LIV HOMAGE TO RUDOLF CARNAP

outstanding visiting professors. Carnap declined bec au se of the loyalty

oath affair. My husband regretted but fully concurred in Carnap's
decision and asked him for permission to quote publicly from his letter
bec au se the matter was still being fought at UCLA.
Carnap had a kind ofinnocent purity and rare integrity. Ina's nickname
for Carnap, 'Angelface', testifies to that. If 1 suggested a little white lie as
a convenient way out of an uncomfortable situation, he would say with a
smile "You have a criminal mind". Re really lived his principle of toler-
ance and took Kant's Imperative seriously: to treat human beings, whether
in one's own person or in that of any other, always as an end, never as a
means only. Carnap would defend anybody whom 1 criticized by trying
to undetstand and explain rationally other people's irrational feelings and
behavior. Re was a good model and when 1 have a problem 1 still catch
myself thinking: 1 must ask Carnap.

Los Angeles City College MARIA REICHENBACH

13

Philosophers, scientific theorists and research workers in the foundations

of science ali Qver the world received the news of Rudolf Carnap's death
in September 1970 with sorrow and distress. The following is an attempt
to give a briefappreciation of Carnap's philosophical achievements.
For the last 35 years, Carnap lived in the United States and thus spent
the greatest part of his creative life there. The course of his scientific
activity began in Vienna, however, where he published one of the first
German language textbooks in mathematical logic as weB as his first
major philosophical work, Der Logische Aujbau der Welt (The Logical
Construction of the World). This book represents a high point of empiricist
philosophy. For centuries, empiricist philosophers had always held that
our concepts and our knowledge rest upon experience. Rowever, in
principle, they never got beyond dogmatic assurances and programmatic
proclamations. Carnap was the first to attempt to carry this program
into action, in his Logische Aujbau. With his mastery of modern logic
he constructed a system of concepts in which ali the concepts of the
factual sciences were derived from a single empirical basic concept, by
chains of definitions. There are two principal grounds upon which he
 HOMAGE TO RUDOLF CARNAP LV

later gave up this constitutive theory of concepts: first, he recognized

that certain concepts, namely the so-called disposition concepts in prin-
ciple cannot be introduced by definitions. Second, he believed he could
demonstrate that the positivist sense-datum language which he introduced
in this book should be replaced by a physicalist language. Nevertheless,
this first work exhibits an altogether extraordinary constructive and
analytical talent. One of the leading contemporary logicians, Quine, once
remarked that it remained a complete mystery to him how Carnap in this
work could have attempted to introduce by means of definitions such con-
cepts that no one would have thought to be definable except in a dream.
Carnap's second major work was the Logisch~ Syntax der Sprache
( Logical Syntax of Language) , which appeared in 1934. One may rightly
say that, with this work as well, Carnap established a milestone in the
history of recent philosophy.ln this work, the so-called linguistic tendency,
already hinted at in the ongoing work in analytic philosophy and meth-
odology of science, is fully realized. Here, Carnap was oriented by the
insight that the task of a scientific philosophy consists in the construction
of formal artificial languages, adequate to the precise reconstruction of
scientific propositions. By contrast to natural languages, these artificial
languages are constructed according to strict rules, which bear only upon
the external (abstract) form of expressions, and therefore may be charac-
terized as syntactical rules.
There are three outstanding features which may be culled from the
embarras de richesses which this work presents. Carnap was the first
philosopher to show the necessity to sharply distinguish between two
levels oflanguage in the logical analysis oflanguage: The Object language,
in which one represents the objects of inquiry, and which is construed by
analysis as an uninterpreted calculus; and the metalanguage, which is an
interpreted language and which is used to talk about the object language.
Carnap was able to show that the confusion of these two levels oflanguage,
a failure to distinguish between object-linguistic and metalinguistic con-
cepts, is responsible for the introduction of certain contradictions, and
that this confusion led even such distinguished logicians and mathemati-
cians as Bertrand Russell and David Hilbert into serious errors.
Another important feature of this work is the distinction which Carnap
introduces between material and formal modes of speech. Carnap shows
that many philosophical errors arise from the confusion of those state-
LVI HOMAGE TO RUDOLF CARNAP

ments which are meant to denote things and their properties with those
statements which in fact denote linguistic expressions. So for example, the
materialist asserts "All corporeal things are constituted by subatomic
elements" whereas the phenomenalist asserts "All corporeal things are
constituted by sensations". There appears to be an irremediable contra-
diction between these two assertions. However, if both of these theses are
translated into the correct formal mode of speech, the first thesis asserts
that all scientific propositions about things are translata bie into physicalist
statements of the sort given, whereas the second thesis asserts that all
propositions about things are translatable into sense-datum statements.
Therefore, it is no longer an issue of contradictory metaphysical assertions,
but rather one concern ing two compatible and rationally discussable
linguistic theses.
However, Carnap's most important contribution in this work is, with-
out a doubt, in that he established that alliogicai rules are to be under-
stood as syntactical rules. The earlier formulation may be stated briefly
thus: Syntactical rules are rules of formation, which tell us how to con-
struct new statements from other statements, as for example, from the two
statements A and B one constructs the statement "lf A then B". In this
earlier formulation, the syntactical rules stiH contain something of con-
tent, namely, they constitute judgments, insofar as they tell us how one
may derive certain judgments as consequences from other judgments.
Against this view, Carnap was able to show that even purely logical
relations between statements can always be derived from syntactical
structures alone. As against the formation rules of grammar, the logical
rules of inference can be seen as transformation rules, which, like all
syntactical rules, take only the external or abstract forms of statements
into account.
Unfortunately, Carnap's work remained unknown to linguists when
it appeared, mainly because he explicitly excluded natural!anguages from
his considerations. Thereby, linguistic theory remained unaware of the
existence of grammatical transformation ru!es (for natural languages).
The transformational grammar ofNoam Chomsky, therefore, was greeted
as a sensation when it appeared irr 1957. As the linguist and !ogician Bar-
Hille! once remarked, Carnap had anticipated the essential insights of
this new linguistic theory 25 years earlier.
The Logical Syntax of Language is Carnap's most difficult work. When
 HOMAGE TO RUDOLF CARNAP LVII

it appeared, only a few Polish logicians, who were working along some-
what similar lines, were able to understand it properly. In fact, it was
the Polish philosopher and mathematician, Tarski, who had the greatest
influence on Carnap, in the works which followed the Logica! Syntax. In
1935, in an important paper on the concept of truth in formalized lan-
guages, Tarski showed how the concept of truth, as well as many other
semantical concepts, could be introduced in a precise way for formal
artificial languages. It was characteristic of Carnap that he not only
grasped Tarski's ideas fully, but immediately set himself the task of
constructing a systematic theory of semantics. Shortly after his resettle-
ment in the United States, he began to publish a ser ies of works on se-
mantics. The conception which he presented in his Logica! Syntax was
not abandoned, but, in a significant sense, elaborated. Whereas he there
conceived of languages only as uninterpreted calculi, he now rounded
out this treatment of formal languages with a theory of interpretation of
sllch languages. What was distinctively new here was that the idea of inter-
pretation was not based on a vague psychological concept of meaning, but
was rather introduced as an intersubjectively controllable and mathe-
matically rigorous concept. Truth and logica! truth are the two funda-
mental concepts of semantics.
Only when these concepts were made precise was it possible to make
any clear sense of the demand for a proper foundation and for the com-
pleteness of a logical system. A logical calculus is complete only when
it can be shown that there is a matching relationship between the syntac-
tical concept of provability, defined for the calculus, and the semantic
concept of logical truth. This view of the necessity of semantic complete-
ness for a logical system is generally accepted today.
The third volume of Carnap's writings on semantics, Meaning and
Necessity, contains some completely new ideas, namely the basis for a
theory of intensional semantics. The earlier works on semantics, almost
exclusively on a foundation of extensional semantics, dealt with the basic
concepts of individual, class, truth-value. In the course ofhis investigations,
Carnap recognized that this conceptual apparatus remained incomplete,
and needed to be extended by a system of intensional semantics. Among
the basic concepts of such a theory are such concepts as the sense or
meaning of names and sentences, the synonymity or meaning-equivalence
of expressions, and the analyticity of propositions. Above alI this fulIy
LVIII HOMAGE TO RUDOLF CARNAP

developed conceptual apparatus requires a logic of modalities, which

bears on the concepts of possibility and necessity, as well as an epistemic
logic, i.e. a logic of knowledge and belief. Carnap ties his analysis to that
of Frege, in the main, but at the same time achieving a unified system-
atization as well as an essential simplification of Frege's ideas. The dis-
cussion of the nature and problems of intensional concepts, initiated at
the publication of Meaning and Necessity, goes on until the present day. In
all the subsequent investigations and in the critical confrontations in this
field, Carnap's book remains the standard work.
In addition to his studies in semantics, Carnap continued to dedicate
himsel\ to problems in the foundations of knowledge in the natural·sciences.
A clear and easily accessible insight into this aspect of Carnap's thought
can be gotten from the very readable work, Philosophical Foundations
of Physics. Among the many new ideas which are introduced here,
let us mention only one especially important one: the two-levels theory
of scientific languages. This theory decisively goes beyond the positi-
vistic view according to which all concepts and statements in the
natural sciences are ultimately based on what is observable and percepti-
bie. Carnap was able to show that in addition to the observational
language, which was the only base language which positivism allowed,
there was also a theoreticallanguage which, in contrast to the observation
language, could be only partially understood. The fundamental concepts
and principles of the theoreticallanguage can only be partially interpreted,
and even then only in an indirect way by means of so-called rules of
correspondence. Most important of all, the meanings of concepts which
occur in a theoretical language, are seen ta be dependent on the theory
formulated in this language. So, for example, not only does modern
physics propose hypotheses concerning elementary particles different
from those of classical physics, but the very meaning of the term 'elemen-
tary particles' is entirely different in each of the two theories.
Carnap's central aim, his theory ofinductive inference, has thus far gone
unmentioned. One can get a general idea of Carnap's position here by
contrasting it with that of his opposite number, Karl Popper. As is well
known, natural scientists often allege that their procedure is inductive, by
contrast with that of mathematicians, which is deductive. If one asks
what this so-called 'inductive' method is, no clear answer is forthcoming.
The philosopher David Hume had already remarked upon the many
 HOMAGE TO RUDOLF CARNAP LIX

difficulties raised by the view that in inductive arguments the conclusion

follows from the premises. Popper considered these difficulties to be
insurmountable, and developed a purely deductive theory of the corro-
boration of hypotheses. According to Popper, we corroborate hypotheses
in the natural sciences by attempting to refute them. When this attempt at
refutation fails, we then say that the hypotheses have been corroborated.
The corroboration of hypotheses and theories consists not in positive
confirmation, but rather in the bare statement of the failure of an attempt
at falsification. All talk of inductive confirmation and of the probability
of hypotheses is therefore nothing but empty chatter, according to
Popper.
Carnap found Popper's skepticism about induction unacceptable. He
was convinced, instead, that it made perfectly good sense to say of a
hypothesis that it was positively confirmed, or that it was more highly
confirmed than some other hypothesis, and that therefore it was entirely
reasonable to order the probabilities of alternative hypotheses in accor-
dance with the relative weight of available evidence. In the course of his
investigations of the precise sense of these new ideas, Carnap made a
remarkable discovery: It became clear that there were strict rules at the
basis of this inductive reasoning quite analogous to the rules of logical
proof in deductive logic. Therefore Carnap conceived his program of
an inductive logic. 2500 years before, Aristotle had begun to bring the
rules of valid deductive inference to the light of day. Analogously,
Carnap set himself the task of precisely formulating the rules of inductive
reasoning. He saw this task as extraordinarily important, because in
contrast to the very small role played by deductive inference in our
private and public lives, much the grea test part of our everyday thinking
involves inferences of an inductive, probabilitic sort. With this conception,
Carnap became the sharpest theoretical opponent of Karl Popper. Where-
as Popper declared induction to be superfluous, and the belief in it
nothing better than a superstition, Carnap believed he could demonstrate
that an inductive logic was at least as legitimate as deductive logic, and
should be placed alongside it.
The difficulties which stood in the way of carrying out this project were
so many and so enormous that anyone but Carnap would have despaired
of continuing with the task. In 1950, he published the first part of his
thought on the theme of induction in the extensive work, Logical Foun-
LX HOMAGE TO RUDOLF CARNAP

dations of Probability. In this work, there was an unusual combination of

original insights and most painstakingly detailed work. StiU, in many
respects the work proved to be inadequate. Carnap therefore set to work
anew.
His further work on inductive logic was interrupted, during the next
six years or so, by his simultaneous preoccupation with three other
projects. The first was the revised edition of his German work, Abriss der
Logistik (Outline of Logistic), requested by his publisher. Since a simple
reworking of the original text was out of the question because of alI the
new results which had been achieved in the meantime, Carnap wrote an
entirely new book, Introductiol1 to Symbolic Logic, which thereafter
became one of the best known and most widely read introductory works
in modern logic. Two things set this book apart from most other intro-
ductory texts: precise treatment of the semantic and syntactic con-
struction of formal languages; and many applications of logic to special
questions in the various sciences.
The second project was a revision of his semantics, in which he wanted
both to take into account the criticisms made of his earlier publications,
and to report on the latest results of his own investigations. This revision
was completed only in manuscript form, and it was circulated among a
smaU group of students and friends.
The third project turned into a gigantic task. In the Schilpp series (The
Library of Living Philosophers), there was to appear a Carnap volume. He
was required not only to give an account of his personal and intellectual
development, as well as a sketch of his views on philosophical problems,
but, in addition, to reply to other scholars who had contributed articles
concerning his work. The large number of contributions and their extra-
ordinary diversity give a most impressive overview ofthe range ofCarnap's
creative work. In some 250 pages, Carnap gave concise and intellectually
concentrated replies to ali of the 26 essays.
In his last 15 years, Carnap dedicated himself exclusively to the con-
struction of his inductive logic. The comprehensive manuscript of many
hundred pages, which he repeatedly exposed to constant criticism, and
which he reworked again and again, became known only to relatively few
colleagues and friends. It has only the slightest resemblance to the above-
mentioned work on this topic. In place of the originallinguistic version,
there is instead a model-theoretic approach by means of which the range
 HOMAGE TO RUDOLF CARNAP LXI

of application of inductive logic is vastly extended. It was Carnap's

tragedy that he did not live to see the publication of this, the magnum
opus of his life's work. He died a few days after sending the proof-read
manuscript to the publisher.
Carnap's theory ofinductive probability led more often to polemic than
to agreement. Still, even his most radical critics always agreed that Car-
nap's analyses yielded significant contributions to special sets of problems.
In the foUowing eleven points, I wiU introduce the topics of Carnap's
researches, ali of which can be recognized as highly significant, independ-
ent of whatever point of view one may have concerning Carnap's overall
project. This explicit mention of these areas of Carnap's research is
important, I feeI, bec au se at least eight of these points have remained
unknown unt il now, and are to be found only in the as yet unpublished
manuscripts. Even in the very recently published book, by L. Krauth, Die
Philosophie Carnaps, these matters could not yet be mentioned. One may
expect that after the details become known, not only will a lively discus-
sion ensue, but the picture we have had of the scientific thinker Carnap
will have to be revised in certain essential respects.
1. The so-called Principle of lndifference of classical Probability Theory
presented itself in contradictory form. AII the same, the view that "There
is something valid in this principle" kept cropping up again and again.
Carnap tried to work out the correct and unassailable kernel of this
c1assical idea, in a series of invariance principles.
2. An essential feature of ordinary man, as well as of the scientist,
is that he learns from experience. By means of the Principle of the Rele-
vance of the Singular Case as well as by the formulation of the so-called
Reichenbach-Axioms, Carnap tried to make c1ear the concept of rational
learning from experience.
3. In introducing the two concepts of credence-function and credibility-
function, Carnap contributed in an essential way to the clarification of the
foundations of personalistic probability theory as well as of normative
decision-theory. Whereas personalists and decision-theorists go no further
than the first concept, Carnap showed that the transition from the first to
the second is strongly motivated, and quite analogous to the transition
from peripheral manifest properties to their basis in permanent disposi-
tions, in the fields of psychology and physics.
4. Carnap goes along with the personalists in the matter of the Justifi-
LXII HOMAGE TO RUDOLF CARNAP

cation of the Axioms of Probability. Yet, i-t remained for Carnap to give an
exact analysis of the relevant concepts of bet and betting-system, as well
as of (strict) coherence, and thereby to give the precision of today's
logical standards to the justificat ion procedures for this measure.
5. It is noteworthy to remark on the way in which Carnap built his
new model-theoretical conception into probability theory. First of aII,
models yield interpretations of entire language systems - (which are there-
fore, as a rule, much more comprehensive than those which can be ex-
pressed in the usual formallanguages) - and such models can be described
by a single two-place metrical function. These models, or functions, are
chosen as points in a probability space. Atomic propositions are construed
as infinite c1asses of models; and the c1ass of propositions is identified
with the Sigma-partide derived from the dass of atomic propositions.
The probability-measure is defined for the elements of the latter.
By means of this ingenious trick, of being able to get interpretation
systems of entire languages from single functions, and of choosing these
as points of a probability-space, modern model-theory, which is an integral
part of mathematical logic, is, for the first time, brought intv connection
with modern mea:JUre-theory and probability-theory in an unrestricted
way.
6. Carnap didn't pretend to have solved aII the problems which come
under the heading, 'Justification of induction'. Nevertheless, his deliber-
ations on these questions contributed much to differentiate the problem,
and to keep distinct the different forms of reasoning which are under
consideration here. Because there is not a single inductive method, but
rather a whole continuum of such Illethods, the justification-problem has
to be divided into weaker and stronger questions, according to Carnap.
The weaker question concerns the justification for the choice of a stricter
subclass (of aII conceivable inductive procedures) which can be charac-
terized by suitable axioms.
The definite article in the phrase 'The justification of induction' is thus
no longer applicable. For in the context of answering the weaker question,
no grounds are given for fixing upon any particular C-function, which
represents a particular inductive method. Rather, there are only grounds
for the adoption of an axiom system for inductive probability. Only
with the answer to the stronger question is one led to the choice of a
particular inductive method. In the attempt to answer the weaker question,
 HOMAGE TO RUDOLF CARNAP LXIII

aII sorts of grounds have to be drawn upon already: inductive intuition

(coherence or strict coherence), pragmatic considerations, a priori con-
jectures on the structure of the world (e.g. those concerning degree of
uniformity), simplicity-considerations, as well as various subjective aspects.
Carnap was convinced that even if one addressed himself only to the
weaker question, one would have to go far beyond what personalistic
probability theory was able to afford. This theory satisfied itself with
the justification of the axioms of probability theory. an Carnap's view,
these provided much too weak a basis for inductive reasoning. His view
was based on the proof of the statement that for any two logically in-
dependent propositions Hand E and for any given real number r, one
can always find a C-function which will fulfill the Axioms, such that
C(H, E) =r. This shows that there is evidentIy always an abundance
of irrational functions which are admissible, which can be successively
eliminated by further Axioms. The intensive search for such additional
axioms and their establishment took up agreat part ofCarnap's researches
in his last years.
7. It is generally recognized that Nelson Goodman uncovered a funda-
mental difficulty with which aII variants of the concept of inductive
confirmation are afflicted (the so-called Goodman paradoxes). Carnap
tried to overcome at least one important aspect of these paradoxes by
his distinction between absolute and relative coordinates. The concept of
absolute coordinates presents an important contribution to the problem :
"How can we talk about individual things?"
8. Traditionally, the procedures by which qualitative or classificatory
concepts are introduced are completely separated and dealt with differ-
entIy from those by which quantitative concepts are introduced. It was
Carnap's achievement to give so generalized an account, in his theory of
families of attributes, that it could yield a unified methodfor tlle treatment
of quality and quantity.
9. The concern with families of attributes led Carnap to the sketch of
a language-independent theory of attribute-space. This is the first attempt
to characterise the topological and metri cal structures of attribute-spaces
by measure-theoretical means. One may expect that the ideas which Carnap
developed here, which are fully independent of his probability-theory,
will be fruitfully developed in their applications to other disciplines, e.g.
sensory psychology and phenomenology.
 LXIV HOMAGE TO RUDOLF CARNAP

10. A strong limiting condition which Carnap had to impose in his

treatment of languages, in his original linguistic version, was loosened
by his concept of meaning-postulates or analyticity-postulates. Carnap
recognized that this conceptwas stiH too weak. Re therefore introduced
phenomenological base-princip/es, which may be characterized, in more
traditional terms, as synthetic a priori propositions. In a bold generalization
of this notion, he conceived of comparative and quantitative synthetic a
priori propositions.
Il. The considerations noted in 6 above would require a good exempli-
fication of the procedures for establishing norms of rational scientific
discourse.
This last point leads to an entirely new assessment of the Carnapian
project for a probability logic. OriginaUy, as remarked earlier, Carnap
was dominated by the view that deductive logic had to be paraUeled by
inductive logic. The basic notion here, namely the notion of partia/logical
deduction turned out to be ambiguous. The same holds true of the notion
which he always used, degree of confirmation. Both of these concepts fade
more and more into the background. If we take this into consideration
together with the innovations noted above, one may then come to see
Carnap's lifework, not as the construction of an inductive logic, but
rather in thefoundations ofpersonalistic probability theory, and o.(normative
decis ion theory, and in the task 0.( making them precise, formalizing them
and rationalizing them further.
If one takes an unprejudiced overview of aU of Carnap's works, be-
ginning with his Logica/ Construction of the World and on to the latest,
as yet unpublished works, one does not arrive at a simple, conclusive
picture. There are many significant thinkers who have done fundamental
work in one or two branches of inquiry. Rowever, in comparison, one
would have to characterize Carnap as a unique phenomenon in the history
of thought, in that he did decisive pioneer work in six complete/y different
fie/ds.
Unique also was Carnap's combination of originality, insightful and
absolute precision, together with endless patience and persistence, charac-
teristics which are indispensable for the solution ofthe toughest problems.
Though Carnap was 45 when he first went to the United States, he
became one of the most influenti al philosopheIs in that country. Agreat
part of the discussion in the U.S. on philosophy of science, on 'the meta-
 HOMAGE TO RUDOLF CARNAP LXV

science of science', has gotten its most recent stimulation and problem-
settings from Carnap. However, this conclusion needs to be clarified
further: When one hears of an influential philosopher in Germany, one
thinks almost always of the founder of a so-called school of thought; and
beyond that, of someone who uses his established social position as a
University professor to take on as students to be advanced through
habilitation and teaching appointments, only those who take over, re-
produce and further develop his own ideas. Carnap had no students in
this sense, nor did he want this sort of student. What concerned him
was the educat ion of independent, critical investigators, who would be
able to contribute to the development of philosophy as an exact science.
The attempt to achieve conceptual clarity and distinctness of thought,
and to broaden the spirit of rational criticism bound Carnap closely to
many philosopher-friends, across all lines of theoretical difference of
opinion (e.g. H. Feigl, E. Nagel, C. G. Hempel, N. Goodman, W. V.
Quine, Y. Bar-Hillel. Karl Popper should be named here as well, certainly.
If one considers Carnap's dispute with Popper in its basic philosophical
terms, it comes to no more than a 'small internal family quarrel', which in
retrospect appears much less dramatic than it appeared at the time. In
their affirmative attitude to a rational philosophy both of these thinkers
were completely at one with each other.)
[mmediately after Carnap's death, obituaries appeared in various
distinguished German newspapers. [n ali of these, Carnap's book, The
Logical Construction of the World was seen as his major work. The later
works were ignored. For anyone who is in the least acquainted with
Carnap's publications, such accounts make clear how terribly provincial
German philosophy has become. Nowhere in these press notices was it
indicated that this work was, after ali, Carnap's Habilitationsschrijt,
written when he was thirty, and which he himself considered as super-
seded for more than forty years.
Carnap himselfillustrated, in his own case, how constant critical debate
and confrontation can lead to progress in philosophy. There were always
many criticisms of his publications; and Carnap himself brought forth
the sharpest and most decisive critical arguments against his own earlier
work. This sometimes led to what would appear to an outsider as para-
doxical situations, such as that in which N. Goodman (in the Schilpp-
volume) attempted to defend some of Carnap's earlier positions, in the
LXVI HOMAGE TO RUDOLF CARNAP

Logical Construction of the World, against Carnap's later formulations.

The quest for objective criticism was also one of the sources of the
many intellectual contacts which Carnap sought and found. When he
presented his theses for discussion at a meeting, he didn't do it in order
to win adherents to his position, but in order to stimulate objections and
proposals for revision and improvement. When 1 presented him with a
written list of questions, at my last meeting with him in 1967, he told me
beaming, that it gave him the greatest pleasure when students and col-
leagues took the occasion of meeting with him to bring along and present
prepared lists of problems and objections.
In his last years, Carnap's creative work was adverseJy affected by two
factors: his failing memory, and trouble with his eyes. Once he remarked
to me, halfin earnest, halfinjest: "When it comes to forgetfulness, 1 break
ali world-records." This pertained especially to his own vis ual work. He
reported how unspeakably tiring it was for him to proof-read his own
manuscript, because he had to turn the pages back again and again since
he couldn't remember whether he had already said something earlier,
or which sort of formalizat ion he had chosen to use earlier. Until the
two successful eye operations in 1969, his eye troubles of later years made
any reading ali but impossible for him.
On the other hand, his capacity for concentration and the acuteness of
his thinking remained unimpaired to the end. 1 got a lively impression
of this also in my 1967 meeting with him. 1 had been carrying on
a discussion with him and a former student for about five hours. He
vigorously decJined my proposal to take a break in the course of the
discussion and to go for a stroll. So we talked non-stop for the whole
five hours, on ali aspects of the philosophy of science, and he developed
his ideas in response to the partly informational, partly polemical ques-
tions which 1 rai sed through this whole time, without the least sign of
tiring.
A conversation with Carnap did not proceed as one who had read any
of his great works might have pres umed it would. Certainly, Carnap was
the great 'formalizer' among philosophers; and one might assume there-
fore that he was also given only to 'abstract formal thinking'. Nothing
could be further from the truth. In fact, he was a thoroughly 'straight-
forward down-to-earth thinker', if one may describe him thus. An accurate
impression of his style of reasoning and exposition can be gotten from his
 HOMAGE TO RUDOLF CARNAP LXVII

Philosophical Foundations of Physics which was transcribed from tapes of

one ofhis introductory seminars. Even in discussions of probability theory,
in the subtIeties of which he was more completely at home than anyone
el se, his exposition never took on a mathematical character. More often,
he would picture the situation in simple model-worlds, as for example,
in a world in which there are only things with different colors; or he
would describe the considerations that a rational diamond-miner would
have to take into account, once he had found a precious stone, and then
had to decide where to look further. Carnap's super-ego always forced
him, of course, to put his thoughts finally in a precise and formalized way.
But the intuition always came first, and precise refinement and formalization
always constituted the last step. He reported to me that he often advised
his students not to try for complete precision in their work, including
their dissertations. He added, laconically: "ApparentIy, my advice never
seems very convincing, since 1 myself violate it constantIy".
The large number offriendships which Carnap made during his lifetime
did not rest only on his philosophical-scientific contacts. Very often it
was his everyday, non-scientific concerns which led to personal friend-
ships. Since the first World War, which he spent mainly at the front,
Carnap stood politically very far to the left. His reaction to the Marxist
critique of philosophical empiricism is worth noting: whereas he critic-
ized traditional metaphysical philosophers very sharply, he did not attack
or criticize the Marxists. More often, he defended the lack of attention
which he and his empiricist friends gave to economic and sociological
problems, on the principle of the division of labor, on the grounds that
he and his empiricist friends happened to come out ofthe natural sciences.
At the same time, he emphasized the hope that these questions would
become far more central for philosophy of science in the future. He also
agreed with the Marxist critics that the contemporary economic and
social order leads to a brutal dehumanization oflife, and must be changed
therefore. His ideal, which he held with cautious optimism, was a scientific
humanism. He hoped that one day the social sciences would develop so
far that it would become possible for man to fashion his social world
correctIy and rationally, free of any ideology. He supported everything
which he thought would lead to worldwide understanding among men.
After the first World War, he studied Esperanto, and enthusiastically
joined the world-wide Esperantist movement, in the belief that a common
LXVIII HOMAGE TO RUDOLF CARNAP

world-Ianguage, especially if well formed, could contribute to under-

standing among men. Re later supported many peace initiatives. Re was
in correspondence with Bertrand Russell. Both philosophers were united
in their pacifism. The war in Vietnam was among the most frightful
experiences for him, in the years before his death.
Carnap's deep understanding ofhuman needs is little known. Whenever
he carne upon instances ofit, he tried to help. Re was also much concerned
with the Negro question. Re was in long term association with and sup-
ported a peaceful Negro organization in Los Angeles, which sought to
help achieve better education and better life conditions for its members.
The last photograph we have of Carnap shows him in the oftice of this
organization, in conversation with various members. Re was the only
white in the discussion group.
WOLFGANG STEGMULLER
 CARL G. HEMPEL

RUDOLF CARNAP, LOGICAL EMPIRICIST

Rudolf Carnap was the leading figure among the originators and the
moving spirits of the stream of philosophical thought known as logical
positivism or logical empiricism. Carnap preferred the latter name: the
appellation 'empiricism' rather than 'positivism' was to set the movement
apart from earlier forms of positivism, and the term 'logical' was to call
attention to the great importance this new empiricism attributed to the
concepts and methods of contemporary logic as tools of philosophical
analysis.
In Carnap's work, logic loomed large not only as a tool, but also as a
subject of philosophical investigation. A great deal of his research was
devoted to problems in logic and metalogic, including the fields of logical
syntax and semantics; the results of this work comprise substantial con-
tributions to deductive logic as well as the most comprehensive and ri-
gorous system of inductive logic yet devised.
The papers of my colleagues on this symposium address themse1ves to
Carnap the logician; let me therefore attempt a brief appreciation of
Carnap the logical empiricist philosopher.
Carnap's work outside the field of logic was devoted almost exclusive1y
to epistemology and the philosophy of science, and his principal contri-
butions to these fields are united by a common Leitmotiv, namely, the
search for ever more careful and philosophically illuminating reformula-
tions, or explications, of the basic idea of empiricism that all our knowl-
edge of the world ultimately derives from what is immediately given to us
in the data of our direct experience.
Stated in these general terms, the idea could be construed as a psy-
chologic-genetic claim concerning the development of man's concep-
tion of the world; but Carnap characteristically presented empiricism as
a systematic-Iogical claim to the effect that alI the concepts suited to
describe the world - and thus, alI the concepts that could ever be required
by empirical science, from physics to sociology and historiography - can
be reduced, in a clearly specifiable sense, to concepts serving to describe

Jaakko Hintikka (ed.), Rudolj Carnap, Logical Empiricist, 1-13. Ali rights reserved.
Copyright ©1975 by D. Reidel Publishing Company, Dordrecht-Holland.
2 CARL G. HEMPEL

the data of immediate experience or observation; and that analogously

aII our beliefs or assertions about the world can be reduced, in cIearly
specifiable ways, to beliefs or assertions concerning the data of immediate
experience.
This wording is still rather vague. Carnap's standards of cIarity and
rigor would require that, at least ideaIly, an explication of the empiricist
thesis should meet the foIlowing conditions: First, it should characterize
a language, with cIearly specified logical structure and semantical inter-
pretation, in which both the findings of direct experience and the state-
ments of empirical science can be expressed; second, it should delimit,
within that language, a basic experiential vocabulary by means of which
the data of immediate experience or observation can be described;
and third, it should give a precise logical characterization of the sen-
ses in which the terms and the statements of empirical science are cIaimed
to be reduci bIe to experiential terms and to experiential statements,
respectively.
Any specific characterization of the two kinds of reducibility wiII
cIearly yield a necessary condition for a corresponding construal of
'empirical significance' for terms and for sentences - a concept which
played a central role in the logical empiricists' early critique of specula-
tive metaphysics, and which continued to inform their persistent attempts
so to characterize scientific knowledge as to ensure empirical ground-
ing and bearing for scientific theories at a minimal cost in logical con-
straints.
The problem of explicating the relationship between empirical knowl-
edge and the immediate data of our experience has thus been given a
linguistic turn, which is characteristic of Carnap's general conception of
the nature of philosophical issues. The problem now concerns, broadly
speaking, certain logical and semantic features of languages suita bIe for
the description of alI empirical phenomena, and thus, of languages suit-
able for the purposes of empirical science.
In his first major work, Der Logische Aufbau der Welt,l Carnap chose,
from among various possibilities he considered acceptable, a phenomenal-
istic basis: he assumed the data of direct experience to be described by a
phenomenalistic vocabulary, and he argued that alI terms required for the
description of empirical phenomena can be explicitly defined on this basis.
In fact, in a striking tour de force, he limited his primitive vocabulary to
 RUDOLF CARNAP, LOGICAL EMPIRICIST 3

one single predicate, standing for the binary relation of remembered

partial similarity between what he calIed elementary experiences. Re-
ducibility of terms was thus claimed in the very strict sense of explicit
definability. This definability thesis for terms immediately implies that
alI empirical statements are translata bie into sentences containing only
the phenomenalistic terms chosen as basic.
The Aujbau was inspired especially by the construction of mathematics
from logic in Frege's work and in Prin cip ia Mathematica and by Russell's
ideas on the definitional reconstruction of the physical world on a pheno-
menalistic basis. What made Carnap's book such an outstanding achieve-
ment - a "dazzling sequel to Dur Know!edge of the Externa! World", as
Quine has called it 2 - is that Carnap did not just offer a general program
for a definitional 'construction of the world', but actually formulated
explicit definitions for a considera bie array of concepts; these belong to
the lowest domain of the constructional hierarchy, i.e. the realm of 'auto-
psychological objects'. The definitional steps intended to lead from this
domain of subjective experience to the public spatio-temporal world of
physics, to other persons and other minds, and on to socio-cultural
phenomena, were characterized in a more sketchy manner.
It has since become clear - not least as a re suit of Carnap's own sub-
sequent research - that the basic theses of the Aujbau are too strong to be
tenable; and for the recent English edition of his work, 3 Carnap wrote a
special preface in which, with characteristic candor, he pointed out what
he considered to be their principal shortcomings.
One of these concerned the thesis of definability. As early as 1935
Carnap pointed out that within a language with an extensional logic,
dispositional properties are not explicitly definable in terms of manifest
characteristics that are symptomatic of them, and he argued that dis-
positional terms could be introduced into the language of science by means
of so-called reduction sentences, which he regarded as affording partial or
imcomplete definitions. In his classical essay, 'Testability and Meaning',4
Carnap developed this idea systematicalIy and in detail. His earlier thesis
of the definability of all empirical terms was replaced by the broader one
that alI empirical terms can be introduced by chains of reduction sen-
tences on the basis of an observational vocabulary, and the translatability
claim for the sentences of empirical science gave way to a correspondingly
weaker one, roughly to thc effect that every empirical statement is at least
4 CARL G. HEMPEL

partiaIly confirmable by evidence sentences of molecular form containing

only observational terms. The earlier construals of empirical significance
in terms of verifiability or falsifiability were modified accordingly.
As for the experiential basis of empirical knowledge, a phenomenalistic
construal had initiaIly recommended itself to Carnap for reasons that
incIuded the apparent promise of providing empirical science with a
secure bed-rock foundation con si sting of phenomenalistic sentences es-
tablished by immediate experience, and thus indubitably true. 5 But Carnap
soon carne to share the view held by Neurath and by Popper that the
idea - or the ideal - of such empirical certainty is iIIusory, that every
empirical statement is open, in principle, to revision; and thenceforward
he preferred to construe the basic evidence for empirical knowledge as
expressed in a 'thing language', by means of sentences in which directIy
but publicIy observable attributes are ascribed to publicIy observa bIe
physical objects.
The resuIting explication of empiricism thus took the form of physi-
caIism, according to which the thing language, or more broadly the lan-
guage ofphysics, affords a unitary language for aII empirical science in the
sense that aII scientific terms can be introduced on the basis of a physical
vocabulary, by means of definitions or - in a weakened version of the
thesis - by means of reduction sentences; and that aII scientific statements
are translatable into, or - in the weaker version - at least confirmable
by, sentences containing only physical terms. For example, psychological
terms like 'angry' or 'excited' were said to refer to bodily states charac-
terized by the disposition to show specific kinds of physical behavior
under specific physical conditions. Carnap regarded it as a long-range aim
of scientific research, however, to replace such dispositional characteriza-
tions by micro-structural ones, which would again be formulated in
physical terms. 6
More recentIy, 7 Carnap pointed out that the construal thus arrived at
does not do justice to the logical character of the terms and sentences of
scientific theories, and he therefore proposed a further liberaIization.
Re characterized the introduction of theoretical terms as being effected by
specifying (1) a set of theoretical axioms containing the ba sic theoreticaI
terms as primitives, and (II) a set of 'rules of correspondence' which afford
an empirical interpretation by Iinking some of the theoretical terms to
observational ones. 8 These interpretative sentences, however, need not
 RUDOLF CARNAP, LOGICAL EMPIRICIST 5

consist in chains of reduction sentences; indeed, Carnap imposed no

explicit general restrictions on their form. In accordance with this con-
ception, Carnap formulated new criteria of empirical significance for
theoretical terms and sentences; these criteria are multiply relativized,
making the empirical significance oftheoretical expressions depend, among
other things, on the axioms and the correspondence rules of the theory in
which they function. 9 This conception, reflecting a deeper exploration of
the logic of scientific theorizing, is a far cry from the simple original
construal of empirical significance as a property of certain terms and
sentences.
In his autobiography, Carnap remarks that even these newer criteria of
significance "are not yet in final form"; but he reaffirms his confidence in
the possibility of further clarification: "1 am inclined to believe", he says,
"that it is somehow possible, even in the wider framework of the theoreti-
cal language, to make a clear distinction between those terms and sen-
tences which are cognitively significant and those which are not".1°
Carnap similarly retained his confidence in the tenability and, indeed, in
the philosophical importance of another idea which had played a promi-
nent role since the initial stages ofIogical empiricism, and which was often
held to mark an advance over earlier forms of empiricism, such as that of
John Stuart MiII: this is the idea that aII cognitiveIy significant sentences
can be sharply divided into those which have empirical content and thus
are synthetic, and those which are analyticaIly true or false; the analytic
truths including as a proper subclass the truths of logic and of mathematics.
Again and again, at various stages in the evolution of his ideas, Carnap
returned to the task of justifying and explicating the analytic-synthetic
distinction; he even proposed an ingenious way of dividing a scientific
theory, with the help of its Ramsey sentence, into an analytic and a
synthetic component. The importance this problem possessed for Carnap,
and the intensity ofthe effort he devoted to its clarification, can be sensed
from the words with wl1ich he introduced his proposed solution: "During
my work on the article ['The Methodological Character of Theoretical
Concepts'] and subsequently,I long searched in vain for a solution to this
problem; more specificaIIy, for a general method for analyzing the total
postulate set [consisting of the theoretical axiomsţtnd the correspondence
rules] into two components: analytic meaning postulates ... for the theo-
retical terms, and synthetic P-postulates ... which represent the factual
6 CARL G. HEMPEL

content of the theory .... 1 believe now to have found a solution for this
problem."ll
The analytic-synthetic distinction and the notion of empirical signifi-
cance, as is well known, have been subjects of considerable philosophical
controversy for the past several decades. Carnap was an active participant
in the debate and sought to meet doubts and objections by suitably modi-
fying his explications, while maintaining his conviction that the explicanda
were reasonably clear and of great philosophical importance. The debate
over this issue is continuing; but no matter what its outcome may be,
there is no doubt that Carnap's efforts to clarify those ideas, and more
generally his search for ever more adequate construals of empiricism, have
considerably deepened and refined our appreciation of the logical com-
plexity of science and of the connections between its branches.
Carnap's method of philosophical analysis might be characterized as
explication by logical reconstruction in model languages with precisely
specified syntactical and semantic features. The standards he set himself in
the use of this method reflect the characteristically clear, precise and
systematic manner in which he sought to express, develop, and defend his
ideas on any subject: in writing as in teaching and in oral discussion,
Carnap evinced a deep faith in the power of reason and of rational critical
argument. It was this search for maximal clarity that made the reliance on
formal logic and artificial model languages attractive, and indeed often
indispensable, to him in tackling philosophical issues.
There could hardly have been a more striking contrast in philosophical
style within the same school of thought than that between Carnap and
Otto Neurath, an influential figure in the earlier stages of logical empir-
icism, whose formulations were suggestive but vague; whose argumenta-
tion, though provocative and often persuasive, tended to be loose, sketchy,
and programmatic; who looked at Carnap's reliance on precise model
languages with respect but deep misgivings; who spoke of the language of
empirical science as a 'universal slang' not governed by precise syntactical
and semantical rules; and whose metaphor of a ship that is forever being
reconstructed on the high seas without benefit of a drydock 12 reflected
his conviction that we cannot start building a scientific language from
scratch and with Carnapian precision because of the ever-present fuzzi-
ness of the ordinary language in which we have to describe our construc-
tion ;13 and that, furthermore, scientific knowledge does not rest on a solid
 RUDOLF CARNAP, LOGICAL EMPIRICIST 7

foundation ofunquestionable truths - a view with which, as noted earlier,

Carnap carne to agree. It is characteristic of Carnap's undogmatic open-
mindedness that despite the striking difference in' intellectual style, he
appreciated Neurath's often shrewd and perceptive judgment, was on
cordial personal terms with him, and repeatedly acknowledged his in~
debtedness to him in his writings.
Carnap was always conscientious and indeed generous in acknowledging
the ideas of others that had influenced his thinking; on the other hand, I
do not know of a single instance in which he complained about not having
received due credit for his ideas or for their priority. About his punctili-
ousness in the matter, there is a story that in a course he gave at Harvard,
Carnap said at one point: "Let A be some physical body, such as a stone,
or a tree, or - to take Russell's example - a dog."
In his commitment to careful formulation and rigorous reasoning, in
his insistence that philosophical theses should be stated and argued for in
ways permitting of objective critical appraisal, Carnap evinced an attitude
that is often said to be the mark of a good scientist; and indeed, Carnap
like other logical empiricists - several of whom had had extensive scientific
training - called for a scientific approach to philosophy, and in the days
of the Vienna Cirele, members of the group often used the appellation
'Wissenschaftliche Weltauffassung' to characterize their conception of
philosophy and philosophizing.
Carnap's predilection for the use of precise modellanguages in phiIo-
sophical explication was clearly akin to the keen interest he took in auxili-
ary internationallanguages, like Esperanto, which in his earlier years he
spoke fluently. These international languages, too, were meant to be
governed by precisely stated general grammatical and semantic rules
without all the usual exceptions, and simple enough to be readily learned
and used; and Carnap felt strongly that the wide adoption of such a
language would be, among other things, agreat help in furthering the
international propagation of scientific ideas. In the preface of his In-
troduction to Semantics,14 for example, he expresses great regret that
Tarski's fundamental work on the semantical concept of truth, published
in Poland in 1933, had remained virtually unknown outside that country
until 1936, because it had been available in Polish only; and he adds:
"This fact, incidentally, confirms once more the urgent need for an inter-
national auxiliary language, especiallyfor scientific purposes." In speaking
8 CARL G. HEMPEL

with Carnap about the subject, one had the impression that he had exam-
ined virtualIy alI of the auxiliary languages that have been proposed since
Esperanto, and it was fascinating to hear him compare their strengths and
weaknesses.1 5 In a conversation about one such Ianguage - 1 believe it
was Lancelot Hogben's Interglossa - Carnap once told me that after
examining it, he had written the inventor, pointing out what he considered
its more important merits and defects; then he added, as an afterthought,
that he had written his letter in that language, to get some prac-
tice.
With that large branch of analytic philosophy which concentrates its
efforts on the analysis of ordinary language, Carnap shared the view that
various philosophical problems have their roots in peculiarities of or-
dinary Ianguage and in confusions about its use; his critique of speculative
metaphysics ilIustrates that point. In fact, Carnap held that in philosophi-
cal explication, the use of symbolic logic and of constructed languages
with explicit syntactical and semantical rules, though "the most elaborate
and efficient method, '" is advisable only in special cases, but not general-
ly," and he considered "the naturalist and the constructionist methods
[as] not necessarily competitive, but rather mutualIy complementary."16
This point is well illustrated, 1 think, by the confluence of ideas shown by
Carnap's physicalistic-behavioristic analysis of psychological concepts,
mentioned above, and the behavioristic approach taken by Ryle in The
Concept of MindP Both proposed dispositional construals of certain
psychological terms; and Ryle's criticism of "the ghost in the machi ne"
is very similar in tenor to Carnap's earlier criticism ofthe idea that behind
the physical structure corresponding to such psychological states as
excitement, there stands an occult property or power - namely, excitement
or the consciousness of it - which it self remains unknowable.1 8 Carnap
describes this idea as involving "a remarkable duplication ... : besides or
behind a state of affairs whose existence is empiricaIly determinable,
another,para/lel entity is assumed, whose existence is not determinable" ;19
and he rejects it after critical analysis.
But much of Carnap's work was devoted to philosophical issues con-
cerning disciplines such as mathematics and physics, which have highly
developed technical languages of their own; and it seems cIear that these
are largely beyond the reach of ordinary-Ianguage analysis. Moreover,
Carnap saw philosophy as aiming at general systematic elucidation rather
 RUDOLF CARNAP, LOGICAL EMPIRICIST 9

than, say, at more casuistic efforts to help individual fiies out of the
particular philosophical fiy bottles in which they are trapped.
Against the use of Carnap's logical-constructionist approach to philo-
sophical problems - especially those not concerning technical scientific
disciplines - it has been argued that ordinary natural languages are too
rich, complex, and fiexible to fit the rigid Procrustean bed of some for-
malized linguistic frame. This objection certainly has a point, but it is not
quite as telling as it may seem.
(1) First, Carnap was far from insisting on just one model of philo-
sophical bed - and that, a Procrustean one: he explicitly countenanced,
and called for, the construction and use ofphilosophical beds of different
kinds and degrees of commodiougness: that was the point of "his prin-
ciple of tolerance". 20
(2) Undeniably, the use of any precise type of model language - to
represent, say, the languageofphysics - involves considerable schematiza-
tion and simplification; but this is true, to some extent, also of any system
of general rules purporting to characterize grammatical features of a
naturallanguage.
(3) Even if philosophy were to limit itself, casuistically, to helping
individual fiies out of their particular fiy bottles, that philosophical activity
or therapy would still have to be informed by general principles. A fiy
trapped in a bottle or a man trapped in a maze might be led out with his
eyes bandaged: he would follow his leader blindly and would eventually
find himself outside, but he would not understand how he had been trapped
nor how he had eventually got out. But there is no analogue to this mode
of physical liberation in the case of a person philosophicaIly trapped in a
linguistic maze. The only way to lead him out is with his eyes open, as it
were: he has to be shown the way out, to use Wittgenstein's phrase; i.e., he
must come to understand what features of the trap Ied to his getting caught
in the first place and how to avoid the same fate in similar situations. And
this always requires insights of a general type, concerning, for example,
linguistic contexts of a certain kind, or language games of a certain sort,
whose general rules are then projected onto the particular case at hand.
(4) As noted above, the precisely characterized languages by reference
to which certain philosophical problems have been studied are often dis-
tinct1y simpler than those required for the purposes of science. For exam-
ple, Carnap's theory of reduction and confirmability, and his vast system
 10 CARL G. HEMPEL

of inductive logic are limited to languages with first-order logic, which

certainly does not suffice for the formulation of contemporary physical
theories. The same remark applies to various studies by Carnap and other
empiricists that deal with the structure and function of scientific theories;
with the qualitative concept of confirmation; with the logic of scientific
explanation, and so forth.
But Carnap often stressed that these studies are intended only as the
first stage in the development of more comprehensive theories, and that
the solutions they offer may well permit of extension to more complex
situations. 1 have heard him remark that EucIid might similarly have
encountered the objection that the basic concepts and principles of his
geometrical theory were much too simple to permit an adequate account
ofthe infinitely rich diversity of geometrical shapes exemplified in nature:
and indeed, the application of geometry to such complex configurations
had to await the development of analytic geometry and of the calculus;
but these powerful extensions did not show that EucIid's theory had been
futile in its simplicity - on the contrary, that theory had been the essentiaI
first step.
(5) Next, an analysis of a subject by reference to a relatively simple
kind ofIanguage may reveal unexpected and quite fundamental difficulties.
This is ilIustrated by the problems that Carnap encountered in attempting
to define dispositional and theoretical concepts; by the paradoxes of con-
firmation; by Goodman's riddle; by various problems concerning induc-
tive inference and probabilistic explanation, which arise even for lan-
guages of rather simple kinds.
(6) FinaIly, it can be of immense philosophical interest to find that what
seems to be a hopelessly difficult problem can in fact be solved for lan-
guages of certain precisely characterized kinds. One cIassical example is
Tarski's method of constructing, for languages with sharply specified
characteristics, a concept of truth that meets his condition of adequacy
and escapes the semantic paradoxes. Another striking iIlustrationis pro-
vided by one of the tasks Carnap tackled in his work on inductive logic,
namely, to give an explicit definition of inductive probability which as-
signs, to any sentence h in a given language, a definite numerical proba-
bility p(h,e) with respect to any non-contradictory 'evidence-sentence' e
in the same language; and in such a way, moreover, that the fuction p
thus defined has the mathematical properties of a conditional probability.
 RUDOLF ,CARNAP, LOUICAL EMPIRICIST 11

The demonstration that this can be done for languages of the kind studied
by Carnap is surely a result of great philosophical significance.
Let me condude with a few words about Carnap the teacher and the
man. 1 had the good fortune to be among Carnap's students in Vienna,
in Prague, and later once more in Chicago. What made me decide, despite
difficulties, to go to Vienna at least for a semester, was the powerful impact
of his AuJbau and of his shorter book, Scheinprobleme in der Philosophie,21
both of which had just recently been published. Carnap's courses differed
strikingly from those 1 had taken with some of the most eminent philo-
sophers in Germany. In my efforts to expand into a coherent text the
notes 1 had taken during their lectures, I often had a sense of practicing an
intricate ritual dance with words whose message was elusive or obscure.
In Carnap's teaching, all was order and light. Ris presentation and dis-
cussion were systematic, lucid, and incisive; he offered careful arguments
for his views and was willing to listen seriously to doubts and objections
from his students. In the traditional seminars, the students were often
expected to expound the ideas of this or that philosopher, while critical
comments on those ideas - not to mention the views of the professor in
charge - were not encouraged. The spirit that prevailed in Carnap's
seminars - and also in those of some of my other teachers, among them
Reichenbach, Schlick, and Waismann - was therefore wonderfully stim-
ulating and exhilarating.
1 have often wished that Carnap had reached a wider audience as a
teacher. Re tended to think of himself as lacking the skill to present
philosophical ideas to beginning students or to a more general interested
public. Yet at the German University in Prague he gave an introductory
course in the history of modern philosophy in which 1 recall hearing some
lectures that dealt most interestingly and lucidly with Kant's First Critique.
But Carnap did feeI most at home with advanced students; and in his
work with them, his strength as an inspiring teacher shone forth most
powerfully.
No one who carne to know Carnap even moderately well could fail
to sense his unusual human stature. Re was arnan entirely innocent of
pretentiousness, self-importance, or pettiness, with a genuine and guileless
openness to the ideas and problems of others, arnan moved by an intense
intellectual curiosity and a deep urge for conceptual order and understand-
ing, but equally inspir.ed by abiding social and humanitarian concerns,
 12 CARL G. HEMPEL

and unfalteringly faithful to his moral convictions. Carnap was a rare

and agreat man.

Princeton University

NOTES

1 Weltkreis-Verlag, Berlin-Schlachtensee 1928.

2 W. V. Quine, 'RusselI's Ontological Development', The Journal of Philosophy 63
(1966) 657-67; quotation from p. 667.
3 R. Carnap, The Logica/ Structure of the World and Pseudoprob/ems in Philosophy.
Translated by Rolf A. George. The University of California Press, Berkeley and Los
Angeles 1967; pp. viii-xi. Further reflections by Carnap on the Aufbau may be found
in Paul A. Schilpp (ed.), The Philosophy of Rudo/fCarnap, Open Court, La Salle, Illinois
1963, pp. 16-20; and in his response, pp. 944-7, to Goodman's essay 'The Significance
of Der Logische Aujbau der We!t', ibid., pp. 545-58.
4 In Phi/osophy of Science 3 (1936) 419-71 and 4 (1937) 1-40.
5 Cf. Carnap in Schilpp, op. cit,. p. 57.
6 See Carnap, 'Psychologie in physikalischer Sprache', Erkenntnis 3 (1932-33) 107-42;
an English translation, with a postscript added in 1957, appeared in A. J. Ayer (ed.),
Logica! Positivism, The Free Press, New York 1959, pp. 165-98.
This articJe affords an excellent example of Carnap's lucid and systematic manner of
presenting, iIIustrating, and supporting a philosophical thesis, of defending it against
objections, and of pointing out its implications for certain special topics - such as the
character of introspective psychology and the 'physicalization of graphology'.
7 See 'The Methodological Character of Theoretical Concepts'. In H. Feigl and M.
Scriven (eds.), Minnesota Studies in the Philosophy of Science, voI. 1, University of
Minnesota Press, Minneapolis, 1956, pp. 38-76.
8 Carnap noted later that the methods he had suggested in the Aujbau for the ascent
from the subjective-phenomenalistic realm to tht: intersubjective realm of physical
concepts could not always yield explicit definitions and were, in fact, more cJose\y
akin to the introduction of theoretical concepts through postulates and correspondence
rules. (Cf. Carnap in Schilpp, op. cit., p. 19 and in the preface to The Logica! Structure
of the World and Pseudoproblems in Philosophy, pp. viii-ix.)
9 See 'The Methodological Character of Theoretical Concepts', Sections VI, VII,
VIU.
10 In Schilpp, op. cit., pp. 80-1.
11 In Schilpp, loc. cit., p. 964. - The proposed method of analysis is set forth in more
detail in Chapter 28 of Cal nap, The Philosophical Foundations ofPhysics (ed. by Martin
Gardner), Basic Books, New York, London 1966. At the end of that chapter, too,
Carnap speaks of his "many years of searching" for a satisfactory answer, and he
concludes: "No difficulties have yet been discovered in this approach. 1 am now confi-
dent that there is a solution and that, if difficultit:s appear, it will be possible to over-
come them." (pp. 272-3)
12 The metaphor occurs in Neurath's articJe 'Protokollsaetze', Erkenntnis 3 (1932-3)
204-14; English translation in Ayer, op. cit., pp. 199-208.
13 In a similar vein, Neurath stressed that scientific procedure "does not depend upon
'exactness' but only upon the permanence of scientific criticism. New ideas of scientific
 RUDOLF CARNAP, LOGICAL EMPIRICIST 13

importance start mostly with vague and sometimes queer explanations; they become
clearer and clearer, but the theories which foIIow wiII stand in time before the door with
aII their new vagueness and queerness." ('Unified Science as Encyclopedic Integration'
in International Encyc/opedia of Unified Science, Volume 1, Number 1. The University
of Chicago Press, Chicago 1938, pp. 1-27; quotation from p. 21. - This essay, with its
markedly pragmatic orientation, reflecting Neurath's historical and sociological in-
terests, contrasts instructively with Carnap's contribution to the same booklet (pp. 42-
62); it is entitled 'Logical Foundations of the Unity of Science;' and takes a purely
logical and systematic approach to its subject.)
14 Harvard University Press, Cambridge, Massachusetts, 1942; p. vi.
15 Section 11 (pp. 67-71) of Carnap's autobiography [in Schilpp, op. cit.], which
deals with the subject of language planning, reflects very clearly the various glOunds
for his interest in international auxiliary languages.
16 Carnap in Schilpp, op. cit., p. 936 and p. 940. These statements occur in Carnap's
response to P. F. Strawson's article 'Carnap's Views on Constructed Systems versus
Natural Languages in Analytic Philosophy' (in Schilpp, op. cit., pp. 503-18).
17 Hutchinson's University Library, London 1949.
18 See 'Psychology in Physical Language', in Ayer, op. cit., p. 173.
19 Ibid., p. 174.
20 Carnap first pronounced this prmciple in Die Logische Syntax der Sprache, Julius
Springer, Vienna 1934. (English translation, The Logica! Syntax of Language, Kegan
Paul; Trench, Trubner & Co., London 1937). He reaffirmed it, in various forms, in
later writings. The important role it played in his thinking is reflected in the repeated
references to it in his 'Intellectual Autobiography' in Schilpp, op. cit.
21 Weltkreis-Verlag, Berlin-Schlachtensee 1928. The volume mentioned as the first
item in note 3 above contains an English translation of this short work as well as of
the Aufbau.
 ANDERS WEDBERG

HOW CARNAP BUILT THE WORLD IN 1928

PREFACE

In his work Der logische Aujbau der Welt, first published in 1928, Carnap
appeared as a philosopher with ambitions no smaller than those ofthe
great system builders. 1 In less than 300 pages he hoped to solve practical-
ly alI the c1assical problems of philosophy, and he thought he could do so
with something of the logical rigor of Principia Mathematica. On more
than one score, Carnap's work invities comparison with Wittgenstein's
Tractatus logico-philosophicus, published seven years earlier. Carnap's
ideas in the Atifbau are in part influenced by Wittgenstein's in the Trac-
ta tus. The world models of both are sketched by reference to the coor-
dinate system of the logic of Frege and Russell. Whereas Wittgenstein
did not wish to say anything about the world beyond what could be said
a priori, Carnap aimed at a kind of inventory of the total contents of the
world. Whereas Wittgenstein speaks in aphorisms, Carnap is the peda-
gogical teacher. The view that the Tractatus is the expression of mystical
vision and the Aujbau that of rationalistic systematics no doubt contains
a measure oftruth. In contrast with the Tractatus, the Aujbau may indeed
appear as a model of c1arity. The more one studies the Aujbau, however,
the more this contrast tends to vanish, the more obscure and ambiguous
the argument ofthis work seems. Since 1 am not aware ofthe existence of
any definitive1y c1arifying analysis of this argument, 1 shall here make a
second attempt to analyze it. 2 My exegesis has taken a critical turn, but its
primary aim is exegesis, not criticism. What is only vaguely hinted at in
the Aujbau 1 have tried to state explicitly. My return to the Aujbau bears
witness to the deep and disturbing impression that the first reading of
this remarkable work made on my youthful mind about fourty years ago.
Roughly speaking, the Aujbau is an attempt to explain certain features
of the World or Reality (the World, or Reality, of Science), or of our
knowledge thereof, by studying a model The world model of the Aujbau

Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 15-53. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
16 ANDERS WEDBERG

is 'phenomenalistic' and even 'solipsistic'. Its basic concepts belong to in-

trospective psychology and, in particular, to the phenomenology of per-
ception. In Section I I shall try to explain the psychological contents of
the model- as far as that is possible, for the Aujbau is lamentably reticent
on this subject. In Section II I shall state some of the formal properties
that the Aujbau seems to ascribe to the model, and in Section III I shall
discuss some properties ofthe languages (for the description ofthe model)
which are used, or hinted at, in the Aujbau. In Section IV I shall consider
the question of the relation between the model and the world. The corre-
lation between the model and the world is intended to be expressed by a
chain of definitions the beginning of which is formulated in the Aujbau.
Hence, Section IV deals with this chain. In Section V I shall make a case
study of the third definition of the chain, that of 'similarity circ1e' (Xhn-
lichkeitskreis). In the light of the previous analyses, I shall in Section VI
reconsider my interpretation of the Aujbau as the presentation of a world
model.
A reader who comes to this essay from the study of the Atifbau may
understandably feeI bewildered when trying to look at the work through
the eyes of my paper. Matters that play a prominent role in the pages of
the Aujbau are not even mentioned here, and things which in the Aujbau
are said in passing, or only implied by what is said, are here considered
as major themes. The reason for this is that I have tried to get at what
seems to me logically fundamental in the Aujbau's web of ideas. Many
ideas are also stated here without sufficient mention of the various
reservations and additiona1 reflections with which they are put forth by
Carnap. This is due simply to a wish to keep the paper short. The in parts
somewhat overpedagogical manner in which the paper is written is ex-
plained by the fact that it was originally composed for a seminar on
Carnap at the University of Stockholm. It is hoped that the perspective on
the Aujbau which the paper expounds will be of use to serious students of
this many-faceted book and, in particular, will help them to articulate
their own perspectives on it.
My debt to Nelson Goodman's penetrating study of the Aujbau in
The Structure of Appearance is, I believe, greater than I am quite con-
scious of myself. What I say in Section V adds little to what Goodman has
already said.
 HOW CARNAP BUILT THE WORLD IN 1928 17

1. THE AUTOPSYCHOLOGICAL BASIS OF THE WORLD MODEL

1.1. The Autopsychological Basis

The world model of the Aujbau is built in accordance with Russell's
simple theory of types. Among its elements there accordingly are certain
'basic elements' (the 'individuals' of Principia Mathematica) which form
the basis for a hierarchy of sets and re1ations. When constructing a world
model we are, on Carnap's view, free to choose the basic elements in
several different ways. In the Aujbau he picks them from the 'autopsy-
chological' realm, ie. his own mentallife. Each reader of the Aujbau is
entitled to think instead of his mentallife. In this sense the Aujbau recog-
nizes a large number of distinct, though similar, world models. Through
a certain procedure, say P, Carnap abstracts from his mentallife a model,
say MCarnap, which he assumes to bear a correlation, say C, with the world
(or, perhaps, with Carnap's knowledge thereoQ. It may then be asked:
For what conscious beings A is Carnap prepared to assume that, if a
model MA is obtained through P from A's mentallife, this model MA
will bear the same relation C to the world (or perhaps to A's knowledge
of the world)? In the preface to the second edition of the Aujbau (1961),
Carnap says that his theory claims validity for "any person (provided that
he has normal senses and that circumstances are not 'particularly unfavor-
able', ",), hence ... [independence] of the accidental selection of his ob-
servations and the course of his wanderings through the world." 3

1.2. The Elementary Experiences and the Autopsychological Stream

The basic elements are what Carnap calls (autopsychological) 'elementary
experiences'. An elementary experience is the totality of everything that a
subject experiences at a given moment of time. My present elementary
experience comprises not only the visual, auditive, olfactory, etc. phenom-
ena that 1 am now conscious of but also my bodily sensations, my feelings
and my thoughts - if thoughts are acknowledged as a separate category
of mental events. 4
The Aujbau hardly contains any well-considered philosophy of mind. 5
It seems, however, that Carnap assumes a sense-datum theory of some
kind. When 1 look at the Royal Palace in Stockholm, the palace itself is
not, on Carnap's view, a part of my elementary experience. On the other
hand, he seems to consider a visual image of the palace as such a part.
18 ANDERS WEDBERG

(In the Aujbau the visual field is said to be spatially two-dimensional,

something the palace certainly is not.) Aiso with respect to our ideas and
thoughts, the Aujbau apparently makes a distinction between their 'ob-
jects' and their 'contents'. Carnap no doubt does not consider Plato in
person to be a part of the elementary experience occurring in someone
who is thinking of Plato. 6
A person's mentallife is taken in the Aujbau to be a continuous stream
which is not supposed to be composed of definite, discrete elements or to
be divisible into elementary experiences in a unique fashion. 7 It is sup-
posed, however, that we can attend to, and make statements about, par-
ticular 'places' in the stream. These places are called 'elementary expe-
riences'.8 In an attempt to interpret these assertions we ought, 1 think, to
distinguish between two possibilities. Either (i) elementary experiences
should be conceived of as strictly momentary cross-sections of the mental
stream. Or (ii) they should be taken to be segments of some, even if very
short, duration. In case (i) it is difficult to understand why we cannot
speak, in an absolute sense, of the set of aU elementary experiences (cross-
sections). This set is then unique - but perhaps not discrete. The Aujbau
contains the thought that the number of elementary experiences is finite.
Possibly Carnap wishes to say that a finite subset, suited to form the
basis for his wor1d model, can be chosen in several ways from the set of
aU elementary experiences. In case (ii) it is obviously conceivable that the
mental stream, like a loaf of bread, can be cut into slices in more than one
way.
At the same time (in §67) that Carnap is inclined to consider elementary
experiences as abstractions of some kind (as mere 'places' in the stream),
he maintains that they are what is "primarily given" both psychologically
and epistemologicaUy.
The set of an elementary experiences occurring in one's own mental
life is symbolically designated by eri. When 1 wish to show that the owner
is a person A, 1 shall write eriA• When speaking, in what follows, of
elementary experiences 1 shall always assume that we are dealing with
experiences of one and the same person unless the contrary is explicitly
stated.

13. Quality Points

The vis ual field that is a part of my present elementary experience is
 HOW CARNAP BUILT THE WORLD IN 1928 19

assumed in the Aujbau to be a two-dimensional spatial configuration of

color points. Each color point is characterized by its two 'local signs',
which together determine its place in the vis ual field, and by its hue,
saturation, and brightness, which together determine its place in the color
solid. 9 In this sense, the color points occurring in the elementary expe-
riences can be thought of as points in a five-dimensional space (the word
'space' here understood in an abstract sense), which may be called the
color space. 10 The color points occurring in a given elementary experience
will then be a certain subset of the color space.
In that space any two points whatsoever have a mutual distance. (Row
the distance is to be measured is left unexplained in the Aujbau.) If the
distance is less than or equal to a certain standard distance, the two points
are said to be similar (ăhnlich).ll This notion of similarity thus involves
the notion of a standard distance. Carnap seems to consider this distance
as small, but otherwise he says nothing about it. (The question as to how
the choice of standard distance affects the constructions of the Atifbau is
disturbing and impossible to answer.) If the points p and p' are similar
and if p occurs in the elementary experience e and p' in e', then e and e'
are said to be part similar (teilăhnlich).12
Color points are a species of the genus quality points. On one level of
thought, in one sense, Carnap considers an elementary experience as a
complex of a number of quality points of various kinds. One kind are
visual color points, another auditive qualities, a third olfactory, etc. Re is
also prepared to recognize emotive, volitive, and intellectual qualities
(quality points). The word 'sense c1ass' is used to denote such a kind of
quality.13 For each sense c1ass, just as for vision, notions of distance and
standard distance are postulated. Two points from the same sense c1ass
are said to be similar if their distance is less than or equal to the standard
distance of the c1ass. The notion of part similarity between elementary
experiences is generalized in an analogous manner. 14

IA. Recollection of Part Similarity

What Carnap calls "recollection of part similarity" (Teilăhnlichkeitserin­
nerung) or, symbolically, Er is a dyadic relation between elementary ex-
periences of one person. (When we wish to show that we are dealing with
the recollection of the person A, it is convenient to write ErA-) The state
of affairs xEry is said to obtain if and only if x and y are found to be
20 ANDERS WEDBERG

part similar when y is compared with a memory image of x. 15 x Er y thus

seems to involve that the four phenomena:
the elementary experience y,
the memory image of the earlier elementary experience x,
the comparison between y and the image of x,
the insight that x and yare part similar,
occur simultaneously. Since y is the totality of what is experienced when
y occurs, the latter three phenomena must, it seems, be contained in y.
How this is to be understood is something of a mystery, bypassed in
silence in the Aujbau.
The Aujbau assumes (in the form of a definition):
(1.4.1) The elementary experiences x and y are part similar if and
only if xEry, or yErx, or x= y.16

If the assumption is combined with the evident statements:

(1.4.2) If x Er y, then x precedes y in time,

(1.4.3) Any two arbitrarily chosen elementary experiences x and y
fulfill exactly one of the three conditions: x precedes y in time,
y precedes x in time, x = y,l 8

we obtain the conclusion:

(1.4.4) xEr y if and only if x is part similar to y and x precedes y in

time. 19

How is this assumption, which is thus implicitly made in the Aujbau, to

be understood? (Or how is its critical premise (1.4.1) to be understood?)
Was the author of the Aujbau ready to defend the empirical hypothesis
that whenever an earlier experience is part similar to a later one, then the
latter is associated with (contains) a corresponding recollection? Or is
(1.4.4) a convention concerning what finite sets of cross sections through
Carnap's (A's) mental stream may be used as domains of basic elements?
Another assumption, which is also made in the Aujbau (likewise in the
form of a definition), is this:

(1.4.5) x is an elementary experience if and only if there is an ele-

mentary experience y such that x Er y or y Er x.
 HOW CARNAP BUILT THE WORLD IN 1928 21

The methodological status of this assumption is as uncertain as that of

(1.4.4).
1.5. The Impossibility of Subjecting Elementary Experiences to a Proper
Analysis
The Aujbau teaches that "the elementary experiences are essentially un-
analyzable units" or that they are "essentially" inaccessible to "proper
analysis." Proper analysis of an object A consists in distinguishing "parts"
in A, or attending to "properties" or "characteristics" of A. The essential
unanalyzability of the elementary experiences thus means that (i) they
have no "parts" and (ii) they have no "properties" or "characteristics."
One e1ementary experience is distinguished from another only by their
different positions in the net of relations where all elementary experiences
are placed Their unanalyzability is a reason why the elementary expe-
riences are chosen to perform the role of basic elements in the Aujbau, or
a confirmation of the correctness of the choice. Priority, psychological
and epistemological, is according to Carnap accompanied by unanalyza-
bility.21
To the essential unanalyzability of elementary experiences corresponds,
it is said, the impossibility to "construct" (konstituieren) Of "properly
analyze" them within the formal languages adopted in the Aujbau. 22
This impossibility depends on the following circumstances: (a) these for-
mallanguages conform to the theory of types; (b) the basic elements of a
type-theoreticallanguagecannot, on Carnap's view, be conceived as sets
or relations; (c) only sets and relations can, properly speaking, be con-
structed in such a language; and, finally, (d) the elementary experiences
are here playing the role of basic elements.
Let us take a c10ser look at assumption (c) which is motivated by
Carnap. To 'construct' an object A in a type-theoretical language is,
according to Carnap, roughly speaking the same as to conceive of A as a
set (relation) and to indicate what entities (ordered n-tuples of entities)
are to belong to the extension of A. If A is taken to be a set, the construc-
tion of A consists in deciding what its elements are, say a, b, c, ... :
A = {a, b, c, ... } .
A type-theoreticallanguage is said to allow also a converse procedure
which is called quasi-analysis. In a quasi-analysis of A we distinguish
22 ANDERS WEDBERG

(some OQ the sets of which A is an element. The basic elements of a type-

theoretical language can be quasi-analyzed in that language but not
constructed or properly analyzed. 23
One may well ask: (i) Is it the absence of 'parts' that is refiected in the
impossibility to construct or properly analyze the elementaryexperiences?
Against this interpretation speaks the fact that according to Carnap sets
do not "consist of their elements as a whole consists of its parts."24
(That an object A is taken to be a set with elements, in the formal lan-
guages of the Aujbau, thus does not imply that A is taken to have parts.
That elementary experiences are not treated as sets in these languages
does not either imply that they are excluded from the possibility of
having parts.) Or (ii) is it the absence of 'properties' that is refiected in the
impossibility of construction? That relation 'set-element', which con-
struction centers upon, has, however, nothing in common with the rela-
tion 'object-property' as conceived of by Carnap. Rather, the latter rela-
tion is parallel to the relation 'element-set', upon which the so-called
quasi-analysis rests. A 'property', it is said, is a monadic propositional
function (Aussagejunktion), and the extension of such a function is what
is called a 'set'. 25
As Nelson Goodman has forcefully argued, one can incorporate in a
formal language a predicate for the 'whole-part' relation with which
'proper analysis' deals, according to Carnap, or some equivalent vocab-
ulary.26
Carnap's ideas about the unanalyzability of the basic elements seem to
be an echo of Russell's doctrine of the simplicity of individuals, which
was so prominent a part of his 'logical atomism'. 27 Carnap himself refers,
1 think erroneously, to the findings of Gestalt psychology as confirming
his view. 28

1.6. Founded Relations

Some relations have a content that we can immediately experience, where-
as others are arbitrary logical constructions. Carnap calls the former
'founded relations'. The set of all dyadic founded relations between ele-
mentary experiences is designated by the symboljund. An assumption in
the Aujbau is:

(1.6.1) Er belongs tojund. 29

 HOW CARN AP BUILT THE WORLD IN 1928 23

I. 7. The Double Role of the Psychological Theory

The psychological theory which was sketched above in (I.IHI.4) and
which to my knowledge was Carnap's own invention plays a double role
in the Aujbau. (a) The two basic notions in the Aujbau's formal system,
'elementary experience' (erQ and 'recollection of part similarity' (Er), ob-
tain their meaning in and through this theory. (b) The Aujbau presents
the beginning of a chain of definitions which is supposed to be, in prin-
ciple, extendible so as to define any legitimate scientific concept. The
eighteen definitions actually stated in the Aujbau are all concerned with
concepts belonging to the present psychological theory.30 Thus the theory
is also a touchstone for the Atifbau's program of definition.
From the point ofview ofpsychology, also the psychology ofthe 1920's,
Carnap's phenomenological theory is of course terribly crude in both of
its opposite aspects, viz. in the analysis of perception as an order of quality
points and in the view that the perceptual field lacks 'parts' as well as
distinguishing 'characteristics'. Carnap's reference to Gestalt psychology
seems especially unfortunate since his theory does not take into account
any of the many phenomena that this school of psychology has brought
to light.

Il. FORMAL PROPER TIES OF THE WORLD MODEL

Let us call the wor1d model of the Aujbau M. 31 1 shall try to summarize
some of the more important formal properties that the Aujbau, explicitly
or implicitly, assigns to M.
The framework within which all Carnap's reasonings in the Aujbau are
conducted is Russell's theory of types in its non-ramified, purely exten-
sional form. 32 1 shall here take the liberty of using a set-theoretical termin-
ology although that is not done in the Aujbau. This deviation from the
language of the Aujbau is materially irrelevant.
The formal property of M to which 1 shall first call attention is its type-
theoretical structure. For the sake of brevity, let the symbolic expression
&(T), as usual, denote the power set of T, ie. the set of all subsets of T.
By Tl X ... X 1'" 1 understand the Cartesian product of Tl , ... , 1'", i.e. the set
of all ordered n-tuples (Xl>"" xn> such that Xl is an element of Tl , .•. , Xnan
element of 1'". Since the general notion of 'the type hierarchy based on
24 ANDERS WEDBERG

the domain (of basic elements) Y' will be useful later, 1 introduce it
already here:
(110) ~(Y)= the least set of 'types' S such that
(i) Y is a type in S,
(ii) if X is a type in S, then 9(X) is a type in S,
(iii) if X 1, ... , X n are types in S, then 9(X 1 X ... X X n ) is a
type in S.
Using this general notion we may state the type-theoretical structure of
M by the simple formula:

(II.l) M=~(er/).

According to standard set theory there is one and only one empty set
which thus will be an element of any type in M above erI. On the other
hand, Russell's type theory assumes the existence of a separate empty set
in each type above erI. The apparent incongruity between theset-theoret-
ical description (111) of M and Carnap's own type-theoretical outlook in
the Aujbau need not, 1 think, worry us. In a type-theoretical language,
such as those Carnap uses to talk about M, the assertion 'I\Ti' I\r',
where /\T is the empty set of type T and /\r that of another type T', is
inadmissible. That I\T is distinct from I\r, hence, cannot be asserted.
Nor can anything be asserted (denied) of I\T that is denied (asserted) of
I\r. We are thus free to interpret '/\/ and 'I\r' as two distinct names
of the one and only empty set. Similar considerations apply to such
derived sets as {I\T} and {I\r}.
In what follows, let us say that an entity belongs to (is an element in)
M if it is an element of a type in M.
Another formal property of M is this:
(II. 2) eri is finite. 33
Besides eri the two notions Er andfund also play a special role in M. We
assign them to their types:
(113) Er is an element of 9(erl x eri),
(II.4) fund is an element of 9(9(erl x eri)).
There are three further assumptions of a formal nature which seem to be
made in the Aujbau and which will now be codified.
 HOW CARNAP BUILT THE WORLD IN 1928 25

(115) The structure <eri, Er) has no automorphism save identity.

In other words, there is no one-to-one mapping f of eri upon eri, save the
identity mapping f (x) = x, such that x Er y holds if and only if f (x) Er f(Y~
This is a precise formulation of a presupposition for the 'relational de-
scription' of M whose possibility is asserted in the Aujbau. According to
this thesis of the Aujbau, each object can be uniquely characterized by its
place in a relation-net. For the Aujbau the relevant net is the Er net. 34
In Section 16 we have already stated the assumption:

(116) Er is an element infund.

(Of course, in view of (11.6), the propositions (11.3) and (11.4) are equiva-
lent.)
Finally, we have to record the following assumption which is implicitly
made in the Aujbau:
(11.7) There is no relation R in fund and distinct from Er such that
the structure <eri, R) is isomorphic with <eri, Er).

In other words, given that Ris an element infund and R#Er, there does
not exist any one-to-one mapping f of eri onto eri such that xRy if and
only if f (x) Er f (y). This signifies that Er can be characterized as the only
relation in fund with such and such a structure. Combining this idea with
the previous assumptions (Ila), (11.5), and (11.6) we find that each element
in eri can be characterized as 'that object which haS such and such a place
in the net of that relation in fund which has such and such as structure'. In
short, each element of eri can be uniquely characterized in terms of fund
and well-known logical notions. Together with (11.1) and (11.2), (115)-
(11.7) are a precise formulation of the 'structuralism' (thesis of the possi-
bility of a 'structural' description of M) which the Aujbau defends. 35

III. LANGUAGES FOR DESCRIBING THE WORLD MODEL

1111. The Type- Theoreticai Structure

Carnap speaks about M in two distinct ways, let us say 'presystematically'
and 'systematically'. The ideas recorded in Section I belong to the pre-
systematic talk. So does also the assumption (11.1), whereas the assump-
tions (1I.2HII.7) can be restated systematically. The systematic talking
26 ANDERS WEDBERG

about M is done with the aid of certain formalized languageso We shall

take a look at five such languages, LcLso The language within which the
Aujbau's chain of definitions is expressed is L 2 or a language that essen-
tially coincides with L 2 Carnap indicates how L 2 may be translated into
0

L 3 The language Ll is not used in the Aujbau, but the Aujbau contains
0

reasonings which apparently demand that M be described through Ll or

that part of Ll which 1 shall call Lso The language L 4 is not used in the
Aujbau either, but L 4 is of interest as one way of giving more tangible
shape to certain ideas which are hinted at in the Aujbauo The languages
LCL3 are constructed according to the principles of Russell's non-rami-
fied, extensionally interpreted theory of typeso The essence of type-
theoretical grammar is a restriction upon the formation of atomic
sentenceso We think of each name and variable as provided with an index
showing its typeo A name of type T designates an element of T. A variable
of type T has T for its rangeo An atomic sentence of the form F(x) is
admissible if and only if F is of the type 9(T) when x is of type T. An
atomic sentence ofthe form F(x l , ooo, x n) is admissible if and only if Fis of
type 9 (Tl X ooox T,,) when Xl is of type Tl , ooo, X n of type T"o
As a consequence of (112) Carnap ought to eliminate the axiom of
infinity from his languages (or their logics), and he has no use for axioms
of choice (the 'multiplicative axioms' of Principia Mathematica)o When
the axiom of infinity is replaced by proposition (112) it becomes a pro-
blem how, in his systematic talk about M, Carnap can have access to the
theory of real numbers on which he relies so heavilyo But the Aujbau is
full of such riddleso

II120 New Primitive Signs in Ll' L 2 , and L3

The languages LCL3 are all applications and extensions of the language
of Principia Mathematicao Whereas Principia leaves it undecided what
objects belong to its domain of individuals, LCL3 identify the domain
with erlo The new primitive signs in Ll are exactly one name for each
element in eri, say 'el', 000' 'e q', and a predicate 'Er' designating Ero In
L 2 'Er' is the only new primitive sign, in L3 'fund' designatingfundo
A description of M through L 2 is what Carnap calls a "material rela-
tional descriptiono" It is material since the new primitive term 'Er' is a
non-Iogical signo A description of M through L3 is what Carnap calls a
"(purely) structural", "(purely) logical," or "(purely) formal" description,
 HOW CARN AP BUILT THE WORLD IN 1928 27

i.e., a description exc1usively in logical terms. In doing so he c1assifies

'funtr as a logical sign. 36

UI.3. Definability and N ameability in Ll' L2' and L3

"According to the central thesis of the construction theory, it is in prin-
ciple possible to fit each scientific object (or concept) into the construc-
tional system."37 This statement (as welI as its many equivalents in the
Atifbau) can probably be interpreted in several ways. Among other things
it amounts, 1 think, to the assertion that each entity in M can be defined in
the languages Ll' L2' and L3' Let us say that an entity is definable in one
ofthese languages ifthe language contains a sentential function with a sin-
gle free variable which is satisfied by this entity and by it only. It is easily
seen that each element of M is definable in Ll' An individual ei is defined
by the function x = ei' Since eri is finite, each type in M is finite. If now
each element in type T is definable, then also each subset of T (element in
the power set of T) is definable. That subset of T whose elements are
defined by the functions gl (y), ... , gk(y) is defined by the function
(UI.3.1) Vy(X(y)-(gl (y) v .. · v gk(y))) ,
where X is the only free variable. 38 In the same way it is seen that each
subset of T1 x ... X T" is definable if each element ofeach type T1 , ... , T" is
definable.
Let us say that an object is nameable in a language if the language con-
tains a name of the object. Ll contains a primitive name for each indivi-
dual. If an iota operator is present in Lt. the set defined by (III.3.I) has
in Ll the name:
(IU.3.2) IX (III.3. 1)
meaning 'the X such that (UI.3.1)'. If an abstraction operator is available
the same set may obtain a name such as:

I
(II!. 3. 3) {y gl (y) v .. · v gk(y)} ,
meaning 'the set of alI y's such that gl (y) v ... V gk(y)'. Given a suitable
iota or abstraction operator, alI elements in M now become nameable in
Ll in similar fashion. ,
With the aid of (II.2), (II.5HII.7), it can be shown that each individu-
al in M is definable also in L 2 and in L3.39 Hence, by the above argu-
28 ANDERS WEDBERG

ment, alI elements in Mare definable in these languages. Given a suitable

iota or abstraction operator, each element is also nameable in them.

UI.4. Axiomatizabilityof Truth in L1' L2' and L3

AII true sentences of L 1 can be axiomatized, and the same holds for L 2
and L 3 •40
That the truths of LCL3 can be axiomatized does not imply that the
author of the Aujbau considered his work as an axiomatic system. When
we speak of an axiomatic system we usualIy have in mind a situation where
somebody at a certain moment of time lists all the axioms or at least
recursively specifies them in syntactical terms. To do that is humanly
impossible with respect to the languages LCL3 if eri is to be alI the ele-
mentary experiences that someone has during his lifetime. Certainly,
Carnap never claimed that it was possibleY
The truths of LCL3 are also decidable (in the abstract meta-Iogical
sense.) Can Carnap's allusion to "the proud thesis" ofthe solubility of all
scientific problems be understood as an alIusion, among other things, to
this decidability?42

III. 5. A List Language L 4

One of the central theses of the Aujbau is the rejection of 'essence' ques-
tions as 'metaphysical'. Carnap sometimes expresses his own 'antimeta-
physical' view by saying that sets and relations are given only through
their membership lists, or that only these lists are of relevance to science. 43
Carnap also makes the 'fictive' assumption that the person A from whose
mentallife the model MA is abstracted constructs step by step the ele-
ments of MA by bringing together their members. 44 These and similar
ideas in the Aujbau make it natural to think of a language L 4 which de-
signates sets and relations just by lists of their members, and which there-
by mirrors the stepwise construction. Although Carnap does not, for
good reasons, elaborate any such language, it is of some interest to con-
sider what it could look like.
Since the number of basic elements in M is finite, each element in M is
a finite construction on their basis. A non-empty set ofbasic elements can
be denoted by the name:
 HOW CARNAP BUILT THE WORLD IN 1928 29

where the names ei.' ... , ejp denote the elements of the set. If we decide
upon an alphabetical order among the names and demand that they occur
in this order in (III. 5. 1), the notation will be unique (admit no synonyms).
A non-empty dyadic relation among the elements can analogously be
given the name:
(III.5.2) {(erll , erI)' ... , (ersl ' ers ) }
where the name lists all pairs of elements between which the relation
holds. By relying on alphabetical ordering, this notation can likewise be
rendered unique, of course. The notation can obviously be extended so as
to cover aH non-empty sets and relations in M. An empty set or relation
may be designated by the symbol { }.
To make the notation more natural we can slightly change the structure
of M. We can replace item (iii) in (II. 1) by:
(iii +) If X 1, ... , X n are types in S, the Cartesian product XIX ... X X n
is a type in S.
Thus finite sequences of elements in M will now themselves be considered
as e1ements in M. Hence expressions like '(ei' ej)' will also name elements
inM.
We do not include variables and quantification in L4' and we assume
that the theory of types, as fas as it is applicable, holds for L 4 . The atomic
sentences of L 4 will have the form A(B), where A and Bare names of
appropriately interrelated types. From these atomic sentences molecular
sentences are built in the usual way.
The logic of L 4 can be made diminutive. In addition to sententiallogic,
L 4 needs only an 'atomic' logic according to which A(B) is a theorem
whenever A is a name of the form {Bl' ... , Bn}, n~ 1, and B is identical
with Bi for some i::::; n, and IA (B) is a theorem whenever A and Bare not
so related. This logic is an axiomatization of aH truths in L4' and in this
sense L 4 is purely 'analytic' and triviaHy decidable.
Through definitions aH the means of expression occurring in LI' L 2
and L3' but not in L 4 can in principle be introduced in L 4 • For instance,
quantification over variables can be introduced thus:

where nI' ... , nk are exactly aH the names ofthe same type as the variable x.
30 ANDERS WEDBERG

There is no need, 1 think, to comment on the human impossibility of

actually using a language like L4' or on the inability of such a language to
convey information beyond set-theoretical trivialities.

III. 6. A 'Molecular' Language Ls

Only 'proper names', i.e., names of particular elementary experiences in
the context of the Aujbau, are in the Aujbau accorded "a relatively inde-
pendent meaning [Bedeutung]." Only they can "in the original use of
signs" hold the place of the subject of a sentence. "However, it proved
advantageous to admit into the subject position also signs for general
objects.... This improper use, however, is permissible only when a trans-
formation into proper use is possible, i.e., if the sentence can be translated
into one or more sentences which have only proper names in their subject
positions." 4S It is further said that any admissible sentence can be trans-
lated into a sentence which "expresses a definite ... state of affairs relative
to the basic relations. In keeping with the tenets of construction theory, we
presuppose that it is in principle possible to recognize whether or not a
given basic relation holds between two given elementary experiences.
Now, the state of affairs in question is composed ofnothing but particular
relation statements ... " 46
The 'proper' formal language for speaking about M would then seem
to be a language whose whole vocabulary consists of the names of
elementary experiences ('el" ... ' 'e q'), the predicate 'Er', and the truth-
functional connectives. This language, which is a part of Lb 1 shall call Ls.
Carnap seems to maintain that the languages Ll' L2' and L3 can in
some sense be translated into Ls. 47 However, no rules of such translation
are given in the Aujbau, nor does Carnap state what properties of senten-
ces he expects to remain invariant under the translation. No doubt he
required of a translation that the sentences which it carries into each other
shall have the same truth-value. But this can hardly be all that a satisfac-
tory method should accomplish. One can only make guesses as to what
further demands he had in mind. 48
In The Structure of Appearance Goodman describes the Aujbau as a
"platonistic" system. 49 He is thinking of such languages as Ll' L2' or L3.
If we assume that Carnap intended a language like Ls to be the basic
language of the Aujbau, the description is misleading. Carnap's require-
ment that an "improper" language which admits also signs other than
 HOW CARNAP BUILT THE WORLD IN 1928 31

proper names into subject position shall be translatable into a "proper"

language has an obvious kinship with the well-known 'nominalistic' re-
duction program proposed by Goodman and Quine. 50

IV. THE DEFINITIONS OF THE AUFBAU

IV.I. The Aufbau as a System of Definitions

The kernel of the Aujbau is the system of definitions whose fust beginning
(the first 18 definitions) is presented in the book but which is intended to
admit extension so that each legitimate concept of science becomes de-
fined therein. The system which we may call D is a sequence:
(IV.l.1) D=(d 1 ,d2 ,d3 , ••• )
where each definition di has the form:
(IV. 1.2) ai=DCbi.
Definiens bj may contain a definiendum ai only if j> i. Whenever this
is the case, we can think of ai as being replaced by bi in bj. The language
which Carnap uses for his definitions is L 2 • After the replacement just
mentioned all the definientia of D will be expressed in the primitive voca-
bulary of L 2 . They all refer to the model M. The definienda can of course
be construed as mere abbreviations for their definientia, but one then
misses the philosophical point of the Aujbau. On the whole the definienda
are supposed to stand for recognized scientific concepts and to be gathered
from informal scientific language. The definitions are intended to esta-
blish correspondences (of some kind or other) between the world of
science and the model M. Let us give to the (never completed) totality of
expressions which is to be definable in D the name V. Thus, the following
assumption is made in the Aujbau:

(IV. 1.3) For each expression a in V it holds that D can be extended so

that a is defined in D.
The system D gives rise to the following questions (among others):

(IV. 1.4) How is V to be characterized?

(IV. 1.5) If a belongs to V and b is an expression from L 2 , how should
a and b be related semantically in order for the definition
32 ANDERS WEDBERG

a = Of b to be admissible into D?
(IV. 1.6) What significance, beyond linguistic convenience, does the
order of the definitions have?
The Aujbau's answer to question (IV.1.4) is that V is the totality of
'scientific' concepts. 51 Simultaneously, however, it is said that most of the
concepts occurring in science are illegitimate (since they do not conform
to the theory of logical types).52 Sometimes even definability in D seems
to be raised to a criterion of the legitimacy of scientific concepts. 53 It is
also said that V is such that alI 'scientific statements', or alI 'empirical
scientific statements', can be translated by means of D into the language
L 2• 54
The Aujbau has a seemingly straightforward official answer to ques-
tion (IV.1.5), viz extensional identity. In reality, however, the Aujbau's
position on this question is exceedingly complicated as we shall try to
show in Sections IV.3-1V.6 and also in the entire Section V. The Aujbau's
vacillation between various ideas on this point is connected with a deep-
going ambiguity in the whole Aujbau enterprise. Concerning the definien-
tia we know that their denotations change with the choice of the person
A from whose mentallife the wor1d model MA is abstracted. Some of the
expressions in V, as conceived of by Carnap, are clearly 'egocentric' with
respect to their denotation, viz alI those which belong to the phenomenol-
ogical description of the 'autopsychological' (e.g., 'erI', 'Er', etc.). How-
ever, are aU the expressions of V assumed in the Atifbau to have an ego-
centric denotation? I shall use the phrase 'the intersubjective interpretation
of v' to designate a view according to which a very large part of the
expressions in V (e.g., those concerned with physical reality) have a re-
latively constant denotation for alI persons who are sufficiently familiar
with the terminology. The intersubjective interpretation of V implies that
different persons describe a common reality with a part of V. What I shall
call 'the egocentric interpretation' implies that each person, so to speak,
'builds his own world' which he does not share with anybody else, and
that the terms in V when used by him denote constituents in this world of
his. Upon this interpretation the denotation of each term in V varies with
the person using the term. In the Atifbau Carnap hardly makes a clear
choice between the two interpretations. It is possible to read many pas-
sages of the Aujbau with the intersubjective interpretation in mind, but
 HOW CARNAP BUILT THE WORLD IN 1928 33

large parts of the book explicitly demand an egocentric interpretation.

Concerning the third question (IV.1.6) the Aujbau says that the order
of the definitions shall mirror the order in which we build (our image oQ
reality in our predominantly intuitive process of knowledge. 5 5 Concepts
defined earlier shall have 'epistemological priority' over concepts defined
later. 56

IV.2. The Thesis of Intersubjectivity

A priori it is well conceivable that the philosophical purpose which the
Aujbau is intended to serve would demand that a given concept a from V
be defined in different manners depending on who the person A is whose
mentallife is described in L 2• If L 2 is interpreted through the model MA'
perhaps a ought to be defined thus:
a=Dff(erl A, ErA)'
where f is a certain logical function. When L 2 is interpreted through M B
(where B is a person different from A), perhaps a ought to be defined
differently:
a = Dfg(erlB , ErB ),
where g is another logical function. The Aujbau is nevertheless based on
an assumption of intersubjectivity which excludes this and which could
be stated about as follows:
(IV.2.1) Exactly the same chain of definitions D serves the philos-
ophical purpose of the Aujbau irrespective of who the person
A is from whose mental life the world model MA' described
through L 2 , has been abstracted, provided only that A has
"normal senses" and that certain "especially unfavorable cir-
cumstances" do not occur.
Something like this assumption is already implied by the fact that Carnap
develops a single chain of definitions in the Aujbau. In the preface to its
second edition he openly states the assumption. The "especially unfavor-
able circumstances"which he then alludes to are suchas would be counter-
instances to the hypotheses on which his definitions are based. 57 We have
already encountered two such hypotheses ((I.4.1), (I.4.5)), and three more
((V.2.2), (V. 3. 1), (V.3.2)) will soon be stated.
34 ANDERS WEDBERG

IV.3. The Thesis of Extensional Identity

Carnap says that in his definitions he applies, or intends to apply, an
"extensional method". This program signifies:
(IV.3.1) If a definition a=Dfb is (or can be made) a part of D, the
expressions a and b have the same extension.
It is not required, however, that a and b have the same 'sense', 'epistemic
value', or 'intuitive content'. 58 The extensional method is justified, Carnap
thinks, by the Thesis of Extensionality:
(IV.3.2) Two expressions with the same extension can replace each
other in any sentence without changing its truth-value ('log-
ical value').59
According to the preface to the second edition of the Aujbau, the descrip-
tion of the extensional method given in the first edition was not quite
satisfactory. Now Carnap wishes to replace (IY.3.1) by:
(IV. 3. 3) If a definition a=Dfb is (or can be made) a part of D, then
a and b have the same extension by necessity.
The word 'necessity' is explained to signify that the identity of extension
obtains "either on the basis of logica! rules or on the basis of laws of
nature". It is also said that (IV.3.3) is what was intended already in the
fust edition although this was not clearly said there. 60 We may call (IV. 3. 1)
the Thesis of Extensional Identity in its weak form and (IV.3.3) the same
thesis in its strong form.
Let us consider what the consequences of this are. Suppose that a and
b have the same extension. From the triviality a=a we can infer a=b
with the aid of the Thesis of Extensionality. Two designations with the
same extension thus designate the same thing, which is also the opinion
ofCarnap in the Aujbau. 59 We thus find:

(IV. 3.4) The definiendum and definiens of a definition in D designate

the same thing.
If we now remind ourselves of (IV. 1.3) we obtain the conclusion:
(IV. 3.5) Each concept or object designated by an expression from V
is identical with an element in the world model MA-
 HOW CARNAP BUILT THE WORLD IN 1928 35

The choice between the intersubjective and the egocentric interpretation

of V now becomes imperative, and (IV. 3.5) seems to require egocentricity.
The sets eriA and erlB (A # B) can hardly be assumed to have anye1ements
in common, and if they do not, it follows that MA and M B do not have
any elements in common, either, except possibly in my set-theoretical
formulation, the empty set and certain sets derived therefrom. Hence, by
(IV.3.5), each (non-empty) expression in V, the empirical-scientific voca-
bulary, will change denotation with its user. If we were to express our-
selves with accuracy, we should therefore have to reformulate the
previous statements (IV.3.I) and (IV. 3. 3) so as to render this relativity of
denotation explicit.

IVA. The Subjective World Picture

In trying to carry out consistently the egocentric interpretation of V one
encounters great difficulties. Suppose that 'e1ephant' is the term from V
we are dealing with, and assume that the following:
(IVA. 1) elephant = Of f (eri A' ErA),
where f is some logical function, is presented as its Aujbau definition. If
the word 'elephant' designates, in accordance with the intersubjective
interpretation, a kind of animal existing in an objective reality and if the
word is sufficiently well-defined, the question as to the extension of the
word has a well-defined meaning. If we were also to succeed in giving
erIA,andErA, and thereby f(erI A, ErA), a definite sense, the problem
concerning identity of extension could, at least in principle, be investigated
with respect to (IV.4.I). If, however, 'elephant' when used by the person A
designates some phenomenon in the 'private wor1d' of A, how is its ex-
tension to be settled?
Actually, Carnap appears to look upon a person's conception of the
wor1d, 'his private wor1d', as a kind of painting, unbounded save for the
boundary set by the person's death. This wor1d picture is composed of
pictures of various entities, physical objects, other people's minds, 'cul-
tural objects', etc. It may, for instance, contain a number of e1ephant
pictures. The raw material for the creation of the wor1d picture consists
of the person's elementary experiences, and from them the picture is made
in accordance with certain rules. There are, for instance, rules which
govern the creation of pictures of elephants. The wor1d picture is gradually
36 ANDERS WEDBERG

completed and enlarged as the person receives new elementary experien-

ces.
If I have understood Carnap right, he thinks that a person's world
picture may be 'normal' or not. In order that his picture of the world be
normal, the person must, I suppose, have had a sufficiently rich and varied
experience, his memory must be fairly good, and he must possess good
enough a mastery of the rules for making pictures. We could, Carnap
suggests, imagine an ideal picture maker who never forgets what he has
experienced and who makes alI the pictures that the rules alIow him to
make on the basis of his past experience. 6
Carnap's language sometimes lefids itself to the interpretation that the
picture making is a purely set-theoretical construction on the basis of past
experience, consisting in the iterated application of the two operations of
forming sequences <Xl>"" xn> and sets {Xl"'" Xn} from given or con-
structed elements Xl"'" Xn (the set-formation subject to the restriction of
type-theory). On the other hand, he also seems to consider this view ofthe
picture making as a convenient fiction If it is to be taken as a fiction, we
may presume that Carnap had in mind some way of correlating (more or
less) one-to-one the pictures that he takes to be parts of the person's
actual world conception with such set-theoretical constructions. The set-
theoretical entity correlated with a given picture may be called the set-
theoretical counterpart of that picture.
If something like this is presupposed, we could speak of the subjective
extension of the word 'elephant' for A at t, meaning the set of aH the
elephant pictures contained in A's wor1d picture at t. We might also
speak of the set-theoretical image of this subjective extension, mean-
ing thereby the set of alI set-theoretical counterparts of elements in the
extension It is now convenient to introduce the symbol erI A,/> signifying
the set of those elementary experiences that A has had at t, and also the
symbol ErA,t, signifying the restriction of ErA to erIA,t. Using this termi-
nology one could then interpret Carnap's Thesis of Extensional Identity in
the following manner. Assuming stiH that (IV.4.1) is the Aujbau definition
of 'elephant', the Thesis can be construed to assert:

(IV.4.2) The set-theoretical image of the subjective extension of 'ele-

phant' for A at t is identical with f (eri A, t, Er A, t) provided
that A is an ideal picture maker.
 HOW CARNAP BUILT THE WORLD IN 1928 37

If A is not an ideal picture maker, the assertion in (IV.4.2) must be

weakened somehow. (If A fails to be ideal only by not making all the
elephant pictures that his experience and the relevant rules allow, the
image will, I suppose, be only a proper subset of f (eri A, l' Er A, 1).)
It would be insensibly rash to assert that what has now been said is
what Carnap actually had in mind when writing the Aujbau. I will not do
so. At most one can say that the above exposition is one natural way to
make some sort of sense out of what Carnap is saying in the Aujbau. I
suspect that Carnap's 'depth of intention' (this felicitous phrase is due to
Arne Naess) was probably rather low on this point.
The manner of looking upon a person's 'conception of the world' that
I have now tentatively read into the Aujbau is certainly very unrealistic.
The very few elephants that I happen to have met and that I have any
'pictures' of nowise exhaust the role played by elephants in what I could
reasonably call my conception of the world. I believe, for instance, that
large numbers of elephants have existed in the past and that quite a few
still survive in Africa, India, and the Far East. These beliefs are not
accompanied by a corresponding number of elephant pictures in some
grandiose landscape painting of the world. Wittgenstein's words, "Ein
Bild hielt uns gefangen" (Philosophical Investigations, § 115), are very
much applicable to the whole train of reasoning contemplated in this
section.
It should finally be mentioned that the author of the Aujbau might ob-
ject to the present discussion that from his point of view it does not make
sense to distinguish, as I am doing, between 'my conception of the world'
and 'the world' itself. In the spirit of some epistemological idealism, he
might claim that when I speak of 'the world' it is my conception of the
world that I am referring to. The reply to this claim is, I think, readily
drawn from the semantic distinctions made by Bolzano, Frege, and their
followers, including a part ofthe author of Meaning and Necessity.

IV.5. The Thesis of Truth-Invariant Translation

In this section we shall cross once more the same territory as in the
previous section, but our point of view and our purpose will here be
different.

(a) Translating from science into the 'Aujbau'. I shall use the phrase
38 ANDERS WEDBERG

'scientific sentence' as an abbreviatiol1 ofthe phrase 'meaningful sentence

constructed from 10gica1 signs, variables of science and expressions
from V' (it is not much more than a phrase !), and I shall introduce the
notion of 'the Aujbau translation of a scientific sentence'. To obtain the
Aufbau translation of a given scientific sentence S we first have to carry
out the following operation:
(IV.5.1) Operation 1: Replace throughout S any expression ai from V
by its definiens bi in the chain of definitions D.
This is the only operation of translation that is explicitly considered in the
Aujbau.
However, unless we a priori assume that the variables of science refer
to the same type hierarchy as those of the Aujbau, we must also replace
the former variables by the latter in an appropriate manner. To simplify
our problem, let us make the - certainly not un-Carnapian - assumption
that science deals with a type hierarchy of the same general structure as
that of the Aujbau. We may call the type hierarchy of science W and its
domain of basic objects (individuals) O. W can then be described by the
formula (cf. (11.0)):
(IV. 5.2) W=Jf'(O).
The names and variables of science are, we assume, supplied with indices
showing their types just as are the names and variables of the Aujbau.
We shall introduce a mapping of the type hierarchy W into the hier-
archy M whereby each type X of W is mapped upon a type X* of M.
This particular mapping will be called 'the adequate mapping of W into
M'. With reference to this notion we introduce a second operation of
translation:
(IV.5.3) Operation II: If the subscript of a variable in S indicates a
type X of W, change it throughout S into a subscript indi-
cating type X* of M, where X --+ x* is the adeq uate mapping
of W into M.
An 'adequate' mapping X --+ X* must obviously satisfy the requirement
that the application of it through operation II turns any type-theoretically
correct statement of science into a type-theoretically correct statement of
the Aujbau. A moment's reflection shows that this requirement is tanta-
 HOW CARNAP BUILT THE WORLD IN 1928 39

mount to the conjunction of the three conditions: (i) &'(X)*=&'(X*);

(ii) &'(X 1 X ••• X X n )* =&'(X! X ••• X X:); (iii) if aj belongs to type X in W,
then bj belongs to type x* in M. It is not self-evident that there exists a
mapping that is adequate in this sense. The chain of definitions D could
conceivably be so awkwardly constructed that condition (iii) could not be
fulfilled together with conditions (i) and (ii). But then D would obviously
be, in a sense, 'unreasonable'. We assume that it is not, Le. we assume:
(IV. 5.4) There exists at least one adequate mapping of W into M.
On the basis of certain assumptions that seem to be in the spirit ofCarnap,
we shall now show that there exists exactly one adequate mapping, and
that hence our talk of 'the adequate mapping' was justified.
If alI a;'s and b;'s were empty terms, Le. (materialIy speaking) if aj = bj =
= empty set for alI i, then aj = bj would be an element of any type of the
form &'( ... ). Condition (iii) in our definition of'adequate mapping' would
therefore not identify anyone specified pair (X, X*). We shall assume
that this situation does not arise. It is convenient to frame the assumption
in terms of a notion of 'substantial entity'. By such an entity we under-
stand an entity (of the kind involved in W and M) which in no manner
whatsoever involves the empty set in its constitution. More formalIy the
notion can be explained as folIows: (a) If an entity is neither a set nor an
n-tuple, it is substantial; (b) If alI the elements of a non-empty set are
substantial, the set is substantial; (c) If alI the terms of an n-tuple are
substantial, the n-tuple is substantial; (d) No entity is substantial save by
virtue of(aHc). We now postulate:
(IV.5.5) The chain of definitions D contains some definition: aj=Dfbj,
where aj and bj stand for substantial entities.
(IV.5.6) The elements of eri, as well as those of 0, are neither sets nor
n-tuples.
The latter postulate entails:
(IV. 5.7) Two distinct types in M, as weB as in W, have no substantial
elements in common.
Consequently, condition (iii) in our definition of 'adequate mapping'
does single out some particular pairs (X, X*), viz. those corresponding to
the substantial pairs (aj, b;). Now, two distinct mappings satisfying
40 ANDERS WEDBERG

conditions (i) and (ii) ofthe definition can obviously not have any couple
(X, X*) in common. Therefore, there cannot exist more than one ade-
quate mapping.
Let ai and bi be two substantial terms from D. Assume ai to belong to
type X and bi to type X*. If, with Carnap, we assume the Thesis of
Extensional Identity, we find that X and x* will overlap substantialIy-
(ie., have a common substantial element). If substantial overlapping
takes place in one couple (X, X*) of our mapping, it can easily be proved
to occur in alI such couples. Consequent1y, O overlaps with 0*. Since the
elements of O are neither sets nor n-tuples and since eri is the only type in
M ofwhich this is likewise true, 0* must be erI. We have thus found:
(IV. 5. 8) 0* = eri, and X and X* always overlap substantialIy.
Non-substantial overlapping between types may be an entirely trivial
feature, introduced by our set-theoretical form of discourse, and does not
recur when the argument is cast in an orthodox Russellian type-theoreti-
cal form. But substantial overlapping is invariant under translation from
set-theory to type-theory: it is a phenomenon that would have come
within the purview of the type-theorist Carnap of 1928.

(b) The invariance of truth-value under the translation. We have estab-

lished the phenomenon of substantial overlapping (IV.5.8), but we have
so far not encountered any assumed facts which force X and X* to
coincide. If they do not coincide, we may have a situation where, e.g., X
contains infinitely many elements while X*, like alI the types in M,
contains only a finite number. The true scientific sentence affirming the
infinitude of X would then be turned into a falsehood when translated
into the Aujbau.
Carnap, however, apparent1y takes for granted that the folIowing
identity assumption holds:
(IV.5.9) X is always=X*, and, hence, W=M.
From this assumption, together with the Theses of Extensional Identity
and of Extensionality, one can easily infer:
(IV.5.l0) A scientific sentence and its Aujbau translation always have
the same truth-value.
This is, in fact, one of the most central theses of the Aujbau.
 HOW CARNAP BUILT THE WORLD IN 1928 41

(IV.5.9) may weH seem a gratuitous assumption. Speculating about

the reason that prompted Carnap to make it, one comes to think of
Principia Mathematica's view that, in some absolute sense, distinct types
cannot overlap. (Substantial overlapping is, we know, overlapping also
within the framework of Principia Mathematica.) Various considerations
pertaining to the philosophy of science may also have played a role.
Above aH, however, Carnap must have been motivated by the kind of
epistemological and/or metaphysical convictions that we have already
discussed in Section IV.4.
(IV.5.10) is never, it is true, introduced quite explicitly as an independent
criterion of the adequacy of the Aujbau definitions. At least in §51,
however, the demand for truth-invariant translation is emphasized so
strongly as to approach the status of independent criterion. It appears
a priori as a more reasonable criterion than the Thesis of Extensional
Identity. Unlike the latter it does not entail identity between the world
of science and the model M.
However, if scientific sentences are assumed to admit such truth-inva-
riant translation into the language of the Aujbau, this also has strong
consequences for the contents of scientific truth. Let a be the name of a
set, and let b be the definiens of a in D. In L 2 we can construct a true
sentence asserting the finitude of the set named by b. Consequently, there
is in virtue of (IV.5.10) a true scientific sentence which asserts the finit ude
of the set named by a. The finitism that rules in the Aufbau's world model
M is thereby imposed also on the world of science.
The thesis (IV.5.10) also entails a kind of isomorphism between the
world of science and the world model. If 'ai is an element of a/ is a true
scientific sentence, then its Aujbau translation 'b i is an element of b/ will
be true, and vice versa. The set-theoretical relation of elementhood thus
holds between objects designated by expressions from V if and only if it
holds between the entities from M designated by the definientia of the
expressions in D.

IV.6. The Thesis of 'Formal Analogy'

In some passages of the Aujbau Carnap seems to imply that the chain of
definitions D is intended to present 'formal substitutes' for scientific con-
cepts, or that the definientia of Dare expected to exhibit a 'formal analogy'
with the definienda. 63 The talk of 'formal analogy' can plausibly be inter-
42 ANDERS WEDBERG

preted in terms of the notion of isomorphism. The idea that isomorphisms

obtain between structures described in the language of science and struc-
tures to be found within M is in itselfvery indefinite: the number ofways.
in which it may be elaborated is literally unlimited. But the idea is of
interest as stiH another point of view regarding the relationship between
model and world.

v. THE DEFINITION OF SIMILARITY CIRCLE

V.I. The Presystematic Notion of Quality Sphere

In the Aujbau's chain ofdefinitions D=(d 1 ,d2 ,d3 , ... ),d 1 is a definition
of eri (cf. (1.4.5)), d 2 a definition of 'part similarity' (cf. (1.4.1)), and d 3 a
definition of 'similarity circle'. I shall here discuss d3 in some detail since
it clearly illustrates the problematic character of the AuJbau definitions
and in addition is rather easy to understand.
In order to see the notion that d3 tries to catch, we must return to
Carnap's presystematic phenomenological theory. In the five-dimensional
color space any two points have a distance, and there is a certain fixed
standard distance (ef. (1.3)). Let d(p, q) be the distance between the points
p and q, and let (j be the standard distance. We may then speak of'quality
spheres' (of standard diameter) in the following sense: .

(Y.I.1) The point set X is a quality sphere if and only if (i) if p and q
belong to X, then d(p, q)~(j, and (ii) if d(p, q)~(j for each
point q in X, then p belongs to X.

In an identical manner the notion of quality sphere can be introduced also

for the senses other than sight.
It is for this notion of quality sphere that d3 is supposed to afford a
'formal substitute'. 64

V.2. The Correspondence Hypothesis

With each point set X of the color space we may associate a set of ele-
mentary experiences f (X) in the following manner:

(V. 2. 1) f(X)=Dr the set ofall elementary experiences in which some

point from X occurs.
 HOW CARNAP BUILT THIE WORLD IN 1928 43
\

When Carnap frames definition d3 he must, as far as I can see, implicitly

assume what I shall call the Correspondence Hypothesis:
(V.2.2) If X and Y are quality spheres and f(X)=f(Y), then X = Y.
This hypothesis is by no means self-evident: those experiences that partake
of some quality from X might be exactly those that partake of some
quality from Y, although X and Y are distinct spheres. Think, for instance,
of a person living on a tiny atoll in the Pacific who, we may assume, never
hears the sound of the breakers (X) without feeling the smell of the ocean
(Y), and conversely. It is easy to imagine any number of similar counter-
instances to (V.2.2). (It could perhaps be argued that the occurrence of
such a counter-instance involves a certain poverty or rigidity in the mental
life of the person whose elementary experiences we are considering. But
the poverty, or rigidity, need be no gre ater than in the just mentioned
example.)
The substitutes for the quality spheres which Carnap hopes to give
through d3 and which he calls "similarity circ1es" are just those sets a. of
elementary experiences for which it holds:

(V. 2. 3) a.=f(X), for some quality sphere X. 65

v,3. The Hypothesis of Extensional Equivalence

The condition (v'2.3) already defines the similarity circ1es, as conceived
by Carnap. It is, however, formulated in the presystematic pheno-
menological theory. The task is now to find another, extensionally equiv-
alent condition which is formulated exc1usively in terms of language L 2 •
When Carnap proposes such a condition he implicitly assumes the fol-
lowing hypotheses:

(V. 3. 1) If (V.2.3), then each elementary experience which is part si-

milar to each element of a. is itself an element of a..
(V. 3. 2) If (i) each elementary experience which is part similar to each
element of a. is itself an element of a., and (ii) any two ele-
ments of a. are part similar, then (V.2.3).
In the Aujbau Carnap is aware that 'unfavorable circumstances' may
invalidate these hypotheses. 66 An elementary experience y might for in-
stance be part similar to each element in a. =f (the quality sphere X),
44 ANDERS WEDBERG

not because a quality from X occurs in y, but because there is another

quality sphere Y which is represented in alI elements of (J. as well as in y.
Then y would not be an element of (J., and a counter-instance to (V.3.I)
would be given. In order to have a concrete example, suppose a person
with a normallife but with a very persistent ache in a tooth. As long as
the ache lasts, alI his elementary experiences will, I imagine, be part simi-
lar. Any number of similarity circles may be proper subsets ofthe set of ex-
periences that are part similar due to the tooth-ache. (If it is claimed that
the existence of such a persistent tooth-ache is an abnormal rigidity of
the mental stream, the answer must be that it is no more abnormal than
just a persistent tooth-ache.)
Nelson Goodman has shown how counter-instances to (V.3.2) might
arise. 67 The antecedent condition (ii) of(V.3.2) is fulfilIed, not only when
(iia) there is some quality sphere which alI elements in (J. exemplify, but
also when (ii.b) such is not the case but for any two experiences in (J. there
nevertheless is a sphere which these two exemplify. When (i) of (V.3.2)
and (ii.b) hold, (V.3.2) fails to hold. Let us call a set fulfilling (i) and
(ii.a) a Carnap set, one fulfilling (i) and (ii.b) a Goodman set. It may per-
haps be instructive to think of these sets as arising in the folIowing man-
ner. (This is possible if the number of elementary experiences is finite, as
Carnap thinks it is, or at least denumerably infinite.) We form a set by
first putting into it some arbitrarily chosen experience and then adding
new experiences, one at a time, each time making sure that the experience
added is part similar to each previously chosen experience. We continue
in this manner as long as possible, and we let the set contain nothing but
the experiences chosen in this manner. Such a set will clearly fulfill the
conditions (i) and (ii) of (v'3.2), and alI sets fulfilling these conditions are
obtainable by this method. In other words, the finished set will be either a
Carnap set or a Goodman set, and alI Carnap and Goodman sets can be
created in this fashion. It is tempting to ask for the probability that the
finished set will belong to the one rather than to the other category. The
question cannot be answered since we do not possess the premises neces-
sary for a combinatorial ca1culation. However, we know that the number
of Carnap sets is at most as great as the number of quality spheres, and
it is difficult to see any general reason why the number of Goodman sets
could not be greater.
The assertion:
 HOW CARNAP BUILT THE WORLD IN 1928 45

(V. 3. 3) If (V.2.3), then any two elements in IX are part similar,

is analytic. Hence the two hypotheses just stated together amount to the
following equivalence:
(V. 3.4) The condition (V.2.3) is satisfied if and only if the two an-
tecedent conditions of (V.3.2) are fulfilled.
The two hypotheses (V.3.1) and (V.3.2) thus together form a hypothesis
of extensional equivalence between the presystematic condition (Y.2.3)
and a pair of conditions formulated in L 2 • This gives us the Aujbau's
definition d3 of similarity circle:
(Y. 3. 5) IXis a similarity circle = Of IX satisfies conditions (i) and (ii)
of (V. 3.2).

V.4. The Methodological Status of the Hypotheses

According to the idea which is most prominent in the text of the Aujbau,
definition d3 should offer a substitute for the presystematic notion of
quality sphere, introduced above in Section V.1. When the hypotheses
(V.2.2), (V.3.1), and (V.3.2) are not fulfilled the definition cannot do this
in the intended manner. Thus d3 is valid only for persons whose elementa-
ry experiences have the properties demanded by these hypotheses. It is
not certain, however, that the line between those persons for whom d3
holds and those for whom it fails to hold is as clear-cut as it might at
first seem. In Sections 1.2 and 1.4 we saw that the choice of cross-sections
from the mental stream of a person A, to serve as the eriA of MA' may
involve an element of convention. The possibility of different choices
seems to obtain also when it comes to deciding on the standard distance
involved in the idea of ErA- Perhaps the hypotheses underlying d 3 are not
meant to rest upon principles which universally and uniquely determine
our choices in these respects. What the Aujbau assumes may be only that
with regard to many human beings it is feasible to define the basic notions
so as to satisfy the hypotheses whereas it is not feasible with regard to
other human beings.
Even if the definition d3 does not match the presystematic concept of
quality sphere it may stiH, Carnap hints, correspond to the 'intuitive' set-
theoretical construction of similarity circles. Here he is apparently assum-
ing that the intuitive concept formation actually is a set-theoretical opera-
46 ANDERS WEDBERG

tion. In §81 we read: "A closer investigation, for which there is no room
here, shows that these disturbances of the quasi-analytic concept forma-
tion [i.e., the counter-instances to the hypotheses] occur only when cir-
cumstances are present under which also in reality the cognitive process,
i.e., the quasi-analysis intuitively performed in reallife, does not lead to
the normal result".
We know that, according to certain assumptions present in the Aujbau,
alI elements of the model Mare definable in the definition language L 2 •
If the idea of similarity circle at alI makes sense, hence it is definable in
L 2 - quite irrespective of whatever weaknesses the definition suggested by
Carnap may suffer from. Unfortunately, this consideration can not be
invoked in support of the Aujbau's program of definition. What Carnap
is looking for, is a definition that shall be suited for incorporation in a
future, 10gicalIy and epistemologicalIy perfected presentation of science.
The type of definitions whose existence is ensured by the general consi-
derations in Section III, are definitions which it is humanly impossible to
state.
FinalIy, if the unanalyzability of elementary experiences is taken seri-
ously, there are no such things as 'qualities' and 'quality spheres' in the
sense of a literally understood presystematic phenomenological theory.
What there is, strictly speaking, is only the hierarchy of sets and relations
resting upon eri as its base. The presystematic theory is therefore legiti-
mate only to the extent that it can be understood as a faron de parler
about the hierarchy. If in some particular case the 'quality spheres' of the
presystematic theory have no identifiable counterparts within the hier-
archy, they are just an illusion. If one adopts this line of argument - and
it is not foreign to Carnap - one may feeI inclined to question whether the
hypotheses can mean any real restriction upon the validity of d3 . If the
Aujbau's similarity circles (in the sense of (V. 3. 5)) are the best replicas of
the presystematlc 'quality spheres' that we can find within the hierarchy,
welI, so much the worse for those presystematic 'quality spheres' that
cannot be reinterpreted as such circles.

VI. A REVIEW OF THE AIMS OF THE AUFBAU

According to the point of view from which this paper has primarily been
written, the aim of the Aujbau is to present a model of the world but
 HOW CARNAP BUILT THE WORLD IN 1928 47

not identical with the wor1d. In the light of the previous analyses, this
interpretation cannot be accorded more than a very limited validity. Does
the Aujbau really wish to give us a model of - the world? Is what the
Aujbau presents thought to be merely a model of something? It seems to
me that Carnap had several different aims in mind when writing the
Aujbau, without realizing that they are different.
In favor of our primary interpretation speaks, for instance, the fact
that Carnap takes the Aujbau system to be only one among a large num-
ber of systems that all could serve the same aim about equally well. The
particular type hierarchy which is the Aujbau's model is based on a do-
main of individuals consisting of Carnap's own elementary experiences.
There is, as we have seen, a large range of other persons whose elementary
experiences are supposed to do just as weB. Type hierarchies possessing
about the same philosophical interest can also be based, Carnap says, on
what Mach calls "elements", or on various kinds ofphysico-mathematical
entities (elementary particles, space-time points, etc.). If the relation be-
tween each of these hierarchies and the wor1d is the same, they cannot be
identical with the wor1d since they are not themselves identical with one
another.
Against our primary interpretation speaks Carnap's program to use an
extensional method in framing the definitions that establish the corre-
spondences between the model and the world. Successful use of the me-
thod implies an identity between the definiendum's designatum, which is
in the wor1d, and the definiens' designatum, which is in the model. Against
the primary interpretation speaks also the fact that, in translating from
science to the Aujbau, Carnap apparently and tacitly takes for granted
that science and the Aujbau deal will exactly the same type-hierarchy. The
primary interpretation is also belied by the spirit in which Carnap
proposes to solve various great philosophical problems (e.g. the mind-
body problem): he then takes for granted that things are just what they
are defined to be in the Aujbau. It is difficult to escape the conclusion that
Carnap did not stop at the wish to construct a model of the wor1d. He
also desired to lay bare the very 'metaphysical essence' of the world.
Another aim of the Aujbau, a third one, is to give a 'rational reconstruc-
tion' of the manner in which a human being arrives at his conception of
the wor1d. This conception is apparently thought of as a partial picture
of the world, composed of pictures of things in the world. In Carnap's
48 ANDERS WEDBERG

reconstruction the thing pictures are entities in the Aujbau hierarchy and
hence built from elementary experiences by iterating, in a varying order,
the operation of forming finite sequences and that of forming sets. In
some passages this set-theoretical process is regarded only as a model of
the actual psychological process, but almost as often Carnap seems prone
to identify the two.
Yet another aim, overlapping with those previously mentioned, is to
show that all scientific concepts can be defined in terms of the concepts
chosen as basic in the Aujbau. The extensional method to be employed in
the defining implies that each Aujbau definition of a scientific notion a by
an Aujbau notion b:
(i) a=Dfb,

involves the assumption:

(ii) Everything that falls under a falls under b, and conversely,
an assumption that will usually be a hypothesis of some kind: Carnap
demands that it be a 'scientific result'. The Aujbau thus advances the
daring (if not absurd) second-order hypothesis that each scientific con-
cept a has a corre1ated Aujbau notion b such that some day or other
science will be able to establish the first-order hypothesis (ii).

VII. CONCLUSION

The ultimate purpose of the Aujbau was to show by means of an example

how certain extremely general philosophical theses might be elaborated
and defended These theses, which assert the unity of science, the merely
structural insight given by science, the extensionality of the language of
science, the fundamental importance of logic for philosophy, the un-
scientific character of metaphysics, etc., together form an early version
of the ideology of logical empiricism Again and again Carnap reminds
his readers that the particular world model and the particular chain of
definitions sketched in the Aujbau are only tentative. It may therefore
seem unfair to discuss critically these things at such length while passing
by the great ideology in silence. Carnap was to hold fast to most of its
theses during his entire philosophicallife, and he was to reformulate them
many times as a result of renewed reflection. They have also been central
 HOW CAR NAP BUILT THE WORLD IN 1928 49

themes of the analytical philosophy of our age. They have thereby, 1

think, become much too wide and too traficked a ground for a brief essay.
One of the fruits of the ideology is the Aujbau, and we have here studied
only this particular fruit.
It shows both the strength and the weakness of Carnap's style of philo-
sophizing. The strength is the grand vision and the desire for formal rigor.
The weakness is the almost dogmatic trust in philosophical 'theses' and
the lack ofwhole-hearted, independent interest in questions which do not
admit settlement more geometrico. Carnap reminds one of the man in
Plato's Republic who has been outside the cave and seen the light - in
Carnap's case: formal logic, pure mathematics, and mathematical phys-
ics - and whose eyes are "filled with darkness" when he re-enters the cave.
Yet it is there that the realities of the human mind are met with, human
experience and language, human knowledge and science, the things that
the Aujbau wants to teach us something about. The advice of d'Alembert
in his Essai sur les elements de philosophie (1759) comes to mind here: A
philosopher ought to "preserve the suppleness of the mind by not always
keeping it bowed over diagrams and ca1culations, and by tempering the
austerity of mathematics by less severe studies: in a word, to accustom
himself to pass without difficulty from the light into the twilight." 68

Djursholm, Sweden
NOTES

1 Carnap's book as well as the theory developed therein will here be referred to as the
Aujbau.
2 1 made an earlier, very faulty attempt in the paper, 'The Logical Construction of the
World', Theoria 10 (1944),216--246. The present paper is a revised and somewhat enlarged
version of a paper in Swedish, 'Hur Carnap byggde vărlden 1928' (1970).
3 Pp. xiii-xiv. Carnap refers here to § 70 and § 72 of the Aujbau where the presuppositions
of so-called 'quasianalysis' are discussed. Cf. Section V below.
4 §67, §85.
5 Hints as to the philosophical psychology Carnap accepts are given especially in § 64,
§65, §67, §68, § 163, § 164.
6 According to §64 the Aujbau makes no distinction between perception, hallucination,
dream, etc. "In the beginning of the system the experiences will be considered solely as
they present themselves; the positing, made in them, of their contents as real Of unreal is
dropped or 'bracketed'; i.e., we shall exercise the phenomenological 'with-holding of judg-
ment' (btOxf]) in HusserI's sense .... "
7 Most probably Carnap would agree that the psychic stream is like the stream coming

from a tap that is sometimes turned on and sometimes turned off. If we do not (with Leibniz)
50 ANDERS WEDBERG

postulate that periods of apparent absence of consciousness (e.g., dreamless sleep) are filled
by unconscious menta! activity, such periods will signify interruptions in the stream.
8 §67.
9 §77.
10 Conceming the five-dimensionality of the space of color points, see especially §80, §81,
§86.
11 § 72, § 80. In § 72, "the fixed maxima! distance still involving similarity" is said possibly
to vary from part to part of the color solid I have disregarded this complication in the
text. On p. 100 the standard distance is characterized as "a certain arbitrarily fixed mag-
nit ude". On p. 80 it is spoken of as "the largest distance that may obtain between two sen-
sation qualities that are still similar ... at that place of the sense modality." The first quo-
tation would seem to make the notion of similarity arbitrary in the sense that it will depend
upon an arbitrarily fixed distance. The latter quotation, on the other hand, gives the im-
pression that the standard distance is dependent upon a pre-existing intuitive notion of
similarity.
12 §77, §80.
13 §85, §86, § 115.
14 §77,§80.
15 § 78, § 108.
16 § 78, § 110.
17 §87.
18 §87, § 120.
19 In The Structure of Appearance, first ed., p. 133, Goodman points out that the reader
of the Aujbau can on the whole disregard Camap's own explanation of Er and interpret
it simply through (1.4.4): the logico-philosophica! machinery of the Aujbau then functions
just as well (or just as badly) as when Camap's explanation is assumed.
20 §109.

21 §67, §68, §71, §162.

22 §68, §69, §71.
23 §68.
24 §37.
25 §28, §33.
26 The Structure of Appearance, fust ed., Ch. II, Sections IV, V.
27 Cf. Bertrand Russell, Logic and Knowledge, p. 175ff., p. 321ff.
28 Cf. the review ofliterature in §67.
29 § 154, § 155.
30 The concepts defined are elementary experience (§ 109), part similarity § 78, § 110), sim-
ilarity circle (§80, §111), quality (§81, §112), part identity (§82, §113), similarity between
qualities (§85, § 114), sense class (§85, § 115), sight (§86, § 115), sensation, simultaneity of sen-
sations (§93, § 116), division (of elementary experience) into sensations, division (of elemen-
tary experience) into qualities (§ 116), place (in the visual field) , identity of place (in the
visua! field) between qualities, neighborhood of places (§ 88, §89, § 117), identity of color
in neighboring places, color, color neighborhood (§90, § 118). The notions that are here
italicized have been discussed in 1.1-1.4. The notion of similarity circle and its definition
will be discussed at some length in Section V below.
31 When convenient, I shall also use the relativized symbol MA'
32 This is a deliberate simplification Camap's attitude towards extensions (sets and ex-
tensional relations) in the Aujbau is as ambiguous as Russell's in Principia Mathematica.
Although the Aujbau is prima facie constructed so as to involve an extensiona! type hier-
 HOW CARNAP BUILT THE WORLD IN 1928 51

archy, Carnap maintains that only the basic elements, the elementary experiences, are
genuine "objects" (Gegenstiinde) with a non-fictive status. Cf. §27-§45. Cf. also Section
1II.6 below.
33 § 180.
34 § 10-§ 16.

35 § 10-§ 16, § 153-§ 155. The assumptions (II. 5)-(11. 7) can be restated as a single assump-
tion:
(11.5*) Letfbe a one-to-one mapping of eri onto erI. Then the condition:
(a) f is the identity mapping,
is equivalent to the condition:
(b) There is a relation R in fund such that f is an isomorphism between
<eri, R) and <eri, E).
36 § 154, § 155.
37 § 121.
38 I use the signs '-,', 'v', '1\', '->', ' ....', '3', "'il', '1', in their usuallogical sense. The expres-
sion' {y I...
y ... }' is used in the usual set-theoretical sense, meaning 'the set of ali y such
that ... y .. .'.
39 Let f(Er, el> ... , eq) be the conjunction of ali true sentences of the forms:

Further, let qi(Er, x) be the sentential function:

Then by virtue of (II.3)-(II.5) ei is defined through qi(Er, x) in L2' and in L3 by:

3R(fund(R) 1\ qi(R, x)).
40 A complete system ofaxioms for LI is obtained if the following axioms are added to a
suita bie logic:
(i) All true sentences of the forms ei Er ej' -, ei Er ej;
(ii) \fx(x=e l v ... v x=eq).
As extralogical axioms for L 2 we may choose:
(iii) 3x l ... 3xq f(Er, Xl' ... , X q );
(iv) 3x l .•. 3xq \fx(x=x l v ... V x=xq ).
(f is here to be understood as in note 39.) In addition to (iv) an axiom system for L3 must
contain axioms specifying which dyadic relations between elementary experiences belong
to fund and which do not. One such axiom might be this:
(v) 3R (fund (R) 1\ 3x l ... 3xq f(R, Xl' ..• , xq)).
41 In my 1944 paper I did not make this point sufficiently dear, something that C. G.
Hempe1 rightly criticized in his review of the paper.
42 § 183 and also § 179-§ 182.
43 §99, § 102.
44 §99-§ 105.
45 From §27.

46 From § 180.
47 When Carnap looked back on the Aujbau in 'Testability and Meaning', he described
52 ANDERS WEDBERG

its language as "molecular," that is as a variable-free predicate-logicallanguage of the first

order (Philosophy of Science 4 (1937), p. 19~ L 5 is molecular in this sense.
48 A method of translation can be obtained by slightly modifying some ideas of Quine's
in his paper, 'On Universals'. It is, however, humanly impossible to apply it: it presup-
poses an amount of knowledge on the part of the translator that a human being can never
achieve and writing-paper of immense size.
49 The Structure of Appearance, first ed., p. 116.

50 Cf. the papers by Quine, and by Goodman and Quine, listed in the bibliography of this
essay. The Aujbau's notion of an 'improper' language does not entirely coincide with Quine's
notion of a 'platonistic' language. Carnap considers a language 'improper' if it admits also
signs other than 'proper names' to subject position A proper name is then thought to be
a "sign designating a definite particular concrete object (e.g., 'Napoleon', 'the moon')" (§27).
A language is platonistic in Quine's sense if the assumed ranges of values of its variables
are not limited to the kind of objects that the proper names of Carnap designate. Lan-
guages may be improper without being platonistic (e.g., L 4 ), and also platonistic without
being improper.
51 ce, e.g., p. xviii of the preface to the second edition.
52 §31.
53 § 52, § 169.
54 § 122, § 119.
55 § 100.
56 § 54-§ 59.
57 ce pp. xiii-xiv of the preface to the second edition.
58 §47-§50.
59 §43-§45.

60 Cf. pp. xiii-xiv of the preface to the second edition.

61 §99-§ 102.

62 §2, §50, *51, §180.

63 §69.
64 §80.
65 §72, §80.
66 §72, §81.
67 The Structure of Appearance, Ch. V, Sec. 3.
68 Oeuvres philosophiques, historiques et litteraires de d'Alembert, Tome 2, Paris 1805, pp.
78-79.
BIBLIOGRAPHY

Carnap, Rudolf, Scheinprobleme in der Philosophie, Das Fremdpsychische und der Realis-
musstreit, Berlin 1928. Reprinted separately, Frankfurt am Main 1966.
Carnap, Rudolf, Der logische Aujbau der Welt, Berlin 1928; third ed., Hamburg 1966.
Carnap, Rudolf, Der logische Aujbau der Welt, Scheinprobleme in der Philosophie, Hamburg
1961. (A reprint of both the previous volumes, with a new preface by the author.)
Carnap, Rudolf, The Logical Structure of the World, Pseudoproblems in Philosophy, trans-
lated by Rolf A. George, Berkeley and Los Angeles 1967. (With the author's new preface
froJllI961.)
Carnap, Rudolf, Abriss der Logistik mit besonderer Beriicksichtigung der Relationstheorie
und ihrer Anwendungen, Wien 1929.
Carnap, Rudolf, 'Testability and Meaning', Philosophy ofScience 3 (1936), 419-471, 4 (1937),
1-40. (Also reprinted separate1y, New Haven, Conn., 1950.)
 HOW CARNAP BUILT THE WORLD IN 1928 53
Carnap, Rudolf, 'Intellectua! Autobiography', in P. A. Schilpp (ed.), The Philosophy of
Rudolf Carnap.
Carnap, Rudolf, 'Nelson Goodman on Der logische Aujbau der Welt', in P. A. Schilpp (ed.),
The Philosophy of Rudolf Carnap.
Goodman, Nelson, The Structure of Appearance, Cambridge, Mass., 1951. Second revised
edition, Indianapolis 1966.
Goodman, Nelson, 'The Significance of Der logische Aujbau der Welt', in P. A. Schilpp (ed.),
The Philosophy of Rudolf Carnap; also in Sidney Hook (ed.), American Philosophers at
Work, New York 1956.
Goodman, Nelson and Quine, W. V., 'Steps toward a Constructive Nominalism', Journal
of Symbolic Logic 12 (1947), 105-122.
Hempel, Cari G., Review of Anders Wedberg's Theoria paper, Journal of Symbolic Logic
13 (1948), 222.
Hillman, Donald J., 'On Quality Classes', Theoria 28 (1962),45-52.
Quine, W. V., 'On Universals', Journal of Symbolic Logic 12 (1947),74-84.
Quine, W. V., 'Designation and Existence', Journal of Philosophy 36 (1939), 701-709.
Quine, W. V., 'Notes on Existence and Necessity', Journal of Philosophy 40 (1943),113-127.
Russell, Bertrand, Logic and Knowledge, London 1956.
Schilpp, Paul A. (ed.), The Philosophy of Rudolf Carnap, The Library of Living Philosophers,
La Salle, Ill., 1963.
Wedberg, Anders, 'The Logica! Construction of the World', Theoria 10 (1944),216-246.
Wedberg, Anders, 'Hur Carnap byggde vărlden 1928' (How Carnap built the world in 1928),
in Logic and Value: Essays Dedicated to Thorild Dahlquist on His Fiftieth Birthday (Filo-
sofiska Studier utgivna av Filosofiska Foreningen och Filosofiska Institutionen vid Upp-
sala Universitet, no. 9), Uppsala 1970, 92-124.
 ROLF A.EBERLE

A CONSTRUCTION OF QUALITY CLASSES

IMPROVED UPON THE AUFBAU

Focusing on the construction of quality classes in Carnap's Der Logische

Aujbau der Welt [1], called the Aujbau for short, we begin with a review
of the basic definitions, continue with a presentation and discussion of
major old and new difficulties concerning these definitions, and end up
proposing an alternative approach which probably leads to all elementary
quality classes (those which are notcoextensional with genuine logical
compounds) while remaining neutral with respect to the basic individuals
one accepts.
Efforts to replace discourse regarding qualities by discourse regarding
similar things have been made by many philosophers for many different
reasons. The view that we need take account only of things which
resemble one another and not also of classes or properties has been
attractive to nominalists. The reduction of classificatory judgements to
comparative ones of only one basic kind has tempted philosophers who
pursued epistemological simplicity and primacy. An ordering of data with
respect to their similarities, whether those data be physical or phe-
nomenal, has been taken to underlie the fonnation of observational con-
cepts. And for various purposes philosophers have long wanted a theory
of 'true' (as opposed to logically contrived) qualities which might just
be the ones whose presence warrants judgements of resemblance.
While the program of constructing a similarity theory of classifica-
tion has had many proponents and opponents, philosophical debates re-
garding it have long remained obscure for lack of a specific theory whose
merits one might discuss. Could one hope to construct, on the basis of
similarity, all empirical properties and relations or only certain epis-
temologicalIy privileged ones? Should one expect that the contemplated
reduction of qualities to similarity preserves meanings, extensions, or
some other structural characteristics? Do the goals of the program de-
mand that the construction should work in alI conceivable circumstances,-
or is it enough if it leads to success only when the distribution of qualities
over things is like that commonly encountered in one's actual world of

Jaakko Hintikka (ed.), Rudolj Carnap, Logical Empiricist, 55-73. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Hol/and.
56 ROLF A.EBERLE

experience? For the basic resemblance relation, should one chose partial
qualitative identity, indistinguishability by direct comparison, just barely
discernible difference, or partial qualitative nearness? Do we need, in
addition to similarity, also similarity in a given respect or to a certain
degree? One appropriately answers such questions by actuaUy con-
structing a resemblance theory of classification while specifying the
goals, the range of applicability, and criteria of adequacy which are to
govern its construction. Carnap's Aujbau contains a brilliant attempt to
carry out this task with aU the precision and attention to detail which its
philosophical importance demands.
Unfortunately there are circumstances in which Carnap's similarity
theory of classification, and thereby the most basic parts of the Aujbau,
seem clearly incorrect and not easily amended. Carnap himself pointed
out and illustrated most ofthese difficulties in the same work. Apparently
he did not feeI at the time that they were disastrous for his program.
Goodman however, especiaUy by his discussion of the Aujbau in the
Structure (short for The Structure of Appearance, [4]), has later con-
vinced many readers that the defects at issue are after aU quite serious and
apparently insurmountable in a particularistic system such as the one
Carnap had set out to construct. Indeed, Goodman has been so successful
in translating for the English reader, summarizing, and criticizing the
lower levels of the Aufbau while at the same time advancing a sophis-
ticated rival theory of his own, that Carnap's contribution may thereby
have seemed refuted and outdated. Furthermore, Carnap himself carne
to prefer physicalism to phenomenalism and never again seriously tried
to work out the details of an improved phenomenalistic theory. Ac-
cordingly, one might well gain the impression that the Aujbau should be
ranked among the interesting but hopeless fiascoes.
We shaU attempt to show, to the contrary, that Carnap's resemblance
theory of classification can be amended so as to avoid most of the objec-
tions which have been raised against it and in a manner which still ac-
cords with the spirit of the Aujbau. Before doing so, however, it may help
to summarize Carnap's basic constructions and to list and briefly discuss
the main difficulties one has to meet.
1. REVIEW OF CARNAP'S CONSTRUCTION OF QUALITY CLASSES

For his basic entities, Carnap chose elementary experiences (Sections 67,
 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 57

109 1 ) which are thought of as total momentary stages in one's stream of

experience. An elementary experience could include all the colors in the
visual field, all the sounds, odors, savors, and tactile impressions which a
person might experience within a moment of subjective time.
As an additional primitive, Carnap employs a binary relation of
remembered similarity. The field of that relation is included in the class of
elementary experiences, and we postulate that the relation is asymmetric
since it should obtain between experiences only if one of them is tem-
poralIy prior to the other.
U sing these primitives, Carnap next defines the notion of part similarity
(Sections 78, 108) as paraphrased by the following:

DE FINITION 1. x is part similar to y if and only if both x and y are ele-

mentary experiences and either x bears remembered similarity to y or y
bears remembered similarity to x or x=y.

It folIows from the definition that part similarity is reflexive in the class
of elementary experiences as well as symmetric. Intuitively, two ele-
mentary experiences are part similar if either they share a common
phenomenal quality (such as a color, a sound, or a location in the visual
field), or if each has a quality of such a sort that the two qualities are
approximately the same (as two shades of color might be which are
adjacent in a discrete color chart).
In addition to part similarity, Carnap also employs the similarity rela-
tion of part identity. Intuitively, two elementary experiences are part
identical if they have a quality in common. Formally, part identity wilI
subsequently be defined in terms which presuppose the notion of part
similarity. Yet, even though part similarity is systematically more basic
than part identity, we are supposed to carry out the constructions about
to be mentioned with respect to both relations (see end of Section 72).
Given any relation of similarity (such as part similarity or part iden-
tity), one defines the notion of a similarity circle relative to that relation:

DEFINITION 2. Sis a similarity circle just in case (1) for alI x andy in S,
x is similar to y, and (2) there is no z such that z is not in S and for all x in
S, zis similar to x.
58 ROLF A.EBERLE

Thus, a similarity circle is a class of pairwise similar experiences which is

'largest' among such classes in the sense that nothing outside the class is
similar to aU experiences in the class. If one's similarity relation is that
of partial qualitative identity, then similarity circles themselves are in-
tended to be extensions of sense qualities (Section 72). But if the similarity
relation is that ofpart similarity, then similarity circles function as mere
auxiliaries in further constructions.
With respect to any similarity relation, Camap next defines the notion
of a quality class. He first tentatively considers and then rejects (in
Sections 80, 81) an explication which nevertheless seems worth stating
and whose precise reconstruction is presumed to be:

DEFINITION 3. C is a quality class if and only if C is a non-empty class

such that
(1) for every S, if Sis a similarity circle and the intersection of
S and C is non-empty, then C is included in S, and
(2) for every elementary experience x, if x is not in C then there
exists a similarity circle S such that x is not in S and C is
included in S.
Condition (1) of this definition provides that quality classes are not
divided by similarity circles which they intersect (as, e.g., the class of
experiences which seem either red or blue might be divided by the circle
of those which seem red). Given the assumption that C is non-empty,
Condition,(2) ofthe definition implies (but is not implied by) the assertion
that'C is largest among those classes which satisfy Condition (1). Thus, the
class of aU experiences which seem both red and sweet would normaUy be
excluded by clause (2), because it would be properly included in the class
of those which only seem red.
Due to the possibility of 'accidental' overlap between quality classes
and similarity circles (which will be illustrated below), Camap (in Sec-
tions 81, 112) prefers a definition which modifies part (1) of Definition 3:

DE FINITION 4. C is a quality class if and only if C is a non-empty class

such that
(1) for every S, if Sis a similarity circle and the result of dividing
the number of experiences which are in both S and C by the
 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 59

number of ones which are in C is greater than 1, then C is

included in S, and
(2) for every elementary experience x, if x is not in C then there
exists a similarity circle S such that x is not in S and C is
included in S.

Clause (1) of Definition 4 says that a quality class is wholIy included in

every similarity circle with which it shares more than half of its members.
This is supposed to capture the idea that a quality class is not divided by
'essential' overlap with any similarity circle. According to Carnap (Sec-
tion 81), clause (2) of Definition 4 is meant to convey that the quality
class C is the largest among alI possible classes which satisfy Condition
(1), though actualIy it neither implies nor is implied by the obvious sym-
bolization of that reading. If one's similarity relation is that of part
similarity, then quality classes, rather than similarity circles, are in-
tended to coincide with extensions of phenomenal qualities. By the given
definitio~s, a similarity circle need not be a quality class and, conversely,
a quality class need not be, nor even be included in, a similarity circle.
Assuming that quality classes had been adequately characterized, one
can then define part identity as membership in a common quality class:

DEFINITION 5. x is part identical with y just in case there is a quality

class C such that both x and y are members of C. '

This condensed summary is sufficient for our purposes. Further develop-

ments on the lower levels of the Aujbau, such as the construction of sense
realms, of vision, and of temporal and spatial order are not of present
concern to us.

II. SOME DIFFICUL TIES REGARDING THE CONSTRUCTION

There are circumstances in which the definitions given in the preceding

section fail to characterize exactly those classes which are extensions of
phenomenal qualities. In describing these circumstances, we shall initialIy
as sume that the similarity relation at issue is that ofpart identity, so that
similarity circles, rather than quality classes, are supposed to correspond
to extensions of qualities.
60 ROLF A.EBERLE

We find Goodman's notation perspicuous in presenting examples:

lower case letters are used to denote qualities, and numbered concatenates
of such letters designate elementary experiences which, irrespective of
order, possess just the qualities designated by those letters. It is also
understood that distinct letters denote different qualities. E.g. by the list

(1) a
(2) ab
(3) b
we mean to convey that our universe of discourse comprises exactly three
elementary experiences, the first of which is uniformly characterized by
the quality a, the second has both of the qualities a and b, and the third
has the quality b alone. With respect to the relation of partial qualitative
identity, 2 is similar to both 1 and 3, whereas 1 and 3 are dissimilar.

A. The 'Difficulty' of Extensionality

Suppose we had the three experiences

(1) a
(2) abc
(3) c
Then, by Definition 2, we would obtain the similarity circ1e {l, 2} of ex-
periences which are a and the circ1e {2, 3} of ones which are c; but the unit
c1ass {2} of experiences which are b is not a similarity circ1e. Suppose we
were to define the notion of a quality extension so that alI similarity circ1es
are quality extensions and the non-empty intersection of any two quality
extensions is again a quality extension. We could then obtain {2} as a
quality extension, one which corresponds to the compound quality of
possessing both a and c. It is due to the fact that b is here coextensional
with that compound quality that we cannot isolate the c1ass of experiences
which are b. But Camap, whose approach was emphaticalIy extensional,
would hardly regard it as a defect of his system if it merely failed to yield,
as a basic quasi constituent, a quality which is coextensional with another
or with a combination of others which his methods do alIow him to con-
struct at least indirectly. Indeed, we mention this difficulty only in order
to sharpen the sense in which this system is an 'extensional' one.
 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 61

B. The Companionship Difficulty

The folIowing example is borrowed from the Structure (p. 160)2:
(1) br (4) g
(2) b (5) bgr
(3) bg
U sing Definition 2, we can construct the class {1, 2, 3, 5} of experiences
which are b and the class {3, 4, 5} of one's which are g. However, the class
{I, 5} of experiences which are r tums out not to be a circle. GeneralIy,
the difficulty will arise whenever one quality is a constant companion of
another. Thus, if everything which is r is b, as it happens in this example,
then the class of alI r's will not be a similarity circle. Camap himself men-
tioned and illustrated this difficulty and even suggested its name in the
Aujbau (Section 70). He felt that it could occur only in unfavorable and
unlikely circumstances.

C. The Difficulty of Imperfect Community

The next illustration is also taken from the Structure (p. 163):
(1) bg (3) br (5) b
(2) rg (4) r (6) g
Whereas the previous difficulty consisted in not obtaining enough ex-
tensions of qualities, the present difficulty consists in getting too many.
For, when Definition 2 is applied to this example, it tums out that the
class {1, 2, 3} is a similarity circle and hence should be the extension of
some simple quality even though no single quality is common to alI mem-
bers of that class. GeneralIy, the difficulty will arise whenever three or
more things are pairwise similar without having any one quality in com-
mon. Camap, in the Aujbau, mentioned a difficulty of this sort (in Section
72), though Goodman is responsible for its name.
In discussing further difficulties, we shall also have to mention the rela-
tion of part similarity. Let us agree that letters which are next to each other
in their alphabetic order, and they alone, refer to qualities which are
approximately the same. Thus, a, b, and care thought of as qualities such
that any experience which has a is part similar to one which has b, and one
which has b is part similar to one which has c, but one which has a is not
part similar to one which has quality c.
62 ROLF A.EBERLE

D. The Difficulty of Accidental Overlap

The example about to be given is modified from one listed in the Struc-
ture (p. 175):
(1) a .. . (3) b.. . (5) c.. . (7) d.. . (9) e.. .
(2) a .. . (4) b.. . (6) c.. . (8) d.. . (10) e .. .
(11) bd.. .

By using Definition 2 '.Vith respect to part similarity, we can distinguish

the simi1arity circles

{l, 2, 3,4, lI} {3, 4, 5, 6, lI} {5, 6, 7, 8, lI}

{7, 8,9,10,11}
If we now use the tentatively proposed Definition 3 of a quality class, then
the intended class {3, 4, Il} of aH b's will not be a quality class, for it over-
laps the circle {5, 6, 7, 8, lI} without being contained in it and thus
violates Condition (1) ofthat definition. It is because ofsuch examples of
'accidenta1' overlap that Carnap favored Definition 4. Indeed, according
to that revised definition, {3, 4, Il} does turn out to be a quality class. 3
However, as Goodman has noted (regarding his slightly different
example on p. 175), the unit class {II} is also a quality class according to
both definitions. In order to exclude that example, Goodman proposed
(on p. 178) to replace the former Clause (2) in Definition 4 by the foHowing
condition:
(2') C is included in some similarity circle and, for every class K,
if K is included in all similarity circles with which K shares
more than half of its members and C is included in K, then
C=K.
Unfortunately, the class C = {5, 6} of all c's will then no longer be a
quality class since properly included in class K = {5, 6, Il} which meets the
specifications of (2 '). 4

E. The Difficulty of Finite Non-Cyclic Similarity Chains

Consider the example
(1) a .. . (3) ab ... (4) b.. . (6) c.. .
(2) a .. . (5) b.. . (7) c.. .
 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 63

It is meant to be essential to this example (but not to the previous one) that
the quality a is at the beginning (or at the end) of the similarity chain, so
that there is no further quality whose exemplars might be indicated to the
left of the experiences 1 and 2 in the above display. This situation could
arise if a were some color at the visible end of the spectrum or some just
barely audible sound. The similarity circ1es, as far as the example has been
elaborated, are just
{l, 2, 3, 4, 5} {3, 4, 5, 6, 7}
and it does not matter whether there are any further circ1es extending to
the right. Ifwe apply either Definition 3 or Definition 4 to test whether the
intended quality c1ass {l, 2, 3} of experiences which are a is one by defini-
tion, we find that it is not. That is so because e.g. 4 is not exc1uded from
any circ1e which inc1udes {1, 2, 3} and hence contributes to violate
Condition (2) ofthose definitions. Furthermore, it does not help to replace
Condition (2) by Goodman's Condition (2') since the c1ass C= {l, 2, 3} is
properly inc1uded in the c1ass K= {1, 2,3, 4} which meets the condition
of that c1ause. This difficulty will arise whenever a quality oecurs at the
end of a terminating similarity chain, and such chains will exist whenever
the number of qualities or experiences is finite and not cyc1ically ar-
ranged.
III. DISCUSSION OF THE DIFFICUL TIES

When he was writing the Aujbau, Carnap was aware of 'unfavorable

circumstances' in which his method of classification would not yield the
intended result. But it seemed to him that such circumstances were un-
likely to occur if one's experiences were sufficient1y numerous and varied
(Section 70), and that if they occurred then ones ordinary process of con-
cept formation would fail as well (Section 81). Thus, Carnap did not ex-
pect his definitions to have their intended sense in all 'possible worlds' of
experiences, but only in ones which an actual person, who has not been
subjected to sensory deprivation or selective stimulation, is likely to find
himself in the long ron. Given his own standards of adequacy, how serious
are the difficulties we have mentioned?
We may grant, 1 think, that the difficulty (C) of imperfect community
is unlikely to occur. But the companionship difficulty (B) is bound to arise
whenever there is a systematic connection between qualities a and b due to
64 ROLF A.EBERLE

which all a's are b's, so that the class of all a's cannot be constructed. Still,
so long as only simple phenomenal qualities are at issue, it is not easy to
think of actuallaw-like connections of that sort. Perhaps it is a law that
some minimal brightness is a constant companion of any color or that
the smell of onion is a constant companion of the taste of onion, so that
colors and onion-taste could not be constructed. But minimal brightness
is hardly a simple phenomenal quality; and if our tongues are fit to discem
only combinations of sweet, sour, salty, and bitter, then perhaps there
is no such quality as onion-taste which might be perceived even if not
accompanied by onion-smell.
Consider next the difficulty (D) of accidental overlap. Recall that
Camap originally required (in Definition 3) that a quality class be in-
cluded in every similarity circle with which it shares one of its members.
Due to that difficulty, he changed the requirement so that a quality class
had to be included in every circle with which it shares more than ! of its
members, the implicit assumption being that accidents cease to be such
if they happen more than half of the time. However, supposing that
accidents can be ruled out on proportionality grounds alone, why
should! be the magic number? Depending upon circumstances, either
! or 1- seem equally plausible cut-off points.
Still using Camap's standards of success, the difficulty (E) of finite non-
cyclic similarity chains seems especially serious because it is bound to
occur no matter how many elementary experiences of no matter how
varied a nature we may have. In order to get at the heart ofthe difficulty,
consider an example which is even simpler than the one previously given:
(2) b
(1) a (4) c
(3) ab
where the chain of adjacently similar experiences begins with 1. The cir-
cumstance due to which the difficulty cannot be overcome in Camap's
system lies in the fact that the experience 2, which is uniformly of quality
b, and the experience 3, which has both of the qualities a and b, bear the
relation of part similarity to precisely the same entities and are therefore
indistinguishable with respect to that relation. In such circumstances
then, and so long as part similarity is one's only primitive, possession of
simple and compound qualities cannot be distinguished regardless ofhow
quality classes may be defined. On the other hand, the experiences 2 and 3
 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 65

can be distinguished with respect to the relation of part identity: every-

thing which is part identical with 2 is part identical with 3, but not con-
versely, which serves to define a qualitative part-whole relation. Ac-
cordingly, we seem to require, as primitives, both notions of similarity.
We need part similarity in order to obtain an ordering of qualities with
respect to their qualitative nearness, and we need part identity in order to
obtain a qualitative part-whole relation. 5
Carnap's own attitude towards the two relations of similarity is some-
what surprising. He recommended (in Section 72) that we initially ignore
the differences between part identity and part similarity and that we con-
struct similarity circ1es and quality c1asses with respect to both relations.
We would then discover that similarity cyc1es which are constructed rela-
tive to part similarity show more mutual overlapping than one' s which are
constructed relative to part identity and could thereby distinguish the two
relations. Yet formally, part identity is not primitive but so defined as to
presuppose part similarity, similarity circ1es, and quality c1asses. This
disparity between the intuitive and the systematic order of construction
seems methodologically suspect. The suspicion grows as we discover that
part identity is so defined that part identical things need not even be part
similar, and becomes overwhelming in view of the difficulty mentioned in
the preceding paragraph. Surely, Carnap should have taken both simi-
larity relations for primitive.
Let us now turn to a method of construction which differs somewhat
from Carnap's and whose formation was stimulated by the following
considerations: In the Structure, Goodman represents simple phenomena1
qualities by atoms, so-called 'qualia', which have the feature that any two
individuals which overlap an atom overlap one another. Now, the rela-
tion of part identity is analogous to that of overlapping, and Carnap's
'quasi constituents' play a role somewhat like that played by Goodman's
'parts'. This suggests that there might be c1asses of experiences which
function in Carnap's system much like atoms behave in Goodman's sys-
tem. To that effect, let us say that a c1ass A of elementary experiences is an
atomic class just in case A is a smallest c1ass satisfying the condition that
any two elementary experiences which are part identical with all members
of A are also part identical with one another. Thus, a c1ass of phenomena
which appear uniformly zinc yellow but differ with respect to their non-
visual qualities might be such a c1ass if any two entities which are part
66 ROLF A.EBERLE

identical with all of these phenomena must be zinc yellow and hence part
identical with one another. A quality class, we can say next, is any largest
class of elementary experiences which includes a given atomic class A
and is such that all ofits members are part identical with all members of A.
Although only part identity is mentioned in these definitions, part
similarity will still be needed in ordering quality classes so as to obtain
quality-stretches, such as yellow, or sense realms, such as sound.
These definitions seem to meet all genuine difficulties we have previ-
ously listed, except the companionship difficulty and a version of im-
perfect community which is quite unlikely to occur. We shall briefly in-
dicate why this is so while assuming that the difficulties about to be
mentioned are 'pure' forms, not also involving the companionship diffi-
culty.
Starting with the last mentioned difficulty (E) of finite non-cyclic simi-
larity chains, the present definition serves to pick out the intended quality
class {1, 2, 3} of that example because the class {1, 2} will be an atomic
class (unless a is a companion of some other quality which we have ruled
out by assumption). Regarding the illustration of the difficulty (D) of
accidental overlap, it now turns out that the wanted class {3, 4, Il} of
all b's is indeed a quality class because (still assuming that companionship
does not occur) the class {3, 4} is an atomic class. On the other hand, the
unwanted unit class {Il} of that example is no longer a quality class or
even an atomic class. Finally, if the difficu1ty (C) of imperfect community
occurs as previously illustrated, then it too is satisfactorily resolved. For
the unwanted class {1, 2, 3} does not include an atomic class, and the
atomic classes are just the three singletons {4}, {5}, {6} each of which
represents just one uniform quality.
So, if we accept the criteria of adequacy mentioned in the Aujbau and
if there are no systematic connections between simple phenomenal quali-
ties, then the present method of construction seems to work satisfactorily
and, 1 think, about as well as any method can work which employs only
part similarity and part identity.
For the convenience of readers who want to see how our definitions
work in the presence of the companionship difficulty, we reproduce here
the former example:

(1) br (2) b (3) bg (4) g (5) bgr

 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 67

Call an elementary experience an atom if its singleton is an atomic class.

The trouble with this example is that there are two different atoms, 1 and
2, which are part identical and indistinguishable, with respect to part iden-
tity, from all other experiences. Since indistinguishable, the two atoms
cannot give rise to different quality classes (that of alI r's and that of alI
b's). ActualIy, the atoms 1 and 2 might stiH be distinguishable with respect
to part similarity: something might be part similar to 1 and not to 2 by
virtue ofthe quality r. If so, we could stiH characterize the r's as the class
of aU those b's which are part similar to that entity (we let the reader
supply the general definition).
For similar reasons, the difficulty of imperfect community can stiU
arise in improbable circumstances. E.g: if the only experiences are:
(1) bg
(2) gr
(3) rb
then each ofthese experienc;es wiH be an atom and indistinguishable, rela-
tive to part identity, from all other experiences in the set. GeneraUy, our
method will fail whenever there are different but indistinguishable atoms.
Camap himself, in 'Replies' [2], acknowledged the need for rather
drastic revisions ofhis former system. Apparently, he had either come to
realize that the 'unfavorable circumstances' in which his construction
failed were more likely to occur than he had formerly thought, or else he
had come to expect more of his system; perhaps that it should work in aU
conceivable circumstances. Now, if our present method delivers correct
results up to indistinguishability of atoms, then further improvements
can apparently be gained in only two ways: either (1) we change the
atoms of the system so as to make them qualitatively less complex
or (2) we revise the basic relations of the system so that they wiH serve
to discriminate between previously indistinguishable atoms. In the
Structure, Goodman chose the former approach. His atoms are qualita-
tively so simple that any two of them which differ at aH can be distin-
guished relative to their relations of overlapping. Camap, in 'Replies',
also seemed to favor a change of basic units. They were to be minimal
concreta, such as temporal slices of color spots (a specific color, together
with a location and time), rather than Gestalt-phenomena like the ele-
mentaryexperiences of the Aujbau. The choice of such units immediately
68 ROLF A.EBERLE

suggests the question whether the old difficulties could not arise again
with respect to whatever qualities remain inseparable quasi-parts of the
new atoms. Perhaps it is a law, for example, that visual sharpness or dis-
tinctness occurs only at the center of one's tield ofvision, so that the com-
panionship difficulty might still arise with respect to visual sharpness and
centrality. One is really safe only if one's basic units contain no more than
one qualitative quasi-part.
For the sake of historical accuracy it should be mentioned that
Carnap, in the 'Replies', proposed not only a revision ofbasic units, but
also an enrichment of th~ conceptual apparatus expressed by primitives.
Indeed, he was willing to use primitives which refer to sensory qualities
(such as red) outright. Of course, doing so avoids all difficulties as well as
the entire program of quality construction which presently concerns us.
Between the two basic approaches for improving the system, that of
chosing new basic units and that of employing resemblance relations of
somewhat greater expressive power, the latter seems to me preferable.
If we set out to make our atoms qualitatively so simple as to avoid all
difficulties, then we seem forced to render our basic units so abstract and
so far removed from one's which might serve in an empirical theory of
perception, that they become formally equivalent to Goodman's qualia.
Even if such entities should not be objectionable merely on the grounds
that they are abstract and contrived, their introduction would narrow the
range of applicability of the system. Instead, we shall try to tind a method
of quality construction which works whether one's basic units be elemen-
tary experiences, Mach's elements, or any other phenomena which it
might be advantageous to regard as epistemologically or psychologically
basic. Preferably, the method should be applicable not only to phe-
nomenal entities, but also to physical objects. These goals, it seems to me,
Carnap himself would have endorsed

IV. A GENERALL Y APPLICABLE METHOD OF

QUALITY CONSTRUCTION

As we have seen, the relations of part identity and part similarity are
so weak that circumstances can arise where qualitatively different entities
remain indistinguishable with respect to those relations. Accordingly,
somewhat stronger similarity relations are needed. We begin with a
 QUALlTY CLASSES IMPROVED UPON THE 'AUFBAU' 69

triadic relation

x exactly resembles y but not z

or, more explicitly, 'in a certain respect, x exactly resembles y but not z'.
InformaUy interpreted, 'x exactly resembles y but not z' shaU be true of
individuals x, y, and z just in case there is some quality Q such that both
x and y have Q but z does not. The judgements corresponding to that
relation consist in fixing on some quality and then both comparing and
contrasting objects with respect to that same quality.
Among quality classes, we pay special attention to those which are not
coextensional with proper logical compounds of other quality classes,
and caU them elementary quality classes. Regarding them, the foUowing
principle holds :

Assumption of Extensionality. There is no elementary quality

Q and no set S of qualities other than Q such that (1) for aU
individuals x, x has Q if and only if x has aU the qualities in S,
or (2) for aU individuals x, x has Q if and only if x has some
quality in S.

Suppose, e.g., that exactly those individuals should be sour which are
also both green and soft or are also either green or soft. Carnap's notion of
a quality, and ours, are such that we do not caU the compound property of
being both green and soft a 'quality', nor the compound property of
being either green or soft. Since we are only concerned with qualities-in-
extension, being sour will be the same as one of those compound proper-
ties. Under such circumstances, we call the quality of being sour 'non-
elementary'. As one may expect, the characterization of non-elementary
quality classes will pose special problems in every extensional system.
We shaU further assume that the number of qualities isfinite. Also, let
us make it part of what we mean by an 'elementary quality' that at least
one individual has it and at least one individual lacks it. We exclude
empty and universal elementary quality c/asses chiefly for reasons of con-
venience. The assumptions we have made so far could serve to detine the
notion of an elementary quality class.
In order to see how our triadic resemblance relation can be employed to
construct quality classes in difficult circumstances, consider once again
70 ROLF A. EBERLE

the example of companionship:

(1) br (4) g
(2) b (5) bgr
(3) bg
The class of alI individuals which exact1y resemble 2 but not 4 will give us
the class {1, 2, 3, 5} which is the elementary class of alI b's. The class of
alI individuals which exact1y resemble 4 but not 2 is just {3, 4, 5}, the class
of alI g's. And the elementary class {1, 5} of alI r's, which we could not
previously isolate, is just the class of all objects which exact1y resemble
1 but not 2. So, if for all individuals y and z we construct the non-empty
classes of all things which exact1y resemble y but not z, then we will ob-
tain all elementary quality classes ofthis example. In addition, we will also
get certain undesirable unions of quality classes which will have to be
thrown out upon construction in order to remain true to Carnap's notion
of a quality class.
Or consider an especially virulent form of imperfect community:
(1) abc (3) cde
(2) bcd (4) dea
Again, in constructing all non-empty classes of the form {x:x exact1y
resembles y but not z}, where y and zare the individuals 1-4, we obtain
alI intended quality classes. We also get some unwanted unions (e.g. if
y = 1 and z = 3) which will have to be weeded out.
There are cases of quality construction where the relation of part iden-
tity is also required. E.g. in the example
(1) abd (3) b (5) d
(2) ace (4) c (6) e
the elementary class {1, 2} of all a's can be obtained as the class of all
things which are part identical with both 1 and 2, but not by the previous
method. Apparent1y, we also need part identity as a primitive. However,
if one is willing to abandon universal qualities, as we henceforth are, then
part identity can be defined:

D EFINITION 6. x is part identical with y just in case there is a z such that

x exact1y resembles y but not z.
 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 71

By extrapolating from examples, we arrive at the general definitions:

DE FINITION 7. C is a pseudo quality class just in case C is non-empty

and for some individuals x and y, either C is the class of aH things which
exactly resemble x but not y, or C is the class of alI things which are part
identical with both x and y.

DEFINITION 8. C is an elementary quality class if and only if C is a

pseudo quality class but not a union of two or more pseudo quality classes
aH of which differ from C.

Thus, we first construct all classes of the form {x: x exactly resembles y but
not z} and one's of the form {x:x is part identical with y & x is part
identical with z}, and then we eliminate aH those which are unions of two
or more other constructed classes. Since we have assumed that the num-
ber of qualities is finite, this procedure should be effective. Furthermore,
one can prove (by strong induction on the number of qualities which an
arbitrary individual may havel that this method will yield exactly the
elementary quality classes in aH conceivable circumstances.
While we can construct aH elementary qualities, how about the non-
elementary ones? Unfortunately, qualities which are coextensional with
genuine intersections cannot be obtained by present methods. Thus,
if the example earlier given of the 'difficulty' of extensionality is felt to
present a real difficulty, then it will remain so for our present approach.
However, we can stiH regain some qualities which are coextensional
with genuine unions, for this reason: Camap's construction of similarity
circles seemed to yield unwanted classes only in circumstances which
exemplified the difficulty (C) of imperfect community. But once we have
elementary quality classes at our disposal, we can specify the conditions
under which that difficulty will arise. If that difficulty should indeed be
the only one, we can say that any union of elementary quality classes
shall again be a quality class provided it is a similarity circle and the ele-
mentary classes of the union cannot be cyclicalIy arranged. If similarity
circles should yield too many classes in other circumstances, then chances
are that we can again specify those circumstances as conditions on ele-
mentary quality classes. This leaves us with unobtainable disjunctive
qualities of only one sort: such as are coextensive with unions of cyclic
72 ROLF A. EBER LE

elementary classes, as in this example:

(1) abx (3) cax
(2) bcx (4) d
It appears then that we can get alI quality classes except the ones which are
coextensional with undesirable candidates, where the condition which
makes them undesirable can itself be specified within the system. 1 rather
think that this is the maximum one can achieve so long as one remains
perfectly neutral with respect to the basic individuals (and does not, e.g.,
make them qualities outright) and so long as one employs only ex-
tensional methods (and does not, e.g., refer to unactualized possible
combinations of qualities).
The triadic resemblance relation we have employed so far is an ex-
tension of pact identity and will yield quality classes in circumstances
where Carnap wanted to represent qualities by similarity circles. How-
ever, we still need a resemblance relation, somewhat like that of part
similarity, which captures some sense of qualitative nearness and gives
rise to an ordering of qualities. To that end, let us use the triadic relation
x resembles y but not z
or, more fulIy expressed, 'in a certain respect, x is approximately like y
but unlike z'. In rough outline, the truth conditions for such formulas
might be given as folIows: We associate with every interpretation a set of
formally identifiable entities which can represent qualities and a binary
similarity relation between qualities which is reflexive and symmetric.
Then we can say that 'x resembles y but not z' shall be true of individuals
x, y, and z just in case there exists a quality F such that x has F and for
some quality G, y has G and G is similar to F, but there is no quality H
such that z has Hand H is similar to F.
1 do not know whether part identity is definable in terms of this new
primitive. To be on the safe side, make it also a primitive.
Now construct elementary quality classes exactly as it was done in
Definition 7 and Definition 8, but replace alI former mention of exact
resemblance by mention of resemblance. So defined, the former difficulty
(D) of accidental overlap no longer occurs. For we can now construct the
formerly elusive class {3, 4, Il} of alI things which are b as the elementary
class ofthings which resemble 5 but not 7. Furthermore, the difficulty (E)
 QUALITY CLASSES IMPROVED UPON THE 'AUFBAU' 73

of finite non-cyclic similarity chains is also blocked: we can obtain the

formerly unspecifiable class {l, 2, 3} of an things which are a as the
elementary class of an objects which resemble 4 but not 6. Although 1
have no proof, working with examples sustains the hope that this method
will yield exactly the elementary quality classes in an similarity chains
which contain more than one similarity circle. It seems that resemblance
serves to construct an elementary quality classes which can be defined
in terms of exact resemblance and will, in addition, give us a qualitative
ordering. Apparently then, a triadic resemblance relation and possibly
part identity suffice to carry out an basic quality constructions which an
extensional system, such as the Aujbau, can hope to attain.

NOTES

1 The sections correspond to equally numbered ones in the Aujbau [1], whether the
German original or its translation are considered.
2 Page numbers associated with the Structure [4] refer to the second, and not to the original
edition.
3 Goodman's original example on p. 175 of the Structure leaves doubts whether the dif-
ficulty involved is essentially that of accidental overlap or that of finite non-cyc1ic similarity
chains. Also, it does not serve to illustrate the kind of difficulty due to which Carnap
changed from Definition 3 of 'quality c1ass' to Definition 4. For, c1ause (2) of those defini-
tions remains the same; and in Goodman's original example, the questionable c1ass K=
= {l, 2, 3, 19} also violates clause (2) of those definitions, due to the fact that e.g. 4 is not
in K but also not exc1uded from any similarity circ1e which inc1udes K.
4 Actually, the class C = {5, 6} of ali c's is not a quality c1ass even according to Definition 3
or Definition 4, since II is not exc1uded from any similarity circ1e which inc1udes C. But that
circumstance, unlike the point we make in the text, seems due only to the 'accident' that
we have not extended the similarity chain of our example sufficiently far.
S In a slightly different context, this point was already made in [3], p. 167.

BIBLIOGRAPHY

[1] Carnap, Rudolf, [Aujbau], Der Logische Aujbau der Welt (1928); second ed. Der
Logische Aujbau der Welt - Scheinprobleme in der Philosophie, Hamburg (1961); transl.
by Rolf A. George with title The Logical Structure ofthe World and Pseudoproblems in
Philosophy, The University of California Press, Berkeley and Los Angeles, 1969.
[2] Camap, Rudolf, ['Replies'], 'Replies and Systematic Expositions', in Paul A. Schilpp
(ed.), The Philosophy of Rudolf Carnap, The Library of Living Philosophers, La Salle,
Ill., 1965.
[3] Eberle, Rolf A., Nominalistic Systems, D. Reidel, Dordrecht-Holland, 1970.
[4] Goodman, Nelson, [Structure], The Structure of Appearance, Harvard University
Press, Cambridge, Mass., 1951; second ed., Bobbs-Merrill, 1966.
 RUDOLF CAR NAP

OBSERVATION LANGUAGE
AND THEORETICAL LANGUAGE*

ABSTRACT. (Original English.) Among the non-logica! constants ofthe language of science
two kinds are distinguished, the observation terms (e.g., 'blue') and the theoretica! terms
(e.g., 'electric field'). The latter terms are introduced, not by definitions, but by postulates
of two kinds, theoretica! postulates, e.g., basic laws of physics, and correspondence
postulates which connect the theoretical terms with observation terms. As Hilbert has ex-
plained, both mathematics and theoretica! physics can in this way be constructed in the
form of uninterpreted calculi. It is here briefly indicated that by this method of construc-
tion also the mathematica! terms have meanings (in a wider sense) assigned to them. The
theoretica! terms obtain at least an incomplete interpretation by means of the corre-
spondence postulates. It is shown how the distinction between analytic and synthetic sen-
tences can be defined also for the theoretica! language.

I. INTRODUCTION

In modem investigations of the language of science it has often proved

valuable to distinguish between observation language and theoreticallan-
guage. The-first contains sentences such as 'This thing is hard, white, and
cold'; the latter such as 'At the space time point with the coordinates
x, y, z, t, the components of the electrical field have such and such
values'. The observation language contains onlyan elementary logic. The
sentences of this language are assumed to be fulIy understood. The
meaning of the theoreticallanguage, on the other hand, always remains
incomplete. It possesses a very comprehensive logic which contains
alI of classical mathematics.
This paper will discuss some questions on the relations between these
two languages in an informal way. The structures of the languages of
mathematics and physics as parts of the theoreticallanguage will be made
clear. Further, it will be made clear how the distinction between analytic
and synthetic sentences, which usually is made only for completely inter-
preted languages, can be extended to the theoretical language. Paul
Bemays has recently made some interesting critical remarks on some
points in the development of logical empiricism. In the present paper 1
shall, among other things, answer some of the questions Bernays has

Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 75-85. AII rights reserved.
Copyright Dialectica and the Rudolf Carnap Trust.
76 RUDOLF CARNAP

raised, and, on some points, present my current position or at least briefly

indicate it.
For brevity 1 shall use the folIowing symbolism, let L be the language
considered here which contains the observation language Lo and the
theoretical language LT as sublanguages. The constants of L will be
divided into logical and descriptive (i.e., nonlogical) constants; the mathe-
matical constants will be regarded as logical. The primitive descriptive
constants of Lo we shall call observation terms or O-terms. These stand
for observable objects or processes, and observable properties and rela-
tions among such objects (e.g. 'Zurich', 'cold', 'harder than'). The
primitive descriptive constants of LT we call theoretical terms or T-terms.
We assume that only terms which are not explicitly definable by the
O-terms will be taken as T-terms. These might be fundamental terms of
theoretical physics (e.g., 'mass', 'temperature').

II. THE LANGUAGE OF MATHEMATICS

Let the structure of LT be such that it contains a type-theoretic logic with

an infinite sequence of domains DO, D 1 , D 2 , etc. D" is called the domain
of the nth level. Each variable and each constant belongs to a definite
level. A variable of the nth level has D" as its range of values and a con-
stant of the nth level denotes an element of D". The elements of DO are
the members of the infinite sequence O, O', O", etc. Officially, no meaning
is given to these logical expressions. Their use follows from the rules of
the language. We shall, however, in order to relate these expressions to
familiar concepts, unofficially regard DO as the domain of natural num-
bers with 'O' denoting the number O, 'O' the number 1, etc. Further, we
will regard D"+ \ for arbitrary n, as the domain of alI subclasses of D".
Thus D 1 is the domain of alI classes of natural numbers, D 2 the domain
of alI classes of such classes, etc. A universal sentence of the form
'(x") ( ... x" ... )' says that ... x" ... holds for every element x" of D".
An ordered m-tuple of arbitrary objects of various domains may be
defined as a class of higher level. An m-place relation can be regarded as
a class of higher level. An m-place relation can be regarded as a class of
m-tuples and a function as a one-many relation.
In this way, L contains expressions and variables for alI objects that
appear in classical mathematics. Numbers of various sorts (natural,
 OBSERVATIONAL AND THEORETICAL LANGUAGE 77

rationals, reals, and complex numbers), relations between numbers,

functions of numbers, and so forth. But we make no ontological assump-
tion about the existence (in a metaphysical sense) of such objects. We only
introduce expressions which follow definite syntactical rules. In other
words, we specify a ca1culus. The syntactical rules, for example primitive
sentences and rules of inference, are formulated in a syntactical meta-
language. The fundamental syntactical rules, such as rules of formation,
specifications of primitive sentences, or axiom schemata, and rules of
inference, specify genuinely performable operations with symbols. For
these rules, a metalanguage with only an elementary logic similar to that
of Lo is sufficient.
Such a structure of mathematics corresponds to the procedures rec-
ommended by Hilbert and carried out by him, together with Bernays.
These methods can also be accepted by those of an intuitionist, con-
structivist, or even a nominalist orientation, since the elementary syn-
tactical metalanguage satisfies the requirements which these orientations
impose on a meaningfullanguage. Further, an elementary semantics can
be formulated in an elementary metalanguage. Beyond this, however,
there is also the possibility of formulating a theoretical syntax and a
theoretical semantics in a richer metalanguage which is in part a theo-
reticallanguage and therefore not completely interpreted. Concepts such
as, truth and logical truth for a language as comprehensive as L can be
formulated only in such a theoretical metalanguage.
Although officially the type-theory of the L language is constructed
only as an uninterpreted calculus, it seems to me that one can regard this
language as meaningful in a wider sense. For instance, formalists as well
as intuitionists agree that a formula such as '2 + 3 = 5' is meaningful. In
order to regard the whole mathematical sublanguage as meaningful, we
must make further assumptions:
(la) Universal sentences over the elements of DO are meaningful.
(1 b) Ifuniversal sentences over any domain of objects is meaning-
fuI then so are the universal sentences over classes of such
objects.
(lc) The conjunction of two meaningful sentences is meaningful;
the negation of a meaningful sentence is meaningful.
The intuitionists would reject this and so would Hilbert and Bernays,
in case 'meaningful' was meant in a strong sense. On the other hand, the
78 RUDOLF CARNAP

majority ofpractical working mathematicians would make these assump-

tions and, as argument for it, would point to the fact that in century-long
work count1ess results on real numbers and sets and functions of such
have been found while relatively few differences of meaning have turned
up. If now, we regard formulas as meaningful which have been in use
by a large group of people for a long time, as a fruitful means of mutual
understanding, then we must acknowledge the formulas of c1assical
mathematics as meaningful in this sense.
On the other hand, it must be admitted that a universal sentence over
c1asses of natural numbers is less c1ear than one over natural numbers
and that the c1arity of our thought, and therewith the certainty of the
insights we have won through going to higher levels, gets less and less.
The antinomies have impressively shown us that even plausible insights
turn out to be invalid at times. In my view, however, it does not folIow
that formulas with higher variables are meaningless but only that our
work with these formulas is uncertain. Furthermore, we know from
GOdel's results that even logic with variables of lowest level, in other
words, the arithmetic of the natural numbers can not be completed by
effective rules and that the incompleteness increases with higher levels.
Even when we, according to Hilbert's suggestion, renounce any c1aim
to insight as to the validity of a mathematical theorem or axiom through
understanding the meaning, and satisfy ourselves instead with regarding
the mathematical system as a mere calculus proving only its consistency,
we know today, again through Gode1, that new deductive means are
necessary, themselves becoming less certain as the calculus becomes more
comprehensive. Weyl felt discouraged and depressed by these results. 1
But there seems to me no sufficient ground for that. In physics it is even
less possible to reach certainty that the basic laws, taken as postulates,
wilI be in accord with alI future observations. Every physicist is always
aware that perhaps tomorrow a lack of correspondence will be dis-
covered. He is prepared ahead of time in this case, to modify his system.
Similarly in mathematics when we tind a paradox we modify the pos-
tulates. There is no ground to despair. In this view I am in agreement with
Hilbert and GOdel. Even Weyl says that we cannot greatly object to it. 2
It must however, emphasize that I see only a remote analogy between
physics and mathematics, namely the impossibility of absolute cer-
tainty. I would not regard logical-mathematical intuition, that is, the
 OBSERVATIONAL ANO THEORETICAL LANGUAGE 79

grasping of a conceptual relation~hip, as an epistemological analogue to

observation of a factual relationship between observed objects, as some
conceptual realists do. 3

III. THE LANGUAGE OF THEORETICAL SCIENCE

From the language of mathematics as above described, the theoretical

language LT is obtained by adding the descriptive T-terms. For the
assertion of facts and the formulation of laws, in any empirical scien-
tific theory, descriptive terms are obviously necessary. In this, it seems
to me that the old saying that the book of nature is written in the lan-
guage of mathematics is quite misleading.
For the introduction of T-terms two new kinds of postulates must now
be laid down. First the theoretical postulates, or T-postulates, which
contain no O-terms but only T-terms, and secondly, the correspondence
postulates, or C-postulates, which contain both O-terms and T-terms.
The T-postulates might be fundamental laws of the scientific theory
concemed. The C-postulates link the T-terms with the O-terms and
thereby give certain empirical, 'operational' rules for the use of the
T-terms (e.g. 'when one body is warmer than another then the tempera-
ture of the first is higher than that of the second') (it seems better not to
speak of 'operational definitions' since a T-term is not explicitly definable
on the basis of O-terms). We assume that the number of both kinds of
postulates is finite. Let T be the conjunction of the T-postulates, C that
of the C-postulates, TC that of both together.
This method of introducing theoretical terms through postulates has
been discussed often in the last fifty years from various standpoints,
especially for terms of theoretical physics. 4 Hilbert has emphasized the
analogy between such a construction of a system of theoretical physics
and the formalist construction of mathematics. Both of them can be
regarded as uninterpreted calculi. As earlier mentioned, however, it is
possible to provide the sentences of a mathematical calculus a meaning,
though not with fully established clarity. And the term and sentences of
the physical system can receive, through their linking with O-terms in the
C-postulates, a meaning that similarity is never complete, since it is al-
ways pos~ible, by adding further C- or T-postulates to strengthen the
meaning. Despite the similarity in this respect, there is, however, a
80 RUDOLF CAR NAP

fundamental difference between the basic mathematical signs and the

physical T-terms, the first are logical constants, the latter descriptive.
Among the many methodological problems associated with the con-
struction of a theoretical system, say for theoretical physics, on the basis
of T- and C-postulates, 1 wish to pick out two and discuss them briefly.
The first concerns the theoretical objects of physics, the second the dis-
tinction between analytic and synthetic sentences in a theoretical system.
The position is often taken that the introduction of theoretical terms
for such things in theoretical physics as forces, fields, partic1es, and the
like involves certain conceptual difficulties. How should we conceive
these things which we can never direct1y observe? How should we con-
struct a general conceptual scheme in which not only the object of an
already given scheme of physics may fit, but also others, perhaps forces,
partic1es, or special objects of an entirely new kind of which we present1y
have no conception but which a physicist might introduce tomorrow?
These difficulties do not enter for the form of theoreticallanguage de-
scribed above. Here only the following two types of objects enter. First,
the things designated by the O-terms, namely the observable things and
their observable properties and relations and second, the mathematical
objects of the types of systems described earlier, namely the natural
numbers, c1asses of these and so forth. 1 cannot carry out the construc-
tion here but only indicate some of the features of the structure through
examples.
We shall assume that the number of all observable objects is either
finite or denumerably infinite and that these objects are numbered in an
arbitrary way (for example, let body k have the number 17). Let us now
consider a physical magnitude whose value for every observable bod)' is
a real number (e.g., the mass of the body). This magnitude can then be
construed in our system as a function f whose numerical value is not
ascribed to the body itself but rather to its number (e.g., the proposition
that body k has mass 5 would be expressed by the sentence j(17)=5'.
The function symbol j' is here, to be sure, a descriptive constant but the
function f is identical with a mathematical function 9 (i.e., for every
n,/(n)=g(n)), which be10ngs to the type system earlier described). (This
circumstance is analogous to the following, which may be easier to see.
Let the constant 'np' be defined as 'the cardinal number of the planets'.
This constant is descriptive, to be sure, but the thing described by it is a
 OBSERVATIONAL ANO THEORETICAL LANGUAGE 81

natural number which belongs therefore to the domain of DO. The

number np is identical with 9, but the identity sentence 'np= 9' is syn-
thetic. We can furthermore introduce a space-time coordinate system in
which each small body at any time point can be assigned an ordered
quadruple of real numbers. Then by generalization, every such quadruple
can be regarded as representative of a space-time point (therefore as an
unobservable theoretical object.) Then physical magnitudes, such as
mass-density, may be introduced, which have a value for every space-
time point e.g., a real number. A function of this sort is construed in our
system by a function F of quadruples of real numbers. Here again, F is
identical with a mathematical function of the same logical type. A
physical particle is now conceived of as a four dimensional space time
region (with very small spatial, and perhaps very large temporal, ex-
tension), inside of which certain physical magnitudes have a charac-
teristic distribution. (An electron in the classical sense is, for example,
characterized by a definite local distribution of electrical charge-density
and mass-density.) These examples suffice to show that it is not necessary
to assume new sorts of objects for the descriptive T-terms of theoretical
physics. These terms designate mathematical objects, for example, num-
bers of functions of numbers or the like which, however, are physically
characterized, namely, so they have the relations to the observable pro-
cesses established by the C-postulates while simultaneously satisfying the
conditions given in the T-postulates.

IV. THE OISTINCTION BETWEEN ANALYTIC ANO

SYNTHETIC IN THEORETICAL LANGUAGES

We call a true sentence logically true, in the narrow sense, when its truth
is already established by the meaning of its logical constants. As a tech-
nical term, (explication) for this concept we take 'L-true'. A customary
definition for this is as follows:
Dl. A sentence Sis called L-true (in the language L), ifit is either
a true sentence without descriptive constants or results from
one by substituting in descriptive constants. We say that Sl
L-implies S2 or that S2 follows logically from Sl when Sl ~S2
is L-true. We say that Sl is L-equivalent with S2 if Sl=S2
is L-true.
We call a true sentence analytic (or logically true in the wider sense)
82 RUDOLF CARNAP

if its truth is determined solely by the meaning of the constants it contains

including the descriptive ones (e.g., 'No bachelor is married'). As a
technical term for this concept we will take 'A-true'. This concept can be
defined on the basis of meaning postulates which represent meaning
relationships which hold among the primitive descriptive constants. 5 We
will here call these postulates A-postulates. Let A be the conjunction of
the A-postulates in a given language, then for this language we lay down
the following definition.
D2. A sentence S is called A-true when it follows from the A-
postulates.
If, for example, 'bachelor' and 'married' were primitive predicates in a
given language and their meanings were such that they logically excluded
each other, then the following sentence might be taken as a postulate
'for every x, if x is a bachelor, then x is not married'. Since the meanings
of the O-terms, according to assumption, are completely clearly under-
stood, laying down of A-postulates for them presents no difficulty in
principle. Let Ao be the conjunction ofthe A-postulates for the O-terms.
In contrast to the O-terms, the T-terms are not fully interpreted. They
have only as much interpretation as is attributed to them through the
postulate-TC but these postulates have a double function. They not only
contribute to the determination of the meanings of the T-terms but they
also set forth the factual content of the theory. This is evident from the
fact that with their help, along with sentences on observable processes,
predictions can be derived of future observable processes. Hence, a C-
postulate or T-postulate cannot in general (aside from certain special
cases which we here leave aside) be taken as an A-postulate. This is one
of the difficulties of defining the concept of A-truth for sentences with
T-terms. 6 On these and other grounds some philosophers have even held
this task to be unsolvable. 7 As far as I know, no solution of this has
previously been given.
It seems to me, however, that there is a way to overcome these dif-
ficulties. The procedure which I will now describe 8 uses an idea from
F. P. Ramsey.9 Ramsey suggested transforming a given theory TC in the
following manner into a sentence R, without T-terms. If n different
T-terms appear then these are to be replaced throughout in TC by n
distinct variables of corresponding types which do not appear in TC.
Then the sentence R is constructed by prefix ing n existential operators
 OBSERVATIONAL AND THEORETICAL LANGUAGE 83

with these n variables. The O-terms appearing in C remain undisturbed.

Ramsey showed that for this sentence R the following holds.
(2) Every sentence without T-terms, which follows from TC also,
foHows from R.
(The converse naturally holds also since R follows from Te.) Therefore,
we can regard R as a representation of the empirical content, or more
precisely, for the observational content of the theory Te. The sentence R
says that there are n mathematical objects which satisfy the postulate TC
or, more exactly, that are related to observable processes by the relation
established by C, and which bear among themselves the relations given
in T. Therefore, Ris, in general, a factual sentence which can be checked
by observation. Ramsey proposed to use the existential sentence R in-
stead of the postulate TC and therefore to avoid the T-terms entirely.
That has certain advantages but also brings with it certain complica-
tions. Therefore 1 retain the T-terms but select difIerent postulates for
the purpose of distinguishing between analytic and synthetic. We lay
down the sentence R as a P-postulate i.e., as a synthetic, e.g., physical
postulate. As A-postulate AT for the T-terms we make the conditional
sentence R::J TC. It can be easily shown that for this sentence AT the
foHowing holds.
(3a) The conjunction R.A T is L-equivalent with Te.
(3b) Every sentence without T-terms that follows from AT is
L-true.
Taking AT as A-postulate for the T-terms (and for the O-terms ap-
pearing in C) seems justified in view of the results mentioned. For, first,
this sentence which contains aH T-terms, together with the sentence R
which contains no T-terms, say just as much as the postulate TC, as
(3a) shows, and, secondly, AT as seen in (3b), makes no factual assertion
about what is observable. AT says only the following: In case the wor1d is
so constructed that any n-tuple at aH of mathematical entities exists
which satisfies TC then the T-terms are to be understood in such a way
that the things designated by them form such an n-tuple.
In the first form of the theory, in which the sentences of TC are taken
as postulates, the sentences R and AT are theorems. In the second form,
in which the sentence is R and AT appear as postulates, the sentences of
TC are theorems. We have here, therefore, two equivalent representa-
tions of the same theory.
84 RUDOLF CARNAP

The A-postulate Ao for the O-terms, together with AT now form the
A-postulate for the whole language L, therefore, we define:
D3. A sentence Sis called A -true in the language L when it follows
from Ao and A T. Furthermore, we may define:
D4. A sentence Sis called P-true in the language L, when it fol-
lows from Ao, AT and R. Therewith, we have reached our
goal, namely, an explication of the concept of an analytic
sentence for a language with theoretical terms.

NOTES

* Translated by H. G. Bohnert from Dialectica 12 (1958), 236-248.

1 Weyl [1], last paragraph.
2 Weyl [2], p. 212.
3 On this, see my reply in [3] to Wilfrid Sellars, with whom 1 agree on this point.
4 For recent publications, see Carnap [2] and other papers in Minnesota Studies (and in
the forthcoming Volume III) and their bibliographies.
5 Meaning postulates are thoroughly discussed in Carnap [1].
6 These difficulties are discussed, for example, by Hempel [1] and [2].
7 For example Quine in [1] and [2].
8 This procedure is discussed in detail in Camap [4].
9 The methodological significance of the Ramsey-sentence R is extensively discussed in
Hempe1 [1].
BIBLIOGRAPHY

Bemays, P. [1], 'Von der Syntax der Sprache zur Philosophie der Wissenschaften',
Dialectica 11 (1957), 233-246.
Carnap, R. [1], 'Meaning Postulates', Philos. Studies 3 (l952), 65-73. (Reprinted in: Carnap,
Meaning and Necessity, 2nd edition, Chicago, 1956.)
Carnap, R. [2], 'The Methodological Character of Theoretical Concepts', Minnesota
Studies [1], 38-76.
Camap, R. [3], 'Replies and Systematic Expositions', in: Schilpp [2].
Carnap, R. [4], 'CarI G. Hempel on Scientific Theories'. (Reply to Hempel [2] in Camap [3].)
Giidel, K. [1], 'Russell's Mathematical Logic', in: Schilpp [1].
Hempel, C. G. [1], 'The Theoretician's Dilemma: A Study in the Logic of Theory Con-
struction', in: Minnesota Studies [2],37-98.
Hempel, C. G. [2], 'Implications of Carnap's Work for the Philosophy of Science', in:
Schilpp [3].
Minnesota Studies in the Philosophy of Science [1], VoI. 1 (ed. by H. Feigl and M. Scriven),
Minneapolis, Minn., 1956.
Minnesota Studies in the Philosophy of Science [2], VoI. II, (ed. by H. Feigl, M. Scriven, and
G. Maxwell), Minneapolis, Minn. 1958.
Quine, W. V. [Il, 'Two Dogmas of Empiricism', Phil. Rev. 50 (1951), 20-43. (Reprinted in:
Quine, From a Logical Point of View, Cambridge, Mass., 1953.)
 OBSERVATIONAL AND THEORETICAL LANGUAGE 85

Quine, W. V. [2), 'Carnap and Logica! Truth', in: Schilpp [2). (Reprinted in part in: S. Hook
(ed.), American Philosophers at Work, New York, 1956.)
Ramsey, F. P. [1), 'Theories', 1929. Published as Chapter IX A in: The Foundations of
Mathematics, London and New York, 1931.
Schilpp, P. A. (ed.) [1], The Philosophy of Bertrand Russell, The Library of Living
Philosophers, New York, 1944.
Schilpp, P. A. (ed.) [2), The Philosophy of Rudolf Carnap, The Library of Living Philos-
ophers, LaSalle, III., 1963.
Weyl, H. [1), 'Mathematics and Logic', Amer. Math. Monthly 53 (1946),2-13.
Weyl, H. [2), 'Review of Schilpp' [1), Ibid., 208-214.
 DAVID KAPLAN

SIGNIFICANCE AND ANALYTICITY*

A Comment on Some Recent Proposals of Carnap

ABSTRACT. The most recent, and surely the most notable, attempt to make precise the
empiricist thesis:
A sentence is meaningful if and only if it is connected with experience
is that of Carnap in his article, 'The Methodological Use of Theoretical Concepts' [1 J.
Carnap's approach is to distinguish between observation terms and theoretical terms. He
then proposes a method of distinguishing 'significant' theoretical terms from 'non-
significant' theoretical terms by means of their connection as given by some theory T with
certain observation terms. The present paper reports two consequences of that proposal.
Given almost any theory T, first there is a definitional extension 1'* of T such that every
theoretical term of T* (including those of T) is significant (according to Carnap's proposal)
with respect to the theory T*; and secondly there is a 'deoccamization' T** of T such that
no theoretical term of T** is significant (according to Carnap's proposal) with respect to
the theory TU. The interest in these two results lies in the fact that definitions, though
ordinarily thought of as adding no empirical content to a theory, seem to have the power
(according to Carnap's proposal) of transforming non-significant terms into significant
ones; and the process of deoccamization (which consists of 'splitting' a theoretical term into
a conjunction or disjunction of two new theoretical terms) which would ordinarily be
thought of as subtracting no empirical content from the theory, seems to have the power
(according to Carnap's proposal) of transforming a significant theory into a non-signi-
ficant one. The possibility of attaining these two results is thought to constitute an in-
adequacy in Carnap's proposal.
Many empiricists have maintained some version of the thesis:
A sentence is cognitive1y meaningful if and only if it is
connected with experience.
The connection, however, appears to have weakened with age. When
the thesis was first suggested, as part of a program to eradicate meta-
physics (or at least that part of it which was considered unconnected), the
required connection amounted to an actual identification of cognitive
meanings with certain experiences. But slogans like, 'the meaning of a
sentence is its method of verification' have gradually faded to requests
that, '(meaningful concepts) be logically connectible with the terms of a
suitably chosen observation basis'.
The danger that metaphysics may thus be readmitted to philosophy
has been compounded by another tendency in recent formulations of the
thesis. They are specified in such a precise way that certain results can be

Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 87-94. AII rights reserved.
Copyright David Kaplan.
88 DA VID KAPLAN

shown to be unquestionably entailed by them. Thus, for example, when

Ayer proposed that "a statement is meaningful, if some observation
statement can be deduced from it in conjunction with certain other
premises, without being deducible from those premises alone", Berlin
was able to point out that any non-analytic sentence S was thereby
meaningful since there is always some observation sentence O, which can
be deduced from S in conjunction with S~O, but cannot be deduced
from S~O alone. 1 (Note that if Sis non-analytic there is some O such
that it is not the case that (~S f- O), but then it is not the case that
(S ~ O f- O).) Since Ayer's criterion had been so c1early formulated he was
forced to either accept that aH non-analytic sentences are meaningful
or drop the criterion. Ayer, by the way, took the latter course and
proposed a new criterion which Church pointed out had the consequence
that if there are three independent observation sentences, every sentence
or its negation is meaningful. This tendency toward rather precise for-
mulations has put the eradicators in a rather weak position vis a vis the
eradicatees (i.e., the metaphysicians). It has been shown that various
specifications of the required connection between sentences and ex-
perience have entailed that an sentences are meaningful, or that nega-
tions of meaningful sentences might be meaningless, or that no universal
sentences are meaningful, or (a'lld this has caused the most embarrassment
lately) that theoretical physics is meaningless!
A metaphysician who was up on the literature in the philosophy of
science might have availed himself of the opportunity to c1aim that of
course his formulations were remote from experience, they are part of
the theoreticallanguage. It appears to be extreme1y difficult to toe that
fine line between the electron and the absolute. The most recent and
surely the most notable attempt in this direction is that of Camap in his
artic1e, 'The Methodological Use of Theoretical Concepts' which occurs
in Volume 1 of Minnesota Studies in the Philosophy ofScience. The purpose
of this paper is to report certain consequences of that proposal.
Camap's proposed definition is relative to a kind of language often
discussed in investigations into the methodology of science. It is divided
into two parts, an observation language Lo and a theoreticallanguage L T •
The observation language contains terms referring to observable proper-
ties and relations, etc., and the theoretical language contains terms
referring to unobservables. Let VT be the c1ass of descriptive (i.e., non-
 SIGNIFICANCE AND ANALYTICITY 89

logical) constants of LT and Va the c1ass of descriptive constants of La.

We also consider a finitely axiomatizable theory in LT and a finite set of
connecting postulates. The theory can be represented by the conjunction
of its axioms, T, where T is a sentence of L T ; and the connecting postulates
can be represented by their conjunction C, which contains terms of
both VT and Va. The sentences of La are usuaUy thought of as com-
pletely understood, thus the realm of insignificance is relegated to V T.
Carnap approaches the problem of cuHing out the significant sentences by
way of the elements of VT which occur in them. Re caUs a sentence
significant just in case aU elements of VT which occur in it are significant.
The previously mentioned thesis is now read:
An element of V T is significant with respect to La> LT' T,
and C if and only if it is connected with La.
The problem is then to specify the connection in terms of La, LT' T, and C.
This is done by means of two definitions:
D 1. A term m is significant relative to the class K of terms, with respect to
LT' La, T and C = DF the terms of K belong to VT, m belongs to VT but not
to K, and there are three sentences, SM and SK in LT and Sa in L", such
that the following conditions are fulfilled:
(a) SM contains m as its sole descriptive term.
(b) The descriptive terms in SK belong to K.
(c) The conjunction SM. SK. T. C is consistent (i.e., not logically false).
(d) (SM.SK.T.C)f-Sa
(e) NOt((SK.T.C)f-Sa )
D2. A term m is significant with respect to LT' La> T and C =DF there is a
sequence M of terms of V T, such that the last element of M is m and each
element of M is significant relative to the c1ass of those terms which pre-
cede it with respect to LT' La> T, and C.
For means of illustration 'let us study a simple L", LT' Y", Vr. T, C.
Va = {J, P, B}, La = the c1ass of aH sentences of first order logic with
identity, containing only elements of Va as non-logical signs. VT=
= {b,j, g, h, m, n}, LT=the c1ass of aU sentences offirst order logic with
identity, containing only elements of VT as non-logical signs.
T=(x) (hx=>jx).(x) (hx=>(bxv "-'gx)).(x) (mx=:nx) ,
C=(x) (Bx=>hx).(x) (Jx=>Jx).(x) (gx=>Px).
90 DA VID KAPLAN

It is easily shown from D 1 that g is significant relative to the nul/ class

ofterms (taking SM as (x) gx, SK as (x) (x=x), and So as (x) Px). Therefore
by D2, g is significant (taking M as the one term sequence (g»). A similar
procedure establishes that f and h are significant. We show that b is
significant relative to {g}, by taking SM as (3x),....,bx, SK as (x) gx, and So
as (3x),....,Bx. Note, however, that m and n cannot be 'connected' to ex-
perience in the sense of the above definitions, since they occur in what
Carnap calls an isolated postulate. No sequence ofterms beginning with
elements of Vo and passing to new terms which occur with some previous
term of the sequence in a conjunct of T or C can ever reach m or n. Ac-
cording to Carnap's definition they are not significant, which accords
well with our intuition.
Let us call T' a definitional extension of T if it is formed from T by the
addition of some definitions for new constants in terms of the primitives
of T. Let VT' and LT' be the corresponding extensions of VT and L T.
Note that definitional extensions of a theory are ordinarily thought of as
having no more empirical content than the original theory.
1 will now show that there is a definitional extension T' of T such that
every element of VT , (including m and n) is significant with respect to
Lo, LT" T' and C.
Let us form T' by conjoining the following four definitions to T.
Definition (x) (d1 x == (mx . (3x) (Ix))).
Definition (x) (d2 x == (mx => (3x) (gx))).
Definition (x) (d3 x == (nx . (3x) (Ix))).
Definition (x) (d4 x == (nx => (3x) (gx))).
It can be immediately seen that d 1 is significant relative to the nuH class,
(taking SM as (x) d1 x, SK as (x)(x=x) and So as (3x) Jx). Thus d1 is
significant with respect to Lo, LT" T' and C. We can next show that d2 is
significant relative to {dd, (taking SM as (x) d2 x, SK as (3x) d1 x, and So as
(3x) Px). Since we have established that d1 is significant, d2 is also. Ifwe
now take SM as (x) mx, SK as (3x) d2 x and So as (3x) Px, we can establish
that m is significant relative to {d2 } and hence, as before, significant.
An analogous procedure would show that d3 , d4 , and n are also
significant. This completes our task since the constants of VT , which were
significant with respect to Lo, LT' T and C will obviously also be signi-
ficant with respect to Lo, LT" T', and C. (The actual details of the above
proofs are facilitated by the use of a few well-known theorems about
 SIGNIFICANCE AND ANALYTICITY 91

definitiona1 extensions, e.g., that the addition of a definition to a set of

premises in primitive notation does not increase the class of derivable
sentences in primitive notation.)
Let us call T'. C' a deoccamization of T. C if it is formed from T. C
either by rep1acing aH occurrences of certain elements of VT by the con-
junction of two new primitive constants of the same type, or by replacing
aH occurrences of certain elements of VT by the disjunction of two new
primitive constants of the same type. Let VT' and LT' be the result of the
corresponding replacements in VT and L T • The idea of a deoccamization
is that a theoretical term is split into two terms. For example aH oc-
currences of some one place predicate P foHowed by a single term IX,
might be replaced by the conjunction plex,. p 2r:x.. Although we would not
look with favor upon such a multiplication of entities beyond necessity, 1
think we would not say that a deoccamization of a theory and its connec-
ting postulates can rob it of empirical content. In fact, any deductive sys-
tematization of the sentences of Lo established by T. C is likewise
established by any deoccamization of T. C. This is easily shown as fol-
lows.
Let 0 1 and O 2 be two sentences of Lo such that
(1) Not(1-0 1=>02)'
(2) I-T.C=>(01=>02)'
We then say that T. C establishes 0 1 => O 2 , But if T'. C' is a deoccamiza-
tion of T. C, it is a substitution instance of T. C. From (2) by substitution
we can derive,
(3) 1-T'.C'=>(01=>02)'
Since the terms being substituted upon do not occur in 0 1 or O 2 , they
(0 1 and O 2) are unaffected. Hence T'. C' establishes 0 1=>0 2,
1 will now display a deoccamization T". C" of a deoccamization
T' . C' of T. C such that no element of VT" is significant with respect to
Loo LT'" T" and C". First,

V T' = {b,fl ,12, gl , g2, h 1, h2, m, n} LT' = the appropriate class

of sentences .
T'=(x) ((h 1xv h2x)=>(j1xv j2x)).
(x)((h 1x V h2x)=> (bx v ",(glx.g 2x))).(x)(mx=nx).
C' = (x) (Bx=> (h 1x v h2x)).
(x) ((flx V j 2 x)=> Jx).(x) ((glX.g 2X)=>Px).
92 DA VID KAPLAN

We now deoccamize T'. C' to obtain:

V T" = {b , fl. 1 , fl.2 , f2.1 , f2.2 "
gl g2
" "
h 1 h 2 m n} ,
LT" = the appropriate class of sentences ,
T" (x) ((h 1x V h2x)=> ((11. l X. fl, 2X) V (p.1X. p. 2 X)).
=
(X) ((h 1xv h2x)=>(bxv ",(glX.g 2X))) (X) (mx=nx).
C" = (X) (Bx=> (h 1x v h2x)).
(X) (((Il. l X. fl. 2X) V (l2.1X. p. 2X))=> JX).
(X) ((glX.g 2X)=>PX).

It is beyond the scope of this paper to offer a proof that no element of

VT" is significant with respect to Lo, LT'" T", and C", but the intuition is
quite straightforward. The basic idea is that those terms which were
shown to be significant relative to the nuH class of terms with respect to
Lo, LT' T and C (namely f, g, and h), have been split in such a way
that if one attempts to reproduce the original argument with respect to
either component, one needs to know that the other component is signi-
ficant. (The second deoccamization was required by the connection
established in the first conjunct of T between hand f) Thus the sequence
of terms referred to in Carnap's D2, can have no beginning.
The former result, that definitions can import significance, suggests
that Carnap's proposal may be too weak; whereas the present result, that
deoccamization can export significance suggests that it may be too strong.
Let us return to the earlier re suIt. It can be generalized under various
sets of hypotheses about Lo, LT' V T, T, and C. One of the weakest 1 have
been able to find is the following. The hypothesis says approximately that
there are at least two descriptive constants in V T that lead independently
to different observational results. To be more precise:

If LT' Lo, V T, T, and Care the appropriate entities, and

(1) Pl E VT,P2E VT, and
(2) there are sentences P1, P2, 01, O 2 such that P1 containsPl as
its sole descriptive constant, P2 contains P2 as its sole descrip-
tiv~ constant, alE L o' O E Lo, and
(3) (Pl' P2· T. C) is consistent
(4) Pl' T. C'r 01, Not [T. C. 'r 01]' and
(5) P2 • T. C'r 0b Not [T. C'r O 2 ], and
(6) Not [Pl' T.C'r O 2]
 SIGNIFICANCE AND ANALYTICITY 93

then there is a theory T' such that T' is a definitional extension

of T and every eligible constant of VT , is significant with
respect to La> LT" T' and C.
I call a term of the theoretical vocabulary eligible if there is some sentence
S which contains the term as its sole descriptive constant and such that
neither S nor its negation is implied by T. C.
Carnap considers in his article a slightly stronger criterion obtained
by a modification of D 1. It has the result that none of the theoretical
constants of our original model Lo, LT' V T' Va' T, C would be con-
sidered significant. However, one can still prove a theorem analogous to
the above but with a slightly more restrictive hypothesis.
Were it not for the fact that philosophers are licensed to put rather
simple observations in seemingly profound settings, one might object that
the foregoing merely indicates that Carnap should have restricted the
application of his criterion to the primitive constants of V T . This objec-
tion appears somewhat stronger when one realizes that Carnap did in fact
restrict the application of his criterion to the primitive constants of VT .
Even this observation does not seem to me to completely rob the theorem
ofvalue, since I, for one, find it of greatest interest to learn that there was
a good reason for something I have done. However one can draw furtner
consequences in view of another proposal of Carnap's on a criterion for
analytic sentences of the theoreticallanguage. This proposal appears in
an article entitled, 'Beobachtungssprache und theoretische Sprache', in
Dialectica for 1958. 2 His proposal has the consequence that if T' has the
form of a definitional extension of T (where the added constants are now
thought of as new primitives so that the criterion is clearly applicable),
then the sentence T' == T is analytic in T'. Again it appears that in a cer-
tain sense, T' has no more empirica! content than T. Thus it seems in-
appropriate for the realm of significance to be increased.

UCLA
NOTES
* Paper delivered at the annual meeting of the American Philosophical Association,
Pacific Division, held at Santa Barbara, December, 1959. © David Kaplan.
1 Throughout the body of the paper, all logical signs with the exception of '~' are used
autonymously. Concatenation is indicated by juxtaposition.
2 For an English translation, see above pp. 75-85.
94 DA VID KAPLAN

POSTSCRIPT 1975

This paper appears exact1y as it was written sixteen years ago. It was
written to be heard not printed (thus the absence of proofs of the general
results), but because ofCarnap's reaction to it, reported in [5J, 1 thought
the original version might have some curiosity value. At the same meeting
at which this paper was presented Carnap began a turn in a different di-
rection with his [6J. But others have pursued refinements of the line
herein criticized (see, for example, [7J and a reply in [8J, and see [9J
and the bibliography therein for refinements of Ayer's original proposal).
My objection regarding definitional extensions was independent1y dis-
covered in 1962 by W6jcicki and is reported in [10]. Rozeboom, whose
own criticisms of Carnap's definition appeared in [11 J, suggested, in a
private communication, a simplification of the argument regarding defi-
nitional extensions which depended on definitions whose definiens con-
tain both theoretical and observation terms, but such a procedure now
seems to me not in the spirit of Carnap's program.

BIBLIOGRAPHY

[1] Carnap, R., 'The Methodological Use of Theoretical Concepts', in H. FeigI and M.
Scriven (eds.), Minnesota Studies in the Philosophy of Science, Volume 1, University
of Minnesota Press, Minneapolis, 1956.
[2] Ayer, A. J., Language, Truth and Logic, Victor Gollancz, London, 1936.
[3] Berlin, 1., 'Verification', Proceedings ofthe Aristotelian Society 39 (1938-9), 225-248.
[4] Church, A., Review of Second edition of [2] The Journal of Symbolic Logic 14 (1949),
52-53.
[5] Kaplan, D., 'A Memory of Rudolf Carnap', in R. Buck and R. Cohen (eds.), Boston
Studies in the Philosophy of Science, Volume VIII, D. ReideJ, Dordrecht-Holland,
1971.
[6] Carnap, R., 'On the Use of Hilberts e-Operator in Scientific Theories' in Y. Bar-Hillel
et al. (eds.), Essays on the Foundations of Mathematics, Dedicated to A. A. Fraenkel
on His Seventieth Anniversary, Magnes Press, Jerusalem, 1961 and North-Holland,
Amsterdam, 1962.
[7] Achinstein, P., 'Theoretical Terms and Partial Interpretation', British J.for the Phi-
losophy of Science 14 (1963-4), 89-105.
[8] Bohnert, H., Review of [7] The Journal of Symbolic Logic 36 (1971),321-322.
[9] Yourgrau, W. and Works, C., 'A New, Formalized Version of the Verifiability Prin-
ciple', Ratio 10 (1968), 54-63.
[10] W6jcicki, R., 'Semantical Criteria of Empirical Meaningfulness', Studia Logica 19
(1966), 75-102.
[11] Rozeboom, W. W., 'A Note on Carnap's Meaning Criterion', Philosophical Studies
11 (1960), 33-38.
 RYSZARD W6JCICKI

THE FACTUAL CONTENT OF

EMPIRICAL THEORIES

This paper is concerned with the distinction between factual and con-
ventional (analytic) truth. The problem is one of the most controversial
one's in the methodology of science, and no solution to it is likely to be
commonly accepted. The proposal for dealing with the issue that 1 am
going to examine here is different from the one given byCarnap (cf. [11],
[12]); nevertheless is belongs to the same philosophical tradition. 1 share
both Carnap's empiricist attitude and his logical orientation in dealing
with problems of philosophy of science.
The fust section of the paper is an expository one. It gives a concise
presentation of both Carnap's solution to the problem of analyticity and
certain further contributions in the field. The second and the third sec-
tions are also of a preparatory character. The main ideas ofthis approach
to the problem of factual truth and analyticity that 1 am going to discuss
here will be presented in the fourth section. The fifth contains a brief
discussion of the notion of terminological convention. Finally the sixth,
which is the last one, shows how Carnap's solution can be reconstructed
within the conceptual framework set up in the paper.

1. CARNAP'S PARADIGM OF SOLUTIONS TO THE PROBLEM

OF ANALYTICITY

The solutions to the problem of analyticity which have been offered by

Carnap and by those who took his approach as a paradigm for their
own investigations derive from a more general account of the logical
structure of empirical theories. The basic assumptions of this account
were established by Carnap in number ofwritings (cf. especially [10] and
[13]). The same or essentially similar assumptions were accepted by a
great number of other philosophers, cf. e.g. Hempel [18], [19],
Przelycki [28], and Tuomela [37]. According to this account empirical
theories are assumed to be formalized within first order predicate cal-
culus. Each such theory () is conceived of as the set oflogical consequences

Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 95-122. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
96 R YSZARD W6JCICKI

of certain sentences selected as the postulates of (). The postulates are

divided into two mutualIy exclusive classes: the set of theoretical pos-
tulates P and the set of correspondence rules (postulates) C. In contra-
distinction to C, the theoretical postulates Pare defined to contain no
observational terms. Thus aH extralogical terms which appear in Pare
theoretical. Denote them by T 1 , ••• , Tm and folIowing the widely spread
custom, call them T-terms. Similarly, denote the observational terms
appearing in the correspondence postulates C by 0 1 , ••. , Om, and call
them O-terms. The notion of T-sentence and that of O-sentence will be
understood in an obvious manner. It should be made clear, however, that
the sentences we shall speak about may belong to the language Lo of the
theory () as well as to its second order strengthening. The sentences which
contain both O-terms and T-terms will be called mixed. By definition, the
correspondence postulates of () are either mixed or O-sentences.
Each O-term is assumed to possess a fixed and welI-defined meaning
(interpretation) established independent1y from the theory (). Carnap
has taken an even more radical point of view: he has demanded the
meanings of O-terms to be independent of any empirical theory whatso-
ever. This radical attitude is not shared, however, by alI his folIowers
(cf., e.g., Sneed [35], Tuomela [37]). The semantic status of T-terms is
entirely different. The postulates of the theory () are the sole source of the
meanings T-terms possess. An interpretation (in general merely partial)
is effected for T-terms by restricting their proper interpretations to those
that establish the truth of the postulates of () given prior interpretations
of O-terms. This account of meanings of theoretical terms is often called
the partial interpretation view, as the meanings do not thus become fulIy
specified.
Given any set of sentences X we shaH define the O-content (empirical
content) of X to be the set of alI O-sentences which are 10gicalIyentailed by
X (this seems to be the most natural definition of empirical content under
the assumption that aH of the evidence available for us can be stated in
O-terms). (For some comments on this definition cf. Glymour [17]). It
should be stressed that we arrive at two different concepts of empirical
content depending on whether second-order O-sentences are assumed to
be included into the O-content or not. Two sets of sentences A and B
which have the same O-content in the 'first-order' sense need not be 0-
equivalent (i.e. their O-contents need not coincide) in the 'second-order'
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 97

sense. (For the proof of this assertion see Przelycki and W6jcicki [30],
cf. also Sneed [35].) Differences between the 'first-order' approach which
can be called proof-theoretic, and the 'second-order' approach which
amounts to the model-theoretic one, and the relevance of these dif-
ferences for the problem of analyticity were discussed in Przelycki and
W6jcicki [29], and, from a purely technical point ofview, in [30]. Here
we shall assume that the factual content comprises all, i.e. both first
and second-order, O-sentences that are entailed by the set of sentences in
question.
In his examination of the problem of analyticity, Carnap confined him-
self to finitely axiomatizable theories. Although fully general methods of
dealing with the problem are known (cf. Przelycki and W6jcicki [29] and
Williams [38]), we shall not go beyond finitely axiomatizable theories
here, and thus we shall assume that the set ofaxioms P u C of () is finite.
We are allowed, then, to replace P u C by a finite conjunction n of its
elements. Consider the second order formula n(t b ... , tn) obtained from
n by proper simultaneous replacement of T-terms by predicate variables
ti' Its existential closure

is known as a Ramsey Sentence. In what follows we shall use R(n) as an ab-

breviation for (1). It is easily seen that the empirical content of R(n) is
exactly the same as that of Pu C. It is also a matter of a straightforward
argument to show that the implication

(2) R(n)--+n,

which from now on will be abbreviated as C(n) and called Carnap sen-
tence, has truth value independent from any empirical evidence. Indeed,
if R(n) is false then C(n) is vacuously true. If, under the given interpreta-
tion of O-terms, R(n) is true, then the truth of n, and thus C(n) as well,
is assured by the requirement imposed on the intended interpretations of
T-terms; note that R(n) states the existence of such interpretations of
T-terms which verify n. We may summarize then our analysis by saying
that C(n1 and consequent1y all sentences it entails, are true by convention,
or equivalent1y, that they are analytic. Since the conjunction R(n) /\ C(n) is
immediately seen to be equivalent to P u C we hafe succeeded in replacing
98 R YSZARD W6JCICKI

the postulates P u C by the purely factual postulate R (n) and the purely
analytic one C(n). The solution to the problem of analyticity we have
presented is just the one that was given by Camap.
By producing a suitable example (cf. e.g. [29]) it can be shown that
there may exist a T-consequence A. of P u C which being logically stronger
than C(n) (i.e. A. entails C(n) but the converse does not hold), stiH satisfies
the folIowing two conditions.
(i) AII O-sentences entailed by A. are logically true.
(ii) The conjunction A. /\ R(n) is equivalent to Pu C.
Since A. is stronger than C(n), it is not analytic. On the other hand, in
view of (i), A. can neither be proved nor disproved on the base of any
empirical data. Is then Camap's solution adequate? Should not we
count A. analytic as welI? An answer to this question essentially depends
on what are the semantic rules selected to determine the meanings of
T-terms. Unfortunately, scientists do not take much care about con-
structing empirical theories in the form of well-defined semantic systems
ready to be examined by logicians and philosophers. If alI that we know
about() reduces to thatP and Care sets ofpostulates ofthis theory, weare
not in a good position to settle conc1usively what are the semantic rules for
T-terms accepted by the relevant group of scienticists.
In our exposition of Camap's settlement we have admitted that saying
that the postulates P u C stipulate the meanings of T-terms amounts to
saying that the following semantic convention (rule) is accepted.
(CI) Whenever possible, T-terms should be interpreted insuch a way
as to establish the truth of the sentences in P u C given prior
interpretations of O-terms.
As a matter of fact, the rule (CI) deviates from the one that was sug-
gested by Camap. It is rather the one that was analysed by Ajdukiewicz
[2] (cf. also Kokoszynska [23]). The rule proposed by Camap is the
following one.
(C2) T-terms should be interpreted in such a way as to establish the
truth of C(n) given prior interpretations of O-terms.
The differences between (CI) and (C2) are rather subtle (cf., e.g.,
Przelecki and W6jcicki [29]) and can be ignored. Infact, if we admit
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 99

that, whenever it proves impossible to find an interpretation of T-terms

that verifies Pu C, we are free in selecting any interpretation of T-terms
whatsoever, then (CI) and (C2) turn out to be equivalent in the sense that
they yield the truth of exactly the same sentences. With the help of either
of these rules, we may establish, making no reference to the meimings of
O-terms, the truth of alI sentences and only those which are logical con-
sequences of Carnap sentence C(n). Thus, if no semantic rule but (CI)
or (C2) governs the meanings of T-terms, the sentence ,1, which has given
rise to our discussion, turns out not to be analytic.
Approving the settlement, we may ask all the same why the following
convention should not be singled out.
(C3) T-terms ought to be interpreted in such a wayas to establish
the truth of A. given prior interpretations of O-terms.
It may be taken for granted that each consistent set of semantic rules
for a given language defines it as a semantic system. Obviously, certain
semantic systems may have no practical value, for example for being too
weak, nevertheless good or bad stiH they do exist. Whenever, what we
shall assume, the set of rules that determine the denotations for O-terms is
consistent, the adding (C3) to this set cannot resuIt in any inconsistency.
This follows from that A. satisfies the condition (ii). Then, there exists a
semantic system in which (C3) is valid, and in this system the sentence A. is,
no doubt, true by convention and hence analytic 1. 1 see no possible fauIt
with this argument and therefore, 1 am not able to share Winnie's con-
viction (cf. [41]) that no answer to the question of what sentences are
analyticly true other than that given by Carnap should be sound.
Winnie's argument in favour of Caniap's solution is based on the
assumption that each analytic sentence has to be empirically vacuous
what implies that we are not allowed to accept rule (C3) until we succeed
in proving that it meets this requirement. Here, you may be confused, for
to say that A. is empirically vacuous seems to mean exactly the same as to
say that A. is empirically equivalent to the empty set. But if the require-
ment of empirical vacuity is understood in this way, clearly it is satisfied
by A. in virtue of (ii). However, Winnie's definition of an empirically
vacuous sentence differs from one proposed above 2. Although, Winnie's
proposal seems to be of considerable interest for methodology of science,
1 wonder why we must not accept semantic rules which establish
100 R YSZARD W6JCICKI

analyticity of sentences being not empirically vacuous in Winnie sense.

If this "must no1" can be motivated at all it can be motivated solely by a
pragmatic argument, i.e. by showing that the semantic systems that
violate it do not serve their purpose.
Irrespective of the eventual settlement of the issue, we certainly.must
admit that as long as we have no general recipe for constructing A,
the rule (C3) (incidentally, (Cl}-{C3) are rather schemata of rules than
genuine rules) cannot be subjected to any wider inquiry. The full gener-
ality is one of important advantages of Carnap's solution. An attempt to
construct a definition of analyticity as general as Carnap's, one still being
not equivalent to the latter, was undertaken by myself in [43] (cf. also
remarks at the end of the Section III of Przelycki and W6jcicki [29]).
1 did not succeed, however, in proving that my proposals result in a solu-
tion different from that given by Carnap. The problem has been settled
recently by Williams in his incisive and elegant paper [38].
Needless to say that all these attempts to overcome some real or
perhaps sometimes plausible shortcomings of Carnap's solution to the
problem of analyticity do not undermine the importance of Carnap's
analyses. On the contrary, the variety of difficult and deep problems that
were raised and solved within the conceptual framework developed by
Carnap and in full accordance with the substantial features of the
paradigm he set up confirms in the most convincing way the exceptional
purport of his work.

2. CERTAIN CRITICAL COMMENTS ON

THEORETICAL-OllSERV A TION AL DICHOTOMY

Carnap's and his followers' approach to the problem of conventional

truth hinges heavily on the assumption that the language of science can
be divided in two by distinction between observational and theoretical
predicates. This assumption was questioned so many times and by so
many authors (cf., e.g., Achinstein [1], Bunge [6], Dalla Chiara Scabia
and Torlando di Francia [15], Popper [27], Putnam [31]) that it seems to
be hardly possible to say something new in the area. Thus, 1 shall con-
fine myself to making a short comment on the issue.
The main trouble with theoretical-observational dichotomy consists in
that the overwhelming majority of extralogical terms involved in scien-
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 101

tific inquiry act depending on where and when they are applied either as
observational or as theoretical. Using certain predicates to assert some
properties of objects of a particular kind and placed in particular ex-
ternal conditions we may arrive at perfect1y observational sentences ready
to be easily and convincingly verified by means of the relevant empirical
procedures. None the less, an application of the same predicates, in
exact1y the same way but in different circumstances, e.g. to some other
objects or perhaps to the same objects but considered in external condi-
tions changed, may provide us with sentences that are perfect1y immune
to any empirical testing and thus that are theoretic. It was admitted by
Carnap and argued by most of his critics that theoretica1 and observa-
tional concepts 'lie on a continuum'. I am afraid that this elaim, although
it seems to be evidentely true, is in a way misleading.
It certainly may happen that the predicates involved in an empirical
theory can be ordered in accordance with the regularly growing 'degree
of observability' we may ascribe to them. It may also happen that some
of the predicates of the theory are 'highly observational' (i.e. in most of
their applications they behave as observational terms) while the re-
maining are 'highly theoretical'. Clearly, while the latter case gives us an
example when the theoretical-observational dichotomy works well, the
former provides us with an example when aH settlements establishing this
dichotomy are disputable. But, it may also happen as well, and I am
afraid that this is typica1 for a great deal of empirical theories, that aH
predicates of the theory are in nearly the same degree observational and
thus they are in nearly the same degree theoretical. If they lie on the
continuum, they unfortunate1y lie in about the same place and aH pos-
sible ways of dividing them into observational and theoretical are com-
pletely unacceptable, for aH of them drasticaHy distort the picture of real
interrelations amongst the terms. Obviously, I assume here, that the
notion of an observational term is not understood in an orthodox
manner, i.e. not only predicates of the every-day language such as 'red'
or 'squere' deserve to be caHed observational but also, for instance,
physical quantities such as 'position' or 'mass', provided that there are
measuring procedures which, in a given domain, enable us to establish
their values. The orthodox formulation of the doctrine of logical em-
piricism seems to possess only historical value.
It is elaimed, sometimes, that the partial interpretation view is false. I
102 RYSZARD W6JCICKI

am afraid however that this elaim is incorrect for the same reasons for
which it would be incorrect to say that Classical Mechanics is false. In a
sense all descriptive theories are false, if you like, for none of them brings us
an absolutely correct account ofphysical reality. From this point ofview
neither Classical Mechanics, nor Relativistic Mechanics are true. On the
other hand both of them provide us with reasonably accurate description
of certain phenomena and the difference between them does not consist iti
that one of the two theories is true while the other is false, but rather in
that the range of applications of Classical Mechanics is considerably
narrower than that of the relativistic one.
The partial interpretation view is not so much wrong as over-
simplified. Its shortcomings affect essentially the range of applications of
possible results of all analyses taking it as a starting point. These short-
comings, however, cannot be overcome by just rejecting the observa-
tional-theoretical dichotomy and denying the validity of the results that
presuppose it. We face the task of working out the fundamentals of a new
approach to the problems we have dealt with and to apprise the old results
with the vantage point of the new paradigm.

3. AN OUT LINE OF SEMANTICS FOR EMPIRICAL THEORIES

Assume, once again, an empirica! theory () to be given. The symbol Le

will be used, as hitherto, to denote the language of (). The set of all sen-
tences asserted within () (postulates, laws etc.) will be denoted by Ae. We
do not require Ae to be deductively closed. Assuming that a fixed set of
rules of inference is singled out as the set of rules of inference of (), we
shall write
(1) A hx,

whenever the sentence (X is provable by means of those rules from the set
of sentences A.
The language Le will be assumed to be the language that can be
described as the language of a mathematical theory 'ile augmented by
certain new symbols FI' ... , Fm call them the specific symbols of (). They
are to play the role of the symbols of certain empirical parameters, i.e.,
quantities and perhaps qualities the theory () deals with. From the
purely formal point of view, however, the symbols Fi can be treated as
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 103

variables (or rather, as they cannot be bound by quantifiers, as uninter-

preted constants) ranging over certain sets of objects examined in '\l1J. For
example the symbol s of position of a partic1e is (from the formal point
of view), a variable representing, say, a curve in Euc1idean space, or,
perhaps, a variable representing a function which maps Cartesian
product P x Re, where P denotes a set of formal objects called partic1es
and Re is the set of real numbers into Cartesian power Re 3 • These are
obviously two different ways of conceiving s, but this is notorious that
we can apply different mathematical formalisms to describe the same
empirical phenomena
Selecting any mathematical interpretation of the symbols Fi results in
defining a so-called standard model structure, sms, for LIJ; we assume
all symbols of LIJ but Fi to possess well-defined and fixed meanings.
Taking for granted that it is known what it means for a sentence of the lan-
guage to be true in an sms, we shall define IX to be mathematically true
if and only if IX is true in all sms's and we shall define IX to be a (mathe-
matica!) consequence of a set of sentences A (to be mathematically entailed
by A) if and only if IX is true in an sms whenever all sentences in A are true
in this structure. If A entails IX we shall write
(2) Alf-IX.
The relations f- and If- need not coincide even if all inference rules ad-
mitted in () are sound in the sense that they allow us to infer IX from A only
if IX is entailed by A. The relation ofprovability may be 'weaker' than that
of entailment and, if we require the notion of proof to be effective, it will
be in general weaker. Se1ecting the rules of inference to be valid in () we
should try, for obvious reasons, to select only the sound ones. Let us
introduce the notion of a well-adjusted empirical theory and let us state
the following condition that characterizes it partially.
(WAl) In order for () to be well adjusted ali rules of inference valid in ()
must be sound.
The condition will be followed by certain furthers, which will be stated
later.
To facilitate our further discussion, let us assume that each term Fi is
to be interpreted in sms's as a function of the form
104 R YSZARD W6JCICKI

where ah"" ak range in each sms over a fixed set of objects P (being dif-
ferent for different sms's) t ranges over a fixed finite interval T of real
numbers (once again T may differ from one structure to another) and
finally x is a real number. The subscript k should be treated as a function
of i, we shall keep it, however, constant 3. Some special conditions can
be assumed to be satisfied by the functions F;'s and similarly some mathe-
matical requirements can be imposed on the objects in P. For instance
some F;'s may be demanded to be twice differentiable, some of the
others may be demanded to take only positive values.
Our assumption, which confines the set of all parameters to be dis-
cussed to the functions of the form (3), limits rather drastidy the area of
our considerations. Losing on generality we gain on simplicity, however.
It may also make us feeI more comfortable when we observe that some
non-trivial empirical theories (e.g. Classical Partide Mechanics) are not
ruled out by the restrictions placed on the F;'s. Furthermore, we may
expect that our considerations might serve as a pattern for similar ones
carried out under certain other, perhaps more liberal, assumptions.
Thus far, the theory Owas considered as a purely formal system, in the
subsequent discussion to be conducted in this section we shall be con-
cerned with empirical interpretations for e. Denote by Ro the set of all
physical (empirical) systems, which comprise the set of all intended appli-
cations of O, i.e. the set of all those physical systems that Ois supposed to
describe. The set Ro will be called the range of O, and the elements of Ro
will be called the domains of 0 4 . What sort of entities are the physical
systems? This question is central in our further discussion. Let me em-
phasise, as strongly as possible, that physical systems should be con-
ceived as fragments of physical reality. They are physical phenomena such
as those that surround us every minute and in every place. A burning
piece of paper, a growing child, a stone falling down as well as such com-
plex phenomena as, say, economical relations in Europe in the XVlIIth
century, or home and social environment of people living in a particular
suburban area, all these physical phenomena are possible domains of
different sciences (theories). This statement, however, has merely the
value of philosophical credo. Adhering to it, I declare myself a realist.
But philosophical attitudes usually provide us merelY with rough ideas
to be grasped intelligibly in further theorizing. In our case we could not
pursue any further unless we say how empirical phenomena can be rep-
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 105

resented by appropriate set-theoretic structures. We certainly may try to

describe relationships holding between scientific theories and the physical
reality in the fuzzy language of philosophical quest for the essence of
knowledge, but we may try also to model these relationships by means of
suitable set theoretic constructions, similarly as, say, physicists model
physical phenomena by means of suitable mathematical constructions.
1 have already examined the problem we are approaching now in a few
papers (cf. especialIy [44], [45]) and what 1 am going to say below will
to some extent coincide with the earlier proposals. The simplest solution
to our problem is furnished by the assumption that each physical phe-
nomenon in the range of () can be conceived as a structure of the form

(4) (P, T, F;, ... , Fn),

where Pare the objects involved in this particular phenomenon, T is a
period of time, and Fi are physical quantities (we shall also call them
parameters) being functions of form (3). Structures of this sort will be
called strict. Once again, we denote terms F; and its possible referents by
the same symbol. Assuming that objects in P meet alI necessary require-
ments and that the time instances in Tare represented in the unique way
by real numbers, we come to the conclusion that each physical structure
of the form (4) defines a standard model structure for (). Let us call the
representation proposed classical or strict. Thus under the classical
representation the range of () turns out to be a set of structures that are
reducts (fragments of a relevant kind) of sms's for Le. To distinguish
amongst different representations of the range of () we shall furnish the
symbol Re with different superscripts. The classical representation will
be denoted by RJ. This way of representing Re is simple and natural. To
my best knowledge, Montague (cf. [25]) was the fust who defined it and
showed how it can be utilized in the methodology of science. It has a1so
been exploited by Sneed in his monograph [35]. 1 should perhaps make
clear at this point that the reason for my calling the approach classical is
not that in my opinion the works of Montague and Sneed, incidentalIy
among the most valuable contributions in the area of methodology of
science, are to be regarded as classical. Rather, 1 propose this terminology
because the classical representation of phenomena enables us to describe
the relations between sentences of Le and domains of () with the help of
the classical Tarski concept of truth.
106 R YSZARD W6JCICKI

The next representation will be called operational. It is termed in this

way to stress that it has much in common with Bridgman's doctrine. 1
had given an account ofthis approach in [42], [44], [45]. Recently 1 have
learned that the same approach was independent1y developed by Dalla
Chiara Scabia and Toraldo di Francia (cf. [14], [15]). Its main idea is as
follows. Let

be a collection of certain specific methods of measurement of quantities

F1' ... , Fn respectively. As we may have many such collections at our
disposal, we shall use a similar notation in alI cases. Thus, e.g., v will
stand for V1' .•• , Vm " will stand for "1' ... ' "n and so forth. Whenever we
shall write Fi, it will be assumed that the values which take FI are to be
assigned to it by means of the method Ili' likewise FI takes its values in
accordance with the results obtained by application of Vi. The notion of a
method of measurement should be conceived in a possible loose way.
What is to be called a method of measurement may consist of different
measurement procedures which can be applied alternatively, moreover it
needs not to be direct in the sense that establishing values of Fi may
require establishing values of certain other quantities, it also may involve
some sophisticated calculations. As a matter of fact we do not enquire
into how a method works, with the proviso that it works well, i.e. the
application of the method does not lead to inconsistent results. It is not
exc1uded, however, that data established by applying different methods
may be inconsistent. ConsequentIy, and at this point we are approaching
Bridgman's ideas, Fi, FI, Fi, ... should be considered as different para-
meters 5. Let us assume that Fi are continuous functions and thus they
can be measured only up to a certain degree of accuracy. Thus a value
which an operational parameter (as we shall caB parameters ofthe form Ff)
assigns to objects to which it is applied is always a real number interval
L1x whose length is determined by the precision of Ili.
You might expect that an operational structure should be defined as a
system ofthe form
(6) (P, T, Fi, ... , F:),
i.e. likewise as a 'c1assical' structure (4) with the exception that Fi's are
replaced by their operational counterparts. However, an operational
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 107

structure will be defined in a more sophisticated manner. The structures

of the form (6) will be called quasi-operational. We certainly can conceive
Ro as a set of quasi-operational structures, thus arriving at a quasi-
operational representation Ri of the range of ().
To transform (6) into an operational structure we have to get rid ofthe
infinite set T. It can be done in the following manner. Start with selecting
a sequence E of momentary events, each of which happens within the
period T. We shall require that no pair of events in E is simultaneous.
Since each physical event lasts, during a perhaps extremely short but finite
period of time, each event e in E represents a subinterval of T. By an
operational structure we shall mean a structure of the form
(7) (P, E, Fi, ... , F:)
with a selfexplanatory qualification that Fi should be interpreted as
functions of the form

whereeEE.
Observe that the k+ l-tuple (al> ... , ake) can be conceived as e-slice
of the k-tuple al"'" ak, i.e. as the latter entity considered at time of e.
Thus, with regard to the possibility of making use of the thing-slice lan-
guage (cf. Carnap [9]), the set E plays scarcelyan auxiliary role.
The range Ro conceived as a set of operational structures will be de-
noted by R:.
Searching for an adequate representation for Ro, i.e. one c1eared of
theoretical assumptions, as it is possible might be continued. We describe
in a very sketchy way the fourth (which stiU does not seem to be the last)
possibility. Each physical quantity can be defined as a homomorphism
from a relational structure into a numerical one. The procedure is
examined in the measurement theory where you can seek for details
(cf. e.g. [22], [33], [34]). By combining together qualitative structures
corresponding to particular parameters Fi we arrive at a structure being
a quantitative representation of a phenomenon in Ro. R: will stand for Ro
taken in the last representation considered.
All structures in R~, i = 1, 2, 3, 4, are assumed to be similar; it is not
so much necessary to make this assumption as convenient. (Two struc-
tures of any sort whatsoever will be said to be similar if they consist of
108 R YSZARD W6JCICKI

the same number of elements, and the corresponding elements can be

c1assified as set-theoretic entities ofthe same sort. Thus, eogo, a two-place
function which takes sets as its values should correspond to a two-place
function which takes sets as its values, a set should correspond to a set
and so fortho This is exactly the same notion of similarity as one applied in
algebra and model theoryo) (The structures similar to those in R~ will be
called the structures ofthe sort i, or briefly i-structureso) When a structure
is considered it is usualIy understood that alI its parameters are welI-
defined, ioeo they are defined for alI their possible applications to objects
of the structureo In spite of that the empirical structures we are going to
consider may contain parameters which are not welI-defined in this senseo
The measurement procedure Jl;, for example, may not alIow us to estab-
lish the value of Ff(al' ooo, ak' e) for a particular selection of objects
al' ak and a particular event (time instance) eo lf Fi is not defined for
000'

alI its arguments it will be said to be partially definedo A structure will be

said to be partially defined if it contains some partialIy defined parameterso
Now we face the task to define the notion of truth related to phe-
nomena in Re and perhaps other semantic notions in terms of which an
account of relations holding between () and Re can be giveno Obviously
in dealing with the issue we shall replace Re by one of its possible set-
theoretic substituteso The solution to the problem will be obtained by
defining for each i-structure f!E; i = 1, 2, 3, a c1ass of standard model struc-
tures, sms(f!E), which are in a sense consistent with f!Eo The structures in
sms(f!E) will be called abstractedfrom f!Eo We shaH discuss aH three cases
separate1yo
lf f!E is a welI-defined structure of the first sort then there is at most
one sms !/ which inc1udes f!E as its part; I remind you that we have as-
sumed that each se1ection of an interpretation for the terms Ft. 000' Fn
defines an sms !/ with the proviso that both the F;'s and the objects in the
set P on which the parameters Fi are defined satisfy certain conditionso
Accordingly we put either
(9a) sms(f!E)={!/}
if f!E defines an sms, or
(9b) sms(f!E) =0 ,
where 0 denotes the empty set.
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 109

If X is a partially defined l-structure, then there exist the set W(X) of all
well-defined structures similar to X and consistent with X in the sense
that whenever a parameter Fi is defined in X for some arguments, then
the value it takes in X on these arguments coincides with the value it
takes in any structure X' in W(X). We put

(10) sms(X) = sms(W(X)) ,

i.e. the sms's defined by partial l-structure X, are the sms's defined by
structures in W(X).
Consider now a structure X of the second sort, say of the form (6).
Observe first that to each such structure one can assign in a unique way
a set Id(X) of l-structures in the following manner. We shall say that a
1-structure X' of the form (4) is an idealization of X, X' E Id (X), if and only
if the following two conditions are satisfied:

(i) X' is well defined,

(ii) Fi(al' ... ' ak; t)E Ff(a l , ... , ak, t),· i = 1, ... , n,

for ali sequences (al' ... , ak' t)for whichFf is defined.

We clearly put:
(11) sms(X) = sms(Id(X)) ,
where sms(Id(X)) is the union of aU sms(X '), X' Eld(X). In some par-
ticular cases it may turn out that sms(El)=0.
The case of X being a 3-structure (say of the form (7)) is a bit more in-
volved. Once again we should start with defining the set Id(X) of idealiza-
tions of X. Assuming that some procedures for measuring time are given
we may assign to each eEE a real number interval T(e). Consider the
2-structure X' of the form (6) defined by the following two conditions:
G) For each sequence (al' ... , ak, t),Ff(at. ... , ak' t) is defined if and only
if for some eEE, tET(e) andFf(al' ... , ak' e) is defined.
(jj) WheneverFr(a l , ... , ak, t) is defined then
(12) Ff(al' ... , ak> t)=F)(at. ... , ak' e),
for this unique efor which tET(e).
Weput
(13) Id(X) = Id(X ') ,
110 RYSZARD W6JCICKI

and finalIy we define sms(X) once again by the Formula (11). Observe
that in alI cases considered it may happen that sms(X) =0.
If X is a 4-structure, sms(X) could be defined along the same lines. Our
definition of structures forming the quantitative representation of Ro is
too sketchy, however, for trying to accomplish such a task.
We are now in the position to formulate the definitions we need. Let X
be an i-structure, i = 1, 2, 3, let el be a sentence of Lo and let A be a set of
sentences of this language.

DEFINITION 1. el is verified by X (el is true in X) if and only if

sms(X)#0 andfor each 9'esms(X), el is true in 9'.

DEFINITION 2. A is confirmed by X (the sentences in A are simul-

taneously confirmed by X) if and only if there is 9' e sms (X) such that ali
sentences in A are true in 9'. 6

DEFINITION 3. el is falsified by X (el is false in X) if and only if

sms(X) #0 arulfor each9'esms(X), el isfalse in9'.

DE FINITION 4. el is well-defined in X (el is empirically decidable in X)

if and only if el is either true or false in X.
Once again the terminology 1 propose is not accidental. The concepts
defined above are intended to be associated with ones that Camap and
Popper discussed in their early writings (cf. [7] and [26]). It is clear for
example, that, in general, laws of empirical theories cannot be verified
by operational structures. It seems, however, to be reasonable to require
that they should be at least susceptible of falsification. The latter require-
ment can perhaps be best pronounced by saying that any hypothesis con-
sidered as a.plausible law should be together with the laws established
so far simultaneously confirmed by alI possible empirical structures.
Clearly, besides similarities, there are also striking differences between
the concepts defined here and those we have mentioned. The most im-
portant consist in that the former are semantic while the latter have been
purely syntactical.
Let me close this section with a short discussion of the folIowing issue.
Consider any set-theoretic representation R~ of Ro. The way in which the
structures in Ro are defined is 'theoreticly laden' in the sense that certain
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 111

properties of the structures are determined with the help of some theoretic
assumptions. For instance we do not exclude that measurement methods
involve certain laws of e, or that our measurement instruments are
deviced in accordance with the laws of the theory. You may wonder
whether it makes sense to test a theory by comparing it with the sţate
of affairs that takes place in a structure defined with the help ofthe theory.
Observe, however, that there is no reason to claim that the theory cannot
be falsified by means of measurement procedures developed on the base
of the laws of the theory. The coherence of the laws of a theory and
measurement data cannot be assured by the fact that measurement
procedures are based on the laws of the theory, simply because the ob-
jectives of measurement procedures are defined by features of physical
reality and not only by the procedures themselves.

4. THE FACTUAL CONTENT OF EMPIRICAL THEORIES

Let us start our discussion of the notion of factual content with defining
the notion of the (full) content of a sentence a of Le, cont{a). The defini-
tion 1 would like to propose is trivially simple.

DEFINITION 5. The content ofa, cont (a) is the set ofali sms's in which a
is not true. 7
One of the immediate and pleasant consequences of the definition is
that the set
(15) {cont{a):a is a sentence of Le}
(the set of all contents of the sentences of Le) is a Boolean algebra. In-
deed, denote the set of all sms's by 1 and the empty set by 0. We easily
see that (we apply standard set-theoretic and logical notation):
(16) cont {a /\ la) = 1,
(17) cont {a v I a)= 0,
(18) cont {a /\ P)= cont (a) u cont{p) ,
(19) . cont {a v P)= cont (a) n cont{p)
(20) cont{la)= l-cont{a)
thus (15) is a field of sets.
The notion of empirical content we have considered in the first section
112 RYSZARD W6JCICKI

of our paper does not have properties analogous to that stated in (16)--(20).
For example, in general, the empirical content of ac /\ f3 cannot be obtained
by summarizing the contents of the components of the conjunction.
As a matter of fact, aside from (16) and (17), none of the identities
(18), (19), (20) holds true for empirical contents. Should they hold true?
1 think they should! Both for intuitive and practical reasons. Intuitive,
since the clauses (16)--(20) seem to be the most natural conditions which
have to be satisfied by the notion of any content whatsoever, practical,
since in order for such a concept to work smoothly when applied the
contents of sentences should possess possibly good formal properties.
Thus we established a standard for the solution sought.
The idea which is central for alI investigations into the problem of
analyticity is that of a 'possible world'. A sentence is 10gicalIy true when
it is true in alIlogicalIy possible worlds (structures for the language). A
sentence is mathematicalIy true when it is true in alI mathematicalIy
possible worlds (standard model structures). But the conventions we ac-
cept need not be concerned only with logical and mathematical con-
cepts. We usualIy accept certain definitions which amount to setting up
some strong conditions to be satisfied by the interpretations of the
defined terms. We may state also certain terminological conventions
which are considerably weaker than the definitional ones. (IncidentalIy,
some logicians and philosophers seemingly do not approve the idea
of such conventions. For example in the most valuable analysis of dif-
ferent senses of the notion of analyticity given by Hintikka [20] no other
conventions than definitional are taken into consideration.)
Let us take it for granted that whatever empirical theory is considered,
the experts in the field will rule out certain empirical structures as a priori
impossible. The idea of an a priori, or shall we rather say theoreticalIy,
impossible structure is vague. The main difficulty in making it precise
consists in that some structures can be specified as theoreticalIy im-
possible for certain reasons which after a more careful examination are
turning out not to be purely linguistic. But to start with, let us assume
that the set of alI structures of a given sort is divided into two. We shall
denote the set of alI theoreticalIy possible i-structures by P~. Before we
shall use the notion of a theoreticalIy possible structure for defining the
main notions we are interested in let us make the folIowing two ob-
servations.
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 113

Note first that it is reasonable to expect that no i-structure fI such that

sms (fI) = 0 is in p~, at least with the proviso that we made no fault with
the selection of a proper mathematical formalism for (J. This observation
gives rise to the folIowing necessary condition for (J to be welI-adjusted
(WA2) In order for (J to be well-adjusted no structure fI in p~ can be
such that sms(fI)=0.
Needless to say that even if (J is well-adjusted we cannot, in general,
as sume that all i-structures fI for which sms (fI) i=0 are in P~.
(WA3) In order for (J to be well-adjusted it should hold true that,'
R~s;;;P~.

Fumished with the notion of a theoretically possible domain we may

define the notion of factual content and that of analyticity in a straight-
forward way. Select i and write sms(P~) for the union of alI sms(fI),
fIEP~; let rt. be a sentence of Lo.

DEFINITION 6. Thefactual content ofrt., F cont (rt.), is the set ofali struc-
tures in sms(P~) in which rt. is not true.

DEFINITION 7. The sentence rt. is analytic (analytically true) if and only

if F cont(rt.)=0.
Obviously both of the defined notions depend on your choice of i. We
shall as sume throughout the section that i is fixed.
You may easily prove the following assertions.

ASSERTION 1. For ali sentences rt., p of Lo:

(i) F cont(rt. /\ 1rt.)=sms(P~),
(ii) Fcont(rt.v 'rt.)=0,
(iii) F cont(rt. /\ P)=F cont(rt.) u F cont(p) ,
(iv) F cont(rt. v P)=F cont(rt.)n F cont(p) ,
(v) F cont('rt.)=sms(P~)-F cont(rt.),
and thus the set of ali F cont(rt.) is afield of sets.
Given a set ofsentences A, write F cont (A) instead of U {F cont(p): PE
EA}.
114 R YSZARD W6JCICKI

ASSERTION 2. If f) is well-adjusted and IX is derivable from A then

F cont{IX)~F cont (A).
Observe that the converse is not, in general, valid. 8 We have, however:

ASSERTION 3. lf f) is well-adjusted then F cont{A)~F cont{B) implies

that for ali f!l' in p~, if B is corifirmed in f!l' then the union Au B is con-
firmed in f!l'. 9
The following then holds true in particular.

ASSERTION 4. Iff) is well adjusted and F cont (A) =0 (i.e. all sentences
in A are analytic) then for all B and all f!l' in p~, B is confirmed in f!l' if and
only if Au B is confirmed in f!l'.

ASSERTION 5. If F cont{f3)~F cont (IX), then 1X-+f3 is analytic.

The converse need not be valid, although c1early F cont{IX-+f3)=0 when-

ever f3 is derivable from IX.

ASSERTION 6. lf IX is analytic then IX is empirically decidable in aU f!l' E P~.

If this trivial consequence of the definitions accepted astonish you,

consider any sentence of the form IX v --, IX. If IX is empirically decidable,
IX v --, IX is empirically decidable as well.

It is often c1aimed that those sentences which cannot be in any cir-

cumstances decided empirically have no factual content. Let us briefly
discuss the issue. It is interesting by itself; besides, such a discussion
should help us in bringing forward the intuitions underlying the con-
siderations conducted.
Apparently, the most plausible explication of the vague idea of em-
pirical undecidability is given by the following definition

DEFINITION 8. A sentence IX is empirically undecidable if and only if it

is well-defined in no structure f!l' E P~.

As an immediate consequence of the definition of an analytic sentence,

Assertion 6, and Definition 8, we obtain the following Assertion.
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 115

ASSERTION 7. Ii oc is empirically undecidable, and f} is well-adjusted then

F cont(oc)#O.

This is a paradoxica1 consequence, for, contrary to phi10sophica1 tradi-

tion, it denies the apparent synonymity of empirical undecidability and
empirical meaning1essness. I grant this objection but aH the same, I am
going to keep aH the definitions unrevised In accordance with the defini-
tions proposed, aH but analytic sentences have factual content even if
they are undecidable. 1 o
Let us, however, discuss a few examples in order to gain a better insight
into the situation. In Classical Mechanics, CM, the position is assumed
to be a continuous function. Let '}' be the sentence of CM stating this
assumption. Thus taking for granted that CM is weH adjusted, aH i-
structures, i = 1, 2, 3, which do not confirm this assumption are ruled out
of P~. If we put i = 1, it turns out that '}' is both analytic and empiricaHy
decidable. In turn, consider the case i = 2. This time '}' becomes em-
piricaHy undecidable. Are you convinced that it has no factual meaning?
I am not. Although it can be neither verified nor falsified in any quasi-
operational structure stiU, given any such structure!!E it states something
which could be pronounced like that: no matter how you improve your
measurement procedures you will not succeed in proving that the posi-
tion parameter is not continuous. Obviously, the case i = 3 yields the same
result as the discussed one.
Consider now Heisenberg's uncertainty princip1e and assume that aH
theoretically possib1e domains shou1d agree with it. As an immediate
consequence we have sms (pJ) = 0, where () stands for Quantum Mechanics
or an appropriate part of it. Obvious1y aH sharp structures disagree with
Heisenberg's princip1e. We have also sms(Pi)=0, for the quasi-opera-
tional structures admit sharp measures for time in aH circumstances. Only
sms(PJ)# O. Consider certain fixed real numbers Xo, yo' Let f: be the
sentence saying that there is a partide q such that a coordinate com-
ponent of q equals X o and at the same time the corresponding component
of momentum of q equals Yo~ The sentence f: denies Heisenberg's principle.
You may daim that whenever Heisenberg's principle is accepted there
is no room for sharp structures at aH, both in the language and in the
metalanguage and thus aH the semantic definitions proposed cannot be
applied in dealing with the situation. 1 grant the objection although 1 am
116 RYSZARD W6JCICKI

not sure that 1 agree with it without any restrictions. Let us then examine
e, applying the conceptual apparatus established in the paper. It is easily
seen that IX is not true in any fEeP:, but obviously it is not false either.
It is an empirically undecidable sentence, and we know already that this
'implies that F cont{e) # 0.

5. TERMINOLOGICAL CONVENTIONS

There are two possible ways of conceiving the idea of a terminological

convention which unfortunately cannot be cleared out separated one
from another. Firstly it can be c1aimed, and it often does, that the
terminological conventions help scarcely in establishing some con-
venient abbreviations which are to make our work more comfortable.
The definition of the unit of, say, length or the definition of velocity may
serve as suitable examples. The terminological conventions, understood
in this way, serve to establish the meanings of the terms which are from
the theoretic point ofview redundant. In consequence, all the solutions to
the problems of analyticity patiently discussed in this paper, besides
reducing to rather trivial settlements, turn out to be solutions to a prob-
lem ofpractically no importance. At least from the author's point ofview
it is a very good reason to look more c10sely at the second possibility of
accounting of the idea of terminological convention.
Under the second interpretation of the notion of a terminological
convention, all, fundamental assumptions determining the conceptual
framework ofthe theory comprise the set ofterminological conventions.
While a convention taken in the first sense, call it linguistic, is supposed
to be free of any factual assumptions, a terminological convention in the
second sense, call it methodological, can be motivated or even based on
certain empirical assumptions. The conventions defining the sets of pos-
sible domains discussed in the previous section were of a methodological
character. Let us consider a simpler example than those discussed earlier.
Let H stand for 'is heavier than' and let W stand for the parameter
'weight'. W will be assumed to be a quantitative parameter of an empirical
theory f), while H will be treated as a parameter of quantitative struc-
tures for f). F orming the conceptual framework for f) we as sume (but if you
do not like the idea, you need not) that H is a quasi-ordering relation, i.e.,
it is'transitive and antisymmetric. This assumption is motivated by certain
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 117

results of experience and thus it cannot be called purely linguistic. Taking

the advantage of it we may establish the quantitative representation W
of Hin the well-known manner. Consider the sentence
(1) Foralla,b, c, t, if W(a,t)< W(b,t) and W(b,t) < W(c,t)
then W(a, t) < W(c, t).
Clearly, it is arithmetically true and thus analytical (we shall assume that
() is well-adjusted). At the same time (1) states the transitivity of Hand
thus it is true in a quantitative structure only if H is transitive in the
structure. Thus the truth of (1) turns out not to be scarcely a matter of
arythmetic, it hinges on our decision to restrict the set of possible do-
mains P: to those in which H is quasi-ordered.
You may say that the example suggests that the difference between
the linguistic and the methodological senses of convention seems to be
rather easy to grasp. Unfortunately, it would be a hasty conclusion. The
history of science provides us with a great deal of 'perfectly' linguistic
conventions which were rejected for factual reasons. For example, the
carbon was defined to be the element whose atomic weight is 16. You are
likely to say: well, the change of the definition was motivated merely by
convenience, we could keep it unmodified if we wished. You are right, we
could. But then, imagine an experiment which proves H to be non-
transitive. Can we keep our assumption that H is a quasi-ordering
re1ation unmodified? Certainly, we can! We can always claim that in this
particular area in which H appears to be non-transitive, the usual
methods of measuring W (or testing whether H holds between a given
pair of objects) are not allowed. It is up to us whether we shall keep our
terminological conventions unchanged and we shall modify our measure-
ment procedures or, converse1y, we shall continue to apply the same
measurement methods as before but we shall modify terminological
conventions. There is no general solution to this dilemma.
It is hardly surprising that our terminological decisions are 'experience
laden' and that their ability to serve their purposes might be questioned
on the base of empirical evidence available. As we know, and inci-
dentally the example we have discussed illustrates this point, even the
selection of a mathematical apparatus to be used within an empirical
theory hasan empirical character. We can also describe a situation (cf.
Dalla Chiara Scabia [19]) which proves that the selection of logic is
118 RYSZARD W6JCICKI

bearing on the physical properties ofthe domains dealt with in the theory.
Apparently then we should not be so alerted about the possibility of
presence of certain empirical factors in our terminological decisions. We
should attempt to base terminological conventions on sound empirical
assumptions rather than to make them sterile from any empirical con-
tent.
We cannot help conc1uding that there is no sharp boundary line
between conventional and empirical components of science. They are
notoriously mixed. Saying this, 1 am merely repeating what Quine has
argued long ago in his famous critical essay 'Two Dogmas of Empiricism'
(d [32]). Nonetheless, in spite of Quine, 1 am convinced that we may
leam a great deal about empirical theories examining them reconstructed
in languages based on explicitly stated terminological conventions. 1 am
afraid that there is no other way that could allow us to grasp in an in-
telligible manner how the conventional components and factual ones
cooperate in establishing the contents of empirical c1aims.
Observe also that there are some methodological issues of remarkable
importance that call for an analysis of the conceptual framework of
empirical theories. As it was pointed out by Dalla Chiara Scabia [15] the
problem of intertheory relations can be examined by comparing the sets
of structures that confirm the collections of asserted sentences of the
theories (here, we mean theories that are supposed to describe the same
c1ass of phenomena). Although 1 am sympathetic with the idea of ap-
proaching the problem in this way, all the same 1 am afraid that a fully
satisfactory solution to it cannot be achieved if we are neglecting the
question of 'meaning variance', one of the most talked about. 11 Thus,
given two theories 8 1 , 82 , representing two stages of development of a
science (or a part of it), we should not confine ourselves to examine what
are the possible ranges of application of these theories but we should also
try to compare the sets POl' P02 taken in certain set-theoretic representa-
tions we have se1ected.

6. CARNAP'S SOLUTION TO THE PROBLEM OF ANALYTICITY AS A

PARTICULAR INSTANCE OF THE APPROACH PROPOSED

Let me c10se the paper by examining what are the assumptions to be

imposed on the examined theory 8 and the set Poof its possible domains in
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 119

order to obtain the solution to the problem of analyticity which coincides

with that proposed by Carnap. The problem can be solved in a rather
straightforward manner. Start first with dividing the specific terms ofthe
language of () into two parts: 01> ... ' Om and T1 ,. •• , Tn • We need not
assume that Le is the first order language, neither we need as sume that
O-terms and T-terms are predicates, they may be symbols of functions
as well. In order not to modify the established definitions we shall assume
that both O-terms and T-terms depend on time (obviously this depen-
dence can be plausible, what would bring us nearer to Carnap's way of
conceiving the specific terms of ()). Thus a l-structure for () is of the form:
(1) (P, T, 01> ... , Om, T1 , ••• , T n ).
Assume now that we accept Carnap's convention, (C2) (certainly you
can consider (CI) as well) thus C(n) is assumed to be a sentence of Le and
presumably an element of Ae.
In accordance with (C2) we define a l-structure of the form (1) to be a
theoretically possible dom ain of () if and only if C(n) holds true in it. It
immediately follows from this stipulation and the Definitions 5 and 6 that
F cont(C(n) = 0), C(n) is factually empty, and thus it is analytic. Assume
that IX is analytic. Clearly IX is entailed by C(n), for ifin an sms(Y') Carnap
sentence C(n) holds true then Y' is a possible domain and hence IX is also
true in Y'. Thus a sentence is analytic if and only if it is a consequence of
C(n). This is exactly the Carnap's definition. To reconstruct in this way
the other Carnap style solution to the problem of analyticity, we have
to proceed in an obvious way. Start with an appropriate terminological
convention and define accordingly the set P: 12. The definition of
analyticity falling under the schema given by Definition 7 will dis-
tinguish exactly the same sentences as analytic as the earlier one.
We have proved then that the approach developed in this paper is
more general than that of Carnap, indeed. Observe also that the doctrine
of observational and theoretical languages rather affects clear under-
standing of the problem of analyticity than helps to solve it. In fact no
assumptions about the nature of 0- and T-terms and their mutual inter-
relations were made in the reconstruction given above and moreover
no such assumptions are needed.
Institute of Philosophy and Sociology
Polish Academy of Sciences
120 R YSZARD W6JCICKI

NOTES

1 Taking the concept of L-truth as an explicatum for the vague concept of analytic truth,
Carnap states the following convention (cf. [8], p. 10). A sentence Si' is L-true in a seman-
tical system S if and only if Si> is true in S in such a way that its truth can be established
on the basis of the semantica! rules of the system S alone, without any reference to (extra-
linguistic) facts.
2 Winnie defines a sentence cx to be empirically vacuous in a theory () if and only if(l) cx is a
theorem of () and (2) if for some P, the conjunction cx A Pentails an observationa! sentence
l' then l' is already entailed by P alone.
3 Let k=max(k(I), ... , k(n)). One can easily redefine each Fi so as make it a function ofthe
form (3). Thus without any lost of generality, we may assume that k(i)=k, for each i.
4 In general Re is not a unit set, and thus empirica! theories should be conceived as theories
which admit many interpretations rather than a singular one. Note however, that this has
nothing to do with vagueness of empirica! terms, although the vagueness results in the
existence ofvariety plausible interpretations of the terms (cf. Przelrcki [28], Williams [40]).
I have discussed the reasons for which empirical theories cannot be, as a rule, considered
as theories of a singular domain somewhere e1se (ef. [42] and [44]). A similar' account of
empirical theories was proposed by many authors (cf. Beth [3], [4], Dalla Chiara Scabia
and Torlando di Francia [15], van Fraassen [16], Montague [25], Sneed [35], Suppe [36]).
5 Since we do not assume that measurement methods are homogeneous in the sense that
each of them consists of the same sequence of elementary operations repeated at different
places and times, ef. Bridgman [5], as a matter of fact the approach proposed here deviates
from the c1assical doctrine of operationalism.
6 In my earlier papers (cf. [42], [43]) a set of sentences A confirmed by flE was called
approximatively true in flE. I must say, however, that at present I am not happy with the
terminology I proposed. Let me mention that Dalla Chiara Scabia and Toraldo di Francia
have defined independently a semantic relation akin to that of confirmation (approximative
truth). Considering singular sentences rather than sets, they propose to call a sentence con-
firmed in flE in the sense of Definition 2 just true in flE. I am afraid that the terminology they
propose is not the best one either.
7 There is another way to make a vague idea of the content of a sentence precise. Namely,
we may identify the content of a sentence with the amount of semantic information it yields
(cf. Hintikka [21 ]). Observe that the notion of a semantic information is a quantitative con-
cept while the factua! content of a sentence as defined here is a qualitative one.
8 In particular F cont(A)=F cont(B) does not imply that A and Bare deductively equiv-
alent. This provides us with another argument in favour of Glymour's solution to the
problem he discussed in [17].
9 Applying the notion of factual truth we may define the relation 'cx is corifirmed by a set of
sentences A, cxeCOIif(A)', as follows: cxeCor!f(A) ifand only if F cont(cx)nF cont(A)#0.
10 Observe that the concept of factua! content examined here is a semantic one, while,
whenever we c1aim that empirically undecidable sentences are deprived of empirical con-
tent we apply the concept in the sense that is to be stipulated in terms of measurement and
testing procedures and thus is a methodologica! one.
11 A very interesting proposa! for dealing with the issue was given by Williams (ef. [38]).
His considerations, although limited to first-order theories, show one of the ways in which
the problem can be approached.
12 The set P: corresponds to the component d in Winnie's (cf. [38]) definition of a semantic
system. Note also that Pe and d play exactly the same role: they are possible worlds
assigned by semantic rules to the language of the theory.
 THE FACTUAL CONTENT OF EMPIRICAL THEORIES 121

BIBLIOGRAPHY

[1] Achinstein, P., Concepts of Science, Ba1timore, 1968.

[2] Ajdukiewicz, K., 'Le Probleme du Fondement des Propositions Ana1ytiques',
Studia Logica 8 (1958), 259-272.
[3] Beth, E. W., 'Towards an Up-to-date Phi10sophy of the Natural Science', Methodos
1 (1949), 178-185.
[4] Beth, E. W., 'Semantics ofPhysica1Theories', Synthese 12 (1960),172-175.
[5] Bridgman, P. W., The Logic o/Modern Physics, London, 1960.
[6] Bunge, M., Method, Model and Matter, Dordrecht, 1973.
[7] Carnap, R., 'Testability and Meaning', Philosophy of Science 3 (1936), 420--471 and 4
(1937), 2-40.
[8] Carnap, R., Meaning and Necessity, Chicago, 1958.
[9] Carnap, R., Introduction to Symbolic Logic and Its Applications, New York, 1958.
[10] Carnap, R., 'The Methodo10gica1 Character of Theoretica1 Concepts', in H. Feigl and
M. Scriven (eds.), Minnesota Studies in the Philosophy of Science, VoI. 1, Minneapo1is,
1956, pp. 38-76.
[11] Carnap, R., 'Beobachtungssprache und theoretische Sprache', Dialectica 12 (1958),
236-248.
[12] Carnap, R., 'Rep1ies and Systematic Expositions', in P. A. Schi1pp (ed.), The
Philosophy of Rudolf Carnap, La Salle, 1963.
[13] Carnap, R., Philosophical Foundations of Physics, New York, 1966.
[14] Dalla Chiara Scabia, M. L., 'Some Logica1 Prob1ems Suggested by Empirica1
Theories', to be pub1ished in Boston Studies in the Philosophy of Science.
[15] Dalla Chiara Scabia, M. L. and Tora1do di Francia, F., 'A Logica1 Ana1ysis of
Physica1 Theories', Rivista del Nuovo Cimento, Series 2, 3 (1973), 1-20.
[16] van Fraassen, B. C., 'On the Extension of Beth's Semantics of Physica1 Theories',
Philosophy of Science 37 (1970), 325-339.
[17] G1ymour, C., 'Theoretica1 Realism and Theoretica1 Equiva1ence', in Boston Studies
in the Philosophy of Science VIn, 276-287.
[18] Hempe1, C. G., 'The Theoretician's Di1emma', in Minnesota Studies in the Philos-
ophy of Science n (ed. by H. Feig1, G. Maxwell and M. Scriven), Minnesota, 1958.
[19] Hempe1, C. G., 'On the Structure ofScientific Theories', in The Isenberg Memorial
Lecture Ser ies 1965-66, Michigan State University Press, 1969, 11-38.
[20] Hintikka, J., 'An Ana1ysis of Ana1yticity', in Logic, Longuage-Games, and Information,
Oxford, 1973.
[21] Hintikka, J., 'On Semantic Information', in Physics, Logic and History, New York,
1970.
[22] Kanger, S., 'Measurement: An Essay in Phi10sophy of Science', Theoria 38 (1972),
1-44.
[23] KokoszyD.ska, M., 'On Deduction', in The Foundation of Statements and Decisions,
Warszawa, 1965.
[24] Kuhn, T. S., The Structure of Scientific Revolutions, Chicago, 1970.
[25] Montague, R., 'Deterministic Theories', in Decisions, Values and Groups (ed. by
N. F. Washburne), New York, 1975, pp. 325-370.
[26] Popper, K. R., The Logic of Scientific Discovery, London, 1959.
[27] Popper, K. R., Conjectures and Refutations, New York, 1963.
[28] Przelrcki, M., The Logic of Empirical Theories, London, 1969.
[29] Przelrcki, M. and W6jcicki, R., 'The Prob1em of Ana1yticity', Synthese 19 (1969),
374-399.
122 R YSZARD W6JCICKI

[30] Przel~cki, M. and W6jcicki, R., 'Inessentia! Parts of Extensions of First-Order

Theories', Studia Logica 28 (1971),83-99.
[31] Putnam, H., 'What Theories Are Not', in Logic, Methodology and Philosophy of
Science (ed. by E. Nagel, P. Suppes, and A. Tarski), Stanford, 1962, pp. 240-251.
[32] Quine, W. V. O., 'Two Dogmas of Empiricism', in From a Logical Point of View,
Harvard, 1961.
[33] Scott, D., 'Measurement Structures and Linear Inequalities', !ournal of Mathe-
matical Psychology 1 (1964).
[34] Scott, D. and Suppes, P., 'Foundationa! Aspects of Theories of Measurement', The
!ournal of Symbolic Logic 23 (1958), 113-128.
[35] Sneed, J. D., The Logical Structure of Mathematical Physics, Dordrecht, 1971.
[36] Suppe, F., 'Theories; Their Formulations, and the Operationa! Imperative', Synthese
25 (1972),129-164.
[37] Tuomela, R., Theoretical Concepts, Wien, New York, 1973.
[38] Williams, P. M., 'On the Conservative Extensions of Semantical Systems: A Con-
tribution to the Problem of Analyticity', Synthese 25 (1973),398-416, reprinted in the
present volume, pp. 142-159.
[39] Williams, P. M., 'On the Logical Relations Between Expressions of Different Theo-
ries', British !ournalfor the Philosophy of Science 24 (1973),357-408.
[40] Williams, P. M., 'Certain Classes of Mode1s for Empirica! Systems', Studia Logica,
forthcoming.
[41] Winnie, J. A., 'Theoretica! Analyticity', in Boston Studies in the Philosophy of Science
[42] W6jcicki, R., 'Semantyczne poj~cie prawdy, w metodologii nauk empirycznych'
[Semantic Notion of Truth in Methodology of Empirica! Sciences], Studia Filo-
zojiczne 3 (1969), 33-48, English translation will appear in Dialectic and Humanism.
[43] W6jcicki, R., 'Postulates and Their Analytica! Consequences', Akten des XlV.
Internationalen Kongresses fur Philosophie Wien 2.-9. September 1968, III, 228-233.
[44] W6jcicki, R., 'Basic Concepts of Forma! Methodology of Empirical Sciences',
Ajatus (Yearbook of the Philosophical Society of Finland) 35 (1973), 168-196.
[45] W6jcicki, R., 'Set Theoretic Representations of Empirica! Phenomena', !ournal of
Philosophical Logic 3 (1974), 337-343.
 P. M. WILLIAMS

ON THE CONSERVATIVE EXTENSIONS OF

SEMANTICAL SYSTEMS: A CONTRIBUTION TO
THE PROBLEM OF ANAL YTICITY

The attempt to distinguish within an empirical theory between a factual

and a conventional component is an important task. It bears especially
on the following methodological issues.
'Cognitive status' of theories: The dispute between conventionalism
and its critics concerns just where this distinction should be drawn. But
equally, if the conventional part of a theory determines the reference of
some of its expressions, the question whether theoretical assertions of
empirical science include non-conventional statements of determinate
truth-value cannot be settled until an exact characterization of the con-
ventional part is given.
Inter-theory relations: It is clear that logical relations between theories-
inconsistency, equivalence, reducibility, etc. - depend critically on intra-
theoretic relations. These are determined for each theory in a way which
depends on the choice of conventional component.
Evidential support: It is to be expected that an adequate measure of
support furnished to theory by evidence should assign the highest value
to statements belonging to the conventional part. It follows that the
measure appropriate to the language of a given theory depends on how
its factual and conventional components are distinguished.
The problem of distinguishing the factual and conventional component
was solved by R. Carnap for finite sets of postulates as follows [4], [5,
pp. 958-66]. Let n be the conjunction of theoretical postulates and
correspondence postulates and let the descriptive constants of n be
distinguished into O-terms and T-terms. Let Rn be the existential closure
of the formula obtained from n by proper simultaneous substitution of
predicate variables for T-terms. Then the factual part of n is identified
with the consequence class of Rn and its conventional part with the con-
sequence class of Rn;:) n. It is understood in addition that a partial inter-
pretation is effected for T-terms by restricting their proper interpre-
tations to those that establish the truth of Rn;:) n given prior interpretations
of O-terms. Carnap made the following expository observations:

Jaakko Hintikka (ed.) , RudoljCarnap, Logical Empiricist, 123-141. AII rights reserved.
Copyright ©1975 by D. Reidel Publishing Company, Dordrecht-Hol/and.
124 P. M. WILLIAMS

CI Every O-sentence entailed by Rn ::::> n is logically true,

C2 The conjunction of Rn and Rn ::::> n is logically equivalent to n.

In virtue of CI the factual content of n is safeguarded in face of the

stipulation that the interpretations of T-terms should verify Rn ::::> n. In
Carnap's terminology, the O-content of Rn::::>n is null whilst Rn has the
O-content of n. C2 indicates that the two components are jointly ex-
haustive of n. Although a stronger component than Rn ::::> n might be
established as conventional, Carnap proposed that any admissible com-
ponent should share with it the properties expressed by CI and C2.
This same problem, in more general form as the problem of analyticity,
has been subjected to extensive inquiry by the Polish logicians working
in response to questions rai sed by K. Ajdukiewicz [1]. This investigation
has been carried out largely from the standpoint of the theory of models.
It has proved to be an especially suitable framework for clarifying the
intuitive ideas involved. Although the approach has led to desiderata
very similar to those proposed by Carnap, it should be stressed that the
intuitions on which they are based differ significantly and that the re-
sulting requirements need not be equivalent. For instance, it has become
apparent that the property of Rn::::>n expressed in CI is neither the only,
nor perhaps the most appropriate, way of making precise the intuitive
idea that this component should be devoid of factual content. Results
obtained in this field were recentIy surveyed by two prominent contri-
butors, M. PrzelC(cki and R. W6jcicki [10]. The present paper takes the
work of these authors as its starting point. The problem with which it
deals may be stated as follows.
Let us as sume the interpreted language Lo to be given. We do not
enquire here into the origins of this language. A set of postulates II is
also given, whose expressions include those of Lo and certain others
besides. A suitable interpreted language LI is to be constructed in which
the sentences of II are to be expressed. The conventional part of II may
then be identified with those of its consequences which are analytic in LI'
The extended language LI is to respect the status of II as a set of postulates
stipulating the meanings of the expressions which do not occur in Lo.
Since the sentences which belong to II are not intended to stipulate
meanings for expressions of Lo but only for the additional expressions,
it appears that the meanings of the former should be preserved in the
 A CONTRIBUTION TO THE PROBLEM OF ANAL YTICITY 125

extension Li> whereas the latter should be understood in such a way as

to satisfy the sentences of il whenever this proves possible.
These requirements will be stated more exactly below and several
systems satisfying them will be constructed. It will then be possible, in
case il consists of the axiom set of an empirical theory or one of its sub-
sets, to compare their implications regarding the methodological issues
mentioned at the outset. In this paper the comparison is restricted to
issues occurring under the fust heading.
We deal here with the so-called elementary languages with standard
formalization under the additional restriction that the set of non-Iogical
constants of each language 2, denoted by V(2), is to be a countable
set of finitary relational symbols. In addition to its non-Iogical vocab-
ulary, each language is equipped with a denumerable set of individual
variables and certain ba sic logical constants including the identity symbol.
The set of sentences of 2 is denoted by Sent(2).
It is assumed known under what conditions a non-empty relational
m
system is said to be a structure (realization, semi-model) for 2 in the
classical two-valued sense, in symbols m E Str(2), and under what
further conditions m is a model of the set 1: S; Sent(2), in symbols
mEMod(1:). When :Yî s; Str(2), Th(:Yî) denotes the so-called theory of
:Yî defined by
Th(:Yî) = {u E Sent(2): :Yî s; Mod ({ u})}.
The set Cn(1:) of logical consequences of I s; Sent(2) is defined by
Cn(1:) = Th(Mod(I) ,
and the set LV(2) of logically valid sentences of 2 by
LV(2) = Cn(rb).
If C is a class of cardinals, a sentence u of 2 is said to be C-valid, in
symbols u E LVc(2), iffor each cardinal c in C, every structure for 2 of
power c is a model of {u}. Similarly, if u E Sent(2) and :Yî s; Str(2),
u is said to belong to the C-theory of :Yî, in symbols u E ThcC:Yî), if for
each cardinal c in C, every structure in :Yî of power c is a model of {u}.
It foIIows that LVa(ft') = T~(:Yî) = Sent(ft'). If.Yî s; Slr(2) we caII the
spectrum of :Yî, written Spec(:Yî), the class of cardinals c such that :Yî
contains a structure of power c. When 1: s; Sent(2), we write Spec(1:) in
place of Spec(Mod(1:».
126 P. M. WILLIAMS

Let mE Str(It') and m' E Str(It") where V(It') s;;; V(It"). The structure
obtained by omitting from m' the relations correlated with members of
V(It") - V(It') is called the contraction of m' to It' and denoted by
m'Iv<!l'). (By m'la we understand the universe of m' also written as
Im'I.) m' is called an expansion of m to It" when m= m'lv(!l'). More
generally, if V(It') s;;; V(It"), :;f" S;;; Str(It'), :;f"' S;;; Str(It"), :;f"' Iv(!l') de-
notes the class of alI contractions to It' of members of :;f"' and :;f" IV(!l")
denotes the class of all expansions to It" of members of :;f".
Definition 1: The triple L = (It', d, ..It) is an (elementary) semantical
system (or interpreted language) if and only if the following conditions
are satisfied:
(i) It' is an elementary language
(ii) d S;;; Str (It')
(iii) d is closed under isomorphism
(iv) ..It S;;; d
(v) ..It # 0
(vi) 19)11 = 19)1'1 for alI 9)1, 9)1' in..lt.
The first component of a semantical system L consists of a collection of
primitive symbols and rules of formation providing the means for de-
scribing, under suitable interpretation, a structured set of individuals. The
second component may be understood as the family of 'possible worlds'
for the system. Not all worlds need be admitted as possible, e.g. those in
which there is a greatest prime number or in which some round things
are square: it is not generalIy to be expected that alI structures for It' are
included in d. Nevertheless, the family d appears to be closed under
isomorphism for most interpreted languages in the intuitive sense. The
presence of (iii), however, is not strictly necessary for subsequent devel-
opments but it simplifies them considerablyl. On the other hand, d may
additionalIy be closed under elementary equivalence or it may frequently
be an elementary class in the wider or narrower sense, e.g. when deter-
mined as the class of models of a set of meaning postulates (q.v. [2], [7]).
There seems to be no conclusive reason, however, for regarding these
further properties as necessary for any serviceable semantical system. It
must be admitted that in fact no general solution to the problem for-
mulated below exists if either is imposed. The question whether systems
constructed in accordance with the specifications to be given, but lacking
 A CONTRIBUTION TO THE PROBLEM OF ANALYTICITY 127

these properties, are suitable for some purpose can only be decided with
reference to the particular purpose at hand.
The third component of L may be understood as, roughly speaking,
the 'actual world' describable truly or falsely in the system. By clause (v)
some such structure must be incorporated in L. The semantical rules of
the system, however, may fail to single out a unique structure as its
object of reference. In this case 1 contains more than one member and
denotations of non-Iogical expressions are only partially determined. All
structures in 1, however, must be admitted as possible - clause (iv).
Moreover, the universe of discourse of the system should be unique -
clause (vi). L is to describe a determinate set of individuals, but a set
whose structure, from the standpoint of L, may be to some extent in-
determinate. The extent to which indeterminacy can be supported depends
on the purpose the system should serve, e.g. allowing the formulation of
an empirical theory [9, Ch. 4]. We shall refer to 1 as the family of
proper structures of L and the denotations they provide for members of
V(2) as their proper denotations in L. The common universe of members
of 1 is called the proper universe of L.
Within the semantical meta-theory of system L = <2, d, 1), a part
ba sed on the pair <2, d) and a part ba sed on the pair <2,1), the so-
called theories of meaning (intension) and reference (extension), c~ n be
distinguished. Fundamental to these are, respectively, the notions of
analyticity and of truth. The sets An(L) and Tr(L) of analytic and true
sentences of L are defined by
An(L) = Th(d)
Tr (L) = Th(JI).
We also say that (j is a false sentence of L, in symbols (j E Fls(L), if and
only if ("" o) E Tr (L). (It follows that there may occur sentences in L that
are neither true nor false.)
Definition 2: If L = <2, d, 1) and L' =<2', d',1') are semantical
systems, L' is a conservative extension of L if and only if
(i) V(!l') s;; V(2')
(ii) d = d'lv(.2')
(iii) I
1 = l ' V(.2') •
A conservative extension is one in which new expressions are intro-
128 P. M. WILLIAMS

duced but the original expressions retain their meanings. Conditions (ii)
and (iii) require that the classes of possible and actual assignments of
denotations to expressions of the original system should be the same in
both systems. It follows that analyticity and truth, restricted to sentences
of 2, also coincide.
Let us now suppose the system Lo = (2o' do, Jt o> to be given,
together with the set il of sentences of an extended language 2 1 • il may
typically be a subset of the theorems of an empirical theory T. With this
in mind, we denote by 2 T the language with non-Iogical vocabulary
V(2 T ) = V(2 1) - V(2 0). Relational symbols in V(2 T ) will be referred
to as T-terms and those in V(2 o) as O-terms. It is assumed that il is a set
of postulates for T-terms in the sense that, as far as possible, these terms
are to be understood in such a way that the sentences of il are true given
the proper interpretations of O-terms. Furthermore, there is assumed
to be no other way in which the proper denotations of T-terms are stipu-
lated. The problem arises of choosing an appropriate semantical system
L 1 = (21) d l' Jt 1>' satisfying the requirements outlined in the introduc-
tion, which is to give meaning to the expressions of the set il. More
exactly, we arrive at this
Problem: Given the semantical system Lo = (.Po, do, Jt o the lan- >,
guage 2 1 with V(20) ~ V(2 1 ), and the set il ~ Sent(2 1 ), to determine
a semantical system L 1 = (21) d 1> Jt 1> satisfying
(P.i) L 1 is a conservative extension of Lo
(P:ii) /\~ E Str(2 1) [~Io E (do (") Mod (il) 10) ~
~ (~E d 1 +-+~ E Mod (ll)]
(p.iii) Jt o 11 (") d 1 ~ Jt 1 •
(We write 10 in place of 1V(.2'o) and 11 in place of IV(.2'I). Similarly, we shall
write IT in place of Iv(.2'T)·)
The meaning of clauses (i)-(iii) is as follows. The first requires that the
meanings of O-terms should be preserved in the extended system. The
second states that amongst the structures for 2 1 whose contractions to
!Ro both belong to do and can be expanded to some model of il, those
belonging to d 1 are to be just those that are models of il. This expresses,
in one interpretation, the idea that wherever possible T-terms should be
understood in accordance with il. 2 The third clause requires that every
member of d 1 whose contraction to 20 belongs to Jt o shall be included
 A CONTRIBUTION TO THE PROBLEM OP ANAL YTICITY 129

in JI 1. This formulates the assumption that proper denotations of T-terms

are stipulated in no other way than by postulates.
Let us first remark that the conditions of the problem require of every
l
solution that Jl 1 = Jlo 1 ('\ d 1 • It follows that the problem reduces to
the determination of class d 1. In certain cases only one such class exists,
namely when II imposes weaker conditions on admis si bIe interpretations
ofO-terms thanalreadyexistinLo, more precisely when do S Mod(II)lo.
In this case II is said to be non-creative with respect to do. Then the
unique solution is given by

di = doi 1. ('\ Mod(II).

From this it follows that Cn(II) S An(L1 ), i.e. that all logical conse-
quences of set II are analytic in L 1 (and, a fortiori, true in L 1 ). In this
case it can be said that II has exclusively the function of providing T-
terms with meaning. In general, however, this is not the case. At least
when II is the axiom set of an empirical theory it has in addition factual
content, when it no longer happens that every member of do can be
expanded to a model of II. In such cases the solution is not unique. Two
solutions are given as follows.
Theorem 1: Let the classes of structures d~ and JI~ be defined by
d~ = {~E Str(2'l):~lo E do A [~Io E Mod (II) Io~ ~ E Mod (II)]},
JI~ = Jlo11, ('\ d~.
Then L~ = (2'1' d~, JI~) is a semantical system satisfying (P.i)-(P.iii).
Theorem 2: Let the classes of structures d~I aud JI~I be defined by
d~1 = {~ E Str (2' 1): ~Io E do A [((~ltI> E (d 01 1 ('\ Mod (II) Iti> ~
ol
~ ~IT E (d 1 ('\ Mod(II»IT) ~ ~Io E Mod (II)ol)~.~
~ ~ E Mod (II)]},
l
JI~1 = Jlo 1 ('\ d~I.

Then L~1 = (2'1' d~I, JI~I) is a semantical system satisfying (P.i)-(P.iii).

The solution of Theorem 1 is essentially that of Przel~cki and W 6jcicki
[1O.p.383] which, for II finite, is equivalent to the solution of Carnap.
Solution L~ is the weakest in the following sense.
Theorem 3: If L 1 = (2'1' d 1 , J( 1 ) is any semantical system
satisfying (P.i)-(P.iii), then d 1 S d~, Jl1 S JI~; consequently
An (L1) S An (L 1) and Tr(L1) S Tr(L 1).
130 P. M. WILLIAMS

L1 and L11 are frequentIy distinct. They coincide, however, if any of

the folIowing conditions holds:
(a) do s::; Mod (II) \0
(b) do n Mod (II) o = 0
(c) l
(do 1 n Mod (II))lr = Str(!l'r).
The situation described in (a), where II is non-creative with respect to
do, has already been discussed. Case (b) is met, for instance, when II is
inconsistent or, more generalIy, if do = Mod(Ao) for some Ao s::; Sent(!l'o),
when Ao u II is inconsistent. Then (P.ii) is satisfied by alI classes d 1
and any conservative extension of Lo for which (P.iii) holds is a solution.
L1 and L~I agree in selecting theweakestgiven by d 1 = d 011, viiI = vii 011.
When do = Mod(Ao) condition (c) fails to hold unless
(d) Cn(Ao u II) n Sent(!l'r) s::; LV(!l'r).
Now many of the specific postulate sets for which the problem of ana-
lyticity has so far been rai sed have in fact satisfied both (c) and (d). For
these sets no new solution is provided by L~I. But at least when II is
identified with the axiom set of an empirical theory it seems that the
eondition (d) frequently fails and with it (e). Then unless either (a) or (b)
holds, L~J generalIy provides an alternative to L1.
Relative merits of the distinct solutions can only be assessed on the
basis of their consequences. Some ofthe more immediate consequences of
L~, L~J and ofa furthersolutionL~II, to bestated below, will beexaminedin
the remainder of the paper. Particular attention will be paid to the sets
An(L 1) and Tr(L 1 ) in the several cases.
It should be noticed first that alI solutions agree on the sets An(L1)
and Tr(L 1) restricted to sentences of !l' o.
Theorem 4: If L t is any semantical system satisfying (P.i)-(P.iii),

An(L t ) n Sent(!l'o) = An(Lo)

Tr(L 1) n Sent(!l'o) = Tr(Lo).
It folIows that differences between solutions in these respects can be
revealed only in the set Sent(!l't) - Sent(!l' o). This consists of 'mixed'
sentences, in which there occur both O-terms and T-terms, and 'pure'
sentences in which, besides logical symbols, there occur T~terms alone.
It is not an easy matter to obtain informative and general results for the
 A CON TRIBUT ION TO THE PROBLEM OF ANAL YTICITY 131

mixed sentences (except those that are equivalent to sentences of !l'o or

!l'T), but results of some interest may readily be established for T-sen-
tences.
Theorem 5:

An(L:) n Sent(!l'T) = Th(sloll n Mod(II)) n LVAI(!l'T) ,

where Al = Spec(sI o n Mod(lI)lo).

(When :Iî S; Str(!l'), we write :Iî for the class of structures for !l' not
included in :Iî, and Spec(:Iî) for the class of cardinals not included in
Spec(:Iî).)
The meaning of Theorem 5 becomes clearer in the special case sio =
= Str(!l' o) when there are no restrictions, e.g. by postulates, on admissible
interpretations of O-terms. Then

An(L:) n Sent(!l'T) = Cn(lI) n LVAI(!l'T)

where Al = Spec(Mod(lI)lo). Thus analytic T-sentences of L1 are ex-

actly those T-consequences of 11 which are valid in all domains with the
power of some structure for !l' o that cannot be expanded to a model of
11. It follows that unless there exists a non-zero cardinal c such that
every structure for !l' o of power c can be expanded to a model of 11, no
sentence of !l' T is analytic in Li unless it is logically valid. Now it seems
that the non-existence of such a cardinal is typical of most situations.
For the condition that such a cardinal should not exist is satisfied if there
occur amongst the logical consequences of 11 sentences of !l' o that are
not {c }-valid for any non-zero cardinal c. But when 11 is a set of postu-
lates of an empirical theory, and especially when it is identified with the
full axiom set, it is by no means unusual for an O-sentence which is
refutable in every non-empty domain to be derivable from 11. In such a
typical situation a sentence of !l' T is analytic in L1 only if it is logically
valid.
In the general case sioS; Str(!l' o), unless there exists a cardinal
CE Spec(sI o) such that every structure in si o of power c can be expanded
to a model of 11, no sentence of !l' T is analytic in L1 unless it is Spec(sI 0)-
valid. Now again it seems that the existence of such a cardinal is far
from usual. Normally a sentence of !l' T is analytic only if it is Spec(sI 0)-
valid.
132 P. M. WILLIAMS

The typical composition of Spec(d o) requires some comment. If the

non-logical symbols of Lo designate observable properties of events or
things, or observa bIe relations between them, Î.e. if Lo is an observation
language in a narrow sense, it may be assumed to satisfy the weakest
requirement of jinitism of Carnap [3], nameIy that the rules of the lan-
guage do not state or imply that the basic domain is infinite. It may be
supposed, on the other hand, that the possibility of an infinite domain is
not excIuded. If it is granted that the cardinality requirements of such a
language rarely discriminate between finite powers, we may expect every
non-zero cardinal to occur in Spec(d o). It folJows then that a sentence
of 2' T is again analytic in L~ only if it is 10gicaIIy valid.
It is cIear, however, that the requirement of finitism is not satisfied in
every case of the enlargement of a semantical system. We may consider,
for example, the case where the non-logical constants of Lo designate
properties of, or relations between, elements of a space-time manifold
whose continuity is secured by a semantical rule. Let us suppose then
that Spec(d o) incIudes some infinite cardinal but no finite one. It foIIows
nevertheless that a sentence U E Sent(2' T) is analytic in L~ only if there
is a finite cardinal co(u) such that u is {c}-valid for aII c ~ co(u). In this
case no sentence of 2' T is analytic in L~ if it has greater content than the
assertion, for some finite co, that there exist at least Co individuals.
It is to be noted lastly that whatever the composition of Spec(d o),
a T-sentence should be regarded as analytic from the standpoint of L~,
only if it is valid in some non-empty domain. It seems, however, that
few of the statements whose conventionality has been urged by several
writers, e.g. each or aII of the laws of cIassical mechanics, the energy
conservation principle, the law of rectilinear propagation of light etc.,
satisfy this requirement. The views of these writers can hardly be main-
tained in connection with the system L~. The solution L~/, on the other
hand, provides some support for the cIaims of so-called (moderate)
conventionalism.
Theorem 6:

An (I!:) n Sent(2'T) = Th(d ol 1 n Mod(ll) n LVA II(2'T)'

where AII = Spec(do) n Spec(d o n MOd(ll)lo).

Let us consider the case AII = 0 occurring when
 A CONTRIBUTION TO THE PROBLEM OF ANALYTICITY 133

(*) Spec(do) s;; Spec(d o Il Mod(II)lo).

Then

and hence
Cn(II) Il Sent(2 T ) s;; An(I![).
When (*) holds aII T-consequences of II are analytic in L~I. In fact when
do= Mod(Ao) for some Ao s;; Sent(2 o),
An(I!:) Il Sent(2 T ) = Cn(Ao u II) Il Sent(2 T ) ,

provided Spec(A o) s;; Spec(A o u II). Now (*) is clearly satisfied if II is

non-creative with respect to do. But it may also hold when II is creative.
Suppose II to be the axiom set of an empirical theory T. Then if (*) holds,
aII theoretical consequences of T may be treated as analytic in accordance
with LiI without in any way diminishing the theory's factual content
with respect to the subsystem Lo. The strength of L~I in this respect is
indicated by the foIIowing partial complement to Theorem 3.
Theorem 7: If L 1 = <21' d 1, J(1) is any semantical system satisfying
(P.i)-(P.iii) for given Lo and II satisfying (*), then diIIT S;; d 1 1T; con-
sequently
An(L 1 )n Sent(2 T) S;; An (J!:) Il Sent(2 T)'
Let us examine more closely condition (*). If do = M od(Ao), (*) is
satisfied if the cardinality requirements of Ao augmented by II are no
stronger than those of Ao alone. Clearly this is so when Ao u II has
models of all powers. But (*) is also satisfied for any set II when Ao ad-
mits only infinite models (or, more generally, when do is closed under
elementary equivalence and includes only infinite structures), provided
that Ao u II is consistent (dol1 Il Mod(II) non-empty). On the other
hand, it has been suggested that when Lo is an observation language in
a strict sense, Spec(d o) may be expected to include all non-zero cardinals.
Then unless Ao u II admits models of aII powers, it is no longer true that
every T-consequence of II is analytic in LiI . Nevertheless, if there is a
sentence (1 of 2 T which expresses, roughly speaking, the extent to which
the cardinality requirements of Ao u II can exceed those of Ao, every con-
ditional sentence of 2 T whose antecedent is (1 and whose consequent is
134 P. M. WILLIAMS

a consequence of II is analytic in L~I. The foIIowing may serve as an

example.
Suppose 'p' E V(2 T) and let 1ti> 1t2, 1t3, be the foIIowing sentences of
2 T:
1t1 = 'fix ( ~ Pxx)
1t2 = 'fIx'fly'flz(Pxy 1\ Pyz:::> Pxz)
1t3 = 'fIx3y(Pxy).

Let 1to = 1t1 1\ 1t 2 1\ 1t3 and suppose that 1to E Cn(II). Then if Ao u II
is consistent (do cIosed under elementary equivalence and .5310 11 n Mod
(II) non-empty), for every sentence 1t E Cn(II) n Sent(2 T)'
r1to:::> 1t" E An (I!:) n Sent(2 T).
This concludes the discussion of the analytic components of systems L~
and L~I.
When Lo, II and L 1 are given satisfying (P.i)-(P.iii) let us agree on the
foIIowing terminology:
II is true iff Cn(II) ~ Tr(L 1)
II is O-true iff Cn(II) n Sent(2 0 ) ~ Tr(Lo)
II is O-false iff Cn(II) n Sent(2 0 ) n Fls(L o) # 0
II is T-true iff Cn(II) n Sent(2 T ) ~ Tr(L 1)
II is T-false iff Cn(II) n Sent(2 T ) n Fls(L 1) # 0.
Then we may ask questions of some methodological interest such as the
foIIowing. If II is O-true, is II true? If II is O-false, is II T-false? Correct
answers to these questions often depend criticaIIy on the choice of L 1 • To
the first question, however, aII solutions provide the same answer, at least
when the power of the proper universe of Lo, mo say, is finite. More
exactly, when the condition
(e) Jl o ~ Mod(II)lo

is satisfied, JI 1 is determined uniquely by (P.i)-(P.iii) as

(f) Jl 1 = Jlol1 n Mod(II).
If (e) holds it is cIear that II is O-true. Conversely, if II is O-true and the
common power of members of Jl o is finite, (e) holds. If (e) holds, it
foIIows from (f) that Cn(II) ~ Tr(L 1 ). Consequently, if II is O-true and
 A CONTRIBUTION TO THE PROBLEM OF ANALYTICITY 135

mo is finite, II is true. Now it seems that condition (e) is often equivalent

to the statement that II is O-true, whatever the power m o may be, at least
when II is the axiom set of an empirical theory T. It appears then that
most solutions accord with the remark of Carnap [6, p. 256]: "To say
that a theory is a reliable instrument - that is, that the predictions of
observable events that it yields wiIl be confirmed - is essentiaIly the same
as saying that the theory is true."3
The condition JI o C;; Mod(Il) 10, however, can by no means generaIly
be asserted especiaIly when II = T, It seems that few empirical theories
of historical importance have been entirely O-true. When (e) no longer
holds the composition of JI 1 and hence of Tr(L 1) is determined differently
by the various solutions. As for An(L1) we consider Tr(L 1) restricted to
sentences of .fi>T. The situation for L~ is as foIlows:
Theorem 8:
Tr(Lf) (") Sent(.fi>T) = Th(Jlo1 1 (") Mod(Il) (") LVMl(..<l'T)
where MI = Spec(Jl o (") Mod(Il)lo).
Since aII structures in JI o have the common uni verse of power mo,
onlytwocasesarise: (i) MI = 0; (ii) MI = {mo}. Case (i), occurring if and
only if Jl o C;; Mod(Il)lo, has been discussed. In case (ii) we have
Tr(L[) (") Sent(.fi>T) = LV{moJ(.fi>T).
Consequently, since only {mo}-valid sentences of .fi>Tare true in Lt a
sentence of .PT is false in L1 only if it has no model of power mo. It
foIlows that if (J E Sent(.fi> T) is a logical consequence of II, (J is never false
in L~ if II has a model of power mo. More exactly, if m o E Spec(Il),
Cn(Il) (") Sent(.fi> T) (") Fls(LD = 0. Now even if II is assumed to be the
fuII axiom set T of an empirical theory, it appears that the condition
m o E Spec(T) is most often satisfied if only because the power of the
proper uni verse wiIl be chosen to be sufficiently large to meet the typical
demands of T when this is necessary. If so, it may be asserted that, from
the standpoint of L~, no empirical theory is ever T-false. Lastly, in case
(ii) we have
JI~ = {9JlEStr(.fi>l):9JlloEJlo}.
It foIIows that the contraction to .fi>T of Jli consists of aII structures for
.fi>T with the proper universe of Lo and therefore that within this universe
the denotations of T-terms are whoIly undetermined.
136 P. M. WILLIAMS

The corresponding properties of LiI are expres sed as foIIows:

Theorem 9:
Tr(Ei) n Sent(2 r ) = Th(.4 0 11 n Mod(JI)) n
n Th M I(d o11 n Mod(JI» n LVM II(2 r ),
where MII = Spec(.4 o) n Spec(d o n Mod(JI)lo)'
Since each of MI, MII is either the nuIl set or the unit set {m o} and
since MI = 0 implies MII = 0, there cases arise: (i) MI = 0; (ii) MI = {m o},
MII = 0; (iii) MII = {m o}' The first, arising when .4o ~ Mod(JI)lo, was
discussed before. In the third case
Tr(E[) n Sent(2 r) = LV{mo) (2 r),
as in case (ii) of Theorem 7. Furthermore T-terms receive the same inter-
pretations. The conditions under which these two cases ari se, however,
differ considerably. Case (ii) of Theorem 7 arises in the normal situation
given by MI = {m o}' The present case also arises when MI = {m o} but
only if, in addition, MII = {mo}. This case occurs only if moi:Spec(do n
n MOd(JI)lo). In a comment on Theorem 8 it was suggested that the con-
dition moi:Spec(JI) is untypical. A similar comment applies here. The
most usual situation for L~I appears to be given by case (ii) when
l
Tr(Ei) n Sent(2 r ) = Th{mo) (do 1 n Mod(ll) n Sent(2 r ),
and then
Cn(ll) n Sent(2 r) ~ Tr(E{).
Assuming again JI = T, it foIIows that theory T, even when O-false, is
stil! T-true in L~I. It should be recalled, however, that when (*) holds aII
T-consequences of JI, thus of T, are analytic in L? and, a fortiori, true.
On the other hand (*) may fail when the weaker condition m o E Spec
(do n Mod(JI)lo) holds (e.g. when m o is infinite but Spec(d o) conta ins
in addition finite cardinals not belonging to Spec(JI)). It is not easy to
judge when this is likely to arise.
A subcase of (ii) - perhaps the most typical- is distinguishable where not
only.4o $ Mod(JI)lo, but in addition .40 n Mod(JI)lo =0 as when II
is O-false. Then, assuming stil! MII = 0,
.4iI = {9R E Str(2 1 ):9Rlo E.4o /\
/\ 9Rlr E (dol 1 n Mod (ll))1 r} .
If for simplicity we as sume do = Mod(Ao) for some Ao ~ Sent(2 0 ), it
 A CONTRIBUTION TO THE PROBLEM OF ANALYTICITY 137

follows that when II is O-false and mo E Spec(A o u II), the denotations

proper to T-terms in L~I are just those assigned to them by some model
of Ao u II whose universe is the proper universe of the system. This
interpretation of T-terms is clearly more determinate than that afforded
by L~ in the same circumstances.
Let us summarize the results obtained so far. It will be assumed for sim-
plicity of exposition that the set of postulates II has been identified with
the full axiom set of a consistent empirical theory T. Solution L~ is the
weakest in the sense of Theorem 3 and therefore in the several respects
discussed. Typically it provides for no analytic sentences amongst the
theoretical consequences of T beyond those that are logically valid.
Correspondingly it affords only the weakest interpretation of theoretical
terms. Indeed, unless T is O-true, denotations of theoretical terms are
wholly undetermined within the proper universe. Even if T is O-false, in
general no theoretical consequence of T is false. The view that " ... if the
conclusions have been falsified, then their falsification also falsifies the
theory from which they were logically deduced" (Popper [8, p. 33]) cannot,
it seems, be maintained for L~ in the sense intended.
According to L~I, on the other hand, all theoretical consequences of
T can be analytic. L~I then accords with the view of theoretical laws as
neither experimentally testa bie statements about facts, nor a priori truths
originating in pure reason, but as conventions for the use of certain
words and expressions. LiI shows how this view can be maintained
without denying the overall empirical character of science. Correspond-
ingly L~I affords a more exact interpretation of theoretical terms and a
more extensive class of determinate theoretical statements including, in
particular, all theoretical consequences of T. Amongst the latter, however,
there occur only true statements even if Tis O-false and even if, unusually,
not all are analytic. As with L~, there can be no false theoretical conse-
quences of T.
L~ and L~I are by no means the only general solutions to the problem
of analyticity. A further solution L~Il, which is in a sense intermediate
between them, will be mentioned. With L~ it shares the feature that, or-
dinarily, no theoretical consequence of II is analytic unless it is logically
valid. Like L~I it assigns partially determinate denotations to T-terms
even when II is O-false. On the other hand, theoretical consequences of
II can be false in L~Il.
138 P. M. WILLIAMS

Theorem 10: Let there be given the systemL o = (.Po,do,.A o), thelan-
guage .P1 with V(.P o) ~ V(.P 1 ), and the set II ~ Sent(.P1 ). Define the
classes of structures di II and .AiII by

diII={~EStr(.Pl):~loEdo 1\ [«~leE«doll (') Mod(1I))IT)le~

~ ~IT E (dol1 (') Mod (ll»)IT) ~ ~Io E Mod (1I)lo) ~
~ ~E Mod (II)]} ,
.AiII = .A01 1 (') di II .
Then LiII = (.P 1 , diII, .AiII ) is a semantical system satisfying (P.i)-
(P.iii). Let us note first that when the condition

(**) Spee(do (') Mod (II) Io) ~ Spee«dol i (') Mod (ll»)IT)
is satisfied, the class diII may be re-described equivalently by
diII = {~E Str(.Pl):~lo E do [(~IT E (doi 1 (') MOd(II))IT
1\ V
v ~Io E Mod (ll)lo) ~ ~ E Mod (II)]} .
Let us abbreviate the expression on the right of the inclusion symbol in
(**) as A III • Now it seems that in many cases the class AIII in fact con-
tains every non-zero cardinal and hence, a fortiori, that (**) will hold.
For if some non-zero cardinal e is absent from AIII, it may readily be
shown that every T-consequence of II is {el-valid. But this situation,
it appears, is not usually met with when II is the axiom set of an empirical
theory. The properties of L1iI will therefore be investigated under the
assumption that (**) holds.
Theorem 11: If condition (**) is satisfied,

An(I.![l) (') Sent(.P T) = Th(d ol 1 (') Mod (II» (') LVAl(.PT) =

= An (Lf) (') Sent(.P T)'
We have, of course, from Theorems 4 and 10:

An(I![I) (') Sent(.P o) = An (Lo) = An(Lf) (') Sent(.P o)'

It is not generally true, however, that An(L1fI) = An(Li).

Theorem 12: If condition (**) is satisfied,

Tr(I![I) (') Sent(.P T) =

= Th(.A oI1 (') Mod(II» (') Th M l«dol i (') MOd(II»IT)'
 A CONTRIBUTION TO THE PROBLEM OF ANALYTICITY 139

Again two cases arise: MI = 0; MI = {m o}. In the first case, where II is

O-true, we have in fact

In the second case

Th(d ol1 Il Mod (il) Il Tr(I!fI) Il Sent (.!l'T) S LV{mo} (.!l'T) ,
and hence, in particular,

It foIIows that there is agreement with Li in this respect also: unless II

is O-true, no T-consequence of II is true in Li11 unless it is {mo}-valid.
On the other hand, whereas a T-consequence of II is false in Li only if
its negation is {mo}-valid, this need not be the case in Li11• Nevertheless,
if a T-consequence of II is to be false in L111 it must satisfy a very strong
condition. It should express, roughly speaking, the full T-content of II
within the proper universe. When (**) holds, the precise situation is as
foIIows:
Cn(ll) Il FlS(I!{I) Il Sent(.!l'T) S
S {-r E Sent(.!l'T): Cn(il) Il Sent(.!l'T) S Cn{mo} ({-r})},

where we write Cn{mo}({-r}) for Th{mo}(Mod({-r})). Although there is no

assurance that such a sentence -r exists, the case Cn(ll) Il Fls(L11I) Il
Il Sent(.!l' T) oF 0, namely where II is T-false is L111 , is not excluded and
may arise as follows. It must be assumed first that .A'oIlMod(ll)lo=0.
This is secured if II is O-false. For simplicity the further assumption
is made that MOd(ll)losd o. (Suppose, for instance, that do is given
by a set of postulates for O-terms that are included in ll.) Then, under
these limitations and always assuming (**),
{-r E Sent(.!l' T): Mod({-r}) = Mod(ll)IT} S
S Cn(ll) Il Fls(L111) Il Sent(.!l'T).
If there is a T-sentence -r whose class of models (as structures for .!l'T)
coincides with the class of contractions to .!l'T of models of ll, -r is a
false T-consequence of ll. (Needless to say, alI sentences answering this
description are equivalent.) In fact it may be shown that provided m o is
140 P. M. WILLIAMS

finite, the existence of a false T-consequence of il is assured if only

Cn(il) () Sent(!l' T) is finitely axiomatizable in !l' T. The conjunction of
axioms is then such a sentence. Now it seems that the restriction to finite
values of mo can frequentIy be relaxed if il is the axiom set of an empirical
theory. When this is so, it follows that if il is O-false and its T-conse-
quences are finitely axiomatizable, il is T-false in L~II.
Finally it should be noticed that given (**) and Jl o () Mod(il)!o = 0,

Assuming for simplicity Silo = Mod(Ao) for some Ao s; Sent(!l'o), it fol-

lows that when il is O-false, denotations proper to T-terms in L~II are
just t~ose families of relations on the proper universe of the system
that are not assigned to them by any model of Ao u il. Again this is a
more exact interpretation of T-terms than L~ affords in the same circum-
stances.
In conclusion, it may be mentioned that the solutions discussed can
be generalized in several directions. It will be sufficient to refer to one
solution of general type whose special cases include L~ and L~I.
Let "Y = {Vj(!l' T):i E I} be a collection of subsets of V(!l' T). Then if
Silf is defined by
df = {~E Str(!l'l): ~!o E Silo 1\ [((~!0 E (Sil o!l () Mod
(il»!", -+ 1\ m!VI(2'T) E (Sil O !l () Mod(il»!VI(2'T» -+
ieI
-+ m!o E Mod(ll)!o) -+ mE Mod(ll)]},
the system Lf based on Silf and the corresponding Jli satisfies (P.i)-
(P.iii). Li reduces to L~I when "Y is composed of just V(!l' T) itself, and
to L~ when the sole member "Y is the empty set. If"Y is an n-fold partition
of V(!l' T), Li is the system which results from n successive extensions
of Lo by introducing at the ith stage the terms of Vi(!l' T) in accordance
with solution II in a natural and invariant way. A characteristic property
of Li is the following. If conditions (*) holds, every logical consequence
of il whose descriptive constants form a sub set of some member of "Y
is analytic in Li. A similar method can be used to generalize solution III.

The University of Sussex

 A CONTRIBUTION TO THE PROBLEM OF ANALYTICITY 141

NOTES

1 In the absenceof (iii), several of the theorems given below hold only under suitable
restrictions. The fundamental Theorems 1,2 and 10, however, hold as weII in the ab-
sence as in the presence of (iii).
2 The first two cIauses correspond to conditions (i) and (ii) proposed in [9, pp. 98, 99].
3 It should be emphasized, however, that the O-truth of II is only a sufficient condition
for its truth in LI if LI is constructed from Lo on the basis of II. Thus if II' is also 0-
true, e.g. II' is O-equivalent to II in a natural sense, but if II' is nevertheless inconsis-
tent with II, II' wiII not be true in the system constructed on the basis of II. Conver-
sely, II wiII not be true in the system constructed on the basis of II'. The set II U II'
being inconsistent, neither II nor II' wiII generaIly be true in the system constructed
on the joint basis.

BIBLIOGRAPHY

[1] Ajdukiewicz, K., 'Le Probleme du Fondement des Propositions Analytiques',

Studia Logica 8 (1958), 259-272.
[2] Carnap, R., 'Meaning Postulates', Philosophical Studies 3 (1952), 65-73.
[3] Carnap, R., 'The Methodological Character of Theoretical Concepts,' in H. Feigl
and M. Scriven (eds.), Minnesota Studies in the Philosophy of Science 1, Minnea-
polis 1956, pp. 38-76.
[4] Carnap, R., 'Beobachtungssprache und theoretische Sprache', Dialectica 12 (1958),
236-248.
[5] Carnap, R., 'Repiies and Systematic Expositions', in P. A. SchiIpp (ed.), The
Phi/osophy of Rudolf Carnap, La Salle 1963.
[6] Carnap. R., Phi/osophical Foundations of Physics, New York 1966.
[7] Kemeny, J. G., 'A New Approach to Semantics', Journal of Symbo/ic Logic 21
(1956), 1-27, 149-161.
[8] Popper, K. R., The Logic of Scientific Discovery, London 1959.
[9) Przel~ki, M., The Logic of Empirical Theories, London 1969.
[10] Przcl~ki, M., and W6jcicki, R., 'TheProblemof Analyticity',SyntheseI9(1969),
374-399.
 JOHN A. WINNIE

THEORETICAL ANALYTICITY

The logical analysis of the structure of scientific theory is a task which

is central to any comprehensive philosophy of science, and in this
sphere no philosopher has contributed more incisively than Rudolf
Carnap. The clarity and rigor of his analyses have always made them
apt candidates for fruitful discussion and criticism, and Professor
Carnap's last proposal dealing with the problem oftheoretical analyticity
is, 1 hope to show, no exception to this rule. For my aim here is to take
advantage of the rigor of Carnap's formulation by developing its conse-
quences in some detail. The results which emerge from such a development
are then employed to overcome some of the more weighty objections
against the possibility of drawing an analytic-synthetic distinction for
theoretical systems.
The solution which Carnap has offered to the problem of theoretical
analyticity derives from his more general account of the logical structure
of physical theory (cf. [1] and [4]). According to this account, the de-
scriptive terms of the theory to be reconstructed are initially divided into
two mutually exclusive classes: the class of observation terms and the
class oftheoretical terms. The basis ofthis division is extra-systematic and,
admittedly, not a clear-cut matter. Carnap sometimes speaks of observa-
tion terms as those which refer to easily ascertainable properties and
relations of middle-sized physical objects, yet he also characterizes an
observat ion term as any term which occurs in sentences of a language
'used by a certain language community', and 'understood by all members
of the group in the same sense' ([1], p.40). Clearly, the two charac-
terizations will not always coincide. Moreover, in his Philosophical Foun-
dations of Physics, Carnap declares that observational and theoretical
concepts "lie on a continuum", and goes on to suggest that the distinction
best be drawn by looking to the linguistic practice of the physicists who
employ the theory which is the object of our reconstruction ([4], pp.
225-28).
So although the general framework of Carnap's reconstructions pre-

Jaakko Hintikka (ed.). Rudol! Carnap, Logical Empiricist, 143-159. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
144 JOHN A. WINNIE

supposes some sort of observational term-theoretical term division, it

does not in itself dictate the extra-systematic criteria to be employed in
making this division. In particular, the mere adoption of the general
distinction does not commit us to drawing this distinction on an ontological
basis with the all-too-common corollary that theoretical concepts are
henceforth ontologically suspect. Moreover, if, like Carnap, we explicitly
disavow any distinction drawn along ontological lines, we may thereby
acknowledge the strength ofthe serious objections which have been raised
against attempts to ontologically ground this distinction (cf. [7]), while at
the same time allowing for the fundamental epistemological and semanti-
cal importance of observation statements within scientific theory. But
this last thesis will not be argued here. 1 Instead, it will here be assumed
that a viable and significant ob.servation term-theoretical term distinction
has been suitably drawn in a way which is compatible with the over-all
character of Carnap's general reconstruction.
The observation term-theoretical term distinction now permits the
following three-fold classification of the sentences of the theory. Those
sentences which contain observation terms as their only descriptive signs
are called 'observatidn sentences'; sentences which contain theoretical
terms as their only descriptive signs are called 'theoretical sentences'; and
sentences which contain both observation and theoretical terms are called
'mixed sentences'. When a theoretical sentence is taken as one of the postu-
lates of a reconstructed theory, it then is called a 'theoretical postulate',
and when a mixed sentence is so-taken, it is said to be a 'correspondence
postulate' of the theory (cf. [1], pp. 42 ff).
Carnap next goes on to propose that a physical theory be reconstructed
in such a way that its postulates become a conjunction of the form T· C
(or, more simply, TC), where T and Care each in turn conjunctions of
theoretical and correspondence postulates, respectively. Whereas the
theoretical postulates (T) merely assert relations between the theory's
theoretical concepts, the correspondence postulates relate the theoretical
and observation terms of the theory, thus providing the theoretical terms
with empirical significance and conferring testability upon the theory as a
whole. Hence the correspondence postulates play a dual role: they both
lawfully relate the theoretical and the observational and at the same time
serve the semantical function of empirically interpreting the theory's
theoretical terms. Moreover, once this dual function of the correspon-
 THEORETICAL ANALYTICITY 145

dence postulates is recognized, it is but a small step to see that theoretical

postulates play this double role as well. For Carnap does not require that
there be a correspondence postulate associated with each one of the theo-
ry's theoretical terms, and thus those theoretical postulates which involve
a term not contained in some correspondence postulate must also serve,
however indirectly, to provide these otherwise-dangling theoretical terms
with empirical significance.
The fact that Carnap does not require each theoretical term to be
associated with a correspondence postulate is generally regarded as one
of the virtues of his proposa!. Previous proposals which have required
either the explicit definability of each theoretical concept in terms of
observational concepts alone, or have demanded the introduction of each
theoretical term by means of reduction sentences, are now recognized to
have been inordinately severe. Nevertheless, it is just the more holistic and
liberal character of this framework that provides a basis for serious and
reasonable doubts as to the possibility of drawing a sharp analytic-
synthetic distinction within this framework. For as Hempel has pointed
out ([6J, p. 703), the nomological and interpretative functions of the
theory's postulates seem to have become so tightly fused as to make their
sharp separation impossible. To hold that a statement of the theory is
analytic is surely to consider that statement as devoid of factual content;
but then we must deny that statement any nomological function within the
theory, and the latter, of course, is in direct conflict with our previous
recognition that each of a theory's postulates serves, to some extent, a
nomological function.
It is this difficulty which appears to have prompted Carnap's proposa!.
If 1 am right in what follows, Carnap's response was successful, and, this
response in addition provides the means to cogently rebut Quine's most
telling objections to the possibility of drawing an analytic-synthetic
distinction within formalized languages. In order to substantiate these
ciaims, let us now consider the details of the Carnap proposal itself. 2
Consider a physical theory t' whose axioms are expressed as a conjunc-
tion of theoretical postulates T and correspondence postulates C. Then
we may write the theory as

where the first conjunct represents its theoretical postulates T, and the
146 JOHN A. WINNIE

second conjullct represents the correspondence postulates C. The Ramsey

sentence of the theory is now obtained by existentialIy quantifying over
each of the theoretical terms of Te in succession. Thus the Ramsey
sentence of Te becomes, in our notation:
(2) R TC = (Et 1 ) ... (Et k )
[(---t 1 , ... , tC-)'(---Ol' ... , Oj; t 1 , ... , tc-)].
As is welI-known, the Ramsey sentence of the theory, although not con-
taining any ofthe theory's theoretical terms, entails just those observation
sentences entailed by the theory Te. This result has led some philosophers,
Carnap included, to construe the theory's Ramsey sentence as expressing
the entire factual content of the theory (see also [10] and [12]).
In keeping with this view, Carnap now proposes that any adequate
solution to the problem of analyticity proceed by re-expressing theory
Te in the form

(3) FTC'A TC ,

where A TC is to be taken as the theory's meaning postulate. The postulate

A TC and alI of its logical consequences are now considered to be analytic
within that theory.
Any specific proposal for making this division must, however, satisfy
the folIowing three adequacy conditions ([3], 965). First, the conjunction
F TC • A TC must be 10gicalIy equivalent to the theory Te; next, the first
component, F TC , must entail exact1y those observation sentences entailed
by Te itself; and finally, the second component, A Tc , must have null
observational content. These conditions may be summarized as folIows:
(CI) FTc'ATc<:>TC,

(C2) F TC =? ( ) if and only if

---Ob--- ---Ob---
TC =? ( ),

and
(C3) if ATC=?( ___ Ob ___ ), then (---Ob---)islogicallytrue.

Carnap now proposes that we take the theory's Ramsey sentence for the
first component (F TC ), and for the second component (A TC ), he suggests
the rather curious sentence
 THEORETICAL ANALYTlCITY 147

henceforth to be called 'the Carnap sentence' of the theory. The theory

Te is now written as the conjunction of its Ramsey sentence and its
Carnap sentence, i.e., in the form

with the analytic sentences of the theory now becoming the class oflogical
consequences of the theory's Carnap sentence. It may now be demonstrat-
ed, as Carnap points out, that this proposal satisfies each of the three
conditions of adequacy previously laid down (cf. A, Thms. 1 and 2, and
Cor. 3).
The above proposal will henceforth be referred to as Carnap's basic
proposal. For Carnap does go on to remark that should we wish to estab-
lish additional sentences S of the theory as analytic, we may, if we like,
conjoin them to the Carnap sentence provided that the resulting state-
ment (eTC. S) continues to satisfy the third adequacy condition (C3) set
down above ([3], 965). Thus suggestion will be referred to as the extended
proposal, and 1 shall soon argue that the very features which make for the
attractiveness of Carnap's basic proposal are those which tell against the
extended proposal. However, before considering these matters in more
detail, let us examine the basic proposal in a preliminary way with a view
to dispelling some of the curiosity which surrounds the Carnap sentence
itself.
1 have previously mentioned the fact that the Ramsey sentence of a
theory entails exactly those observation sentences entailed by the theory
Te. This fact has led some writers to the view that a physical theory might
best be regarded as asserting no more or less than what is asserted by its
Ramsey sentence. At times, the motives for such a claim are tied in with
an instrumentalistic approach to theoretical entities (cf. [12]), but this
need not be the case. Professor Hempel, for example, has written that
Ramsey's method "is perhaps the most satisfactory way of conceiving
the logical character of a scientific theory ... ", while at the same time
disavowing an instrumentalistic construal of the Ramsey sentence ([5],
p. 85; cf. also [7]).
Now although a theory Te logically entails its Ramsey sentence, the
con verse does not, in general, hold as well. Hence in order to assert the
equivalence of a theory and its Ramsey sentence, it suffices to maintain
that the Ramsey sentence in some sense entails the theory Te. It is just
148 JOHN A. WINNIE

this that Carnap proposes when he suggests that we consider RTC -+ Te

to be analytic. For notice that, by virtue of the fact that the theory Te
logically entails its Ramsey sentence, it follows at once that the Carnap
sentence is logically equivalent to asserting the equivalence of a theory
and its Ramsey sentence, i.e.,

Thus Carnap's proposal may aptly be characterized as a formalized version

of the philosophical thesis that a theory and its Ramsey sentence are
equivalent.
Still another way of viewing Carnap's proposal is to consider its se-
mantical aspects. An analytic sentence ought to hold in aII possible worlds
i.e., in aII possible models of the theory. Since the Carnap sentence is to
be taken as the analytic postulate of the theory, we are thereby, in effect,
excluding aII structures in which the Carnap sentence lails to hold from
the sphere of possible models. Moreover, since the Carnap sentence is of
the form R TC -+ Te, the structures thus excluded are those in which the
theory's Ramsey sentence holds while the theory Te fails. This is, of
course, just what we should expect upon equating the theory to its
Ramsey sentence.
This semantical characterization of the basic proposal is reminiscent
of Carnap's earlier 'meaning-postulate' approach to analyticity. There,
aII state-descriptions in which some meaning postulate failed to hold were
excluded from the c1ass of possible models of the theory which incorpo-
rated these postulates (cf. [2], p. 226). However, these models are so
excluded merely on the basis of a list of sentences dubbed 'analytic'
within that theoretical reconstruction. Thus, as Professor Quine has
pointed out ([8], pp. 23ff), the lack of generality which attaches to this
procedure hardly helps to alleviate any qualms about the possibility of
drawing a non-arbitrary analytic-synthetic distinction in this manner.
Herein lies an important way in which Carnap's last proposal differs
from its predecessor. For c1early, a general account of theoretical ana-
Iyticity is now provided which is applicable to any appropriately recon-
structed theory, and which yields a c1ear determination ofthose sentences
which are to be considered as analytic within the theory. Furthermore,
unlike the syntactical notions of'postulate' and 'Iist of meaning postulates'
the c1ass of analytic sentences singled out by the basic proposal is in-
 THEORETICAL ANALYTICITY 149

variant under distinct but logically equivalent axiomatizations of a phys-

ical theory r. For let T l C l and T2 C2 be logically equivalent axiomatiza-
tions of a physical theory which utilize non-equivalent theoretical and
correspondence postulates (i.e., Tl~ T2 and Cl~C2)' Since T l C l and
T 2 C 2 are logically equivalent, so are their respective Ramsey sentences,
and, as a result, their Carnap sentences are also logically equivalent (i.e.,
(R TtCt -+ T l Cl)_(RT2C2 -+ T2 C2 »). Hence the class of analytic sentences
is the same for both axiomatizations, showing that the arbitrariness which
attaches to the notion of 'postulate of r' does not also atta~h to the notion
of 'analytic in r' (cr. [8], p. 35).
Yet, it must be admitted that a defense of the basic proposal on grounds
of generality alone does not go to the heart of Quine's objection. For a
proposal which urged us to construe, say, every sentence of a theory as
analytic would, for ali its generality, be absurd. And it is clear that Quine
considers the lack of generality which characterizes the 'list approach'
to be a symptom of the underlying arbitrariness of any distinction so-
drawn. Thus in order to adequately deal with Quine's objections we must
provide a rationale for taking the Carnap sentence and its consequences
as analytic, rather than so-taking some other set of sentences of the
theory.
Herein lies a fundamental difficuity for the extended analyticity propo-
sal. Recall that, according to the extended proposal, we may add meaning
postulates so long as the result continues to satisfy the third adequacy
condition, i.e, so long as the resuiting extended meaning postulate entails
no contingent observation sentence. However, the following results in-
dicate that the extended proposal is far too liberal, and that its acceptance
yields a distinction which is to agreat extent arbitrary.
Let us call a sentence properly analytic if it is analytic but not merely a
logical truth. According to both the extended and bas-ic proposals, ali
logical truths will be analytic, and thus what is of interest here is how the
properly analytic sentences compare under each of these proposals. Now
it can be shown (cf. A, Cor. 9) that no theoretical sentence of a theory is
properly analytic under the basic proposal, i.e., only logically true theore-
tical sentences follow from the Carnap sentence alone. 3 Thus, should we
take, in addition to the Carnap sentence, any set ofthe theory's theoretical
consequences as meaning postulates, we would be genuinely extending
the class ofthe theory's analytic statements, rather than merely appending
 150 JOHN A. WINNIE

sentences already entailed by the Carnap sentence. But it can also be

shown that if we were to append any conjunction of the theory's theoreti-
cal statements to the Carnap sentence, the result will in ali cases satisfy
Carnap's three adequacy conditions (cf. A, Th. 5, Th. 6). Thus, according
to the extended proposal, all or none ofthe theory's theoretical statements
may be justifiably construed as analytic. The decision as to just which
theoretical sentences to so choose must then remain a matter ofarbitrary
choice.
While these considerations provide grounds for the rejection of the
extended proposal, they do not, of course, afford reasons for the accept-
an ce of the basic proposal, and clearly we must go beyond the three
adequacy conditions set down by Carnap in order to provide such a
defense. As a preliminary to this task, the notion of an observationally
vacuous sentence of a theory will now be introduced.
A well-known feature of logical truths is that they occur inessentially
or vacuously as premisses in any valid argument. In other words, any
correct derivation which utilizes logical truths in its premisses remains a
valid argument (although not necessarily a correct derivation) when the
logically true premisses are dropped. Now an observationally vacuous
sentence of a theory plays a role with respect to the derivation of the
theory's observational consequences which is analogous to the role played
by logical truths in ali derivations. Observationally vacuous sentences are
those which function inessentially in any deduction of an observational
consequence from the theory. This is what is stated a bit more precisely
by the following definition.
(7) S is observationally vacuous in TC = df.
TC=;.S, and for any SI such that TC=;,SI' if
(SI ·S)=;.(---Ob---), then SI =;.(---Ob---). (cr. A, D. 7ff).

Before applying this notion to the analyticity problem, let us first consider
its relation to the similar, but weaker, notion of observational uncreativity.
Recall that a sentence is said to be observationally uncreative, or to have
null observation content, if all its observational consequences are logically
true. Carnap's third adequacy condition thus amounts to requiring that the
sentence taken as a theory's meaning postulate be an observationally
uncreative sentence. As we have seen, the Carnap sentence is observa-
tionally uncreative, but so is its conjunction with any theoretical conse-
 THEORETICAL ANALYTICITY 151

quence ofthe theory. Now although any observationally vacuous sentence

of a theory is also observationally uncreative, the converse does not hold
in general (cf. A, Cor. 10). To see this, note that for a sentence S to be
observationally uncreative, it suffices that S in isolation yield no contin-
gent observation sentence. But for a sentence S to be observationally
vacuous, it is necessary that S, even in combination with other consequences
ofthe theory, always play an inessential role in the deduction of contingent
observation sentences from the theory.
The observational inutility of such senteces leads at once to the con-
clusion that, under any reasonable construal of 'confirmation' or 'dis-
confirmation', sentences which are observationally vacuous in a theory
can be neither confirmed nor disconfirmed by observational results. A
negative observational outcome cannot be attributed in any way to the
incorrectness of an observationally vacuous premiss, since the same
prediction results when the observationally vacuous premiss is dropped
from the deduction. For the same re,ason, a positive observational outcome
cannot be held to confirm an observationally vacuous premiss.
Since immunity from experimental confirmation or disconfirmation is
surely an essential ingredient of the pre-systematic notion of analyticity,
it is then reasonable to propose that we supplement Carnap's adequacy
conditions by adopting the additional requirement that the meaning
postulate of the theory be observationally vacuous, i.e.,
(C4) A TC is observationally vacuous in TC. 4
At first sight, it might seem that this requirement is too strong, and could
result in logical or mathematical truths as being the only qualifiers for
analyticity (cf. [6], p. 705). However, it turns out that such fears are
unjustified. For it can be shown that (and this is the main result of this
paper) the Carnap sentence of a theory and its logical consequences con-
stitute exactly those statements which are observationally vacuous in thal
lheory. (cf. A, Cor. 12). Thus a theory's Carnap sentence, and only its
Carnap sentence, satisfies the added fourth adequacy condition. Let us
now assess the consequences of this result for the general problem of
theoretical analyticity.
First of alI, the source of the difficulty with Carnap's extended proposal
is now apparent. By allowing meaning postulates in addition to those
entailed by the Carnap sentence, we are thus allowing sentences which
152 JOHN A. WINNIE

are not observationally vacuous to be considered as analytic. But such

sentences will function essentially in the deduction of some contingent
observation statement from the theory, and could thus be reasonably
said to be confirmed or disconfirmed by the appropriate observational
outcome.
While this consideration tells against the extended proposal, it stems
from a result which provides grounds for maintaining that Carnap's
basic proposal successfully meets the charge of arbitrari ness levelled by
Quine against earlier approaches. For the immunity of the Carnap sen-
tence (and its consequences) to both confirmation and disconfirmation
provides the needed rationale for holdingjust these sentences to be analytic
within the theory.
Finally, we come to the difficulty rai sed by Hempel concerning the
dual role played by the postulates of a theory within the framework of
Carnap's general reconstruction. The difficulty, recall, was the following.
Since each of a theory's postulates serves both a nomological and an
interpretative function, no single statement of the theory can serve merely
an interpretati ve function, i.e., be an analytic statement.
The way in which Carnap's basic proposal resolves this difficulty is as
unexpected as it is simple. For it can be shown that no postulate of the
theory Te is a consequence of the Carnap sentence of Te, i.e., each of
the postulates of Te becomes synthetic under Carnap's proposal (cf. A,
Cor. 13).5 Thus the difficulty rai sed by Hempel is overcome by shifting
the locus of the analytic-synthetic distinction from the theory's postulates
to ·its consequence class in general, thereby allowing for the nomological
role of these postulates. The analytic ingredients of the postulates emerge
at a deductively lower le veI than the postulates themselves; they are, so
to speak, filtered out from the postulates by means ofthe Carnap sentence. 6
Indeed, it can be shown that the properly analytic sentences of a theory
must be mixed sentences ofthat theory, but mixed sentences which are not
correspondence postulates of the theory (cf. A, Th. 2, Cor. 9, Cor. 13).
The above results indicate the ability of Carnap's proposal to deal
with objections to the analytic-synthetic distinction which many writers
have found extremely compelling, and perhaps conclusive. But this
proposal is also of interest by virtue of the light it throws upon the rela-
tion ofthe Ramsey sentence method to the instrumentalism-realism debate.
As was mentioned earlier, Carnap's proposal can be seen as a formalized
 THEORETICAL ANALYTlCITY 153

version of the philosophical thesis that a theory and its Ramsey sentence
are equivalent. Now it has been rightly pointed out (cf. [5] and [7]) that,
by virtue of the existential quantification ingredient in the Ramsey sen-
tence method, we are no less existentially committed to theoretical en-
tities (classes and relations) by the Ramsey sentence than we were at the
outset. The above results provide added support for the correctness of
this view. For recall that a consequence of Carnap's proposal is that each
of the postulates of TC, and thus each of the theoretical postulates of TC,
become synthetic under this account. 7 The factual status thus assigned
to each of the theoretical (and correspondence) postulates of a theory TC
is thus quite in keeping with a realistic interpretation of TC, even when
the theory TC is held to assert no more or less than its Ramsey sentence.

APPENDIX: THEOREMS AND PROOF-SKETCHES

1. The Language L

(i) descriptive signs: (a) observation predicates: O)' O 2 , •.. , Oj; (b) the-
oretical predicates; T), T 2 , ••• , Tk •
(ii) logical signs: (a) connectives: "',., v ,-+, ==; (b) quantifiers and
variables: (1) individual: x, (x), (Ex), X), (x)), (Ex)), etc.; (2) predicate:
t, (1), (Et), 1), (t)), (EI)), etc.
(iii) syntactic variables (a) (---Th---), (---Th---)), etc. - sentences of L
containing only theoretical predicates. These are called 'theoretical sen-
tences'; (b) (---Ob---), (---Ob---)) etc. - sentences of L containing only
observation predicates. These are called 'observation sentences'; (c)
(---Ob, Th---), (---Ob, Th---)) etc. - sentences of L containing both
theoretical and observation predicates. These are called 'mixed sentences';
(d) S, SI' S2' etc. --sentences of L.

II. L-Concepls of L
(i) possible models of L:
Def. M= < U, U 1 , U 2 , U 3 , ... , U m , O)' O 2 , ••. , Oj; T), T 2 , •.. , Tk > is
a possible model of L iff:
(i) U is any non-empty class.
(ii) Ui (1 ~i~m) is the set of ali sets of ordered i-tuples of members of
U, where m is the degree ofthe predicate ofhighest degree occurring in L.
154 JOHN A. WINNIE

Intuitively, the Uhs are the range of the predicate variables of L.

(iii) Oi (1 ~ i ~j) is a set of ordered n-tuples of elements of U, where n
is the degree of the predicate Oi'
(iv) ~ (1 ~i~k) is a set of ordered n-tuples of elements of U, where n
is the degree of the predicate of ~.
The notion of a sentence S of L being true in a possible model is assumed
defined in the customary manner. As is usual, a sentence S will be called
L-true (in sym bols, '=> S') iff S is true in all possible models of L. A
sentence SI logically implies S2 (SI =>S2) if and only if every model of
SI (i.e., every possible model in which SI is true) is also a model of S2'
SI and S2 are said to be L-equivalent (SI ~S2) ifand only if(iff) SI =>S2
and S2 =>SI' In the following, the standard results involving the logical
relations among the sentences of L are freely used. These include the
following: (1) S=>S; (2) if =>S, then for any SI' SI =>S; (3) if SI =>S2
and S2 =>S3, then SI =>S3; (4) if SI =>S2 and =>SI, then =>S2; (5)
(SI VS2)=>S3 iff both SI =>S3 and S2=>S3, etc.

III. The Carnap Sentence

D.l. WhereŢ=Ax(---OI' 02, ... ,Oj' TI ,T2, ... , Tk;---),thenR'=df.
(Et t ) (Et 2 ) ••• (Et k ) Ax(---Ol' O 2 ,,,,, 0j; ti' (2"'" t k ; ---).
Th.l. Ţ=>(---Ob---) iff R'=>(---Ob---).
Proof: Cf. [12], 291ff.
As an immediate corollary of D.I, we also have:
Cor.l. Ţ=>R'.
The Carnap Sentence of Ţ is now defined as follows.
D.2. C'=df. R'--+Ţ.
Cor.2. C'~(R'=Ţ).
Proof: Since, by Cor. 1, Ţ => R " it follows that => (Ţ --+ R '). Hence
(R'--+Ţ)~(R'--+Ţ)·(Ţ--+R'). Thus by D.2., C'~[(R'--+Ţ)'(Ţ--+R')],
and so C'~(R'=Ţ).
Cor.3. Ţ~(R'·C').
Proof: (a) by Cor.l, Ţ=>R', and, trivially, Ţ=>(R'--+Ţ), i.e., Ţ=>C'.
Hence Ţ=>(R"C'), (b) By D.2, (R"C')~[R"(R'--+Ţ)], and thus, by
modusponens, [R'·(R'--+Ţ)]=>Ţ must obtain. Hence, (R'·C')~Ţ.
Cor.4. Ţ=>C'.
Proof: Immediately from Cor.3.
D.3. Sis analytic (A-true) in Ţ=df. C' =>S.
 THEORETICAL ANALYTICITY 155

D.4. S is properly analytic in r = df. S is analytic in r and S is not

L-true.
Cor.5. If=>S (S is L-true), then Sis A-true in r.
Proof: follows immediately from D.3 and the fact that any statement
entails an L-true statement.
Cor.6. Sis A-true in r iff r=> S and "-' R' => S.
Proof: By D.3, Sis A-true in r iff C' =>S. By D.2., C =>S iff (R' -+ r)
=> S, i.e., iff ("-' R' v r) =>S, i.e., iff "-' R' =>S and r =>S.

IV. Observation Statements and fhe Carnap Sentecne

Th.2. If S is an observation statement, then S is not properly A-true in
any theory r.
Proof: By D.3 and D.4, this amounts to showing that if C· => (---Ob---),
then =>(---Ob---). Assume that Ct=>(---Ob---). By Cor.4, r=>C" and
thus r=>(---Ob---). AIso, by Th.l, R'=>(---Ob---) as well. But since
C=>(---Ob---), by Cor. 6, "-'Rt=>(---Ob---). Thus R t and ,,-,R t each
entail (---Ob---), and so =>(---Ob---).
D.5. Sis Ob - uncreative =df. If S=>(---Ob---), then =>(---Ob---).
Cor.? If S is A-true in r, then Sis Ob-uncreative.
Proof: Assume that S is A-true in r and S=>(---Ob---). By D.3,
C =>S, and thus C =>(---Ob---). Hence by D.4 and Th.2, =>(---Ob---).

v. Theoretical Stafements and the Carnap Sentence

D.6. r has Ob-content = df. There is a first-order observat ion statement
(---Ob---)1' such that r=>(---Ob---)1 and ~(---Ob---)l.
Cor.8. If r has Ob-content, then ~R t.
Proof: Suppose that r has Ob-content. Then, by D.6, there is a first-
order observation statement, say (---Ob---)l> such that r => (---Ob---)1
and =f> (---Ob--- )1. By Th.l, Rt => ( ---Ob-- -)1 also, and since =f> (- --Ob---),.
it follows that =f> R t.
Th.3. if (---Ob---)1 and (---Th---)1 are both satisfiable, first-order
sentences, then: (i) if (---Ob---)1 =>(---Th---), then =>(---Th---), and
(ii) if (---Th---)l =>(---Ob---), then =>(---Ob---).
Proof: This is a standard result offirst-order logic which relies upon the
fact that the pairs of sentences involved have no descriptive terms in
common. For a proof, see [11], 169.
156 JOHN A. WINNIE

Th.4. If T is consistent and has Ob-content, then if C T => (- --Th- - -),

where (---Th---) is a first-order sentence, then =>(---Th---).
Proof: Assume that C T => (---Th---). Since T has Ob-content, by D.6,
there is a first-order observation statement, (- - -Ob- - -)1' such that
T=>(---Ob---)l and ;,>(---Ob---)l. By Th.l, RT=>(---Ob---)l and thus
- (---Ob---)l => - R T. Now, by Cor.6, - R' => (---Th---), and since
- (---Ob---)l => - R" we obtain: - (---Ob---)! => (---Th---). Since
;'>(---Ob---)l' it follows that -(---Ob---)l is satisfiable. Hence, from
Th.3 (i) it follows that =>(---Th---).
Cor.9. If T is consistent and has Ob-content, then no theoretical
statement is properly A-true.
Proof: Immediately from D.4 and Th.4.
Th.5. If T is consistent, and T=>(---Th---), then C T • (---Th---) is
Ob-uncreative.
Proof: Suppose that [C'·(---Th---)]=>(---Ob---). Then CT=>[(---Th
---)-+ (---Ob---)], and by Cor.6., -R' =>[(---Th---)-+ (---Ob---)].
Now by Cor.4, T => C T and by assumption T=> (---Th---). Hence T=>( C T.
(---Th---»), and since [CT·(---Th---)] =>(---Ob---), T=>(---Ob---). By
Th.l, it now follows that RT=>(---Ob---) also, and thus, trivially,
R T =>(---Th---) --+ (---Oh---». But we have already shown that _ R T =
[( ---Th- --) -+ (- - -Ob- --)], and thus => (- --Th- - -) -+ (- - -Ob- --), i.e.,
(---Th---) => (---Ob---). Since Tis consistent, and T => (---Th---), (-- -Th-- -)
is satisfiable. Hence, by Th.3 (ii), => (- - -Ob- --).
Th.6. IfT=>(---Th---), then T_RT·[CT·(---Th---)]
Proof: As in Cor.3.

VI. Observational Vacuity

D.? Sis observationally vacuous in T=df. T=>S, and for any SI' if
T=>Sl' and (Sl·S)=>(---Ob---), then SI =>(---Ob---).
Cor.IO. If S is observationally vacuous in T, then Sis Ob-uncrea-
tive.
Proof: If S is observationally vacuous in T, then from D.? it follows that,
in particular if (p v -p)·S=> (---Ob---), then (p v -p)=>(---Ob---), i.e.,
=>(---Ob---). Hence the result, by D.5.
Th.? C is observationally vacuous in T.
T

Proof: By Cor. 4, T=>C'. Suppose next that T=>SI and (Sl·C')=>(---

Ob---). Hence, by D.2, [SI· (R T -+ T)] => (---Ob---), i.e., [SI· (- R' V T)]
 THEORETICAL ANALYTICITY 157

::;. (---Ob---), i.e., [(SI' '" R <) V (SI "r)]::;. (---Ob---). Hence (SI' '" R <)
::;.(---Ob---).
Since -r::;.C< and -r::;,SI' -r::;,(C<'SI), and thus (since (C"SI)::;'
(---Ob---)) it follows that -r::;.(---Ob---). And from Th.1 we now have
that R<::;.(---Ob---).
But since (SI' '" R <) =~+--Ob---), and R·::;. (---Ob---), we see that
[(SI' "'R<)V R<] ::;.(---Ob---), i.e. [(SI V R')'(R<v "'R<)] ::;.(---Ob---)
i.e., (SI V R<)::;.(---Ob---). Hence SI ::;.(---Ob---). Thus the result,
from D.7.
Cor.11. If S is A-true in -r, then Sis observationally vacuous in -r.
Proof: Suppose that S is A-true in -r. Then, by D.3, C' ::;.S. Let
-r::;,SI and (SI·S)::;.(---Ob---). Then by D.7, we need now show that
SI ::;.(---Ob---). Since (SI·S)::;.(---Ob---), S::;,(SI ~(---Ob---)). Since
C<::;.S, we then have that C<::;,(SI ~(---Ob---)), Le., (C'·SI)::;.(---Ob
---). Hence, from Th. 7, SI ::;.(---Ob---).
Lemma 1: If C' f;> S and -r::;. S, then '" R <f;> S.
Proof: Immediately from Cor. 6.
Th.8. If C <f;> S, then Sis not observationally vacuous in -r.
Proof: Assume that C' f;> S.
Case 1: -r f;>S. Then the result from D.7 immediately.
Case 2: -r::;.S. By Cor. 1, -r::;.R" and thus -r::;.(S~R<). Now
S' (S ~ R <) .::;. R <. Hence by D.7, in order to show that Sis not obser-
vationally vacuous in -r, it suffices to show that (S ~ R <) f;> R '.
Thus suppose that (S ~ R <)::;. R <, Le., ( '" S V R <)::;. R '. Then '" S::;. R <
and thus "'" R ' ::;. S. Since C' f;> S and -r::;. S, this contradicts Lemma 1.
Cor.12. Sis A-true in -r iff S is observationally vacuous in -r.
Proof: Immediately from Cor. 11 and Th. 8.

VII. Concisely Formulated Theories

Let -r be expressed as a conjunction of postulates Pl' P2 ' •••• Pn'
D.8. Pi (l ~i~n) is an Ob-redundant postulate of -r=df. Pi ·P2 • ...
... ·Pn::;.(---Ob---) iff PI· ... ·Pi-I·Pi+I· ... ·Pn::;.(---Ob---).
D.9. Pl' P2 • ... • Pn is a concise formulation of -r=df. No postulate
PiCI ~i~n) of-r is Ob-redundant.
Th.9. If PI • P2 ' •••• Pn is a concise formulation of -r, then no postulate
of -r is observationally vacuous in -r.
Proof: Let Pi be a postulate of a concise formulation of -r. From D.8
158 JOHN A. WINNIE

and D.9 it now follows that for some (---Ob---), Pi-[P1 ••••• Pi - 1 • Pi+ 1· ...
... ·PJ=(---Ob---), yet [Pl· ... ·Pi-l·Pi+l· ... ·PJ=(---Ob---). The
result now follows immediately from D.7.
Cor.13. If Pl' P2 ' .••• Pn is a concise formulation of r, then no postu-
late of r is A-true.
Proof: Immediately from Cor.12 and Th.9.

Indiana University

NOTES

1 For such arguments, see [9], 88ff.

2 In the body of what follows, the results obtained are stated without proof. Theorems
and an outline of their proofs are inc1uded in the Appendix, and the references (e.g.,
'(A, Th. 2a)') in the main body of this paper are to this Appendix. I am indebted to
Irving Copi and Alberto Coffa for their comments, criticisms, and suggestions concern ing
these resuIts. Any errors which may remain are, of course, my own responsibility.
3 This resuIt, and many of those to follow, is proven under the assumption that the
theory TC is consistent and has observational content.
4 Since (C4) implies (C3), we can merely replace (C3) by the stronger (C4) and obtain
the same effect.
5 This resuIt holds on (trivial) condition that TC is a concise formulation of the theory
(cf. A, 0.8).
6 This is more than mere metaphor. For in order that S be analytic, the Carnap
sentence requires that: (i) TC~ S; but then goes on to require that (ii) "" RTC~ S
(cf. A, Cor. 6).
7 Unless they are logically true or observationally redundant. Indeed, no theoretical
statements which are consequences of TC are properly A-true.

BIBLIOGRAPHY

[1] Carnap, R., 'The Methodological Character ofTheoretical Concepts', in Minnesota

Studies in the Philosophy of Science, VoI. I (ed. by Feigl and Scriven), Minnesota,
1956.
[2] Carnap, R., 'Meaning Postulates', in Meaning and Necessity, 2nd. ed., Chicago,
1958.
[3] Carnap, R., 'Replies and Systematic Expositions', in The Philosophy of Rudolf
Carnap (ed. by P. A. Schilpp), Open Court, 1963.
[4] Carnap, R., Philosophical Foundations of Physics, Basic Books, 1966.
[5] Hempel, C. G., 'The Theoretician's Dilemma', in Minnesota Studies in the Philoso-
phy of Science, VoI. II (ed. by Feigl, Maxwell and Scriven), Minnesota, 1958.
[6] Hempel, C. G., 'Implications of Carnap's Work for the Philosophy of Science',
in Schilpp [3] above.
[7] Maxwell, G., 'The Ontological Status of Theoretical Entities', in Minnesota
Studies in the PhilosophyofScience, VoI. III (ed. by Feigl and Maxwell), Minnesota,
1962.
 THEORETICAL ANALYTICITY 159

[8] Quine, W. V. O., 'Two Dogmas of Empiricism', in From a Logical Point of View,
2nd. ed. Rev., Harvard, 1961.
[9] Quine, W. V. O., 'Epistemology Naturalized', in Ontological Relativity and Other
Essays, Columbia Univ., 1969.
[10] Ramsey, F., 'Theories', in his The Foundations of Mathematics, 3rd. ed., Kegan
Paul,1954.
[lI] Robinson, A., Introduction to Model Theory and to the Metamathematics ofAlgebra,
North-Holland Publ. Co., 1968.
[12] Rozeboom, W., 'The Factual Content of Theoretical Concepts', in Feigl and
Maxwell, [7] above.
 ANDERS WEDBERG

DECISION AND BELIEF IN SCIENCE

Comments on Rudolf Carnap's Views in 'Empiricism, Semantics,
and Ontology'

I. INTRODUCTION

In 'Empiricism, Semantics, and Ontology', 1 Rudolf Carnap is primarily

interested in showing that 'ontological' existence questions are irrelevant
to the acceptability of scientific theories. In the course of discussing this
central theme, he touches a number of difficult problems in the psychol-
ogy of science, such as the role and nature of 'decision' and 'belief. The
problems are such that I feeI it would be rash of me to try to advance any
solutions. The purpose of these comments is the limited one of giving an
exegetica1 and frank1y critica1 discussion of Carnap's views.
The reader of these comments is advised to prepare himself by first
studying Carnap's essay. However, I shall here briefly indicate his line
ofthought:
The problem that Camap faces is this. By many of its adherents, 'empiricism' is interpreted
as involving a denial of the existence of certain kinds of entities, in particular of so-called
'abstract entities', like properties, classes, relations, numbers, functions, propositions, etc.
Simultaneously, it is a fact that many scientific theories are formulated as theories about
abstract entities of one kind or another. Such is, e.g., the case with mathematics and
physics. A scientist, who embraces the present kind of 'empiricism', is thus put in a
quandary. One way out, sometimes tried, is to declare the theories in question, or the most
suspect parts of them, mere uninterpreted calculi.
What Carnap wants to convince us of could perhaps roughly be summarized as follows:
(i) 'Ontologica!' questions as to the existence of a kind of entities are 'non-cognitive',
'non-theoretical', 'metaphysical', 'nonsensica!'. (ii) 'Empiricism', when rightly understood,
takes no stand on any such questions. From the point of view of true 'empiricism', 'nomi-
nalism' and 'platonism' are, e.g., equa!ly devoid of sense. (iii) Nor does the acceptance of a
scientific theory, such as, say, some branch of mathematics, involve any 'ontologica!'
assumption of the existence of the entities dealt with. (iv) To accept a theory, or rather the
'framework' of a theory, is a practica! decision, a choice of an instrument, which as such
needs no theoretical justification whatsoever. (v) Without committing ourselves on "on-
tological questions", we are entitled to regard, e.g., the arithmetic of natural numbers as a
theory about numbers, and not as a mere uninterpreted calculus. 2 In addition we may
perhaps list the following tenet: (vi) The entities of a framework have - so to speak - no
properties but those which the framework attributes to them. 3
In order to convince us of these points, Camap presents (in very schematic outline, alas)
a theory conceming the role of what he calls 'frameworks'. When formalized, a scientific

Jaakko Hintikka (ed.), RudolfCarnap, Logical Empiricist, 161-181. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
162 ANDERS WEDBERG

theory such as, e.g., the arithmetic of natural numbers, may be considered based on a frame-
work. The framework contains two kinds of rules, viz., rules goveming the formation of the
sentences of arithmetic, and rules saying under what circumstances an arithmetical sen-
tence is to be accepted or rejected. The decision to accept a framework may be made with
a certain ulterior purpose in view, e.g., the purpose of expressing and communicating factual
knowledge. The decision is more or less wise, depending upon how well the use of the
framework serves the given purpose. Only beliefs or assertions are true or false. Since a
decision is not a belief, it is inappropriate to ask whether it is true or false. In particular, the
acceptability of a framework does not depend upon a previous answer to any 'ontologica!'
questions. Whether or not there 'really are' natural numbers, is, e.g., irrelevant to the
acceptability of the arithmetical framework.
Actua11y, whţ:n the question: 'Do natural numbers exist?' is asked in advance of the
acceptance of an arithmetical framework, the question is metaphysical nonsense. It is a
genuine theoretical question only within an already accepted arithmetical framework.
Only within a framework can meaningful questions be asked and 'meaningfu!' ('cognitive',
'theoretical') 'beliefs' or 'assertions' be expressed.
When we have accepted a framework, within which the sentence:
5 is a number
is acceptable, we may, in an unproblematical fashion, go on and accept an enlarged frame-
work, where the sentences:
'5' designates 5,
and:
'5' designates a number,
are acceptable. Having accepted the framework of an arithmetical theory, we are rather
bound to regard the theory, not as a mere calculus, but as a theory about something, viz.
about numbers.
In Sections II and III, I shall briefly discuss the nature of a framework as
described by Carnap. In Section IV, I ask myself what, in Carnap's
terminology, it means to 'accept' a given framework, with reference to
Carnap's contention that such acceptance needs no theoretical justifica-
tion. At the end ofSection IV, I arrive at a first hypothetical interpretation
of 'acceptance'. In order to test this interpretation, I inquire, in Section V,
into the aims of Carnap's discussion of acceptance. The result of this
inquiry forces me to dismiss, in Section VI, the first interpretation. In
Sections VII and VIII, I discuss two other hypotheses, none of which I find
satisfactory. In Section IX, I make a fourth and last attempt to find a
satisfactory interpretation of the acceptance spoken of by Carnap. This
interpretation involves a theory concerning the nature of belief. In Sec-
tion X, I shall consider what I take to be Carnap's views on frameworks
and their semantics. In Section XI, I review Carnap's arguments for the
'non-theoretical' nature of 'ontologica!' questions and statements. In
 DECISION AND BELIEF IN SCIENCE 163

Section XII, I try to show that Carnap's position - in spite ofhis own ex-
plicit denial - is a kind of (hidden) nominalism. In the concluding Sec-
tion XIII, I briefly state what seems to me to be the sens moral ofthis study
of Carnap's 'Empiricism, Semantics, and Ontology'.

II. FRAMEWORKS

Bya 'framework' Carnap seems to understand a certain kind oflanguage,

or - to be more exact, perhaps - a certain kind of set of devices within a
language. Carnap envisages a type-theoretical 'rational reconstruction' of
'the language of science' where, e.g., 'observable things and events' are
assigned to one type, the natural numbers to another, 'propositions' to a
third, 'thing properties' to a fourth, and so ono Let L be the type-
theoretical language, obtained by Carnap's reconstruction of 'the lan-
guage of science'. We should think of L, not as a linguistic structure that
is fixed once for alI, but rather as a structure to which new types may be
added and from which types already introduced may be eliminated
To add a new framework to L is to add a new type to L. Thus, in adding
a new framework, first we add to the vocabulary of L a new type of
variables, possibly also new constants substitutable for these variables,
and new predicates with argument-places reserved for the new type of
signs. Further, in a welI-known fashion, we extend the application ofthe
formation rules of L to the new signs, and we extend the rules of deduc-
tion of L to the new sentences thus created. When the new framework is
intended for use within empirical science, we may also give 'rules of
evaluation' according to which "results of observation are evaluated ... as
confirming or disconfirming evidence" for the new sentences (p. 22 [207]).
By these rules the framework is tied up with what Carnap calIs "the con-
tent of our experiences" (p. 24 [208]). The rules of evaluation are, of
course, different in nature from the rules of deduction previously hinted
at. Whereas, in Carnap's terminology, the rules of deduction are, I
presume, purely 'syntactical', the rules of evaluation are not so.
I think it is important to notice that Carnap has a type-theoretical
language-structure in mind. Call1ap does not, I think, wish to assert that
"our common scientific language" actualIy is divided into logical types,
stilliess that it is divided just into those types that he mentions as examples
in his discussion. What he asserts is only, 1 think, that one could give a
164 ANOERS WEOBERG

type-theoretical "rational reconstruction" of "our common scientific

language". Already here, however, 1 feeI that great caution is neces-
sary. For Carnap's discussion of ontological existence questions, it is not
irrelevant how the types are distinguished. Nor is it irrelevant what 'ra-
tional reconstruction' of the rules of logic we choose. If, say, angels
were made into a separate type and we were to choose one possible re-
construction of logic, the sentence: 'There exists an x such that x is an
angel' would turn out to be logically provable. If, on the other hand, we
include possibly existing angels within the type of, say, 'observable things
and events', the same sentence is, as it should be, turned into a risky
hypothesis. To discuss 'our common scientific language' through the
medium of a type-theoretical 'rationhl reconstruction', thus, may not be
without its dangers. The answer to a given question may weU depend upon
what principle of type-theoretical partition we choose.

III. THE RULES OF ACCEPTANCE ANO REJECTION

Concerning the form of the rules, governing acceptance and rejection,

Carnap says very !ittle. A reader who is familiar with his semantical
writings naturally suspects that these rules are just what Carnap elsewhere
caUs 'semantical rules' of a 'semantical system', in particular 'rules of
truth [and of falsehood]'. Nowhere, however, does Carnap explicitly
identify the rules of acceptance and rejection with semantical rules. Sev-
eraI considerations speak against the assumption that he intends such
identity. According to one of Carnap's (Tarski-inspired) ideas, semantical
rules of truth should be such that any sentence of the form :
x is true if and only if p
is valid, where 'x' is replaced by the name of a sentence and 'p' is replaced
by that sentence itself, or by a 'translation (into the metalanguage)'
thereoL IT the rules of acceptance and rejection for a framework F were of
this kind, nobody could ever formulate the rules of F who had not
already accepted either F itself or a 'translation' of F. Nobody, hence,
could ever be in the position to decide, for the first time, to accept ex-
plicitly the rules of a radically new framework F. Nor, ifwe had already
accepted a framework F, could a formulation of its rules of acceptance
and rejection offer any but technicallogical problems (such as were fust
 OECISION ANO BELIEF IN SCIENCE 165

solved by Tarski in his investigations into the concept of truth). But

Camap argues as if we sometimes could decide to adopt the rules of a
radically new framework, and also as if the statement of the rules of an al-
ready accepted, familiar framework could offer problems of a different
order. (Rules to the same effect can, of course, be formulated in many
different ways, and some ways can be philosophically more c1arifying
than others. It couldbe that what, on Camap's view, is problematic is the
philosophically c1arifying formulation of the rules of a familiar .frame-
work.)
My impression, whatever it is worth, is that Camap thinks ofthe rules
of a framework as specifying the conditions under which a sentence is
to be accepted or rejected, in 'observational' and/or 'syntactical' terms.
They do not (or: do not necessarily), like 'rules of truth', offer us
synonymous equivalents for the sentences concemed. In particular, they
are not-like the 'rules oftruth' in some ofCamap's formulations - basep
on 'rules of designation' (or of 'intension and extension'), assigning given
entities as 'designata' (or as 'intensions' and 'extensions') to expressions.
If they were, we should have to accept the existence of those entities before
we could accept the rules as significant. To formulate the rules of a given
framework is a task that Camap assigns, not to 'semantics', but to (pure)
'epistemology' (p. 22 [207J).
In a consistent framework, the set of 'acceptable' sentences obviously
must not overlap with the set of 'rejectable' sentences. But how great may
be the intermediate set of those sentences that are neither acceptable nor
rejectable? Camap says nothing about this question. Perhaps he is in-
c1ined to leave a wide range of possibilities open.
Finally, are 'acceptability' and 'rejectability' always absolute proper-
ties, like 'truth' and 'falsehood', or 'provability' and 'refutability' in a
calculus, or can they be matters of degree, as they would seem to be in
large parts of the empirical sciences? About this nothing is said in
Camap's essay, but 1 presume that he is in favor of the latter altemative.
The linguistic processes, initiated by the acceptance of a framework,
do, in Camap's presentation, have an obvious similarity with processes
in a machine. Let us think of a machine that prints sentences on two strips
ofpaper, those printed on the one being 'accepted' by the machine, those
printed on the other being 'rejected'. (If 'acceptability' and 'rejectability'
are matters of degree, ranging, say, from 1 to 0, we can think of the
166 ANDERS WEDBERG

machine as printing sentences with markers, from 1 to O, on one strip of

paper.) The machine is constructed in such a way that it can receive
'information' from the outside ('observe' its milieu), and that such infor-
mation influences its action. Certain sentences it accepts (rejects) inde-
pendent1y of any information: theyare 'demonstrated' ('refuted') a priori
by the machine. Other sentences are accepted (rejected) only upon the
reception of certain kinds of information: these sentences constitute the
aposteriori 'reports' of the machine. The rules of a framework, as con-
ceived by Camap, describe, 1 think, the functioning of such a machine in
terms of information received and syntactical sentence-structure.

IV. TO ACCEPT A FRAMEWORK. FIRST INTERPRET A TION

A framework may be 'accepted' or 'rejected' by a given person. To accept

a framework, Camap says, is "nothing more" than "to accept rules for
forming statements and for testing, accepting and rejecting them" (p. 23
[208]). Here, there are two acceptances involved, the acceptance of the
rules, and the acceptance (or rejection, as the case might be) of sentences
demanded by the rules. Camap does not say much about the nature of
either of these acceptances. For the understanding of his argument,
however, understanding of the senses of these acceptances is essential.
The explicit rules, contained in the framework as a part of the rational
reconstruction of scientific language, may, in Camap's opinion, corre-
spond to previous non-verbalized habits (p. 22 [207].) Ifwe focus our at-
tention, not on the reconstruction, but on what is being reconstructed,
perhaps to accept a rule is essentially to be in the habit of acting in a man-
ner that is correctly described by the rule. A person can obviously be in
such a habit without consciousness of the rule. In particular, he may be
in the habit without ever having made a decision to get into it. Unless we
are mathematicallogicians and familiar with type theory, we are, in fact,
not in the position knowingly to decide to accept the rules of a type-theo-
retical framework. Being himself a mathematical logician, however,
Camap most of the time thinks of the acceptance as taking place by a
decision. In so far as he is aiming at a description of normal scientific
procedure - in Section IV 1 shall discuss his aims - this is somewhat un-
fortunate. What is true about a decision to get into a habit, need not be
true of the habit itself.
 OECISION ANO BELIEF IN SCIENCE 167

To accept a rule can, a priori, mean to believe that the rule is correct.
1 may, e.g., accept the rule that every statement found in the Bible
should be accepted in the sense that 1 believe this rule to be correct, i.e.,
1 believe that every such statement, as a matter of fact, is true. Obviously,
Camap does not intend the acceptance of the rules to be understood in this
way. The decision to accept a framework is, Camap insists, a practical
decision which does not need any theoretical justification. Let me quote
a few of his many similar statements:

"The decision of accepting the thing language [one of the frameworks mentioned in
Camap's essay], although itself not of a cognitive nature, will nevertheless usually be
infiuenced by theoretical knowledge, just like any other decision conceming the acceptance
of linguistic or other rules. The purposes for which the language is intended to be used, for
instance, the purpose of communicating factual knowledge, will determine which factors
are relevant for the decision. The efficiency, fruitfulness, and simplicity of the use of the
thing language may be among the decisive factors. And the questions conceming these
qualities are indeed of a theoretical nature" (pp. 23-24 [208]).
"The acceptance cannot be judged as being either true or false because it is not an
assertion. It can only be judged as being more or less expedient, fruitful, conducive to the
aim for which the language is intended. Judgments of this kind supply the motivation for
the decision ofaccepting or rejecting the framework" (pp. 31-32 [214]).
"This acceptance [of a framework] is not in need of a theoretical justification (except
with respect to expediency and fruitfulness), because it does not imply a belief or assertion"
(pp. 35-36 [218]).

Camap here argues that only beliefs or assertions are true or false, that
decisions are never beliefs, and that, hence, decisions are neither true nor
false. This may be correct - at least, 1 feei no urge to debate the point.
(Presumably, Camap would likewise argue that habits are never beliefs
and that, hence, habits are neither true nor false. This may likewise be
correct, although the correctness hangs on a rather thin terminological
thread.) However, Camap's further contention that a decision (and pre-
sumably also a habit) is not in need of "theoretical justification (except
with respect to expediency and fruitfulness)", is not at alI convincing.
, Suppose that 1 make a decision (or am in the habit) to believe statements
to be true or false in accordance with certain rules. Does not the theo-
retical justification of such a decision, or habit, depend upon the truth-
values ofthe beliefs? "Even in this case", a defender of Camap would per-
haps say, "the decision can be judged only by its expediency in view of
some ulterior aim. If your aim is to become happy, the decision is ex-
pedient, if it makes you happy. If the aim is to reach true beliefs, it is expe-
168 ANDERS WEDBERG

dient to the extent that the ensuing beliefs are true. But in neither case is
any 'theoretical justification' needed". But such a line of argument seems
to me like mere quibbling. Let us agree on the terminological point that
a decision or a habit to entertain certain beliefs is theoretically justified if
and only if those beliefs are true. 4 H ence, there are theoretically justified,
and theoretically unjustified, decisions and habits.
Since Carnap asserts that the decision to accept a framework does not
need any theoretical justification, it might seem to follow that this decision
cannot be the decision to entertain any beliefs. But, then, in what sense do
the rules of the framework demand of us that we should 'accept' and
'reject' sentences in a given manner? What does the acceptance, or rejec-
tion, of a sentence mean? The natural interpretation seems to be that to
accept a sentence is to believe it to be true (or to believe what it asserts)
and that to reject it is to believe it to be false (or to disbelieve what it
asserts). With this interpretation, however, the decisions (and habits) of
which Carnap speaks would seem to be such as are in need of a theoretical
justification. Is it, then, Carnap's view that the acceptance or rejection of
a sentence is not connected with any belief as to ils truth-value (or as to
the reality of what it asserts) ?
1 can, of course, 'accept' or 'reject' sentences without any such belief.
When working out some symbolic problem on the blackboard, 1 may be
said to 'accept' those symbol configurations that 1 leave on the board and
to 'reject' those that 1 erase, should they happen to be written down by
mistake. Can it be Carnap's view that this manifest procedure of leaving
and erasing sentences has its counterpart in science at large, that when a
scientist accepts or rejects a sentence this is essentially a complex generali-
zation of what is done on the blackboard? If that were his view, we could
understand why he says that the decision to accept a framework does not
need theoretical justification. Does Carnap hold such a view? In order to
see this question in its right perspective, we ought to inquire into the aims
of Carnap's statements about the acceptance of frameworks and of
sentences.

v. CARNAP'S AIMS

It might eventually be possible to formulate what Carnap thinks about the

acceptance of frameworks in a comprehensible statement roughly of this
 DECISION AND BELIEF IN SCIENCE 169

form:

(E) The person P accepts the framework F, if and only if

... P ... F ....

The formulation of (E) is impossible, 1 think, unless we first formulate

what Carnap thinks about the acceptance (rejection) of sentences in a
comprehensible statement of similar form:

(F) The person P accepts the sentence S, if and only if- P - S - .

So far 1 have asked myself what, according to Carnap's intentions, we

should put on the right side of Equations (E) and (F). Now, 1 want to ask
for the 'purpose' of the equations. A priori there are agreat variety of
purposes that a statement such as (E), or (F), might be intended to serve.
1 shall here review only those which seem most pertinent.
(1) The statements (E) and (F) might be intended merely as a declara-
tion concerning the sense in which Carnap intends to use the phraseology
"to accept a framework (sentence)". - (E) and (F) can plausibly be inter-
preted as being, among other things, such dec1arations, but they can not
very well be mere1y that. If they were, nothing of philosophical interest
would follow from them.
(2) (E) and (F) might be interpreted as also being a dec1aration con-
cerning Carnap's own scientific attitude. When he himself 'accepts' a
framework or a sentence it is, Carnap might wish to say, just in the sense
of statements (E) and (F~ The statements might, in Carnap's opinion, be a
characterization of what he himself is doing in such cases. In Section IV
ofhis essay, the section entitled 'Abstract Entities in Semantics' Carnap
actually discusses his own semantical theories, as exposed in Introduction
to Semantics, and Meaning and Necessity. To Ryle's criticism that
Carnap's semantics involves a 'Fido'-Fido-principle and that this is "a
grotesque theory", Carnap rejoins: "Grotesque or not, Ryle is wrong in
calling it a theory. It is rather the practical decision to accept certain
frameworks" (p. 36 [218]). Here Carnap undoubtedly intends to make an
autobiographical assertion. Carnap acknowledges that earlier writers on
semantics, such as John Stuart Mill, Frege and Russell, may be open to
Ryle's criticism, but he dec1ares his own position to be immune against it.
Thus, it seems, Carnap is here saying that (E) and (F) are a correct
170 ANDERS WEDBERG

description of his own scientific attitude, although they may not be an

adequate description of the attitudes of John Stuart MiIl, Frege, and
Russell.
(3) Carnap probably considers his own attitude, as described by (E) and
(F), to be so to speak the 'rational' scientific attitude to a framework.
Hence, I think that we might interpret (E) and (F) as being, in addition to
a description of Carnap's own attitude, a proposal, or an exhortation, to
the world of scientists to adopt the same attitude. An aim of Carnap's
essay is to show how a scientist may embrace a framework, which
countenances abstract entities, without exposing himself to the charge of
deserting 'empiricism' for the heresy of 'platonism'. Perhaps, then, what
Carnap says to the scientists is this: "Ifyou 'accept' your abstract theories
merely in the sense of (E) and (F), then the charge can not justly be made
against you".
(4) Suppose thatit were a very rare occurrence that a scientist accepted
a framework merely in the sense of (E), or a sentence in the sense of (F).
What Carnap demands, would then amount to a very drastic reform of
the scientific spirit. Although I cannot prove it, I have a strong feeling that
Carnap does not here think of himself as such a reformer. It seems to me
that, between the lines, he is saying something like this: "When you
accept, and operate within, a framework, e.g., one concerning abstract
entities, you always do, inter alia, what (E) and (F) explain. In addition you
may do a lot of other things, perhaps mumble to yourself certain non-
sensical metaphysical phrases, entertain various imaginations, and so ono
But these additional activities of yours are irrelevant to 'science' as such.
All that is 'scientific' in your acceptance is described by (E) and (F)". I do
not find such a statement very clear. In fact, I think that it involves rather
a misuse ofthe word 'science'. To me this word has a vague meaning as a
collective designation for certain activities, persons, methods, books,
techniques, theories and so ono In the above statement it is used quite
differently, and I do not see clearly how it is used. But still I think, that
the above statement expresses something that Camap intends to assert.
Thus we have found reason to think that the equations (E) and (F), in
addition to being nominal definitions, are a declaration concerning
Camap's own scientific attitude, an exhortation to the world of science to
adopt the same attitude, and an attempt to grasp the very 'scientific es-
sence' of the acceptance of frameworks and sentences.
 DECISION AND BELIEF IN SCIENCE 171

VI. TO ACCEPT A FRAMEWORK. THE FIRST INTERPRETATION

FOUND INADEQUATE

In Section IV we found that the decision, or habit, to accept and to reject

sentences in accordance with the rules of a framework would eschew the
need of theoretical justification if 'to accept', and 'to reject', a sentence
is to be understood as not involving anş belief. If this were actually
Camap's intention, then, in view ofthe aims ofhis discussion, he would be
implicitly saying something like this: "When I myself accept or reject a
sentence of a scientific theory, I do this without entertaining any belief (or
making any assertion). Scientists, I advice you to do likewise. Actually,
the occurrence of beliefs or assertions is extrinsic to the scientific essence
of acceptance and rejection of sentences." I do not think that we have the
right to impute such a paradoxical position to Camap. If beliefs and as-
sertions were to be banished from science, where would they find a
refuge? In Camap's terminology in the present essay, 'belier, 'assertion',
'theoretical question', 'cognitive statement', 'theory' are alI closely
interrelated terms. Without the existence or possibility of 'belier and
'assertion', there are no 'theoretical questions', no 'cognitive statements',
no 'theories'. If science were freed from beliefs and assertions, it would
also be freed from theoretical questions, cognitive statements, theories.
There are, it is true, certain ut1erances of Camap's that, at least on a
superficial reading, could be tinderstood in this direction. I quote once
more Camap's rejoinder to Ryle: "Grotesque or not, Ryle is wrong in
calling it a theory. It is rather the practical decision to accept certain
frameworks" (p. 36 [218]). What Ryle called the 'Fido'-Fido principle is
clearly not a decision but certain theorems occurring within Camap's
semantics. I take Camap's rejoinder to mean that, being the outcome of a
practical decision to accept a certain framework, these 'theorems' do not
embody any 'theory'. The logical consequence would seem to be that
Camap's entire semantics is not a 'theory' and, since his semantics is
supposed to be on a par with other scientific theories, that no scientific
theories are 'theories'. One even gets the impression that Camap's seman-
tics is not a 'theory' just because it is free from 'metaphysics'. Would,
then, 'metaphysics' be the only home left for belief and theory, when
they have been driven out from science? However, I am sure that we
should not press such somewhat loosely formulated statements too hard.
It is clear that Camap's central position is rather the opposite of that
172 ANDERS WEDBERG

which they could be taken to indicate. Belief, assertion, theoretical

questions, cognitive statements and genuine theories occur only within
science, never in 'metaphysics'. The previous acceptance of a suitable
scientific framework is a conditio sine qua non for belief, assertion,
theoretica1 questions, cognitive statements and theories.
We may hence safely conclude, it seems, that 'to accept' and 'to reject'
a sentence is not, in Carnap's opinion, necessarily a belief-free procedure.
Probably he would not say that, whenever we accept or reject a sentence in
the sense of equation (F), this involves a be1ief. He would, however, I
take it, say that acceptance or rejection, in the sense of (F), frequently
involves belief.
If that is so, we see that Carnap's assertion that the acceptance of a
framework needs no theoretical justification must be misleading. To
accept a framework is to make a decision, or to be in a habit, of accepting
and rejecting sentences in accordance with certain rules. The acceptance
of the framework is theoretically justified only to the extent that the
beliefs, involved in accepting and rejecting sentences according to the
rules, are true. When a person accepts a framework, we are entitled to ask
whether this is theoretically justified or not.

VII. TO ACCEPT A FRAMEWORK

A SECOND INTERPRETATION FOUND INADEQUATE

A procedure may be said not to 'need' any theoretical justification if the

justification is self-evident, or if the procedure, so to speak, justifies itself.
Let us inquire whether Carnap might have meant that the acceptance of a
framework always theoretically justifies itself. How could such self-
justification possibly occur?
One possibility that presents itself is this: A framework is essentially a
set of rules, determining the various categories of expres sion in a lan-
guage, plus a set of 'semantical' rules, assigning 'meanings' to the ex-
pressions or giving them an 'interpretation'. Under the interpretation,
certain sentences turn out true and others turn out false. The rules of
acceptance (rejection) merely enjoin us to accept (reject) what is true
(false). When I let my linguistic habits conform with the rules of a frame-
work, in this sense, the sentences I use will express genuine 'beliefs' or
'assertions' and, in addition, these beliefs will be true.
 OECISION ANO BELIEF IN SCIENCE 173

We have already, in Section III, found reasons why it is improbable

that Camap thinks of the rules of a framework as semantical rules. If
those reasons are sound, we may dismiss this second attempt to inter-
pret his doctrine.
VIII. TO ACCEPT A FRAMEWORK:
A THIRO INTERPRET A TION REJECTEO

The rules that a framework lays down for the acceptance and rejection
of sentences may be thought of as stating under what circumstances a
given sentence is an A-sentence (acceptable) or an R-sentence (to be
rejected). We can also imagine that, in accepting the framework, I decide
to classify a sentence in one of these ways if and only if such classification
is entailed by the rules. In such a case the rules can be saidto constitute
my definitions of the notions of an A- and an R-sentence. Written out in
the conventional manner, these definitions would run something like this:
S is an A-sentence=df. The rules (together with results of
observation) entail 'S is an A -sentence'.
S is an R-sentence = df. The rules (together with results of
observation) entail 'S is an R-sentence'.
Whenever I correctly apply the rules of the framework, my classifica~
tion of sentences into A- and R-sentences is automaticalIy right, in view
ofthese definitions. We may say that my beliefthat a given sentence is an
A-sentence, or that it is an R-sentence, is automaticalIy true whenever
that beliefis obtained in accordance with the rules ofthe framework.
This model indicates one way in which the acceptance of a framework
could be said theoreticalIy to justify itself. However, can the model be
considered as a generalIy valid picture of what acceptance and rejection of
sentences within a scientific theory amounts to? I think noL AlI the
beliefs occurring in our model are beliefs about the classification of sen-
tences. It seems patently absurd to maintain that alI beliefs in science are
(or ought to be) beliefs of that kind. The scientists do have, and are en-
titled to continue to have, beliefs about the atoms, the stars, Alexander
the Great, the religion of Tibet, and so ono Nor are their beliefs alI ofthe
simple form: 'x is a y'. (An attempt to interpret the scientist's beliefs in
accordance with our model meets many other difficulties which, however,
I shall here pass by.)
174 ANDERS WEDBERG

If Camap thinks of frameworks as being se1f-justifying systems of the

present kind, his interpretation of scientific theories in terms of frame-
works is gross1y inadequate.

IX. TO ACCEPT A FRAMEWORK: THE FIRST INTERPRETATION

RESUMED. THE NATURE OF BELIEF

Is there some other and more promising sense in which the acceptance of
a framework might justify it self? The following is mere1y an ad hoc at-
tempt vague1y to indicate such a sense.
If a person properly speaking, entertains a belief, he is, we may as-
sume, doing two things: (i) having a certain 'accepting' or 'rejecting'
attitude toward some sentence, and (ii) giving that sentence a 'cognitive
meaning'. The accepting (rejecting) attitude toward a sentence does not,
in itself, constitute any genuine be1ief, we assume. Even the 'meta-
physician' has this accepting (rejecting) attitude toward his 'metaphysical'
sentences, but his 'metaphysics' is not a genuine belief, only a quasi-
belief - the reason being that he does not give his 'metaphysical' sentences
any 'cognitive meaning'. To find out whether a person gives a sentence a
cognitive meaning or not, we must observe his habits of 'acceptance' and
'rejection'. If the sentence is part of a language system, that the person is
using, and if, in accepting and rejecting the sentences of this system, the
person obeys certain 'dominant habits' (that could be codified in the
form of rules), then and only then does he endow the sentence with a cog-
nitive meaning. A be1ief, that conforms with the dominant habits (rules),
is 'true', whereas one, that breaks the dominant habits (rules), is 'false'.
Since the nature of 'cognitive meaning', 'belier, 'truth' and 'falsehood' is
such, a system of dominant language habits, that constitutes 'acceptance
of a framework', justifies itself.
Little dialectical skill is needed to show how vague and unsatisfactory
is the theory, presented in the previous paragraph. How, e.g., should we
distinguish between a person's dominant habits, that are the standards of
cognitive meaning, truth, and falsehood, and his possibly conflicting er-
roneous habits? Take a person who is in the habit of accepting everything
that is said in the works of the German existentialist Heidegger. Camap
would not, 1 imagine, like to say that this habit justifies itself, or that it is a
part of a system oflanguage habits that, as a whole, justifies itself. Instead,
 DECISION AND BELIEF IN SCIENCE 175

I imagine, he would have to say that this particular habit is largely in-
consistent with the person's dominant language habits (that to some
extent are the same in all the members of the German language com-
munity). But how should the intended distinction be given a clear em-
pirical meaning? When pursued in this direction, Carnap's ideas are,
incidentally, seen to touch topics that occupied the interest of the late
Wittgenstein and that play so prominent a role in Philosophical Investiga-
tions: "What does it mean to play a game in accordance with certain
rules?", "What does it mean to break a rule of a game?", and so ono
Perhaps there are pertinent senses in which the acceptance of a frame-
work may be said to contain its own theoretical justification, senses that
are essentially different from any that I have been able to think of.

x. A FRAMEWORK AND ITS SEMANTICS

Ifwe have accepted a framework F, we may also, Carnap insists (p. 33ff.
[216ff.]), accept what for the moment I shall call its 'associated Carnap-
semantical' framework S (F). In doing so, for each expression of certain
kinds in F, we add its quotation (or some equivalent device). Thus, ifthe
numera15 occurs in F, we may in S(F) introduce the new expression '5'.
We may also introduce specifically semantical words with appropriate
rules. E.g., we may introduce in S(F) the word designates and lay down a
rule to the effect that each sentence of the form:
(i) (write here the quotation of an expression of kind K in F)
designates (write here the expres sion itself)
is to be accepted. According to this rule for instance the sentence
(ii) '5' designates 5
is to be accepted. We may also introduce in S(F) the word true and rules
for its use such that each sentence of the form:
(iii) (write here the quotation of a sentence in F) is true if and only
if(write here the sentence)
is to be accepted. Thus, the following sentence is to be accepted:
(iv) '5 is greater than 2' is true if and only if 5 is greater than 2.
Etc. etc.
176 ANDERS WEDBERG

If someone accepts the framework of numbers, Carnap says (p. 35

[217]), then "he must acknowledge" that numerals designate numbers.
"Generally speaking, if someone accepts a framework of entities, then he
is bound to admit its entities as possible designata. Thus the question of
the admissibility of entities of a certain type or of abstract entities in
general as designata is reduced to the question of the acceptability of
those entities" (ibidem). When we have once accepted a framework F,
to accept the associated Carnap-semantical framework S(F) is, ob-
viously, in Carnap's opinion, an unproblematical step. Concerning the
use of any framework, one may inquire whether it is expedient or fruitful
for the purposes at hand. Carnap appears to imply that the question as to
fruitfulness of S(F) is essentially reducible to the question as to the fruit-
fulness of F itself.
Because of the so to speak automatic transition from F to S(F), if I
accept a framework F, I will also have to acknowledge F as a 'semantical
system'. If Fis a framework accepted by a person P, somebody e1se than
myself, and I ask myself whether Fis a 'semantical system', my answer
will be positive provided that I myself already accept or am prepared to
accept For a system 'equivalent' to F. 5 If, on the other hand, I cannot
bring myself to accept either For any system equivalent to F, 1 shall refuse
to consider F as a 'semantical system'.
If the foregoing paragraph is a correct account of Carnap's position,
and if, in addition, my last attempt in Section IX to interpret Carnap's
views on the relation between frameworks and beliefs is correct, a some-
what paradoxical situation arises. Assume that Fis a framework, ac-
cepted by a person P, other than myself, and that I do not myself accept F
or anything equivalent to F. By the line of thought of the foregoing para-
graph, we find that I shall then correctly refuse to consider F as a seman-
tical system. Simultaneously, however, by the line of thought presented
in Section IX, 1 cannot refuse to acknowledge that the statements made
by P in accordance with F express genuine cognitive beliefs. Thus, I
shall have to distinguish between (i) frameworks that have the character
of semantical systems (in Carnap's sense), and (ii) frameworks the sen-
tences of which express genuine cognitive beliefs but which, nevertheless,
are not semantical systems. This consequence, I believe, runs counter to
a strong trend in Carnap's philosophy.
From an unsophisticated point of view, it would seem that to the ex-
 DECISION AND BELIEF IN SCIENCE 177

tent that Fis an uninterpreted calculus ('accepted' or not), S(F) is likewise

an uninterpreted calculus, and that adding S(F) to F does not in any way
change the status of F.

XI. THE STATUS OF ONTOLOGICAL QUESTIONS

AND STA TEMENTS

A central topic of Carnap's essay is the status of ontological questions

and beliefs. His attitude toward such questions and beliefs is somewhat
difficult to grasp.

(A) An ontological (or metaphysical) question such as the classical one:

'Does the external thing world exist?' is, Carnap says, not a theoretical
question. Questions and statements, as mere sequences of words, can
hardly be divided into 'theoretical' and 'non-theoretical', in any philo-
sophically relevant sense. What Carnap means, must be that a question or
statement, as a sequence of words, together with the manner in which it
is being used by a certain person, is theoretical or non-theoreticaL What
criterion does Carnap use in making the distinction?
In one place, Carnap argues that a question is non-theoretical, because
it has not been formulated in "the common scientific language" (p. 25
[209]). "The common scientific language" appears to me to be a myth,
and a philosophically dangerous one at that. It somehow involves an
appeal to established authorities, an appeal which in this case is question-
begging.
In another place, Carnap infers the non-theoretical nature of a thesis,
viz., the thesis of the reality of the thing world, from the presumed fact
that it can not be formulated in "any ... theoretical language" (p. 23
[208]). This argument is question-beggi~g.
The same thesis is denied theoretical nature also on the score that "it
cannot be formulated in the thing language" (ibid.). On some chosen
understanding of the thesis and some chosen construction of "the thing
language", this impossibility will, of course, obtain. But here, too, the
argument is question-begging. What means of expres sion we include in
"the thing language", will depend upon what we think worthwhile to
say.6
The main trend of these arguments of Carnap's can be summed up in
178 ANDERS WEDBERG

the statement, that ontological or metaphysical questions and assertions

are not quite at home within established science. But why are they not
theoretical? There are, to be sure, many theoretical questions yet
awaiting to be acknowledged in established science.

(B) Another line ofthought is closely connected with Camap's views on

frameworks.
Exactly the same form of words, e.g., 'Numbers exist', or 'There are
numbers', may be used both by the scientist and by the metaphysician.
When used by the scientist, it does express a theoretical assertion, a belief.
When used by the metaphysician, in his metaphysical mood, it does not.
What the metaphysician means, is, Camap says, "a question prior to the
acceptance" of the framework of arithmetic. This seems to be the very
kemel of Camap's distinction betwetm scientific beliefs and metaphysical
quasi-beliefs. It is only through its place in a framework that a sentence
acquires cognitive meaning. If the sentence is tom loose from its frame-
work and taken in isolation, it lacks such meaning.
As evidence for the assumption that the metaphysician isolates his
sentences from any framework, Camap might, 1 imagine, point to the
fact that metaphysicians notoriously disagree and that one cannot off-
hand see any way of preparing the way for universal agreement. The
disagreement and the difficulty, nay impossibility, of reaching an agree-
ment is due, Camap might say, to the fact that the metaphysicians have
no definite framework in mind, no fixed rules that determine under what
circumstances their sentences are to be accepted or rejected.
Actually, any worried metaphysician or ontologist could consider
Camap's doctrine of frameworks as a we1come means of defending him-
self against antimetaphysical critique. With elementary logical skill, he
can formulate the 'framework' of his own theory. Disagreement
between metaphysicians A and B can be looked upon as disagreement as
to the (metaphysically, or otherwise) most expedient choice of frame-
works. The doctrine of frameworks must be supplemented with much ad-
ditional doctrine in order to become a weapon in a fight against meta-
physics and ontology.

(C) Perhaps there is some truth in Carnap's view ofmetaphysics and on-
tology as a 'non-cognitive' use of language. But it is certainly a very
 OECISION ANO BELIEF IN SCIENCE 179

schematic and inarticulate view. Each so-ca1led ontological or meta-

physical question and statement ought to be studied and judged separa-
tely. Nay, each writer's own use of a so-called metaphysical question or
statement should be investigated by itse1f. Frequently, we shall no doubt
find that a so-called metaphysical sentence, as used by author X in book Y,
is ambiguous, vague, and in other ways deficient. But these qualities are
a matter of degree. It is not probable that a clear-cut dichotomy, such as
Carnap wishes to introduce, between theoretical and non-theoretical uses
of a sentence would be interesting and fruitful. It is also a fact that an
unprecise usage can often be made more precise in theoretically in-
tere sting and fruitful ways. Carnap himself has, as is well-known, made
interesting contributions to the 'explication' of obscure concepts. It
would be an unfortunate policy always to dismiss as nonsense what is not
sufficiently clear. How do we know that the unclarity of certain so-called
metaphysical questions is not sometimes of the kind that calls for ex-
plication rather than shutting of eyes and ears?
There are questions with a perfectly clear sense which are extremely
difficuIty if not impossible to settle. How can we know that the difficuIty
of settling certain so-called metaphysical questions is not due, at least in
part, to similar causes?

XII. CARNAP'S HIOOEN NOMINALISM

By the logic of his own argument, Carnap is forced to declare the 'on-
tological' question concerning the existence of abstract entities a non-
cognitive pseudo-question. The affirmative answer: 'Abstract entities do
exist', and the negative answer: 'Abstract entities do not exist', are from
a theoretica1 point ofview equally meaningless. 'Nominalism' is as much
metaphysical nonsense as conceptual 'realism' or 'platonism'. Neverthe-
less, I suspect that, in an important sense, Carnap is what would normally
be called a 'nominalist'.
In Section III, we have already suggested that the linguistic processes
within a framework can be likened to the functioning of a machine that
'accepts' and 'rejects' sentences, in part upon information received from
outside. Turning from this allegory to the human use of language, what
corresponds in human beings to the external information? Carnap says
very little about this. As an empiricist, however, he uses such phrases as
180 ANDERS WEDBERG

"observations" (p. 22 [207]), "empirical investigations" (p. 22 [207]),

"the content of our experiences" (p. 24 [208]), "immediate data" (p. 38
[220]), "the facts of experience" (p. 29 [212]), and the like. What, then;
in Carnap's view, is the scope of 'our experiences' ? Obviously, he is in
the first place thinking of our sense experience. Re seems to have in mind
especially the type of observations that are made in the experimental and
observational sciences, physics, chemistry, biology, astronomy, and so
ono But do the experiences, involved in such observations, exhaust the
whole range of human experience ? Carnap himself mentions (pp. 37-39
[219-220]) the philosophical discussion as to whether we ever "immedi-
ately experience" abstract entities. Carnap finds the affirmative view to
be "very dubious psychology" (p. 39 [220]), but he does not definitely
reject it as false. 7
Suppose, now, for the sake of argument, that we actually do (in some
sense) experience abstract entities. Such experiences could then form one
source of information, in addition to the kind of sense experience of
which empiricists like Carnap primarily think. Concerning a theory of
abstract entities, we could then ask whether it is true to our experiences
of such entities.
The whole tenor of Carnap's essay indicates that he does not believe
in this possibility. A theory of abstract entities should, in Carnap's opin-
ion, be judged entirely on its merit as an instrument for dealing, ultimate-
ly, with something else, with "results of observations", such as those
made in the experimental sciences. It is difficult to conceive any other
reason for this view than a deep-seated conviction that neither do we have
any experience of abstract entities nor do such entities at alI exist In so far
as this conviction does not concern empirico-psychological matters
where 'philosophy' has no authority, it is, by his own argument, reduced
to a piece of metaphysical nonsense that he is prevented from acknowl-
edging. The views expressed in Carnap's essay seem, in fact, to be a very
efficient method of suppressing important philosophical convictions into
the unconscious and thereby making them impossible to debate.

XIII. SENS MORAL

Carnap's discussion in 'Empiricism, Semantics, and Ontology' revolves

around (i) the nature of so-called 'ontological' existence questions, in
 OECISION ANO BELIEF IN SCIENCE 181

particular questions concerning the existence of 'abstract' entities, (ii) the

nature of 'scientific' theories, in particular scientific theories dealing with
'abstract' entities, and (iii) the interrelation between the questions and
theories. The kind ofvague, schematic assertions ofwhich Carnap's essay
abounds are, 1 feeI, far removed from any genuine and philosophically
clarifying study of those diverse phenomena in our wor1d that can be
subsumed under the two (vague and perhaps not necessarily exclusive)
titles 'ontology' and 'science'.
NOTES

1 First published in (1) Revue Internationale de Philosophie 4 (1950), 20--40. Reprinted in

(2) Semantics and the Philosophy of Language, A Col/ection of Readings, ed. by L. Linsky,
111. 1952. Aiso reprinted in (3) R. Carnap: Meaning and Necessity, 2 enlarged ed., Chicago
1956, pp. 205-221. Conceming the latter reprint Camap says (in a footnote on p. 205):
"1 have made here some minor changes in the formulations to the effect that the term
'framework' is now used only for the system of linguistic expressions, and not for the sys-
tem of the entities in question." In the present paper, page-references are given both to
(1) - the pagination ofwhich is reproduced in (2) - and to (3). Thus, a reference 'p. 22 [207])'
indicates that the passage quoted or alluded to is found on p. 22 in (1) and on p. 207 in (3).
2 I cannot quote any straightforward statement of Camap's in support of (v). However,
the whole tenor of the introductory section of his essay seems to imply (v).
3 This is a remarkable view on which, however, I shall not comment in this paper.
4 Ifthe demand for 100% truth should appear too stringent, let us introduce the notion of
'x% theoretical justification', requiring x% truth.
5 A simple case of the appropriate kind of 'equivalence' would be this: two frameworks can
be transformed into each other by a one-to-one correlation between their supplies of
'simple signs'.
6 In another place, Camap gives still another reason for the same contention. The question

whether the extemal thing world exists or not, "cannot be solved because it is framed in a
wrong way. To be real in the scientific sense means to be an element of the system; hence
this concept cannot be meaningfully applied to the system itself" (pp. 22-23 [207]). Here
Camap seems to assume that there is such a thing as 'the scientific sense' ofthe word 'real',
and that when the word is taken in this sense, the sentence: 'The wor1d of things is real'
becomes meaningless. 'The scientific sense' is nothing but a new myth, and the introduction
of this myth constitutes a new appeal to established authorities. Even if there were such a
thing as 'the scientific sense' of a given term, there may still exist any number of other
legitimate senses that could be attached to the same term.
7 This is, by the way, an awkward position for one who holds Camap's general views. If the
question whether abstract entities exist is a pseudo-question, mustn't also the question
whether we experience such entities be a pseudo-question? Carnap could, of course, answer
that he conceives the latter question as being put within a proper framework. But would not
that be merely to dodge the difficulty?
 HERBERT G.BOHNERT

CARNAP'S LOGICISM

Carnap's philosophy of mathematics was logicist to the end of his life

in 1970. It was not a straightforward extension of earlier views of Frege
and Russell. It differed sufficiently, indeed, that some recent writers have
referred to it as conventionalism. 1 But Carnap himself thought of it as
logicist, and within the Frege-Russell tradition. 2 Difficulties that beset
earlier forms oflogicism, controversies over analyticity, and new develop-
ments in foundational studies have resulted in the mature form of
Carnap's logicism being widely misunderstood and underappreciated.
In this paper 3, I propose to review the development of Carnap's out-
look, clarify various points of misunderstanding, and argue for its basic
viability and value.
1. DEVELOPMENT

The development of Carnap's logicism may be divided into four phases

for preliminary purposes. I shall call them the (1) classical, (2) syntactical,
(3) semantical, and (4) theoretical phases. The classical phase may be
thought of as coming to a climax in 1930, the year of the famous sympo-
sium at which he was the spokesman for logicism, along with Heyting for
intuitionism, and von Neumann for formalism. 4 The syntactical phase
may be thought of as exemplified by [4], i.e., with its zenith in 1934-37. 5
The semantical phase may be thought of as having its beginning in 1939
with [5], but not as having any actual ending. In speaking of a fourth,
theoretical phase, I refer to an emergent, overall viewpoint in which a
certain underlying, longheld, pragmatistic-experimental attitude pro-
vided a unifying setting for earlier attitudes (yet without becoming a form
of pragmatism or of Quinean empiricism, but retaining a genuine claim
to being logicist).

(1) The Classical Phase

The early inftuence in this phase were Frege and Russell, of course.
According to Carnap's own account, his 1910-14 classes with Frege were

Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 183-216. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
184 HERBERT G.BOHNERT

important, but it was his reading of Russell in 1919-1921 that made logic
the central instrument of his thought. Yet, a rereading of Frege in 1920
persuaded him simultaneously of the ana1yticity of logic and mathema-
tics. Thus he never shared Russell's shifts between a viewing of a system
of logic as (1) a source of certain know1edge, (2) a hypothesis to be con-
firmed by its mathematically evident consequences, or (3) a body of
factually empty tauto10gy (in 'a sense broadened from Wittgenstein's).
The 1ast shift was an emotional b10w to Russell. But for Carnap, Wittgen-
stein's claims of tauto10gy were simp1y a stimu1ating clarification of
Frege's ana1yticity view. The initial net effect on Carnap of the Frege-
Russell infiuence was thus what 1 shall call strong logicism, namely, the
view that (1) mathematics can, and can best be, construed as a deductive-
definitional development of logic, and (2) logic, and hence mathematics,
is analytic. 6
This strong logicism was refined, modified, and weakened before the
climax of Carnap's classical phase. It was refined by the explicit separa-
tion of geometry, physically interpreted, from mathematics (through the
infiuence of Sch1ick and Hahn). It was modified by a shift from the inten-
sional standpoint of Frege and Russell to a thoroughgoing extension-
alism, which obviated Russell's reducibility axiom and ramified types in
favor of simple types (through the infiuence of Ramsey). 7 It was weakened
by the concession that the axioms of infinity and choice could not be
regarded as truths oflogic. Carnap treated them, as Principia Mathema-
tica already did, as hypotheses. His logicism at that time, then, was of the
more modest, conditional kind which asserted that while the concepts of
mathematics were reducible by definition to logical opes, the truths of
mathematics were deductively reducible only to logic plus axioms of
infinity and choice. 8
To critics of logicism the reasons for retreat from the strong to the
conditional claim may seem obvious, but before proceeding it may be
well to recall a bit more closely how they appeared to logicists, and to
Carnap in particular, at this juncture. For example, hesitation about
assuming an infinite number of abstract entities was not a central factor,
per se. Frege's use of abstract entities was limited not by ontological
doubts but by the appearance of paradox. Russell's theory of types,
which involved decisions as to what abstract entities did or did not exist,
was initiated, in part, to ensure consistency. Thus, although Frege and
 CARNAP'S LOGICISM 185

Russell often spoke platonisticalIy, their treatment of abstract entities

was often auxiliary to the needs of logico-mathematical reasoning. Their
concern for the needs of such reasoning, by contrast, was with respect of
its application to concrete (Le., non-abstract) reality. This focus on the
concrete became especially explicit when Russell's system of types was
interpreted as restricting the. basic universe of discourse to concrete
individuals, leaving alI abstract entities restricted to higher type-Ievels,
where they appeared as attributes, so to speak, rather than as the primary
subject-matter of a body of knowledge. The concrete interpretation,
though not formally necessary, seemed essential to the logicist trend of
thought, which sought to base mathematics not on any independent
realm of abstract objects but, as remarked, on its application to non-
mathematical reality.91t was, therefore, only when the theory oftypes in
this interpretation appeared to require an axiom of infinity for concrete
individuals (by banning any such argument for abstract entities as was
offered in RusselI's Principles) that logicists, including Carnap, felt
compelled to reduce their claim. 10
Although the 'conditiona!' view above may be characterized as a
classically clear variant of logicism, it should be borne in mind, in anti-
cipation of things to come, that the papers of 1930 and 1931 showed
influence from both intuitionism and formalism. This reflected the fact
that Carnap regarded those papers not as defending logicism against
rivals, but as steps toward a synthesis of alI three views. Carnap's
affinity for formalism was especialIy pronounced. This affinity was to
acquire special significance with the (almost simultaneous) appearance of
G6del's results.

(2) The Syntactic Phase

The appearance of Syntax marked a dramatic change. To some, indeed, it
seemed a "surrender to formalism".l1 There was reason for such an
impression. Some of the characteristics of the Frege-Russell tradition
were explicitly put aside and some of Hilbert's attitudes were explicitly
adopted. The system of Syntax was presented, in fact, as not belonging
wholIy to either tradition, but as providing a system which "fulfills,
simultaneously, the demands of both formalism and logicism" Y While
significant continuity with the Frege-Russell tradition became more
apparent in later phases, much of what may be called, distinctively,
186 HERBERT G.BOHNERT

Carnap's logicism had its most elearcut inception in Syntax.

Although motivations for the change were instilled early in the
elassical period, as will be remarked upon presently, the change between
the 1930, 1931 papers and Syntax appeared abrupt: The syntax (forma-
tion rules) of two languages, I and II, was given along with their deductive
apparatus (transformation rules).13 Although this metalinguistic specifi-
cation, for each language, was caHed its logical syntax, the borderline
between logic and mathematics within each language was gone. This was
explicitly intended to be in accord with Hilbert's view that logic and
mathematics should be constructed simultaneously.14 Accordingly, the
universe of 'individuals' of both languages consisted of the natural
numbers and the deductive apparatus ineluded forms of the Peano
axioms. To be sure, only the syntax appropriate for such an interpreta-
tion was given, but it was not a syntax that seemed appropriate for an
interpretation in which members of the universe of individuals were
concrete objects in the logicist sense indicated in the preceding section.
In particular, the earlier logicist reservations with respect to an axiom
of infinity appeared not to apply, since for Language I and II, such
reservations seemed overridden as a matter of syntax. (For Language II,
the axiom of choice also was made part of the transformation rules.) For
Language I such postulation might appear at first glance as part of a
specific mathematical undertaking. It was a first-order system with a
weakened deductive apparatus meant to illustrate, and to be in accord
with, constructivist precepts. But Language II was a powerful, type-
theoretic system in which aH elassical mathematics, aH science, and
indeed, aH cognitive knowledge was to be formulable. It was here, then,
where an infinity of individuals appeared to be proelaimed by linguistic
fiat for such a language, that Carnap most appeared to have foresaken
logicism for conventionalism or formalism. Yet Carnap's elaim for his
languages, especiaHy II, was that they satisfied the elaims not only of
formalism but oflogicism. 15 Since breaks with the logicism of the elassi-
cal phase were undeniable, such a elaim could be meant only for some
essential part.
The essential part consisted, for Carnap, of two logicist theses which
he held to be still supported by Syntax. There was, first, the (somewhat
generalized) analyticity thesis, that the truths of mathematics and logic
are analytic (whether or not mathematics is distinguished from, or re-
 CARNAP'S LOGICISM 187

duced to, logic). Secondly, the view that the basis for interpretation of
mathematical concepts and assertions was found only in their application
to non-abstract, non-mathematical reality (the application view) was still
taken as fundamental, despite alI appearances to the contrary.
Both analyticity and application would appear to involve meaning, of
course, and so to be beyond the reach of syntax. To see how the analyticity
and application views were meant as still maintained, we may recall
various points alternately about each.
With respect to application, the formalistic emphasis was misleading.
Although application would appear to require interpretation, and seman-
tics was explicitly avoided in Syntax, informal references to meaning
occur throughout Syntax. The focus on formalizing syntax did not, in
fact, stem from a view that mathematics could or should deal only with
uninterpreted calculi. 16 It stemmed, rather, from a methodological
viewpoint which implied that only syntax could be usefully formalized (a
viewpoint we will return to presently). Accordingly, the syntax ofthe two
languages provided for the syntax needed for an interpreted language -
by admitting descriptive signs in addition to logico-mathematical ones. 17
With respect to analyticity, a first question to face might be: What
should be expected of an analyticity concept defined for sentences of a
formalized language?18 For Carnap there were two: (1) a sentence's
being analytic should be determined solely by the rules oflanguage; (2)
an analytic sentence should be without factual content. In this phase,
'rules of language' meant only syntactical rules, including definition of
terms belonging to the language as part of the transformational syntax.
Furthermore, lack of factual content of an analytic sentence meant only
that it was to have no factual sentence (as syntactically defined) as a
(syntactical) consequence. While this purely syntactical formulation was
later to be modified, we may fust take note of the difficulties which the
syntactical formulations encountered in the (then) recent results of Godel.
GOdel had shown that the (formation-rule) syntax of Principia Mathe-
matica-plus-Peano (or, equivalently, plus PM's infinity axiom) yielded
sentences which were neither provable nor refutable from its axioms and
rules of inference, and that for any additional, effectivelydefinable set of
axioms, in the given vocabulary, there would still be such sentences.
Furthermore, the result appeared to hold for any consistent, effective
concept of proof.
188 HERBERT G.BOHNERT

One who already accepted the 10gicist view that the purpose of the
logico-mathematical apparatus of a language was only to implement the
articulateness ofthat language might have accepted Godel's results with
mere disappointment - disappointment that a certain maximal goal for
that implementation was unreachable. 19 But a critic of logicism could
react differently. If some logico-mathematical sentence was not resoluble
by the rules of language, though still true or false, the suspicion might
arise, so to speak, that it was synthetic. More embarrassing for logicism
was the fact that Godel's example of a syntactically indeterminate
sentence was obviously true. While one might argue, with some reason,
that it was still somehow 'true in virtue of meaning', it was synthetic
and apriori as far as the syntax of that language could determine. Carnap
therefore took a difficult and not decisively successful course. He dejined
synthetic as neither analytic nor contradictory and set out to define a
syntactic concept of analyticity so strong that it would yield a proof that
every sentence which contained only the logico-mathematical primitives
of his language was either analytic or contradictory (hence not synthetic).
In his syntactical sense, he succeeded. But since the needed analyticity
had to be far stronger than Godel's provability, Carnap was led to
introduce powerful rules of inference (called consequence rules, or
c-rules) in addition to the more usual rules for derivability (called d-rules
by Carnap.) These were, however, so highly non-effective that Kleene
spoke of them as affording only a "conceptual completeness" to the class
oflogico-mathematical sentences. In fact Kleene (who was not at the time
acquainted with Tarski's work on truth) pointed out a simplification of
Carnap's definition of analytic for Language II which may be seen in
retrospect to parallel Tarski's definition of truth, so that although
Carnap's c-rules referred only to syntactical entities - numerals instead
of numbers - Carnap's definition did not really make 'analytic' more
syntactically determinate than 'true' was. 20
In retrospect, this already suggests an explicit turning to Tarskian
semantics, especially since thetruth of Godel's sentence follows directly
from any semantical truth definition which satisfies Tarski's adequacy
condition (given GOdel's other assumptions), and hence from the seman-
tica/ rules of the language. But such considerations may be postponed
untiI we consider Carnap's semantical phase.
Leaving the detaiIs of Syntax for a moment it may be well to comment
 CARNAP'S LOGICISM 189

briefly on the impact of G6del's results on logicism in a more general

sense. Orie aspect of both logicism and formalism that upset many
mathematical observers was the prospect that, if successful, logicism-
formalism would produce a single system that would turn aH future
mathematical effort into routine spade work. Hilbert clearly had assumed
the in-principle decidability of aU mathematical questions. And while
Frege and Russell tended to speak mostly of post hoc justification of
mathematical discovery by logic, Wittgenstein more boldly indicated the
general implications of logicism by remarks such as "there can never be
surprises in logic". Therefore GOdel's result provided a sense of relief in
such mathematicians, even though it also put certain limits on what
might be proven, even for arithmetical statements.
For philosophers, G6del's result was a final capstone to a dissatisfac-
tion some had already been feeling with the logicists' treatment of the
term 'analytic'. In the narrower sense, 'analytic' meant far more than just
necessarily true in virtue oflanguage rules (of any sort). It meant obviously
necessarily true to any who understood the language. Frege had already
noted this attitude in Kant and commented on its unreasonableness
(p. 6 in Grundlagen). But it persisted, even among supporters oflogicism. 21
Carnap, from very early, had argued against such expectations on the
reasonable grounds that "it is possible to be clear about the rules ...
without at the same time being able to envisage alI their consequences and
connections".22
Other philosophers had extra-mathematical reasons for opposition to
logicism, even granting the broader sense that Frege and Carnap pre-
ferred. Therefore when even this broader sense was shown by G6del's
result to be either insufficient or in need of a further, non-effective,
broadening of Syntax, it seemed to many to spelI the end of logicism.
Without yielding anything to the misapprehensions that underlay
some anti-Iogicist attitudes, it may be condeded that G6del's results did
seem to call for some modification of the logicist thesis. Carnap himself
was not quite content to accept alI that G6del's result implied. Despite the
fact that his Language II was a single system whose rules rendered aH
truths of arithmetic and analysis analytic, the non-effectiveness of the
rules led him to envision a hierarchy of progressively stronger systems
in which the region of the non-effectively analytic was correspondingly
reduced. He was led, in fact, to the remark (p. 222 [4]) that any arith-
190 HERBERT G.BOHNERT

metical sentence which is irresoluble in one system can be resolved in a

stronger, but conceded that "mathematics cannot be exhausted by any
one system; it requires an infinite series of ever richer languages". While
the thought prompting this formulation may have had merit, it was open
to a trivializing reading, as Kleene remarked, in which an irresoluble
sentence S becomes resoluble in a richer language in which S is added as
a new primitive sentence. Carnap's remark, although open to criticism,
and not strictly necessary to the structure of Syntax, does suggest a cer-
tain refinement of the basic logicist claim that mathematics can be re-
duced to logic. In particular, the vision of a single, 'final' system of logic
would be explicitly replaced by an acknowledgement of a possible in-
completability of logic, reflected in a weakened ordering of quantifiers,
Le., changing from "There is at least one system of logic which entails
alI mathematical truths" to "For every system of mathematical truths
there is at least one system of logic under which the mathematical truths
can be given an interpretation which makes them analytic".
Such a wording stiH invites trivialization, of course, by the possibility
of arbitrary, ad hoc liberalizations of what a logic is. Nevertheless, in its
stress on the analyticity of some interpretations of any logic-mathematics
it may suggest one kind of justification for mathematical truth that
warrants further consideration - a possibility I hope to make apparent.
Returning now to the consideration of Syntax, it wiH be recalled that
an analytic sentence was expected not only to be determined by the rules
of language but also to lack factual content. Here again, Godel's results
had an impact that may best be considered before considering factual
content per se. G6del's second important result made the consistency
of a logical system containing number theory unprovable in that system.
Its consistency could be proved only in a system which was stronger and
whose own consistency would then be open to question. Since alI sen-
tences of an inconsistent system would have an abundance of content,
factual or otherwise, a claim of analyticity for any sentences of a given
system clearly depended on the assumption of consistency for that sys-
tem. Carnap gave consistency proofs but admitted their lack of persua-
siveness because of the stronger assumptions he had had to use, in accord
with G6del's results.
It has been suggested that this presence of a small but ineradicable
possibility of inconsistency in the logic of any arithmetic shows an un-
 CARNAP'S LOGICISM 191

certainty in any allegedly analytic sentence, and that this proves that no
formal explication for 'analytic' is possib1e. 23 To this I wou1d rep1y that
since Kant took arithmetic as his prime example of a priori, necessary
truth, it would seem that "as certain as arithmetic" might reasonab1y be
taken as satisfying the demands of Kant's explicandum for certainty. If
not, I could only say that Carnap's analyticity is at least an interesting
approximation to the inexplicable.
Supposing the consistency of arithmetic, then, we may proceed to look
more c10sely at the way Carnap thought ofhis system as being interpreted
so that an axiom of infinity would be without factual content. Several
different approaches, in fact, underlie discussions in different parts of
Syntax. 24
The approach that was first taken is not the one which is easiest to
understand. It appeared in connection with Language I in which the
descriptive predicates were pictured as of first level, taking the numerals
as arguments. The familiar illustration was 'Blue (3)'. The meaning of
this was to be suggested by the paraphrase 'Position 3 is blue'. It called
for a rather special viewpoint to make the discrete, one-dimensional,
but colored, wor1d which was envisaged into a convincing example.
Though of interest, we need not pursue it.
A more vivid discussion is given for Language II onpages 149-151. The
set of primitive, descriptive predicates and functors (function-symbols) of
an illustrative 1anguage of science is pictured as inc1uding terms which
a scientist might ordinarily ascribe to a space-time point, to a region, or
to an object occupying a region, such as 'electric charge density',
'the ec1iptic', or 'p1anet'. In Language II they are ascribed to quadrup1es
of real numbers, or to sets of them, where the real numbers correspond
to physically meaningful space-time coordinates in some reference sys-
tem. Since the basic universe for Language II consists of natural numbers,
as it does in Language 1, the reals are constructed at a higher level (in
one of the familiar type-theoretic constructions). Therefore the type-
theoretic level of the descriptive terms are at appropriately higher levels.
Thus a term such as 'horse' (whether primitive or reductionistically de-
fined) would not be predicated direct1y of a space-time region occupied by
an animal but rather of a set of real quadruples corresponding to the
points ofthat region. For present purposes we can leave aside questions
of just how quadruples of real numbers (or their syntactical representa-
192 HERBERT G.BOHNERT

tions) are made to correspond to points in space-time with the two

refiections that (1) astronomers do fit coordinates to physical points by
reference to individual stars, and that (2) names for extended physical
individuals such as stars may occur in Language II, as if denoting
specific sets of real quadruples. InformalIy, one can see that it is possible,
in principle, that such a language could have a very rich vocabulary con-
taining names of ordinary spatio-temporal things, and predicates appli-
cable to them, alI such terms being at the indicated higher type-theoretic
levels, and that discourse much like that of ordinary discourse could be
carried on, undisturbed by the logical level of the terms used. In par-
ticular, one need not suppose that use of a proper name of a physical
continuant necessarily presupposes any knowledge of even one real-
quadruple of the region-set occupied by the thing named. Such considera-
tions would need to enter only to the extent that questions of time and
place do in ordinary discourse. (Although I have spoken, as Carnap
does, as if everything were referred to one coordinate system, it is easy to
imagine modifications which would permit normal use of relative
reference systems.) N ow the first point of this stage setting is to provide a
visualizable background for the claim for which I now wish to argue:
Whatever physical representation of the universe that is made in a
language whose logico-mathematical structure is that of Language II,
is independent of the cardinality of the Peano numbers (although the
manner of representation would depend upon the physical-descriptive
vocabulary and how that vocabulary was related to the logico-mathe-
matical framework). For example ifthe physical world were found to be
bounded in extent, 'large' quadruples would have only some conven-
tional, zero, or singular value for alI physical magnitudes. If it were found
to be discrete and finite, physical magnitudes would turn out to have
significant values only at the finite, discrete points. Needless to say,
physical discoveries can reveal any descriptive vocabulary to be in-
adequate or inappropriate. But such an eventuality is neither hindered nor
facilitated by the denumerability of the natural numbers. For example,
suppose it were found desirable to abandon a given descriptive vocab-
ulary and to inaugurate a new vocabulary suitable to an entirely different
physical dimensionality. This could require a descriptive vocabulary
located quite differently in the type-theoretic structure. The exact shift
needed would be determined by the cardinalities needed and their loca-
 CARNAP'S LOGICISM 193

tion within the structure of Language II, but the resulting physical pic-
ture would not reflect those logico-mathematical features peculiar to the
language representing it, i.e., Language II. Indeed, the logico-mathe-
matical framework of Language II, with its denumerable basic signs, its
capacity for representing desirable universes of discourse of any di-
mensionality of mathematical structure (not requiring transfinites beyond
the reach of type theory) thus served, in effect, as a sort of blank sketch
pad on which could be drawn whatever representation of factual reality
might be found best at any time.
In the foregoing, 1 have confined myself to Language II (going a bit
beyond the wording of Syntax at times but, except for the last remark,
folIowing Carnap's explanations as given in answer to questions).
However, the basic point can be made for other languages. For example,
an essentialIy similar argument can be given for a language of science
couched in the framework of set theory. A common form of infinity axiom
of set theory postulates the existence of a set con si sting of sets which are
built up solely from the null set by iterated membership, in a way which
ensures denumerability. Now, suppose that in order to formulate this
language of science, a set of physical individuals is postulated as a special
set of 'urelements' to which the descriptive vocabulary is primarily to
apply. It needs little argument to see that the existence of the null set
and its infinite train will have no necessary implications for the cardinali-
ty of the set of physical individuals. Nor would there be any converse
implication. While there might be an inadequate supply of physical things
to write enough physical inscriptions, nothing in this language, or in
Language II, need imply the existence of an inscribed numeral for every
number. Indeed, one of the admirable aspects of Syntax is the way in
which Carnap exploits G6del's arithmetization to separate alI questions
of syntax from questions about inscriptions. It would seem then that such
infinity axiom is not disconfirmable by physica1 fact, let alone con-
firmable.
If true, it would quite evidently have to be true in alI possible physical
worlds and be without factual content for any. A platonist might argue
that it asserts a synthetic truth about a non-physical but 'welI-determined'
reality; and a nominalist might reply that it asserts a synthetic falsehood
about that reality. But for Carnap such an issue was necessarily moot and
hence pointless. If it were to be regarded as true at alI it could only be on
194 HERBERT G.BOHNERT

the basis of a convention so to construe the meaning of the epsilon of

membership as to be consistent with its truth. Indeed, the existence
of the null set, to begin with, would appear to be assertible only on such
a ground.
rhi~ train of thought on set theory does not explicitly appear in
Syntax, of course, but it may be taken, I think, as indicative of a central
part of Carnap's willingness to regard some mathematical postulation
of infinity as analytic. Though not conceding analyticity, set-theoretic
platonists act like logicists, in choosing to erect their infinity on the null
set; i.e., they also appear to have striven to render their axiom free ofim-
plication for physical reality and immune to empirical revision. To that
extent, at least, they appear to agree with Carnap as to what is desira bie
in such an axiom.
While the above argument for immunity from factual disproofmay be
regarded as the reason for Carnap's accepting the permissibility of
making an infinity axiom analytic, it gives no positive reason for doing so.
It leaves it as a matter of decision, and hence as a question of value. Its
value for Carnap at this point was c1early not that of substantiating a
c1aim to reduce mathematics to logic, since the two were merged. A more
obvious value c1aim might appear to be found in its value to mathe-
matics, and to the description of the physical world. Without denying
such value it may be noted that to base a decision as to the analyticity of
infinity on it would pose a question: i.e., it would be tempting to suppose
that utility in description of the physical world might reflect some syn-
thetic truth about physical reality. It is significant, accordingly, that
Carnap specifically cites, instead, its utility for general syntax. That is,
he points to its utility in specifying the syntax of whatever language may be
found to have utility in description of the world. It is in keeping with this
viewpoint, that the type-theoretic special syntax for both Language I
and II takes Peano number theory, or its syntactical form, as accounting
for the total 'population' of the (bottom level) universe of 'individuals'
for both languages, thus immediately fitting them, following G6del, for
the specification of arithmetized syntax. This feature is immediately em-
ployed, in fact, in providing the arithmetized syntax (formational and
transformational) of each language in itself(except for the stronger c-rules
which still require an external, though not dissimilar, metalanguage).
Such languages, it is important to note, need not be forced to undergo
 CARNAP'S LOGICISM 195

the 'conceptual revolution' of a change of universe of discourse by a

change in empirical scientific theory. There would stiH be revolutions in
descriptive vocabulary, but the logico-mathematical framework for a
new vocabulary could stiH be given within the old language, which could,
indeed, act as a general syntactic metalanguage for alI subsequent
languages of science, with or without similarly Peanoesque structures.
It may be conceded that a set-theoretic frame might have provided an
increase of generality, especially in the light of the later acknowledged
values of adding semantics to metatheory, but Carnap's motivation at this
point and its rational appeal may at least be glimpsed. (It may be of some
interest to note that the Peano-based structures in Syntax would appear
to be a uniquely satisfying way of acquiescing in Benacerrafs protests
against the superfiuous 'truths' which accompany constructions of the
natural numbers in set theory or in type theory while continuing to ac-
count for their application by giving them systematic location within a
10gicalIy unified language of empirical science.)
So far, we have dealt with the question of interepreting the languages
of Syntax only to the extent of saying that interpretation was intended,
although any actual semantics was excluded. Now we may look first, and
briefiy, at the syntactic surrogate Carnap offered in place of semantics,
and then, more closely, at the way he actualIy conceived of his languages
attaining interpretation. These later considerations may provide some
insight into why semantics was originalIy excluded from Syntax, and later
into his underlying attitude toward semantics itself.
The syntactic parallel to semantics is given in section 62 of Syntax,
headed 'The Interpretation of a Language', pp. 227-233. The inter-
pretation of a language was simply to consist in its translation into an-
other by use of a syntactical metalanguage common to both. This was
later dropped in favor of a Tarskian translational semantics, with which,
indeed, the Syntax account already had features in common. One such
features was a regress of metalanguages and an unavoidable relativity to
what Quine was later to call a 'background language' at some point. 25
The key to the deeper conception mentioned above occurs only at one
point in this section, though it occurs by implication at other places.
Near the outset of section 62 there is mention oflearning to understand a
language by the "purely practical method which is employed in the case
of quite small children and at the Berlitz School of Languages". Though
196 HERBERT G.BOHNERT

mentioned only to be dismissed, such methods loomed large in Carnap's

thought. In classroom discussion, when pursued by questions as to how
the 'ultimate' metalanguage carne to be understood, Carnap would ex-
pand at length on how languages carne to be learned by children - first by
pointings, then by context, correctings, and by increasing use of informal
metalinguistic talk (in a manner not unlike Quinean regimentation).
On a few occasions, this was extended into a revealing discussion of how
Language II, with its apparently non-intuitive number-based syntax,
would be learned by a child in a society which spoke such a language. As
remarked earlier, such a language could have a fuIl descriptive vocab-
ulary at an appropriate type-Ievel, and a child would learn the language by
the usual pointings, etc., without references to the 'grammar' of type-
levels just as English is learned without explicit reference to the grammar
of English. In school, counting, the use of arithmetic, etc., would also be
acquired as at present. The elementary grammar of descriptive lan-
guage, also, could be given with little reference to type-Ievels. At a more
sophisticated level, learners inquiring into the nature of numbers would
be told of Frege and Russell or of set-theoretic treatments of number.
At a more advanced level yet, the full type-theoretic, Peano-based
structure of their native language (II) would be given the learner, along
with its various advantages (its permanence under empirical change, its
use as a permanent general syntax language, its unified, factuaIly empty
basis for arithmetic and analysis, etc.). At this point, the learner (the
class urged) might demand to know what the basic Peano numbers were,
since they were not the Frege-RusseIl, or other finite cardinals. Here, the
'internal' answer would be: the numbers. That answer would be re-
peated, furthermore, in' any of the metalanguages, syntactical or trans-
lation-semantical, to which explicit appeal could be made. When the
learner (and the class) persisted, and was not satisfied by acknowledging
the lack of clear intelligibility of 'external questions', Carnap first took a
mock-reverent platonistic stance about the eternal, ideal status of the
numbers, though emphasizing the empirical meaninglessness of such
an attitude. When this too was rejected, Carnap remarked that if the
learner were very sophisticated, it would be suggested that they were
ideal entities in the sense in which the point at infinity in projective
geometry is an ideal element, i.e., as a useful fiction, though more
provably non-committal. This, too, it was insisted, was only an informal,
 CARNAP'S LOGICISM 197

not quite 'cognitive' response to a not fulIy intelligible 'externa1' question,

just as was the earlier p1atonistic stance. Neverthe1ess, 1ater deve1op-
ments suggest the "pragmatic fiction" rep1y to be c10ser to Carnap's
operant attitude.
The foregoing discussion revea1s, among other things, the attitude
which Carnap took toward meaning in his syntactic phase. Meaning was
(as it stiH is primari1y, for Quine) a matter ofthe psycho1ogy ofthe users of
language. Far from sharing Wittgenstein's early picture theory of lan-
guage, Carnap saw the syntactical structure of a language as only slowly
and imperfectly acquiring (psychological) meaning for its users. In
particular, the psychological growth of understanding involved no
direct, instinctive grasp of the logical syntax of the language, but
required 'regimentation', as Quine has described. Discussing the process
at alI could only be done with impressionistic language and metaphor, so
that it was natural for Carnap to speak ofmeaning "seeping" from some
terms to others. 26
Furthermore, since no amount of regimentation could, even in
principle, eliminate alI possible misunderstanding (even possibly gross
ones) because of the relativity to the regress of metalanguages, neces-
sarily terminating somewhere, it seemed to involve an unrealistic,
"metaphysical" omniscience to speak, semanticalIy, of some language-
independent reality to which the signs of a syntactically specified lan-
guage referred. 27 This, then, was the methodological outlook that
prompted the focus on syntax as the only usefully formalizable aspect of
language. 28
Summarizing the discussion of the syntactical phase, then, we have
seen something of its 'pure syntax' motivation, the reason for the
Peano-based form of language used, the way it could be thought of as
acquiring meaning, the way in which the infinity axiom could be regarded
as content1ess (and in that sense open to interpretation as analytic,
whether or not the Juli scope of 'analytic' was proof-theoretically
reachable) and how the attempt so to interpret the infinity axiom was
basically tied to its value in giving the syntax of languages of factual
science. Thus despite alI appearances, the motivation, and even (in a
transformed way) the results, stiH were largely logicist in character.
198 HERBERT G.BOHNERT

(3) The Semantical Phase

While the ideas to be developed in this phase had long been taking form,
in readings of Frege and Wittgenstein, and in discussions in the Vienna
Cirele and with GOdel, real work in this area was initiated by conversa-
tions with Tarski and by the study of Tarski's historic [23]. AIthough
Tarski's work is now valued by logicians less for its method of inter-
preting and defining truth for a given language than for its opening up of
new fields to metatheoretic investigations by making interpretations of
syntactical systems into elearcut mathematical objects to be quantified
over, its first importance for Camap was in the first mentioned ac-
complishment. Especially important to Camap was the fact that Tarski's
procedure could provide a definition of truth for a language of empirical
science, with respect to its intended interpretation as specified in a
semantic metalanguage, provided that the metalanguage had adequate
descriptive power. Thus Tarski's method bid fair to be, as Tarski elaimed,
a logical formulation ofthe ancient 'correspondence theory oftruth', and,
for Carnap it opened the way to a more straightforward treatment of
factual content (and lack of it) and to supplying the semantic dimension
commonly felt for analyticity. By enlarging the sense of the phrase
"true in virtue of the rules of language" to include "semantical rules",
it seemed to permit construing analyticity as "true in virtue of meaning"
in a more literal way.
The new approach meant treating meanings in other than an ex-
elusively psychological sense, but fears of metaphysics were allayed by
Tarski's cogent arguments to the effect that in his straightforward, corre-
spondence treatment of semantics, to speak of what a sentence said or
what a word stood for was no more metaphysical than using the object
language itself. Camap was sufficiently reassured, in fact, that in his
own semantical formulations he chose to speak of designat ion rules such
as "'blau" designates blue', though the effect was almost equivalent to
Tarski's use of satisfaction. Carnap did not abandon psychological
meanings in favor of the objective designata of semantics, of course. In
discussing how a language with a specified semantics was leamed, he
simply treated the leamer's gropings as a gradualleaming ofthe designa-
ta, rather than of meanings in a less formal sense.
Camap's early work in this phase [5], [6] shows his development of
semantics in keeping with the straightforward motivations indicated.
 CARNAP'S LOGICISM 199

But the development of the more general, model-theoretic aspects of

semantics soon gave rise to new and difficult problems for analyticity, as
opposed to a narrower concept of logical truth. The distinction between
the two concepts was sharpened by the fact that the conceptual scheme of
semantics seemed peculiarly adapted to the earlier logicists' conception
oflogic as being concerned only with the forms of sentences as framed in
syncategorematic signs. Indeed, in terms of the categorematic-syn-
categorematic distinction, we might say that under one common semantic
treatment a sentence was a logical truth just in case it was true under
any appropriate assignment of extensions (roughly speaking) to the
categorematic signs. According to this conception the Peano axioms of
Languages I and II were obviously not logical truths, and their claim to
analyticity was given no support from any general conception of seman-
tics. Even definitions, when expressed as universal biconditionals in an
object language, were'challenged by Quine, though such sentences had
been taken as paradigms of transparent analyticity, even by non-
logicists. 29 Though the problems posed for definitions might be disposed
of, it was soon to be eclipsed by a more serious one when Carnap decided
to include meaning postulates as analytic. 30
To better understand Carnap's attitudes in the matter, it will be well
to recall some general aspects of his out1ook and methods during this
period.
To begin with, in order to develop his new semantical ideas sys-
tematically he tended to focus on simple languages. And because of the
way in which semantics lent itself to the syncategorematic view of logic,
he tended to revert to the older view himse1f. In fact it may be relevant
to recall that in his classes in the early forties, his case for logicism gave
special prominence to the fact that first-order-Iogic-with-identity could,
in effect, ascribe cardinal numbers to finite classes (described in such a
language) by means of numerical quantifiers. 31 To be sure, he still taught
type-theoretic logic, and presently provided it with a semantics, by use
of an informal, set-theoretic metalanguage [7]. But the outlook was
closer to the classical than to the syntactical phase. Indeed, there was a
gradual return to an intensional outlook not unlike that of Frege and
Russell. This last aspect is of sufficient importance for what will follow
that it calls for somewhat closer attention.
In part, the return to intensionalism was due to Carnap's studies in
200 HERBERT G.BOHNERT

modal logic [8]. But it seems to have been due also to an aspect of
Tarskian semantics which many lost sight of in their concern with
model-theoretic inquiries. 32 An interpretation of a set of predicates is
commonly thought of simply as an appropriate assignment of extensions
to the predicates; and truth is defined relative to that assignment. Even
when truth is defined for an empiricallanguage of science with respect to
its intended interpretation, the intended interpretation (in a Tarskian
semantics) is commonly viewed as consistingjust of one such assignment.
Quine has called Tarskian semantics a theory of reference rather than of
meaning largely for this reason (130-138 [19]). Although there is nothing
intensional in the truth definition itself, Carnap has emphasized that in
defining truth for an empiricallanguage, the extensions are most naturally
assigned by use of descriptive words such as 'cow', the extension ofwhich
is a matter of empirical fact. The interpretation and the resulting sense of
'true' itself could vary significantly if the terms used in the metalanguage
to assign the extensions were replaced by different but coextensive words.
Tarski's original specification that the semantic metalanguage contain
"a translation of every expression of the language studied" (e.g. p. 172
[23]) is left informal, but it seems at least open to the reading Carnap put
upon it, namely that the translation should, in some sense, be the same in
meaning. Carnap may have originally understood this meaning in the
same informal, psychological sense which he had in mind in Syntax.
But Tarski's arguments for the non-metaphysical safety of semantics in
general, may have extended, in Carnap's mind, to permitting talk of the
objectivity of the property Blue mentioned in "'blau" designates Blue'.
In any case, the difference between the two approaches to semantics led
Carnap to propose a distinct bifurcation in terminology - distinguishing
a model from an interpretation. 33 With this brief recollection of some
Carnapian attitudes during the semantic phase, we may return to the
questions posed by semantics, in its 'syncategorematic' aspect, for
analyticity as opposed to 'strict1y logical' truth.
Before considering the problem ofmeaning postulates more generally,
we may consider the easier case of definitions, viewed as a special kind
of meaning postulate. Although Quine's arguments on the matter go
beyond the merely model-theoretic aspect mentioned, it is well to recall
precisely why this aspect seemed to Carnap to pose no problem for
analyticity, just as it had not for presemantics logicists. As noted earlier,
 CARNAP'S LOGICISM 201

a definition expressed as a universally quantified biconditiona1 of a certain

sort, is not true under all appropriate assignments of extensions to its
categorematic signs. But a theory is common1y thQught of as forma1ized in
a language with a fixed vocabu1ary of primitive terms, and given a seman-
tics by fixing ranges from its sorts of variab1es and giving interpretations
for its primitive terms, (the interpretations, and some of the ranges,
typically being thought of, by Carnap, as being given by descriptive
terms of the metalanguage, when the theory is one of empirical science).
A sentence of the theory in primitive terms is logically true just in case
it is universally valid. But to require universal validity of a sentence con-
taining defined terms, to be logically true, is to treat the defined terms as
primitives in an enlarged language whose semantics is no longer com-
plete and for which truth is no longer defined. New, but indeterminate,
possibilities appear to be contemplated which the original theory has no
chance of explaining. The fact that from a deductive standpoint defini-
tions do act like axioms added to the theory furthers the impression that
definitions must act semantically in the same way (or that no reason
can be given why they should not act the same way). But a well-known
ploy to avoid this loss of coherence, often emphasized by Church, is
to regard alI definitions as extra-systematic abbreviations for 'genuine'
expressions of the theory. This has the merit not only of not adding to the
extensional possibilities contemplated by the theory, but of preserving
the 'meanings' of the original descriptive terms of the metalanguage -
whether those meanings are thought of as objective intensions or thought
of merely in an informal, psychologica1 sense.
Although unwie1dy, its possibility in principle would seem to render
unobjectionable the rule that a sentence of the theory is 10gicalIy true if
and only if it is true in every domain under every appropriate assignment
of extensions to its primitive predicates, where for each such assignment
to primitives, the assignments to defined predicates are exact1y those
determined by the definitions, for that assignment to the primitives. It
is, after alI the assignment of differing extensions to words with fixed
intensions that permits such assignments to be regarded as yielding
characterizations of other 'possible wor1ds' rather than merely as varying
(extensional) interpretations of a syntacticallanguage form. Such assign-
ments of extensions to defined terms would permit the only really in-
telligible extension of the concept of other possible wor1ds, to sentences
202 HERBERT G.BOHNERT

containing defined terms, in that it permits the meanings of such defined

terms to be thought of as fixed by the definitions in a way in which un-
controlled assignments to defined terms would not. To amplify the use
of the suggestive metaphor of logical space which Carnap took over from
Wittgenstein, 1 would emphasize that it is only with respect to its intended
interpretat ion that a language defines the logical space of possible worlds
describable by that language. Le., for an interpreted language, analytic
truths (or logical truths in the broader sense) should be those which are
true no matter what is the case, in so far as what is the case is describable
in that language. If now, we define logical and/or analytic truth for a
given language as truth under any assignment of extensions in which the
(object language) definitions are true, we are not limiting the 'genuine'
logical space of that language, despite the fact that from the purely
syncategorematic framework of general semantics it would appear that
way.
Turning now to the problem of non-definitional meaning postulates,
we see that although the ploy of keeping definitions extra-systematic is
no longer available, the technique of requiring of a logical-analytic truth
only that it be true under alI appropriate assigmnents of extensions which
make a given set of sentences true is still available. It is appealing on
grounds very similar to those mentioned. But it raises difficult questions
as to when the limitation of apparent logical space is merely appropriate
to meaning and when it forecloses genuine possibiIity.
Much useful work has been done on the assumption that the borderline
can be drawn, and many persuasive arguments have been offered to the
effect that it could be. 34 But for purposes of discussing Carnap's logicism
we need not argue that the issue has been resolved in complete generality
(for alI types of meaning postulates and alI sorts of vocabularies), but
consider only the question of meaning postulates laid on certain, ap-
parently categorematic, signs which Carnap would wish to reckon as
belonging to the non-descriptive, i.e., logico-mathematical, part of a
language of science.
Taking least controversial cases first, we may note that identity has the
form of a dyadic predicate but that it is treated as part of the logic of
theories in 'standard form'. A special clause in the recursive definition
of universal validity appropriate to such theories has the effect of
granting universal validity to alI sentences true in alI interpretations in
 CARNAP'S LOGICISM 203

which '= ' is interpreted as identity. If the logic of identity used does so by
imposing first order axioms on identity, such axioms do not 'implicitly
define' identity (without the special identity-semantics), as shown by the
possibility of non-normal interpretations. Hence, they are meaning pos-
tulates. If the 'logic of identity' is deductively treated by way of rules of
inference, the semantic validity of inferences deductively made must
still be ensured by the identity semantics mentioned. 35 Identity was
explicitly defined in type-theoretic logic, of course, and so seemed
acceptably logical under the syncategorematic view, as maintained in
that context. But Quine has made a well-known case that type-theory is
set theory in disguise. Under that view, identity is defined in terms of
another categorical dyadic relation, epsilon, which Quine has argued
belongs to non-analytic mathematics. Nevertheless, Quine, and just
about everyone else, is willing to regard 'real' identity as logical (despite
the value of treating it simply as axiomatized, for many foundational
purposes).
A next, fairly transparent, case is that of elementary class inclusion. By
'elementary' here 1 mean applied only to classes of individuals (i.e. as
Quine's virtual classes [20J or as in Carnap's compound predicates [7J).
Today we would regard elementary inclusion as logical because it is con-
textually eliminable as short-hand for sentences of quantificationallogic.
But recall that class-algebra antedated quantification theory. Was it
improper to call inclusion and class algebra logical until (more syn-
categorematic) quantification theory appeared? To be sure, the inclusion
sign and Boolean algebra have many non-Iogical interpretations and the
logical one was not completely evident. (Even Schroder was criticized by
Frege for interpreting inclusion too much like part.) The point is that
among all possible interpretations there was one which was 'purely
logical'.
With this thought in mind, 1 turn to the question as to whether there
might not be a 'purely logical' interpretation of epsilon. Quine has argued
that type-theory's use of predication confuses the logical notion of
predication, expressed by ordered juxtaposition or by a syncategorical
copula, with the dyadic membership relation. The charge is especially
striking if we limit ourselves to a monadic, extensionally construed type
theory. One could, however, pursue the opposite view, namely that
dyadic epsilon is an improperly conceived abstraction from the notion
204 HERBERT G.BOHNERT

of predication. To begin with, epsilon has a certain naturalness when

one wishes to disregard the meaning of what is being predicated and
study the logic of the purely extensional aspect of predication. It soon is
seen to have the added attractiveness that it permits study of hypo-
thetical predications that no finite or denumerable vocabulary could
make, but for which the 'logic' seems the same. Soon, however, it comes
upon limitations inherent in the fact that what it is extrapolating from is
not a relation in the categorematic sense. Not only can it not replace
predication all together, on threat of an immediate regress, as
Reichenbach has ably pointed out (pp. 320-1 [22]), but (meaning?)
postulates must be laid upon it to keep the faulty analogy from leading
to absurdities that would be ruled out by the logical grammar of predica-
tion. (Carnap once expressed his fear of the possibility of metaphysics
arising in semantics not as fear of talk e.g., of propositions per se, but
of reification of propositions, by which he meant predicating things
of them that could meaningfully be predicated only of things, i.e., the sort
ofmetaphysics he welcomed the theory oftypes ruling out. (p. 71 [10]).)
The suggestion that 1 am making is that one might reconceive epsilon as
syncategorematic, though stiU as a purely extensional abstraction from
the original predication, and distinguished from it, syntactically and
semantically. Some form of set theory (not under the same constraints
predication, let alone those of type theory) might then be reconstructed
as consisting of the (syncategorematically) purelY logical truths of this
new 'set-predication'.36
But now suppose that no syncategorematic reconstrual of epsilon is
ever found, must it then be non-Iogical? Suppose, indeed, that, unlike
elementary c1ass inc1usion, no such reconstrual can be found. Might we
not stiH impose meaning postulates that we regard as logically true?37
While my final argument that we can, and should, may wait for Part II, 1
would like to mention here an interesting argument which he earlier
gave me, as to the analyticity of almost any set theory. 1 am not sure how
seriously Carnap intended it. It did not arise in the 1968 conversations,
but 1 stiU think it provides an interesting backdrop for them.
Carnap presented his case as being one which would make the analy-
ticity of set theory c1ear to an extreme platonist. Godel's platonism has
often been said to be informally epitomized by his reference to "some
well-determined reality, in which Cantor's (continuum) conjecture must
 CARNAP'S LOGICISM 205

be either true or false".38 But Carnap's platonism, was in a sense more

extreme than Godel. Abstract entities of every consistently imaginable
(or unimaginable) sort existed, and every cpnsistent set ofaxioms had
many interpretations in the realm of abstract entities. The question was
not one of truth but of singling out. If one theoriest wants to com-
municate to another theorist what he means by the words and sentences
of his theory he must often say that he means a given term in a way in
which it would be true to say so and so, (i.e., lay down a meaning pos-
tulate). Even for descriptive terms, pointings would not always suffice
but would also have to be supplemented by meaning postulates. For
example, 'between' may be roughly understood ostensively. But Tarski's
sense of 'between' (which includes end points so that b can be between b
and c) and Veblen's sense (which exc1udes end points) could only be dis-
tinguished by postulates. ('Between' is primitive in both systems.) Yet
the different postulates do not appear to disagree as to fact,but as to
meaning. Each postulate makes a selection from possible meanings. 39
In the case of the continuum hypothesis, there was not in Carnap's
argument, just one well-determined reality, but many. Any set theory
which left the question open simply did not completely single out which
of the many c10sely related concepts of set, c1ass, or epsilon was meant.
By the time enough conventional meaning postulates were laid on the
terms to single out their one intended interpretation in the realm of
abstract entities, anything that could be said about the selected subrealm
would be analytic, in virtue of the meaning postulates. Even before
enough was said, in fact, whatever postulates were given would have
to be regarded as meaning-narrowing, i.e., as analytic and not as factual.

(4) The Theoretical Phase

In discussing the theoretical phase of development in Carnap's logicism,
I wish merely to p'oint out ingredients, so to speak, and leave the final
outlook to emerge in Part II. I shall speak first of general attitudes, then
of certain developments that spurred their development,and then of
their bearing On certain issues and their expression in Carnap's publica-
tions.
To begin with, Carnap always had a certain urgency toward decision
and progress in the use of language. Perhaps the best way of conveying
what 1 mean by this is by reference to a vivid word-picture of Quine's.40
206 HERBERT G.BOHNERT

When 1 was a child 1 pictured our language as settled and passed down by a board of
syndics, seated in grave convention along a table in the style of Rembrandt. The picture
remained for a while undisturbed by the question ofwhat language the syndics might have
used in their deliberations, or by the dread of vicious regress.
For Quine the picture referred to an illusionary past. For Carnap such
a scene would refer, impressionistically speaking, to an urgently needed
near future. It is high time, we seem to hear him urging, after centuries
of naturally evolved ambiguity, for men to sit down together and (using
naturally evolved language as well as they can) agree on clearcut rules.
Only with respect to such rules can man hope to become clear as to what
is substantial and what is illusory in argumentation. For Carnap, the
language-reconstructionist, it was perfectly natural to speak of a sen-
tence as being analytic according to meanings just now assigned or con-
templated - more so, indeed, than to speak of its being analytic with
respect to some more naturally accrued, but vaguer, meaning.
Secondly, Carnap had a significantly ambivalent attitude toward the
powers of man's intuition, both in general and as applied to matters of
logic and mathematics. On the ·one hand, he would base conjectures on
his own intuition as to what was or was not analytic when there was
nothing else to go on, but he was deeply skeptical of the complete reliabil-
ity of alI such intuition, except as a starting point and guide. On the other
hand, he had great hope for man's ability to use formalism to construct
an instrument whereby he could double-check intuitions by making
them explicit and reducing the size of the steps of inference involved.
Indeed, he saw in formalism an instrument capable of leading man over
intellectual chasms where his intuition seemed to fail altogether. Carnap's
generation, after all, had been made acutely aware of the powers of
mathematics to handle geometries of space-time which defied common
intuition. These, then, were among the general attitudes which under-
lay his early stress on syntax and his doubts of semantics and meta-
physics, and which were never quite absent even during the most en-
thusiastic part of his semantic phase.
These attitudes were reinforced, in the direction of what 1 am calling
the theoretical phase, by certain developments in his approach to ques-
tions concerning empirical knowledge. These need to be mentioned here
because his logicist philosophy of mathematics, like Quine's non-
logicist one, held that a full interpretation of mathematics can only be
given within the context of empirical science, and the epistemological
 CARNAP'S LOGICISM 207

questions bore on the logical framework of that context. A crucial point

was the transition from conceiving of a formalized empirical science
as one having, in principle, only observational descriptive primitives to
one containing also non-observational 'theoretical' primitives. 41 With-
out going deeply into the matter, we can immediately note certain
consequences for Carnap's attitude toward Tarskian correspondence
semantics. Previously Carnap had been able to envision a close
correlation (possible in principle) between the objective designata as-
signed by his semantics to words and their meaning as understood in a
more informal psychological sense. As remarked earlier, he pictured a
learner of a semantically specified language groping to bring his
psychological understanding of a word into accord with its objective
designatum With respect to theoretical terms he thought for a while
of such a learner being guided by reduction sentences (meaning postulates
in which theoretical terms were linked to observational ones), but there
seemed nothing psychological or determinate to be guided toward. In
attempting to give a semantic definition of truth for such a language, one
was faced with unsatisfying alternatives.
On the one hand, one could give a straightforward correspondence
truth definition by using a metalanguage which could translate the
theoretical terms into theoretical terms of its own. But then Tarski's
original reassurance to Carnap (that to say a sentence was true was no
more metaphysical than to assert the sentence itself) was less re-
assuring. Or one could regard the language as a whole as incompletely
interpreted, with truth as a primitive term of the metalanguage. In the
latter case, Carnap found himself driven back to speaking of meaning
seeping from obserV'ational to the theoretical terms, in the older, psychol-
ogistic way. Without going further into the matter here, I need merely
mention that it not only revived and reinforced his earlier reservations
about semantics, but it inaugurated a more definitely 'theoretical' or
hypothetical attitude toward it. Semantics stiU appeared of great value
in explicating what we meant, or appeared to mean, by such words as
'designate', 'analytic', as it also was for concepts of information content,
modalities, and logical probability. But a meta-statement about the truth
of an object-Ianguage statement which contained theoretical terms as-
sumed something of the same theoretical character as the statement itself.
There had long been, indeed, a recognition of a necessary relativity to
208 HERBER T G.BOHNER T

language of many conceps inc1uding analyticity, as earlier remarked.

But he finally made his point most explicitly in connection with seman-
tics - in making his distinction between internal questions and external
questions and extending it so as to hold also for semantic metalanguages
as well as object languages [9]. But the basis for the attitude is already
c1ear in Wittgenstein's stress on what could not be spoken of, with com-
plete intelligibility, about language. Although Carnap regarded Syntax
as a refutation ofWittgenstein's extreme c1aim that a language could not
meaningfully talk about itself at all, much of Wittgenstein's attitude
remained - first with respect to all semantics, and then with respect to
some ultimate semantical metalanguage. One of Quine's complaints
against Carnap's separate definitions of analyticity for separate lan-
guages was their lack of providing a general concept of analyticity for all
languages. More recently, Quine's inquiries into linguistic relativity have
come to grant a certain ineffable status to logical truth, if not to analy-
ticity, that is more harmonious in spirit with the Wittgenstein-Carnap
acceptance of relativity to language (though neither of the latter would
have seriously spoken of it as ontological relativity).
Finally, before leaving the theoretical phase per se, it will be well
to consider one of Carnap's last, and one of his fullest, discussions of the
bearing of the theoretical term problem on analyticity [11]. What 1 wish
to remark on here is not his specific proposal with respect to analy-
ticity - which has to do with the Ramsey sentence conception - but rather
on the fact that it is formulated in connection with a language of the form
of Language II of Syntax - and with much ofthe same syntactical motiva~
tion. This language was conceived of as having a semantics, though now
of a more theoretical character.
A particularly noteworthy aspect of the Language II form, in the light
of developments in the semantic phase, is the additional dimension
which could be added to the c1aim that the Peano numerals, as individual
constants, are 'purely logica!'. 1 have already remarked on their lack of
factual consequence in both a syntactical and common-sense-set-
theoretical way. But there is another way which may be of interest to
investigators of 'possible world' semantics as well as workers in mathe-
matical foundations. A question which has arisen in connection with
possible-world semantics concerns the question of the identification of
individuals in different possible wOrlds in which they may have radically
 CARNAP'S LOGICISM 209

different properties. One answer (which 1 can not impute to any specific
ut1erance of Carnap's but which seems implicit in some of them) is that
in any of Carnap's coordinate languages, such as Language II, the coor-
dinates are not to be thought of as correlated with any fixed set of
individuals, external to language, which retain their identities in other
possible worlds, but rather as an unchanging, logically provided back-
drop, framework, or 'sketch pad' on which various possible worlds may
be represented (somewhat in the way that many configurations of facts
may be represented in a computer by putting different bit-patterns in the
'fixed-individual' registers). This conception is in harmony, with a num-
ber of utterances in the Tractatus when interpreted in the way Carnap may
have. In these passages there is much emphasis on the ability oflanguage
to represent only the form of reality (a view having considerable em-
phasis on Carnap's early extensionalism) and that the 'objects' spoken of
in a language are simply adjuncts to the description of that form. 42
U nder this conception, individual constants were hence to be reckoned
as syncategorematic, in a generalized sense, and hence as providing
simply part ofthe 10gica1 framework oflanguage. Le., they, and variables
ranging over them, were purely 10gica1 signs. Whether there were to be a
finite, denumerable, or non-denumerable number of them, was a matter
for decision on non-factual grounds.
To summarize the overall tendency of the theoretical phase, impres-
• put it that Carnap's original trust in the poten-
sionistically, we might
tialities of formal construction was prompted to reassert itself, and
to take precedence over reliance on semantics, despite its explicative
potentialities (even with respect to analyticity). Indeed, in a tacit, opera-
tional sense, semantics became a formalized adjunct to the vast syn-
tactical network he had envisioned in Syntax. The language learner carne,
psychologically, to understand the formally prescribed semanticallocu-
tions in the same way he was pictured in the syntactical phase as coming
to understand the descriptively used object language and its syntactical
metalanguage. 43 But the syntactical power of the Peano-based 'mother
tongue' that enabled it to provide the syntax of new descriptive sub-
languages, also enabled it to do the same for semantical metalanguages
containing new theoretica1 descriptive terms, ar, indeed, ta formulate
semantical metalanguages according ta quite different, new semantical
conceptions. Le., syntactical structure returned, at least partially, ta
210 HERBERT G.BOHNERT

providing the underlying spine of the claim of analyticity at least for

mathematical truth.
II. OUTLOOK

In my final 1968 conversations with Carnap, I wanted to know not only

whether he stiU held in general to a logicist viewpoint, which he quickly
assured me he did, but to know, in particular, how he regarded Quine's
charges that type-theoretic logic was not logic but "set-theory in dis-
guise" and, how, as a logicist, he could assimilate the many-faceted
developments in set-theory, especially of the bifurcation into Cantorian
and non-Cantorian forms of set-theory. I pointed to the overall impres-
sion of repeated retreats by logicism in the successive views of mathe-
matics from (1) that ofbeing deducible from basic logicallaws and defini-
tions, to (2) having a (non-definite and questionable) syntactic deter-
minability, to (3) having a semantic, but not syntactic determinateness, to
(4) being not fully semantically determinate but true by meaning-
postulate. And I pointed to the surprises that seemed in store for those
who would impose meaning postulates on the epsilon of membership.
Carnap replied that he had early been attracted to set theory because
ofhis extensionalist viewpointand had thought ofusing it before adopting
the type-theoretic approach of Syntax, (as remarked p. 33 [21]). Re stiH
thought set theory could be given an analytic interpretation. But as a
basis for logic and mathematics as applied to the empirical sciences he
remained attracted by something like type-theory's classification ofmany
expressions as non-sentences rather than as odd sentences which were
ruled false via an accumulating mass of set-theoretical conventions.
Furthermore, set theory no longer seemed adequate as a complete
logico-mathematical setting for a generallanguage of science in view of
the increasing importance he gave to the value of non-extensionallan-
guages in science. Type-theory in its extensional form couldbe interpreted
set-theoretically, as Quine claimed. But it need not be thought ofthat way.
With respect to set-theory itself, however, he took the view that any
theory with a predicate-constant could be thought of as true or false in
accordance with differing empirical and mathematical interpretations.
But in the way in which he understood epsilon (which he took to be the
same as that in which mathematicians do) the most familiar set theoretic
axioms seemed analytic, in the informal sense of true-in-virtue-of-
 CARNAP'S LOGICISM 211

meaning, aside from details which could be regarded as minor points of

convention. Re said that he regarded set and member as simple concepts
and felt that after a few initial misconceptions he, as many others, could.
c1aim he understood them as c1early as humans can understand any con-
cepts. But he stressed that 'understood' must not be taken to mean some-
thing like 'able to visualize alI instances of - a conception already
abandoned for large finite collections, let alone infinite ones. Nor should
'understood' be taken to imply a capacity to foresee alI implications of
proposed principles involving the understood term, or to be sure of their
mutual consistency. Though simple in themselves, the concepts of set
and member were of such vast application that we must check on how we
suppose them to apply in previously uncontemplated areas by con-
structing and studying many formal systems. With respect to the bi-
furcation between Cantorian and non-Cantorian set-theories, he said
that one could only wait and watch developments, with respect to what
could be thought of as analyticalIy true in virtue of the understanding that
we had of set and membership. With respect to the construction of a
formallzed language of empirical science, however, we need not be that
hesitant. First of alI, no such language can hope, in any case, to encom-
pass the incompletability of set theory (even though he would tend to view
the rightness of developments in it, informalIy, as analyticity in accord
with the 'simple' meanings of set and member, rather than as truth in
some empirical or some metaphysical sense). But for a formalized lan-
guage of empirical science we should have no trouble in providing it with
a logico-mathematical structure adequate to its needs, and of whose
analyticity (and consequent fact-neutrality) we could be confident of.
To be sure, it could always happen that we might be mistakenin a given
case, but he stressed that the changes we would be led to make in such an
intended logico-mathematical structure would always be consequences of
our inability to foresee an the logical consequences of our assumptions,
and not because an assumption about empirical (or platonic) fact was
found in error. It would not be easy to ten the difIerence always, but he
felt it important, for the sake of c1arity, to try.44
Camap's logicism, in part, then, may be thought of as (1) an urged
proposal that for any given language of science we seek to formulate its
semantics to ensure, as far as we can, that its logico-mathematical ap-
paratus makes no commitments as to physical fact, and (2) an accom-
212 HERBERT G.BOHNERT

panying claim that this seems possible to do, in that nothing in the
development of mathematics so far rules out the possibility of such non-
commital, or analytic, interpretations. In keeping with his concern with
concrete applicability, he saw great methodological value in endeavoring
to maintain a factually empty logic and mathematics clearly separated
from the factual burden of any physical theory, not only to focus the
factual burden (rather than spreading it over mathematics as well), but to
make clear what the content is of any sentence in a given language. 45
As for the ultimate ground for any logical or mathematical truth, he
saw nothing it couldbe other than analyticity, in the developing, emergent
sense, indicated.
The progress of mathematics will not be greatly hindered by lack
of agreement on the ultimate ground of its truth, nor facilitated by agree-
ment. But analyticity, however obscure, seems hardly more obscure than
mathematical intuition or platonistic inner vision. Empiricism has a
sturdy appeal, but Quine's diffuse, holistic empiricism-in-suspension is
hardly more than a recognition of possible needs for revision. Since
Carnap's view also concedes possible needs for revision, the difference
between them is slight, as Carnap has often emphasized (pp. 67, 915-922
[21 ]). What difIerence there is is primarily in Carnap's advocating an at-
tempt to maintain a distinction between two kinds of revision in the
interests of clarity.
My closing suggestion, then, is that the question to ask about Carnap's
10gicism is not whether it is true, but whether it can be made true for
specific languages and theories, and, if it can be, whether it may not be
desira bie to attempt it.
NOTES

1 E.g., C. Parsons, p. 199 [18]. Convention plays an important part in logicism, of course;
and Carnap has acknowledged early infiuence by Poincare's conventionalism, (p. 15 [21]).
But he would have been distressed at his final view being categorized as conventionalism, a
view too easily caricatured. He was aware that some of his formulations, especially in [4]
had invited such construal, and he made repeated attempts to set matters straight, e.g.,
in pp. 26-29 [5]. It would seem inconsiderate, if not unjust, to overrule these reasoned
disc1aimers by simple appellation.
2 This assertion rests on discussions over many years but especially on an intensive three-
day discussion in the summer of 1968, two years before his death, in which his philosophy
of mathematics was a principal topic.
3 This is an expanded version of a paper of the same titie delivered on August, 1971 at the
IVth International Congress for Logic, Methodology, and Philosophy ofScience, Section 4,
 CARNAP'S LOGICISM 213

in Bucharest, Romania. Various de1ays prevented publication in the proceedings of the

Congress, so that only an outline appeared in its program. The material indicated in that
outline appears here mainly in the second section of Part I and in Part II. I wish to thank
Professor Hintikka for his steady encouragement and unusual understanding.
4 [2] and also his 1931 [3].
5 Although it will be sometimes convenient to think of the phases as time periods, they
could more realistically be thought of as differing sets of attitudes, the time period of any
one set being simply that in which it had its most explicit expression. The inceptions of all
of them were early and their effects persisted.
6 Church, in his 1960 reconsideration oflogicism [12], distinguishes between a weak and
a strong form. He takes the thesis of the strong form to be that "mathematics can be
obtained entirely from pure logic without the introduction of additional primitives or
additional assumptions" (pp. 181). From the context, Church means this to include the
view that logic, and hence mathematics, is analytic, though as it stands, his formulation is
one Russell could have accepted before accepting the inclusion.
7 Carnap's extensionalism is somewhat masked in his 1930, 1931 papers, as well as later,
by his frequent use of the term 'property' which he regarded as (extensionally) synonymous
with 'class.'
8 This conditional logicism is not the weak form of logicism first remarked upon in
Church [12] (which is simply the recognition of the priority of logic as the means for ob-
taining consequences from mathematical postulates), nor is it the related view, which I
shall call deductivism, which identifies mathematics with the drawing of consequences from
any postulates (cf. Veblen pp. 4, 5 in [26]). Rather it is essentially the logicist view that
Church himself concludes with in [12].
9 This trend of logicist thought, which I will call the application view, is expressed fre-
quently. For examples, see Camap's remarks on Frege, p. 48, and Beth's remarks on Frege
and Camap, p. 474 in [21].
10 Shortly after the 1930, 1931 papers, Camap came to the view that his purely extensional
view permitted him to regard the axiom of choice as analytic (ef. p. 47 [21]) but discussion
of this point may be postponed.
11 Thus Beth, p. 476 [21].
12 p. 326 [4].
13 In keeping with Camap's conception of syntactical transformation rules I shall oc-
casionally use the adjective 'syntactical' where more recent literature might use 'proof-
theoretic' .
14 See Camap pp. 325-327 [4]. In particular, an older logicist view of logic as being con-
cemed with syncategorematic signs, which lack independent meaning, but provide logical
form was swept aside by Camap, p. 26 [4], with the remark that the distinction between
"expressions having independent meaning" and "synsemantic" ones "may be regarded
as more siguificant from the psychological than the logical viewpoint". The issue was to arise
more pointedly later, of course.
15 p. 326 [4].
16 In his otherwise penetrating review [16] of Syntax (about which more shortly), Kleene
occasionally seems to take this to be what Camap is proposing.
17 More precisely, each ofthe two languages was, in effect, aframework for languages in
that although the logico-mathematical primitives of each were fully specified, the descrip-
tive vocabulary was left open.
18 Posing this question may suggest too strong1y that Camap's primary interest at this
point was in explication, when it was more in forging a valuable linguistic concept.
214 HERBERT G.BOHNERT

19 After ali, GOdel himself, in his 1931 paper (reprinted in [25]), appeared to accept a
modification of the conditiona!logicism earlier mentioned. According to the modification,
ali theorems of arithmetic and analysis which can be derived from the Peano second order
axioms by use of the 'pure logic' of PM can be derived without the Peano axioms from
pure PM plus axioms of infinity and choice. His use of the Peano axioms rather than the
PM constructions was, he noted (footnote 15 p. 599), only for convenience. (This weakening
use of 'theorems' in stating logicist theses is sometimes found where the weakening effect is
clearly not intended.)
20 In his review [16). I am indebted to Professor Kleene for clarifying several passages in
the review, and for making available to me a copy of his 1939 talk 'On the term "analytic"
in logica! syntax' (in which he amplified the simplification mentioned), and also for calling
my attention to Carnap's review (JSL 5,157-8) of an abstract ofthat talk, which was never
published itself (due to a copy's loss in World War II). Carnap refers again to Kleene's
simplification on p. 247 [6).
21 Carnap quotes Schlick (p. 101 [4]) thus: "In the case of an analytic judgment, to under-

stand its meaning and to see its a priori validity are one and the same process."
22 He thus summarizes his point (p. 102 [4]).
23 This point was urged by my colleague Herbert E. Hendry in a debate on analyticity in a
1969 Michigan State University Philosophy Colloquium. I am indebted to him in genera!
for his stimulating arguments against analyticity and his helpful remarks on some points in
this paper.
24 In his Autobiography (pp. 47-48 [21]), Carnap remarks that although he sought an
analytic interpretation for an infinity axiom during his Vienna period he "did not achieve
complete clarity" on the question, but that later he found "severa! possible interpretations"
that were suitable.
25 It is of some interest to note not only Camap's commendation of Dubislav for em-
phasizing the relativity to language (p. 44 [4]), but also a Quine-like emphasis of the
holistic views of Duhem and Poincare (p. 318 [4]).
26 The idea, though not the word 'seeping' itself, is clearly seen on p. 142 [4).
27 Carnap's earlier mentioned view of human intuition being unable to foresee ali the
consequences of the rules it chooses expresses itself at times in a willingness to admit the
remote possibility of gross error in even the most basic parts of logic.
28 As mentioned earlier, much of the methodologica! out1ook underlying the syntactic
phase formed early in the classica! phase. The Tractatus, for example, despite its unaccep-
table picture theory oflanguage, had many impressive anticipations of it, as in proposition
3.33.
29 Thus Frege's early logistic claim that mathematics was derivable from "general logica!
laws and definitions" had not previously been challenged on the score of definitions.
30 Meaning postulates were primitive sentences meant to be analytic though not of
definitional form nor ofpurely syncategorematically logica! vocabulary. Peano's postulates
could be examples, although Carnap's concern with meaning p6stulates arose in connec-
tion with certain difficulties in inductive logic pointed out to him, in 1951, by both J.
Kemeny and Y. Bar-Hillel (independent1y).
31 This and related facts about numerica! quantifiers, e.g., that ali instances of the natural
laws of addition are similarly forthcoming for non-overlapping sets, still seem undeniable
arguments for at least some limited form of logicism in the earliest, simplest sense. Ex-
plicitly first-order numerica! quantifiers are apparently due to Tarski [24], although their
definientia had earlier been used to define particular numerals as second level predicates.
32 Although Donald Davidson vividly refocused attention on it in his [13], and has
continued to do so.
 CARNAP'S LOGICISM 215

33 pp. 902-3 [21]. Since 'model' is frequently used for an interpretation or a structure in
which a set of sentences is true, the distinction might better have been drawn between an
extensional interpretation and an intensional (or at least descriptively given) interpretation.
34 See Kemeny [15] for such an argument and for references to earlier publications. For
references to work done in probability, information theory, etc., on the assumption that
the distinction can be made, see [21].
35 It is worth noting that in presemantics days, rules of inference were also viewed in some-
what the same light as meaning postulates - namely, as forcing (psychological) meaning on
the 'syncategorematic' logica1 connectives and quantifiers. Carnap was fond of pointing
out the practical importance of seeing the rule of Modus Ponens in action to a child's early
understanding of 'if'.
36 These speculations on a syncategorematic 'set-predication' are not Carnap's, but only
prompted by refiecting on his occasiona1 heavy stress in speaking of the purely logica!.
37 From the beginning of the syntactica1 phase Carnap never took the syncategorematic
view oflogic or of semantics, too seriously. The problems with meaning postulates that most
absorbed him were those involving descriptive terms. Logic and mathematics provided the
factually neutral framework for language whether or not it used 'categorematic' signs.
This view (of logic at least) seems shared by the increasing number of logicians who treat
erstwhile syncategorematic signs categorematically (e.g., for whom 'connectives' and
quantifiers become (constant) functions), e.g., Church, Henken, Shoenfield, etc.
38 'What is Cantor's Continuum Problem?' reprinted in [25]. See pp. 263-264.
39 The 'between' example was not Carnap's, but is cited only to illustrate his attitude
toward meaning postulates.
40 In the Foreword, p. xi, of Lewis's [17].
41 The transition began with his 1936 'Testability and Meaning', reference to which may
be found in [21] along with references to later deve1opments.
42 2.17 What the picture must have in common with reality in order to represent
it. .. is its form of representation.
2.022 It is dear that however different from the real world an 'imagined world
may be, it must have something in common with the real world.
2.023 This fixed form consists of the objects.
2.0271 The object is the fixed, the existent; the configuration is the changing, the
variable.
43 That this whole, semantically extended metasystem might be regarded as an important
meaning-narrowing context for the object language, and indeed that it could be explicitly
unified with the object language and knowledge expressed in it in one, overall Ramsification,
I have argued elsewhere.
44 My summary of the 1968 conversations ends at this point. I have omitted technical side
discussions, but otherwise it is as close as notes and memory permit.
45 This methodologica1 value, not only for mathematics, but for analyticity in general,
has, of course, been eloquently spelled out by a wide variety of writers, from strong ad-
herents to the formalized language approach, such as Kemeny and Martin, to strong
partisans of the naturallanguage approach, such as Grice and Strawson.

BIBLIOGRAPHY

[1] Benacerraf, P. and Putnam, H. (eds.), Philosophy of Mathematics, Prentice-Hall, 1964.

[2] Carnap, R., 'The Logicist Foundations of Mathematics', in [1] translated from
Erkenntnis (1931).
216 HERBERT G.BOHNERT

[3] Carnap, R., 'Die Mathematik als Zweig der Logik', BIătter fur deutsche Philosophie
4 (1930), 260-285.
[4] Carnap, R., The Logical Syntax of Longuage, New York, 1937, translated and ex-
tended from German edition, 1934. (henceforth Syntax).
[5] Carnap, R., Foundations of Logic and Mathematics, Chicago, 1939.
[6] Carnap, R., Introduction to Semantics, Cambridge, Mass., 1942.
[7] Carnap, R., Introduction to Symbolic Logic and Its Applications, New York, 1958,
translated from German original of 1954.
[8] Carnap, R., Meaning and Necessity, Chicago, 1947.
[9] Carnap, R., 'Empiricism, Semantics, and Ontology', Revue international de Philoso-
phie, 15 Jan. 1950,20-40.
[10] Carnap, R., Logical Foundations of Probability, Chicago, 1950.
[11] Carnap, R., 'Beobachtungssprache und theoretische Sprache', Dialectica 12 (1958),
236-248.
[12] Church, A., 'Mathematics and Logic', in Logic, Methodology and Philosophy of
Science (ed. by E. Nagel, P. Suppes, A. Tarski), Stanford, 1962.
[13] Davidson, D., 'True to the Facts', Journal of Philosophy 66 (1969), 748-764.
[14] Edwards, P. (ed.), The Encyclopedia of Philosophy, New York, 1967.
[15] Kemeny, J., 'Analyticity versus Fuzziness', in Form and Strategy in Science (ed. by
J. Gregg and F. Harris), Dordrecht-Holland, 1964, pp. 122-145.
[16] Kleene, S., Review of [4], Journal of Symbolic Logic 4 (1939),82-87.
[17] Lewis, D., Convention: a Philosophical Study, Harvard U. Press, Cambridge, 1969.
[18] Parsons, C., 'Mathematics, Foundations of in [14J.
[19] Quine, W. V., From a Logical Point of View, Cambridge, Mass., 1953.
[20] Quine, W. V., Set Theory and Its Logic, Cambridge, 1963.
[21] Schilpp, P. (ed.), The Philosophy of RudolfCarnap, Open Court, 1963.
[22] Reichenbach, H., Elements of Symbolic Logic, New York, 1947.
[23] Tarski, A., 'The Concept of Truth in Formalized Languages', translated by J. H.
. Woodger in Logic, Semantics, Mathematics, Oxford, 1956. Originally published in
Polish 1931, German 1936.
[24] Tarski, A., Introduction to Logic, Oxford 1941, originally published in Polish 1935,
and in German 1936.
[25] van Heijenoort, J., ed., From Frege to Godel, Cambridge, Mass., 1967.
[26] Veblen, O., 'The Foundations of Geometry', in Monographs on Topics of Modern
Mathematics (ed. by J. W. A. Young), New York, 1954.
[27] Wittgenstein, L., Tractatus "Logico-Philosophicus, London, 1922.
 JAAKKO HINTlKKA

CARNAP'S HERIT AGE IN LOGICAL SEMANTICS

One of the most genuine tributes one can pay to any thinker who has
already passed away is to be able to say that his work does not have to
be given a special consideration as a kind of venerable museum specimen
but can be discussed on its own merits as if its author were still among
us. This tribute we can pay in full measure to Rudolf Carnap's work in
logical semantics as in other areas, and I am sure that it is the way in
which Carnap himself would have preferred to have his work remembered.
In this chapter, I shall therefore not shy away from those aspects of
recent discussions of semantics which might at first appear to by-pass
Carnap's work or even to stand in an opposition to it. An important
additional reason for doing so lies in the fact (which I shall try to argue
for) that much of this recent work in semantics is, appearances notwith-
standing, an outgrowth of Carnap's ideas or consists of attempts to solve
the important problems Carnap rai sed in semantics. Much of the credit
of his successors' work is thus due to Carnap.
What, then, is crucial in Carnapian ideas? It was once said by David
Kaplan that Carnap's Meaning and Necessity - the book 1 will mostly
concentrate on - represents the culmination of the golden age of (logical)
semantics.l This age, if 1 have understood Kaplan correCtlY' is supposed
to extend from Frege to Carnap, and to be characterized by that familiar
contrast which in its several variants has been known by such labels as
Bedeutungen vs. Sinne, references (or nominata) vs. senses, or extensions
vs. intensions. In Meaning and Necessity Carnap uses the last pair of
terms. 2
Carnap's work in MN and elsewhere may very well seem to be the end
product of this tradition. The importance of the extension-intension di-
chotomy to him is amply shown by the table of contents of MN. It reads,
in part:

1. The Method of Extension and Intension.

5. Extensions and Intensions.

Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 217-242. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
218 JAAKKO HINTIKKA

60 Extensions and Intensions of Sentenceso

90 Extensions and Intensions of Individual Expressionso
11. Extensional and Intensional Contextso

220 L-determinate Intensionso

230 Reduction of Extensions to Intensionso
280 Frege's Distinction between Nominatum and Senseo
290 Nominatum and Sense: Extension and Intensiono
---------.
And even in sections whose titles do not sport the terms 'extension' and
'intension' these two concepts loom Iargeo For instance, the fourth of
Carnap's five chapters is devoted almost exclusively to the question of
how the distinction is to be accommodated in one's metalanguageo
MN represents the Fregean tradition also in that Carnap emphasizes
the primacy of intensions over extensions, to the point of speaking of a
reduction of extensions to intensions (MN, Section 23)0 In his own ab-
stract of Section 27 we likewise read: "000 a semantical rule for a sign
determines primarily its intension; only secondarily, with the help of
relevant facts, its extensiono" This goes back in some form or other to
Frege who said in so many words that "in the conflict between extensional
and intensional Iogicians 1 am taking the side of the Iattero In fact 1 do
hold", Frege continues, "that the concept is Iogically prior to its ex-
tensiono" 3
The intensions we need in a Carnapian semantics include such old
friends of philosophers as propositions, properties (as distinguished from
the classes they determine), and individual conceptso Needless to say,
their primajacie philosophicaI importance could not be greatero The post-
ulation of such intensionaI entities has been claimed by Carnap's critics
to violate the standards of enIightened scientific empiricismo The most
important of these critics is Wo Vo Quine0 4 It has been made beautifully
clear by Quine himself how much of his philosophy of Ianguage can be
understood as a reaction to Carnap's semanticso 5
This is not a place to try to adjudicate the whole of Carnap-Quine
exchangeo It seems to me fair to say, nevertheless, that Quine has spotted
certain weak spots in Carnap's position, at Ieast insofar as MN is con-
cernedo This is not decisive, however, for 1 shall argue that Carnap's ideas
 CARNAP'S HERITAGE 219

aHow for developments which serve to solve Carnap's difficulties to an

incomparably greater extent than the critics, and the philosophical com-
munity at large, have so far acknowledged. At the same time, these devel-
opments show that the strict intension-extension contrast is far too narrow
a framework for a realistic semantics, and that Carnap's own ideas can
easily be extended so as to widen this framework essentially. These, at
any rate, will be the main theses of the present chapter. If 1 am right,
Carnap was not the last Mohican of Fregean semantics, based on the
extension-intension contrast, but rather the first and foremost herald of a
new epoch of possible-wor1ds semantics.
Usually, Carnap's critics have focused on what the critics c1aim amounts
to the unobservability or perhaps rather inevitable empirical underdeter-
minacy of intensions. To put their point very briefly: We cannot ever
hope to find out for sure what the intensions are that underiie a heathen
tongue - or our own idiolect, for that matter. AH the speech disposition
of the speakers of any given language are compatible (according to
Quine) with the postulation of more than one set of incompatible inten-
sions.
These criticisms seem to me somewhat premature - as premature in
fact as the views under criticism. We shaH return to the problems of
empiricism and observability later. Meanwhile, 1 want to emphasize that
the true weakness of Carnap's position in MN is not the non-empirical
character of his main semantical concepts. EspeciaHy in some of the papers
appended to the second edition of MN, Carnap in fact presents plausible
arguments to show that his concepts do carry an empirical import. The
reason why these arguments have not swayed more philosophers than
they have done is not so much due to the arguments themselves. It is due
to the fact that the crucial intensional concepts themselves were not
analyzed far enough by Carnap in MN. They were not developed in a way
which would have created a viable framework for Carnap's own argu-
ments and for the further development of his semantics, e.g., so as to
allow a natural and convincing treatment of belief-sentences. What is
missing in MN are not reasons for the empirical character of intensional
concepts but rather aH penetrating analyses of them in some more infor-
mative and more easily operationalizable terms.
Whoever deserves the credit for the analyses which we just found want-
ing in MN, they are found in a full-fledged form in what is often called
220 JAAKKO HINTIKKA

possible-wor1ds semantics. Its main outIines wiII emerge from a compar-

ison with Carnap's position in MN.
The historical fact which one cannot but find absolutely fascinating
is that in MN Carnap carne extremely close to the basic ideas of possible-
wor1ds semantics, and yet apparentIy did not formulate them, not even to
himself. The conceptual framework deve'oped in the first chapter of MN
is that of state-descriptions. It is in terms of state-descriptions that Carnap
defines aII his crucial concepts, such as those of range, L-truth, L-equi-
valence, identity of intensions, and so on. 6 TechnicaIIy, this almost
amounts to a possible-worIds semantics. AII that Carnap had to do here
was to take a good hard look at his state-descriptions and to ask: What
are they supposed to be descriptions of in some realistic, down-to-earth
sense? One natural answer is that they are descriptions of the different
possible states of affairs or courses of events (in short, 'possible worIds')
in which a speaker of the language in question could conceivably find
himself and which he could in principle distinguish conceptuaIIy from
each other. From this answer it is only a short step to the crucial idea
that the rules for using the language wiII have to be shown - in principle -
by the way a weII-informed speaker would use it in these different cir-
cumstances according to the rules, i.e., by the extensions which the ex-
pressions of the language would have in those several 'possible worIds'.
This is aII we need to arrive at the basic ideas ofpossible-worIds semantics.
It is especiaIIy tantalizing to see that Carnap in fact says in so many
words in MN that his state-descriptions "represent Leibniz' possible
worIds" (p. 9). In his intellectual autobiography,7 Carnap likewise
mentions Leibniz' possible worIds as one of the original guiding ideas
of his distinction between logical truth and factual truth (p. 63). What
is missing is thus apparently only an insight into the possibility of using
these possible worIds for the purpose of analysing the intensionaI objects
Carnap in fact leaves unanalysed in MN.
The move from MN to possible-worIds semantics is closest at hand in
the case of intensions of sentences, i.e., in the case of propositions. The
class of those state-descriptions in which a sentence OS' is true is called
the range R(S) of oS' (MN, p. 9).8 According to Carnap, this range
determines the identity of the intension I(S) of oS' in the sense that
I(S) = I(S') iff R(S)=R(S') (MN, p. 23, Def. 5-2). In the same sense, R(S)
determines the proposition expressed by 'S'o What, then, is more natural
 CARNAP'S HERITAGE 221

than to define this proposition as R(S) or as something closely related

to R(S), such as the characteristic function ofthe set ofworIds described
by the members of R(S)? In the latter case, propositions wiII be functions
from possible worIds to truth-values. This definition is most natural
because it is natural to say that to understand a proposition is to know
what restrictions its truth places on the worId. Such restriction is precisely
what the membership in the set of worIds described by the members of
R(S) amounts to. Here we have a good example ofthe kind of structural
analysis of fundamental semantical notions which possible-worIds
semantics enables us to carry out but which I found missing from MN.9
This kind of structural insight is not restricted to intensions of sen-
tences. Historically, perhaps the most crucial question is what can be said
of the intension l(i) of an individual expression 'i'. 8 In MN, Carnap never
formulates explicit criteria for the identity of the intensions l(i) and l(i')
of two individual expressions oi' and 'i" comparable with his criteria for
the identity of the intensions of sentences. Implicit in his discussion (see
e.g. p. 40) is nevertheless a criterion -according to which l(i) = l(i') iff
'i = i" is true in every state-description. By the same token as in the case
of the intensions of sentences, this naturally leads us to identify the intens-
ions of individual expressions (dubbed by Carnap individual concepts)
with functions that for each possible worId W pick out a member of some
domain of individuals (or with some essentially equivalent entity). (This
domain must obviously be thought of as depending on W, if we waht to
have a flexible, presuppositionless treatment of the situation.10) Thus if
1(i) is the intension of 'i'; the function 1 must be thought of as having
a second argument, too, and thus as being of the form l(i, W), where W
is the possible worId in which we are considering the reference of oi'.
In fact, one's interpretation of the intensions of sentences as sets of
possible worIds (or, essentially equivalently, functions from possible
wor1ds to truth-values) tends rather strongly to prejudice the case for
a similar treatment of individual expressions. As I have pointed out on
several occasions,11 if one introduces the modal operators 'N' ('necessar-
ily') and 'M' ('possibly') and formulates the truth-conditions in the most
natural manner imaginable, merely requiring the substitutivity of identity
for atomic expressions, then it can be shown that be criterion of substi-
tutivity (say of'a' and 'b' for the case of just one layer ofmodal operators)
is the truth of 'N(a = b)'. According to the treatment of sentences, how-
222 JAAKKO HINTIKKA

ever, this is true iff 'a = b' is true in alI (relevant) possible worlds, i.e.,
iff 'a' and 'b' pick out the same individual from each of these possible
worlds. This is the analogue to Carnap's explicit criterion for the identity
of the intensions of two sentences, and in the same way suggests the
identification of I(i) with a function from possible worlds to their respec-
tive domains of individuals.
What is remarkable in this analysis of individual concepts is how very
closely it comes to the intentions (with a 't'!) of modern semanticists from
Frege on. Frege said that the intension (Sinn) of a name must include
more than just its reference. It must also include the way in which this
reference is given (die Art des Gegebenseins, 'Sinn und Bedeutung', p. 26
of the original edition 12). Now the functional dependence which this
phrase "way of being given" clearly means can - and must - be spelIed
out by specifying how the reference depends on everything it might depend
on, which in the last analysis is the whole possible world we are dealing
with. 13 (Of course this does not preclude that it depends only on certain
particular aspects of that world!) But this is precisely what the function
I(i, W) gives us. Here, possible-worlds semantics therefore folIows as
closely as one can hope in Frege's and Carnap's footsteps. I cannot but
find it very strange that it apparently never occurred to Frege that to
speak of "die Art des Gegebenseins" is implicite to speak of a functional
dependence of a certain sort. There does not seem to be an inkling of
this idea in his writings.
Clearly predicators can be dealt with in the same way as individual
expressions. Their intensions wiIl be functions from possible worlds to
sets of n-tuples of the members of their domains, or some similar entities.
This completes my sketch of the step from MN to possible-worlds
semantics. The step is so short that it is not surprising to find a report
according to which in his unpublished work Carnap did take something
essentialIy tantamount to it. Richard Montague reports in his paper,
'Pragmatics and Intensional Logic',14Iast paragraph, that "Carnap had ...
proposed in conversation that intensional objects be identified with func-
tions from possible worlds to extensions of appropriate sorts ... ". In fact,
in addition to conversations, Carnap's 'Replies and Expositions' in the
Schilpp volume contain a sketch of what he caUs "translation of a modal
language into an extensional language" (pp. 894-6). Apart from minor
technical differences, this 'translation' is to alI practical purposes an out-
 CARNAP'S HERITAGE 223

line of a model-theoretical treatment of intensions, with what Carnap

called models playing the role of possible worlds. In fact, propositions
are in so many words 'represented' as classes of models, certain other
intensions as functions from models to suitable specifications of the prop-
erties of their individuals, and the necessity of a proposition amounts
to its truth in all possible worlds. In brief, we seem to have here a full-
fledged possible-worlds semantics explicitly outlined by Carnap. Yet this
impression is definitely misleading. Carnap has most of the basic technical
ingredients of a possible-worlds semantics right there in his hands, but he
does not know what to do with them philosophically and interpretation-
ally. His notion of a model is not that of a possible world, for he is,
e.g., allowing descriptive predicates to be arbitrarily re-interpreted in a
modeI.15 In a different possible world surely those and only those things
are to be called red that are red there. Hence the interpretation of des-
criptive predicates must be assumed to be constant between different
possible words. This is nof required of Carnap's models, however.
Montague reports that, according to Carnap's verbal suggestions, too,
"possible worlds [are] identified with models".16 In other words, possible
worlds were not thought of by Carnap as the real-life situations in which
a speaker might possibly find himself, but as any oId configurations -
perhaps even linguistic - exemplifying the appropriate structures. As we
shall see, it is this apparently small point that precludes Carnap from
some of the most promising uses of possible-worlds semantics.
Although possible-worlds semantics thus may be said to be (in some
respects) a natural and perhaps even fairly small further development
beyond Carnap, it nevertheless puts the whole of the classical Frege-
Carnap semantics into a radically new perspective. Here 1 shall only
comment on three aspects of the new perspective. (1) First, the new
semantics opens the door to the treatment of a large class of philosophi-
cally interesting notions, thus answering a number of Carnapian ques-
tions. (2) Second, 1 shall argue that possible-worlds semantics shows con-
clusively the insufficiency of a semantics which is primarily ba sed on
the intension-extension distinction. (3) Third, 1 shall suggest that possible-
worlds semantics perhaps points to a way of removing the objections
which Quine and others have raised against intensional concepts because
of their alleged unobservability, empirical vacuousness, behavioral non-
specificity or because of some similar defect.
224 JAAKKO HINTIKKA

1. Already in MN17 Carnap put his semantics to work for the purpose
of spelling out the logic of modalities ('necessarily' and 'possibly'). It is
not always appreciated sufficiently that this made Carnap into the first
modal logician to employ semantical methods.
The details of Carnap's modal logic are rather predictable, and need
not detain us here. If Carnap had formulated his point in the suggestive
terminology of 'possible worlds', ali that is really involved in Carnap's
modallogic (apart from the treatment of individuals, their existence, and
their uniquen~ss) is the old idea that necessity means truth in every
possible world and possibility truth il'} at least one possible world. Once
again, the necessity of dragging along ali intensions as unanalyzed entities
leads Carnap to a lengthy discussion of how we ought to address them
in our metalanguage.
Carnap's failure (in MN) to analyze his intensional concepts seems to
be a partial reason for a much more serious oversight, however, than his
worry about a bunch of somewhat scholastic problems concerning one's
metalanguage. The point is perhaps made most forcefully in a somewhat
technical-sounding jargon. When propositions become functions from
possible worlds to truth-values and individual concepts functions from
these worlds to members of their respective domains of individuals, all
sorts or interesting conceptualizations can be reached by restricting the
domains of these functions (in the relation-theoretical sense of domain)
to subclasses of the class of alI possible worlds.
The first major novelty in the subsequent technical development of the
semantics of modallogics was in fact the idea that not alI possible worlds
are on a par. Given a world W, only some possible worlds are relevant
alternatives to W. Then necessary truth of a sentence in W has to be
characterized as its truth (truth simpliciter) in ali the alternatives to W,
and its possibility a fortiori as its truth in at least one alternative to W.
The first heady discovery in this area was that by imposing simple restric-
tions on the alternativeness relation we obtain the semantical counterparts
to ali of the most important axiomatic systems of modal Iogic. 18 Their
semantics is (with one exception) unobtainable in the simpIe-minded
Leibniz-Carnap assumption of the parity of ali possible worlds.
This does not seem to affect Carnap's immediate purpose, for he was
trying to explicate the notions of logical necessity and logical possibility.
For them, it is natural to argue, ali worlds are equal: what is necessary
 CARNAP'S HERITAGE 225

or possible in one is likewise necessary or possible, respectively, in any

other. Hence Carnap's modallogic seems to be unobjectionable as far it
goes, and the advantages of the alternativeness relation appear primarily
technical.
This is not the whole story, however. The most important phiIosophicaI
uses of alternativeness relations are for the purpose of studying certain
notions in which Carnap was interested, especiaIly the notion of belief.
The use of an alternativeness relation makes it possible to accommodate
such notions within possible-worlds semantics.19 In fact, the interpreta-
tion of the alternativeness relation itself is exceptionaIly clear in this case.
(It turns out that it has to be relativized to a person.) Worlds alternative
to W (with respect to a person a) are then worlds compatible with every-
thing a believes in W. Understanding the concept of belief wiII then
become tantamount to mastering this particular kind of alternativeness
relation (relation of doxastic alternativeness). How close this comes to
our actual ways with notions like belief can perhaps be seen by pointing
out that to know what a believes (say, in the actual world) is clearly
very close to knowing which possible states of affairs or courses of events
are ruled out by his beliefs and which ones are compatible with it. This,
of course, is just what the alternativeness relation specifies. Our analysis
thus constitutes an important step beyond Carnap in the analysis of the
concept of belief.
Carnap was apparentIy prevented from analyzing the concept of belief
in this way by the very same peculiarity which made us say above that
he never reached full-fledged possible-worlds semantics, viz. by his failure
to interpret his models as genuine possible worlds, i.e., real-life alterna-
tives to our actual world. This does not matter as long as one is merely
studying the notions of logical necessity and logical possibiIity. It already
begins to matter if we are interested in analytical necessity and analytical
possibility, for here arbitrary reinterpretation wilI destroy those relations
of synonymy (or whatnot) which do not reduce to the formal truths of
logic. This may perhaps be handled by means of explicit meaning pos-
tulates, but no comparable trick has much appeal in the case of belief.
Hence the step from Carnap's "translation of modal language into ex-
tensional language" to possible-worlds semantics, small though it might
seem, makes aH the difference in the world to our analysis of belief. By
the same token, it enables us to undertake similar analyses of several of
226 JAAKKO HINTIKKA

the most important philosophical concepts, including knowledge, mem-

ory, perception, obligation, etc.
One marriage of the problem of belief to possible-worIds semantics
does not solve aII the problems concerning belief-sentences Carnap dis-
cusses in MN. However, it opens the door to new developments in this
direction. Carnap's own terminology enables us to describe the situation
succinctIy. Carnap called an expression intensional iff the identity of
intensions (L-equivalence) is a necessary and sufficient criterion of sub-
stitutivity in that expression. (For a more accurate definition, see MN,
p. 48.) Carnap pointed out that beIief-expressions are not intensional in
this sense. In fact, the faiIure of intensionalityis here twofold. If 'i' and
'i" are individual expressions, I(i) = I(i') is neither a necessary nor a
sufficient condition of substitutivity. Carnap's discussion in MN is
addressed soIely to the problem created by the Iatter fact. What is at
issue here is of course the fact that even if 'i' and 'i" are L-equivalent
(logicaIIy equivalent), a rational believer may very well be unaware of
their equivalence, and an interchange of 'i' and 'i" may therefore make
a difference to his beliefs. For this reason, we need for the purpose of
anaIyzing the concept of belief a relation which is (at Ieast sometimes)
stronger than logical equivalence. In the Schilpp volume (pp. 897-900)
Carnap in effect calls equivalence cIasses with respect to the former
(stronger) relation senses, those with respect to the latter intensions. This
does not alone help us very much, however. The main problem here is
the characterization of the new, stronger relation.
This problem is not automatically solved by possible-worId semantics
but remains a problem there. However, gradually we seem to be getting
even this probIem under contro1. 20
Carnap's own attempted solution to this problem was in terms of what
he called intensional structure. 21 Roughly speaking, two expressions have
the same intensional structure iff they are built up in the same way of
logically equivalent unanalyzed parts. The intensional isomorphism of
S and S' is proposed by Carnap as a sufficient criterion for the logical
equivalence of 'a believes that S' and 'a believes that S". This solution
appears to me ad hoc, however, untiI some general theoretical reasons
are given why it is just differences of intensional structure that essentially
tend to obscure our insights into logical interrelations of sentences. For
it was precisely this faiIure of us humans to be 'Iogically omniscient' that
 CARNAP'S HERITAGE 227

causes the failure of intensional isomorphism to be a sufficient condition

of substitutivity. Hence the formal restrictions on substitutivity ought
to refiect those structural factors that are principalIy responsible for the
failure of'logical omniscience'. However, ifthe question is put in this way,
it seems to me clear tbat more interesting candidates for this role can be
suggested. 22
Let us leave this half of the problem and retum to the failure of L-equi-
valence to constitute a necessary criterion of substitutivity. What the
fault-finders uniformly overlook is that this part of the problem is beauti-
fully solved by possible-worlds semantics. According to this semantics,
'i' and 'ilO are interchangeable in discussing a's beliefs iff they pick out
the same individual in alI the possible worlds we have to consider here.
These worlds, in turn are ali the worlds compatible with what a be-
lieves. Hence the identity of the references of 'i' and 'ilO in these worlds
means that a believes that i = i'. But if so, quite obviously 'i' and 'ilO are
interchangeable in discussing a's beliefs, provided they are consistent.
Hence possible-worlds semantics at once leads to the right condition of
substitutivity, thus carrying the analysis of an important problem of
Camap's essentially further.
The reason why the identity of intensions is not a necessary condition
for substitutivity here is nicely brought out by the fact that f(i) = f(i')
means that the functions which pick out tbe references of 'i' and'i",
respectively, coincide on the whole set of possible worlds, while the truth
of
'a believes that i = ilO
in a world W only requires that they coincide on the much smaller set
of altematives to W.

2. This brings us already toward my second main point. It is that possible-

worlds semantics conclusively shows the insufficiency of a semantics based
solely on the distinction extension-intension or Bedeutung-Sinn. This
distinction is alI right, but it just does not do the whole job nor even one
of the most important parts of the job that a satisfactory semantical
theory must do. Hence, the classical Frege-Camap semantics is very seri-
ously incomplete, notwithstanding its closeness in some respects to pos-
sible-worlds semantics.
In order to see what the problem is, it may be useful to try to have
228 JAAKKO HINTIKKA

an overview of the aims of the Frege-Carnap semantics. For our present

purposes, the relatively unproblematic part - the part to which Frege in
fact paid less attention - is the semantics of purely first-order (quanti-
ficational) notions. Onlya small selection of the problems concerning it
were taken up by Carnap. (Some of them are in fact very naturally
suggested by the basic ideas of possible-wor1ds semantics, but 1 shall not
discuss them here.) The problems 1 want to focus on here are due to the
failure of our expressions to behave in modal contexts (in the wide sense
of the word in which 'propositional attitudes' like belief are also consid-
ered modal notions) in the same (relatively) unproblematic way as in
first-order contexts. Now the locus classicus of the Frege-Carnap seman-
tics is of course Frege's paper 'Sinn und Bedeutung'.23 The very first
question Frege asks in this paper concerns the behavior of identities like
'i = i" vis-a-vis the notion of knowledge. More generally, to explain the
failure of the substitutivity of identity in modal contexts is obviously one
of the basic tasks of any satisfactory semantics.
The basic answer Frege-Carnap semantics gives is that what matters
in modal contexts are not the extensions of one's expressions but rather
their intensions. At first blush, this seems quite wrong-headed, for the
right criterion of substitutivity in, say, belief-contexts (doxastic contexts)
is certainly not the identity ofthe intensions (Sinne) ofthe intersubstituted
expressions. As was already noted, we do not need to have I(i) = I(i') in
order to have 'i' and 'ilO interchangeable in discussing a person's beliefs.
Hence Frege's answer to his own first and foremost question seems to
be seriously amiss. Likewise, it was already indicated that Carnap failed
to say very much of interest about substitutivity in belief-contexts in
terms of his theory of extensions and intensions.
However, here possible-wor1ds semantics rushes to the rescue of Frege
and Carnap. When the intensions of (say) individual expressions are ana-
lyzed as functions from possible wor1ds to the members of their respective
domains of individuals, it becomes c1ear that intensions are after all essen-
tially involved in the substitutivity conditions. The only new thing that
happens in belief-contexts is that it is not the identity of these unrestricted
functions that matters, but rather the identity of their restrictions to a
certain sub set of the set of all possible wor1ds. (Typically, it is the set of
wor1ds compatible with everything someone believes.) The same account
is seen to work for many other propositional attitudes.
 CARNAP'S HERITAGE 229

Hence the Frege-Carnap semantics does come close to giving the right
answer to the question of substitutivity conditions in modal contexts,
although their own formulations did not speli out the matter quite fully.
It is instructive to notice how our treatment of the substitutivity problem
was made possible by the insight into the relation of the possible worIds
to the notion of belief via the doxastic alternativeness relation. Here the
tremendous advantages that accrue from the insignificant-Iooking step
from models to possible worIds are beginning to tell. Both our primafacie
objection to the Frege-Carnap treatment of substitutivity and the simple
answer to it would have been impossible to formulate without this step.
Among other things, Frege's first and foremost problem would have re-
mained unsolved as a consequence.
However, the substitutivity problem is not the only one here, and those
logicians who have tried to make it into the only major problem in inter-
preting mod al logic have only succeeded in clouding the issues. The sub-
stitutivity problem is a paradigm problem caused by the failure of the
usual identity laws in modal contexts. Another set of problems is created
by the failure of quantificational laws in these contexts. The paradigmatic
problem here is to account for the failure of existential generalization,
i.e., of many inferences of the form
(EG) F(a), therefore (Ex) F(x).
where 'F(x)' contains modai operators.
Possible-worIds semantics at once yieIds a natural explanation. The
individual expression 'a' may pick out different individuals in the different
possible worIds we have to consider in 'F(a)'. If so, the truth of 'F(a)'
does not give us any opening for maintaining that 'F(x)' is true of some
particular individual x, as '(Ex)F(x)' claims. Hence (EG) is not valid in
general.
It is also seen at once (at least roughly) when (i.e., on what additional
conditions) (EG) is valid. It is valid iff 'a' picks out one and the same
individual from ali the different possible worIds as a member of which we
are tacitly considering a in 'F(a)'. What these worlds are can be read from
'F(a)', and it turns out that the requisite uniqueness condition can even be
expres sed by a suita bie sentence of our modallanguage. 24
Precisely how this happens is an interesting question, but it need not
concern us here. Our main interest lies in the fact that in order to make
230 JAAKKO HINTIKKA

sense of the reasons for the failure of existential generalization as well

as of the conditions of its success we have to be able to cross-identify,
that is, to say of a member of (the domain of individuals of) one possible
world that is or is not identical with a member of another. (For we had
to say that 'a' picks or does not pick the same individual in different
possible worlds.)
The interpretational aspects of cross-identification offer alI sorts of
problems. However, the overriding fact is c1ear enough, and sufficient for
our purposes: oross-identification must make objective sense.
To see what this means, consider a 'world line' connecting the 'embodi-
ments' or perhaps better the 'roles' of one and the same individual in
alI the different possible worlds. These members of the different possible
worlds may be thought of as being picked out by a function. This func-
tion is of the nature of an intension in the sense that it is of the same
logical type as those functions which serve as intensions of individual
expressions. Let us calI a la Carnap functions of this kind individual
concepts. The objectivity of cross-identification then means that a subc1ass
of the c1ass of alI such special individual concepts has to be objectively
given, viz., the c1ass of those special functions which define world lines
of one and the same individual.
Clearly, it will be a proper subc1ass of the c1ass of all individual concepts,
for obviously not any old function which picks out an individual from a
number of possible worlds picks out the same individual from alI of them
in any conceivable sense of identity. Let us calI the narrower c1ass of
those functions that do so the c1ass of individuating junctions.
When an explicit semantics is developed,25 it turns out that individ-
uating functions, or, rather, their restrictions to certain sets of possible
worlds, are the main ingredients of the truth-conditions for quantified
sentences. They are the most important entities we have to quantify over
in these truth-conditions.
Now it is obvious tbat the c1ass of individuating functions cannot be
defined in the sole terms of the c1ass of individual concepts. As far as 1
can see, it cannot be reduced in any other sense, either, to the c1ass of
individual concepts.
What this means is that a semantics which only recognizes the whole
unanalyzed set of individual concepts as a primitive idea will be incapable
of formulating satisfactory truth-conditions for quantified sentences in
 CARNAP'S HERITAGE 231

modal contexts and also incapable of explaining the failure of existential

generalization in modal contexts. The classical Frege-Carnap semantics
is a case in point, give or take a few minor qualifications. Hence their
type of semantics is insufficient for deaIing with quantified modallogics.
More specifically, it is incapable of dealing ~ith the other paradigmatic
puzzle case in this field, viz. the failure of existential generalization. No
wonder Quine has been unhappy with Carnap's semantics, for unIike
Frege he has explicitly considered the problem of existential generaIiza-
tion in modal contexts over and above the problem of substitutivity of
identity. Small wonder, too, that Quine has directed his main attack
against the idea of quantified modallogic.
From one important point of view, the classical Frege-Carnap seman-
tics is thus seriously incomplete, requiring an essentially new conceptual
element in order to be able to deal with the problem which more than
perhaps anything else has been the bone of contention between Carnap
and his critics, viz. the problem of combining quantification and modality.
To put the main point in a nutshell, the Frege-Carnap semantics explains
the behavior of identity in modal contexts (and propositional-attitude
contexts), but not the behavior of quantifiers in such contexts. The dif-
ference between the two problems is almost Iike a quantifier-switch. In
the case of identity, the problem is to tell when two singular terms pick
out the same individual in each possible world (of a certain sort). In the
case of quantification, we have to ask when one and the same singular
term picks out the same individual in ali possible worlds (of a certain
kind). Only the second problem involves cross-identification between
possible worlds. For this reason, it does not reduce to the first.
At the same time, possible-worlds semantics supports Carnap against
his critics in the crucial matter of the possibiIity of using intensional
concepts in a way which makes it possible, e.g., to construct a semantics
for quantified modal logic. Admittedly there are furtl1er problems here
which may bring out the bite of some of Quine's specific criticisms. 26
However, on the level at which most of the Carnap-Quine controversy
has been carried out, possible-worlds semantics is not only an outgrowth
of Carnap's ideas but also their partial vindication.
This conclusion is so important that it deserves a few further comments
and a few supplementary arguments in its favor. It is not only the case
that the dichotomy extension-intension requires some supplementation
232 JAAKKO HINTlKKA

in order to be workable. What is even worse for those dichotomizers

who stiII try to rely on the contrast, the introduction of individuating
functions messes up thoroughly the neat intuitive contrast between ref-
erences (extensions) and meaning entitites (intensions) which is one of
the apparent attractions of the Frege-Carnap semantics. For the position
of individuating functions in the alleged dichotomy of references and
meanings is hopeIessly ambivalent.
On the one hand, individuating functions constitute a subclass of the
class of those paragons of intensionality, individual concepts. Moreover,
they serve to solve one of the main problems for the treatment of which
meaning entitites (intensions) have usuaIly been introduced in the fust
place.
On the other hand, what individuating functions do is to give us the
individuals which serve (albeit in some cases only potentiaIly) as the
references of our individual expressions. Almost the only reason, it may
be suggested, why we have to deal with such functions here is that we
have to keep an eye on more than one possible world and hence to keep
track of our individuals - the very normal unexciting kinds of entities
that inhabit our actual universe - in these different worlds. The technical
counterpart to this (essentially correct) intuitive view is that the main role
of individuating functions (or suitable restrictions of them) is to supply
the entities one's quantified variables range over in modal contexts (more
accurately, when we quantify into a modal context) precisely when we
insist on quantifying over normal, down-to-earth sort of individuals in
the normal 'objectual' sense of quantification (to use Quine's terminol-
ogy).27
The details of the truth-conditions can easily be spelled out, but they
are not our concern here. They have been spelled out - to some extent at
least - elsewhere. 28
The role of individuating functions and/or their similarity with inten-
sions is sometimes overlooked in possible-worlds semantics. Sometimes
this semantics is developed by postulating a class of individuals (possible
individuals, if you prefer) which then simply show up or fail to show up
in the several possible worlds. This procedure, though entirely justified
for many purposes, is seriously oversimplified, however. 29 It hides the
processes by means of which cross-identifications are actually carried out
and which may rely on many things besides the individuals themselves,
 CARNAP'S HERITAGE 233

such as the structure of each of the two possible worlds in question and
comparisons between them. However, the postulation of possible individu-
als is not only oversimplified pragmatically. It is also oversimplified se-
mantically. The behavior of individuating functions can in principle be
such that the 'manifestations' of individuals they connect cannot simply
be appearances of one and the same individual.
For one thing, it has been argued that world lines can split when we
move from a possible world to another. 30 Although this particular point
is controversial, it seems very hard to rule out all splitting altogether.
What is more important, in some situations we have two difIerent
classes of individuating functions in operation at one and the same time. 31
Such a situation cannot be do ne justice to by simply speaking of a given
class of (possible) individuals. The functional character of individuating
functions, and hence their similarity with intensions, has to be recognized.
We simply cannot save the traditional dichotomy by considering individu-
ating functions as unproblematic dramatizations of the identity func-
tions.

3. While possible-worlds semantics thus demonstrates a major insuffi-

ciency in the traditional semantics which operates with the intension-
extension contrast, it seems to me that in a deeper sense Carnap's work
in semantics will perhaps be only enhanced by this insight. It seems to me
that there are many suggestions and ideas in his writings which will be
thrown into a sharper relief by the perspectives which possible-worlds
semantics opens. This semantics may perhaps even be said to be closer
to the spirit of Carnap's semantical ideas than the traditional intensions-
extensions contrast.
The main new perspective that opens here consists of certain increased
prospects of convincingly and systematically demonstr:ating the empirical
and perhaps also behavioristic import of both intensional concepts and
also of propositional attitudes such as belief, even when they are used
non-extensionally. Quine is undoubtedly right in emphasizing that the
two are apt to stand or falI together. Hence it suffices to discuss the notion
of belief.
In MN (pp. 53-5), Carnap proposes as the first approximation toward
interpreting belief-sentences the following paraphrase of 'John believes
that D':
234 JAAKKO HINTIKKA

(B) 'John is disposed to an affirmative response to some sentence

in some language which expresses the proposition that D'.

This analysis suffers from several difficulties. 32 Among them there are the
following:
(i) There is no guarantee that under (B) belief is invariant with respect
to intensional isomorphism (see above) as Carnap assumed. (John might
respond differently to two intensionally isomorphic sentences.)
(ii) Interpretation (B) leads to problems whenever John understands
a langua~e incompletely or wrongly. (He might assent to a sentence ex-
pressing the proposition that D thinking that it expresses something else.)
(iii) An explication along the li nes of (B) is inapplicable to unver-
balized and perhaps unverbalizable beliefs (e.g., the beliefs of a dog).
(iv) In the form (B), Carnap's criterion is largely inapplicable, because
it presupposes that the applier knows which sentences express which pro-
positions in different languages - and also in one and the same language.
Finding this out easily leads to considerations ofthe beliefsofthe speakers
ofthe languages in question. Hence (B) ought to be reformulated in terms
of John's responses to 'D' itself, not to its synonyms or L-equivalents.
The source of alI these difficuIties (except the first one, which becomes
spurious as soon as one gives up the belief in intensional isomorphism
as the touchstone of substitutivity in belief-contexts) is Carnap's reliance
in (B) on John's responses to certain sentences. In this respect, an entirely
different procedure is suggested by the possible-worlds analysis of belief.
Knowing what John believes means on this analysis knowing which pos-
sible worlds are compatible with his belief and (by implication) which
ones are not. In order to explain what it means for John to believe some-
thing one thus has to explore what this dichotomy between two different
kinds of possible worlds (in relation to John) amounts to. Now it clearly
lies close at hand here to explain it in terms of John's different reactions
to the two different kinds of worlds.
In brief, the idea is this: put John suddenly in a world incompatible
with his (current) beliefs, and he will react in one way. Put him in a world
compatible with his beliefs, and he will evince a different reaction. John
will then believe that D if he exhibits the tirst reaction in no possible world
in which it is the case that D.
This suggestion is of course oversimplitied. However, it is neither trivi-
 CARNAP'S HERITAGE 235

ally unrealistic nor subject to the difficulties which bothered Carnap's

analysis. Nor does it necessarily violate reasonable standards of empiricism
and observability. In short, it seems to open a much more promising line
of thought than the analysis of belief in terms of responses to sentences.
For instance, the problem of attributing unverbalised beliefs to people
and even animals (cf. (iii) above) reduces to the much more general
problem of spelling out the responses which distinguish doxastic alterna-
tives from other possible worlds. In fact, a dog's beliefs ar~ likely to pre-
sent a much simpler case vis-a-vis this general problem than the beliefs
of us humans, for in the case of dogs it is clear that we do not attribute
beliefs to them on the basis of what we think of as going on inside their
minds but on the basis of their characteristic behavior when a belief turns
or fails to turn out to be true.
Likewise, the problem of linguistic mistakes (cf. (ii) above) presents no
difficulties in principle. A person believes that p quite apart from his
responses to any particular sentence synonymous with 'p' if and only if
his reactions to worlds in which 'p' is true and to those in which it is false
exhibit the appropriate difference. Such a difference may even obtain be-
tween two complementary classes of possible worlds which are nof the
ranges of 'q' and 'not-q' for any 'q' in some given language L. Then a
believes a proposition which is not expressible in L.
Moreover, insofar as the a's different responses to worlds compatible
and incompatible with his beliefs can be spelled out, we have an explica-
tion of the notion of belief which even satisfies some of the stringent
methodological canons apparently adhered to by Quine. For this belief
will be more or less on a par with any old dispositional concept, and Quine
explicitly admonishes us to "remain free to alIow ourselves one by one
any general terms we like, however subjunctive or dispositional their
explanations".33
True, Quine seems to be completely happy with dispositional terms
only when the dispositions in question are believed to be somehow tied
to the physical structure of the objects they are dispositions of. 34 As long
as we are not required to spelI out this structure, however, I do not see
that this desideratum isnot satisfied in the case at hand, given some fairly
reasonable view of the physiological basis of our beliefs. (Cf. here Carnap's
remarks on the intensions of a robot in 'Meaning and Synonymy in Nat-
ural Languages'.)
236 JAAKKO HINTIKKA

Moreover, Quine's desideratum is hardly a reasonable one in the first

place, if it is intended to imply that some unique structure is present in
aH cases of the dispositional concept. Surely computer scientists can legit-
imately speak of the software of computers without committing them-
selves to a particular way of realizing them in actual hardware. Statements
like 'such-and-such an item of information is stored in the memory of a
computer' may have a weH-defined and unambiguous sense even if the
kind of memory involved is left completely at large. Hence it appears
that Quine's reservations about dispositional terms do not constitute valid
reasons for denying the possibility of explicating the concept of belief
along the lines suggested by possible-worlds semantics.
Similar remarks can be addressed to intensional concepts proper in-
stead ofthe concept ofbelief. They, too, can perhaps be analyzed in terms
of an informed language-user's behavior in different possible worlds.
I find it rather strange that the promising new opportunities that are
opened here by possible-worlds semantics for the philosophy of language
have not been explored or commented on by philosophical semanticists.
In fact, virtuaHy the only extant discussion that can (almost) be fitted
into this framework is Carnap's own. In his highly interesting paper
'Meaning and Synonymy in Natural Languages' (MN, second ed., pp.
233-47),35 Carnap envisages a procedure of empirically determin ing not
only the extension but also the intension of a predicate. At first blush,
it looks rather analogous to the explication of belief sketched earlier, and
it seems to me that it is basically very much in the same spirit. The
extension of a predicate to a speaker is the c1ass of actual objects to which
he would apply the predicate. In order to get from this to intensions,
Carnap says, we only have to take into account also 'possible cases'.
It is here that differences come in. Carnap is thinking of people's re-
actions to possible objects or kinds of objects-these locutions are actual-
ly used by him - rather than possible worlds. Thus he relies on the dubious
notion of a possible individual, which in any case greatly restricts the
applicability ofthe procedure, for often the applicability of (say) a general
term to an individual depends on other things besides this individual it-
self. (One and the same possible f1ea can be a big f1ea in one world and
not a big flea in another, even if its size remains the same, depending on
the size of the other f1eas in the two worlds.)
This is connected with the fact that Carnap aIIows in his formulations
 CARNAP'S HERITAGE 237

the consideration of a respondent's teactions, to verbal descriptions of

non-actual but possible cases. On our explication, we are dealing with
dispositions to respond in certain way to possible situations, perhaps
ultimately including the whole 'possible world' in question. Only in this
way wiIl our intensions of predicates be of the right logical type. Once
again, Carnap fails to interpret 'possible worlds' realistically and to use
them systematically as a tool of his semantics and/or pragmatics. This
failure has probably been especially unfortunate in the present context,
for Carnap's overt reliance on language has apparentIy suggested to
critics a covert reliance on mental entities of some sort or other. The
possible-worlds reformulation makes it at once cIear that, however much
we have to rely on counterfactual considerations here, this does not
necessarily imply reliance on mental entities or other non-behavioristic
factors.
In spite of its shortcomings, 1 therefore find it exceedingly puzzling that
Carnap's paper has not made a greater impact on the philosophical com-
munity. The probable reason for this seems to be that most philosophers
have not realized how extremely demanding - not to say unrealistic -
the standards of observability are that such critics of Carnap as Quine
have been presupposing. In his recent formulations, Quine operates with
the notion of "the totality of possible observations of nature, made and
unmade" and "the totality of possible observations of verbal behavior,
made and unmade". 36 He cIaims that those notions that cause the inde-
terminacy of intensions, prominently including the notion of belief, are
underdetermined by these totalities of observations. On our explication,
however idealized it may be, this simply is not the case. The extensions
which a speaker would pick out from aII the different possible worlds wilI
determine (what he believes to be) the intension of the predicate, for this
intension is the function that determines these extensions as a furtction of
the possible world in question. This is a place in which the possible-worlds
analysis of intensional notions (and of the notion of belief) turns out to
have a powerful methodological thrust. Critics like Quine have probably
felt that somehow the procedures envisaged by Carnap, even when ampli-
fied along the Iines 1 have sketched, wiIl be powerless to exhaust the con-
tent of such apparentIy mental entities as beliefs and intensions. On the
possible-worlds analysis, the very idea of a possible world serves to gua-
rantee this kind of exhaustiveness. There cannot be any conceivable im-
238 JAAKKO HINTIKKA

port to a feature of someone's beliefs which does not show up in his

attitudes to some conceivable world or other.
The true explanation of this discrepancy between Quine and our recon-
structed Carnap is the wide gap between what Quine would count as
"possible observations" and what most other philosophers, including
Carnap, would presumably include under this heading. There is an ambi-
guity in Quine's notion of "the totality of possible observations". It would
naturally be taken, it seems to me, to refer to observations one could have
made had the course of events been different, Le., observations madein
certain different 'alternative' possible worlds. However, Quine's latest
explanations 37 show that he means possible observations of the actually
realized course of events, Le., observations that could have been made in
this actual world of ours. This is so restrictive an idea that it tends to
cast doubts on the admissibility of any dispositional terms, including the
ones Quine is himself using (e.g., 'stimulus meaning').
Furthermore, it is clear that Quine does not admit counterfactual con-
cepts referring to the past, for otherwise there would not be any problem
of separating the effect of past information from the influence of mean-
ings in people's linguistic behavior. (Cf., e.g., Word and Object, pp. 62-3.)
For in order to spelI out this crucial difference we only have to specify
what someone's behavior would have been if the information (stimula-
tions) he received earlier had been different. AII this helps to explain the
contrast between Quine's vie~s and our reconstructed Carnapianism, but
it also shows how extremely rigid a standard of empirical significance has
been presupposed by Carnap's critics and how little persuasion the
criticisms therefore are apt to produce when their basis is fully under-
stood.
Another possible reason for critics' dissatisfaction with Carnap's dis-
cussion in 'Meaning antl Synonymy in Natural Languages' is that the
pragmatical suggestions he makes there are not tied in any natural way to
his semantics. For instance, Carnap does not in fact define the intension
of a predicate as a class of possible objects (or perhaps of kinds of possible
objects), as his discussion of how to find empirically the intension of
predicate seems to presuppose. 38 This is one of the many places where
Carnap's failure to analyze his intensional concepts can be used against
him with a vengeance. When this failure is corrected, however, Carnap
turns out to be on the side of the angels, it seems to me.
 CARNAP'S HERITAGE 239

Needless to say, tremendous difficulties remain for a possible-worlds

semanticist in his attempts to demonstrate the empirical and possibly even
behavioral cbaracter of beliefs and intensions. For instance, 1 tend to
think myself that the totality of possible worlds with which one has to
operate here is a highly dubious notion, however legitimate it may be to
consider particular possible worlds one by one or even certain restricted
sets of possible worlds. For another thing, there clearly is no unique,
easily characterizable response which would separate worlds compatible
with someone's beliefs from those incompatible with it. Rather, belief
must somehow be construed as a theoretical term. Furthermore, 1 have
not said anything constructive in the present paper about the problems
due to the failure of the identity of intensions to be a sufficient criterion
of substitutivity in belief-contexts.
But even so, it seems to me t~at possible-worlds semantics overwhel-
mingly suggests that Carnap was on the right track. It makes the weight
of his reply to Quine (that is what 'Meaning and Synonymy in Natural
Languages' essentially is) felt in a new way, and it puts the onus of pro-
ducing specific criticisms much more on Carnap's critics than has been
recognized in recent discussion. What is even more important, it suggests
new constructive, empirical approaches to the pragmatics of beliefs and
intensions. As such, it amounts to an important partial vindication of
Carnap vis-a-vis his critics, and shows the power of his ideas to inspire
and to guide further development of the studies to which he himself al-
ready contributed so much.

NOTES

1 Rudolf Carnap, Meaning and Necessity, University of Chicago Press, Chicago,

1947; second ed., with additions, 1956. In the sequel 1 shall refer to it as MN.
2 A distinction is made by Carnap between intensions and senses in 'The Philosopher
Replies' in The Philosophy of Rudolf Carnap, P. A. Schilpp (ed.), Open Court, La
Salle, Illinois, 1963, especially pp. 897-900. The reasons for the distinction, which does
not matter at this stage of our study, will be mentioned later.
3 GottIob Frege, 'Kritische Beleuchtung einiger Punkte in E. SchrOders Vorlesungen
uber die Algebra der Logik', Archiv fur systematische Philosophie 1 (1895) 433-56
esp. p. 455.
4 See W. V. Quine, 'Carnap on Logical Truth' in the Schilpp volume (note 2 above),
240 JAAKKO HINTIKKA

pp. 385-406 and Quine's other writings on the philosophy of language and of logic
since 1941.
5 See Quine's commemorative note on Carnap in Boston Studies in the Philosophy of
Science, VoI. 8, Roger Buck and Robert S. Cohen (eds.), D. Reidel Publishing
Company, Dordrecht, 1971.
6 Meaning and Necessity, first chapter.
7 Schilpp volume (note 1 above), pp. 3-84.
8 As usual, 1 am treating (for simplicity) such placeholders for sentences as 'S' and
such placeholders of individual expressions as 'i' (see below) as iT they were themselves
sentences or individual expressions, respectively. I shall also let quotes be absorbed
into such functions as R and 1 (for the latter, see below).
9 The role of possible-worlds semantics in 'providing structural analyses of various
intensional concepts has been stressed especially forcefully by Richard Montague.
See, for instance, 'Pragmatics' in Contemporary Philosophy - La philosophie contempo-
raine, VoI. 1, R. Klibansky (ed.), La Nuova Italia Editrice, Florence 1968, pp. 102-22;
'On the Nature of Certain Philosophical Entities', The Monist 53 (1969) 159-94; 'Prag-
matics and Intensional Logic', Synthese 22 (1970-71) 68-94.
10 The tirst to carry out systematically this liberalization seems to have been Saul
Kripke.
11 See, for instance, 'Existential Presuppositions and Uniqueness Presuppositions' in
Jaakko Hintikka, Models for Modalities: Selected Essays, D. Reidel Publishing
Company, Dordrecht, 1969, pp. 112-47, and 'Individuals, Possible Worlds, and
Epistemic Logic', Nous 1 (1967) 33-62.
12 Zeitschrift fiir Philosophie /Ind philosophische Kritik, Neue Folge, 100 (1892)
25-50.
13 In R. M. Martin, Logic. Language and Metaphysics, New York University Press,
New York 1971, pp. 59-60, it is objected to this point that the difference in "the expres-
sions that do the referring" suffices as the relevant difference between the ways in which
different objects (or the same object) can be given. This objection is surely completely
foreign to Frege's intentions, for senses were for him non-linguistic entities. In 'Sinn
und Bedeutung', p. 27, he emphasizes that Sinn is independent of language and can be
shared by different expressions in one and the same Ianguage. For another thing,
Frege's very first puzzle about the epistemic difference between the identities'a = a'
and 'a = b' would have been vacuous on Martin's view.
Instead of 'the way of being given' we could also say 'how the reference is given'. It
turns out that the analysis of the relevant how-expression requires in general the
consideration of several possible worlds, just as happens in the possible-worlds seman-
tics. See, for instance, my survey 'Different Constructions in Terms of the Basic Epis-
temological Verbs' (reprinted as Chapter 1 of the present volume).
14 See note 9 above.
15 There are no restrictions to rule this out in Carnap's characterization of a model
in the Schilpp volume, pp. 890-1.
16 'Pragmatics and Intensional Logic' (note 9 above), p. 91. In general, Montague
emphasized (before any one else 1 know of) clearly and appropriately the crucial dif-
ference between possible worIds and models.
17 Meaning and Necessity, Chapter 5.

18 The first to put forward this idea explicitly in print was Stig Kanger; see his dis-
sertation Provability in Logic (Stockholm Studies in Philosophy, VoI. 1), Stockholm,
 CARNAP'S HERITAGE 241

1957. The same discovery was made independently by others, especially by Saul
Kripke.
19 In his paper, 'The method of Extension and Intension' in The Philosophy of Rudol!
Carnap (note 2 above), pp. 311-49, Donald Davidson already pleaded persuasively
for a uniform treatment of intensional contexts and belief-contexts in logical seman-
tics.
20 See my paper, 'Knowledge, BeIief, and Logical Consequence', Ajatus 32 (1970)
32-47, and the literature referred to there. (This paper is reprinted as Chapter 9 in Jaakko
Hintikka, The Intentions of Intentionality and Other New Modelsfor Modalities, D. Reidei
Publishing Company, Dordrecht, 1975.
21 Meaning and Necessity, pp. 5(H)4.
22 Cf. 'Knowledge, BeIief, and Logical Consequence' (note 20 above).
23 Note 12 above.
24 I have tried to examine the conditions on which it is valid in several papers, most
fuIly in 'Existential Presuppositions and Uniqueness Presuppositions' in Jaakko Hin-
tikka, Models for Modalities: Se/ected Essays, D. Reidel Publishing Company, Dor-
drecht, 1969, pp. 112-47.
25 A sketch is found in my paper, 'Semantics for Propositional Attitudes' in Models
for Modalities, pp. 87-111.
28 Some important further problems are discussed in Jaakko Hintikka, 'The Semantics
of Modal Notions and the Indeterminacy of Ontology', Synthese 21 (1970) 408-24,
reprinted as Chapter 2 in Jaakko Hintikka, The Intentions of Intentionality and Other New
Models for Modalities, D. Reidel Publishing Company, Dordrecht, 1975.
27 See W. V. Quine, Ontologica/ Re/ativity and Other Essays, Columbia University
Press, New York, 1969, pp. 63-67, 104-8.
28 See, e.g., the papers referred to in notes 24 and 25 above.
29 See 'The Semantics of Modal Notions and the Indeterminacy of Ontology' (note 26
above).
30 For the systematic background of this problem, see my Mode/s for Modalities,
pp. 130-3, 140. For an argument for alIowing splitting, see Gaii Stine, 'Hintikka on
Quantification and Belief', Nous 3 (1969) 349-408.
31 See my papers, 'On the Logic of Perception' in Perception and Personal Identity,
N. S. Care and R. H. Grimm (eds.), The Press of Case Western Reserve University,
Cleveland, 1969, pp. 140-73; reprinted in Models for Modalities, pp. 151-83. and
'Objects of Knowledge and Belief: Acquaintances and Public Figures', Journal of
Philosophy 67 (1970) 869-883; reprinted as Chapter 3 in Jaakko Hintikka, The Intentions
of Intentionality and Other New Models for Modalities, D. Reidel Publishing Company,
Dordrecht,1975.
32 They are discussed perceptive1y by Barbara Hali Partee in her contribution to
Approaches to Natural Language: Proceedings of the 1970 Stanford Workshop on
Grammar and Semantics, Jaakko Hintikka, Julius M. E. Moravcsik, and Patrick
Suppes (eds.), D. Reidel Publishing Company, Dordrecht, 1973, pp. 309-336; reprinted in
this volume in pp. 243-270. 1 am greatly indebted to Mrs Partee's paper.
33 Word and Object, The MIT Press, Cambridge, Mass., 1960, p. 225.
34 Cf. op. cit., p. 223.
35 First published in Philosophical Studies 7 (1955) 33-47.
36 See W. V. Quine, 'RepIies' in Words and Objections: Essays on the Work of W. V.
Quine, Donald Davidson and Jaakko Hintikka (eds.), D. Reidel PubIishing Company,
Dordrecht, 1969. (See p. 302.)
242 JAAKKO HINTIKKA

37 See 'On the Reasons for Indeterminacy of Translation', Journal of Philosophy 67

(1970) 178-83. In order to speli out fully the logic of the situation, we would have to
use relative modalities in the sense of Hilpinen and distinguish observations that are
possible for us humans to make (Ieaving the rest of the world intact) from observations
that are possible simpliciter (observations made in a possible world which differs from
the actual one also in its non-human aspects). Cf. Risto Hilpinen, 'An Analysis of
Relativised Modalities' in Philosophical Logic, J. W. Davis, D. J. Hockney, and W. K.
Wilson (eds.), Synthese Library, D. Reidel Publishing Company, Dordrecht, 1969,
pp. 181-93.
38 Meaning ami Necessity, pp. 236-40.
 BARBARA HALL PAR TEE

THE SEMANTICS OF BELIEF-SENTENCES

1. INTRODUCTION

Bar-Hillel in 1954 suggested that formal semantics as developed by such

logicians as Tarski and Carnap had achieved insights and developed
approaches which linguists might profitably make use of for the analysis
of naturallanguage. The long delay in taking up Bar-Hillel's suggestion
has stemmed in part from the rejection by some linguists (notably
Chomsky, 1955) of the claimed relevance offormal to natural semantics,
and in part from the preoccupation of linguists with the more tractable
syntax and phonology of natural language to the almost total exclusion
of serious attention to semantics. Within the last few years, however,
linguists have begun to be more concerned with semantics, and to give
more than lip service to the principle that semantic considerations should
have equal weight with syntactic ones in evaluating competing theories of
grammars. 1 The present study is a preliminary investigation into the
mutual relevance of some formal semantical notions developed by Carnap
and the natural-language syntactic theory developed by Chomsky.
The problem around which this study revolves is the analysis of sentences
whose main verbs take as objects or complements (a linguists' distinction)
sentences or propositions (a philosophers' distinction), and in particular,
the question of how closely the meaning of such a sentence is tied to the
linguistic form of the embedded clause. The verb believe is of central
historical importance for such an investigation, because it was the analysis
of belief-sentences that led Carnap to the important notion of intensional
isomorphism, but it seems worthwhile to investigate as wide a range of
such verbs as possible, because they show great variation in relevant
semantic behavior.
1 will begin with Carnap's notion of intensional isomorphism, and then
discuss belief-sentences in the light of his proposals. Then 1 will back-
track to make some informal remarks about the Chomskyan notion of

Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 243-270. AII rights reserved.
Copyright © 1975 by D. ReMel Publishing Company, Dordrecht-Holland.
244 BARBARA HALL PAR TEE

deep structure, and to suggest how it might be applied to such sentences.

Since it will be readily apparent that neither intensional isomorphism nor
deep structure provides satisfactory notions of equivalence for belief-
sentences, the problem will then be to suggest alternatives. A number of
other predicates will be examined before any generalizations are suggested.
2. CARNAP'S NOTION OF INTENSIONAL ISOMORPHISM

The notion of intensional isomorphism was introduced by Carnap in

Meaning and Necessity (1947) in an attempt to handle some problematical
aspects of the semantics of belief-sentences. In this section, with apologies
to philosophers and logicians, 1 will recapitulate some preliminary features
of Carnap's semantical systems, leading up to his statement of the
problem of belief-sentences and his suggested approach to its solution.
2.1. Equivalence and L-Equivalence

For Carnap, the specification of a semantical system S typically includes

syntactic rules of formation, semantical rules of designation for the de-
scriptive constants of the system, and semantical rules of truth for sen-
tences. In Carnap's examples the metalanguage is English, and the
rules of designation for the individual constants and predicates are trans-
lations into English; likewise the sample rule of truth for atomic sentences
simply requires that "the individual to which the individual constant
refers possesses the property to which the predica te refers." The fact that
English is taken as the metalanguage, with its own semantics presupposed,
raises immediately the question of the relevance of this kind of semantics
to the analysis of naturallanguage, but for the moment we will brush this
problem under the rug.
Two sentences A and Bare said to be equivalent in a semantical system
S if they are both true or both false in S, i.e., if A == B is true in S. Two
sentences A and Bare L-equivalent in a semantical system S if A == B is
L-true in S, i.e., A == B holds in every state description (possible world) in
S. The notion of L-truth (in S), intended as an explication of necessary or
analytic truth, amounts to truth which can be established on the basis of
the semantical rules (of S) alone. The concepts of equivalence and L-
equivalence are extended in a natural way from sentences to other desig-
nators such as predicates and individual expressions.
 THE SEMANTICS OF BELIEF-SENTENCES 245

2.2. Extension and Intension

Two designators are said to have the same extension in a semantical system
S if they are equivalent in S. Two designators have the same intension in S
if they are L-equivalent in S.
The extension of a sentence is taken to be its truth-value; of a predica te,
the c1ass it designates; and of an individual expression, the individual to
which it refers (taken as a special individual such as the null set or the
number zero in case the individual expression is a description whose
uniqueness condition fails to hold).
The intens ion of a sentence is taken to be the proposition expressed by
it; the intension of a predicate is the property it designates; the intension
of an individual expres sion is what Carnap calls an "individual concept. "
These terms are made precise by the condition for sameness stated above.

2.3. Extensional and Intensional Context

Two expressions occurring within a sentence are said to be interchangeable
if substitution of one for the other preserves the truth-value ofthe sentence.
If substitution moreover preserves the intension of the sentence, the two
expressions are said to be L-interchangeable.
A sentence A is called extensional with respect to a certain occurrence
of B within it if the occurrence of BI in A is interchangeable with any
expression B 2 which is equivalent to BI. BI is then said to occur in A in an
extensional context.
A sentence is intensional with respect to a certain occurrence of a sub-
expression if (a) the sentence is not extensional with respect to the sub-
expression and (b) the subexpression is L-interchangeable with any
L-equivalent expression. The subexpression is then said to occur in an
intensional context.
For example, a sentence constructed with any of the standard connec-
tives "', v, ::), etc., is extensional with respect to its components. A
sentence constructed of a predica te letter and an individual constant is
extensional with respect to both the predicate and the individual constant.
On the other hand, a sentence constructed with a modal operator such as
the necessity operator is not extensional, but is intensional, with respect
to the expres sion within the scope of the modal operator.
246 BARBARA HALL PAR TEE

2.4. Belief-Sentences and Intensional Isomorphism

In first-order predicate logic, all sentences are extensional with respect
to all their subparts; when the modal operator for necessity is added, all
sentences are either extensional or intensional with respect to their sub-
parts. But not all contexts are either extensional or intensional, and
among the important exceptions are sentences about beliefs. Carnap's
argument for the nonintensionality of belief-sentences leads from a con-
sideration of examples like the following:
(1) John believes that D
(2) John believes that D'.
Carnap invites us to take as object language S a part of English that
includes the predicator believes that and some mathematical terms, and
to take John's responses to questions about his beliefs as acceptable
evidence for his beliefs. Then one can find some L-true sentence for which
John professes belief, which can be taken as D (e.g., "Scott is either
human or not human"). On the other hand, as Carnap says, "Since John
is a creature with limited abilities, we shall find some L-true sentences in
S for which John cannot profess belief." Then we take as D' some such
sentence. Then D and D', both being L-true, are equivalent and L-
equivalent; yet since (1) is true and (2) is false, D and D' are. neither inter-
changeable nor L-interchangeable in the context of (1). Hence the belief-
sentence is neither extensional nor intensional with respect to its sub-
sentence D.
Carnap proposes the notion of "intensional isomorphism" as a possible
way to capture a relation sufficiently much stronger than L-equivalence to
guarantee interchangeability in belief contexts. Two sentences are inten-
sionally isomorphic if they are constructed in the same way out of elements
that are L-equivalent down to the smallest units. (Carnap gives a fuller
definition though not an exact one, since an exact one would require a
fully specified semantical system or systems within or between which
isomorphism could be defined.) Carnap then suggests that the sentence
'John believes that D' in S can be interpreted by the following semantical
sentence:
(3) [15-1]. "There is a sentence Si in a semantical system S'
 THE SEMANTICS OF BELIEF-SENTENCES 247

such that (a) Si in S/ is intensionally isomorphic to 'D' in

S and (b) John is disposed to an affirmative response to Si
as a sentence of S/." (1947, p. 62.)
Then if D and D' are two intensionally isomorphic sentences, (1) and (2)
are equivalent in S and furthermore L-equivalent in S since their equiva-
lence follows from [15-1J, which is a rule of S.
Carnap suggests that his analysis of belief-sentences might be regarded
as a first step in the logical analysis of propositional attitudes. To extend
the analysis to include terms about doubt, hope, fear, surprise, etc., would
simply require the development and refinement of other dispositional
notions analogous to the notion of "disposition to assent to a sentence"
suggested for belief-sentences.

3. PROBLEMS ABOUT BELIEF-SENTENCES

A number of distinct though interconnected problems arise in the analysis

ofbelief-sentences and other sentences about propositional attitudes. The
one 1 take as fundamental to the semantics of belief-sentences is the
question ofwhat kind of equivalence between two sentences will suffice to
make them L-interchangeable in a belief-context. The question is funda-
mental in that an answer to it would provide an important step toward the
specification of conditions for the synonymy of two belief-sentences. An
examination of some objections to Carnap's solution will lead us to the
question of what counts as evidence for the truth of a belief-sentence and
the question of whether the object of believe should be construed as a
sentence or a proposition.
3.1. The Substitution Problem
Mates (1950) constructed a counterexample to Carnap's analysis, using
an argument quite similar to Carnap's argument (stated above, 2.4) for
the nonintensionality of belief-contexts. Mates considers sentences like
the following:
(4) Whoever believes that D, believes that D.
(5) Whoever believes that D, believes that D'.
where D and D' are abbreviations for two different sentences which are
 248 BARBARA HALL PARTEE

intensionally isomorphic (or synonymous by any other explication).

According to Carnap's account, if D and D' are intensionally isomorphic,
then so are (4) and (5). Hence (4) and (5) would be L-equivalent. But as
Carnap (1954) acknowledges, "However, while [4] is certainly true and
beyond doubt, [5] may be false, or, at least, it is conceivable that somebody
may doubt it." The argument holds for any relation between D and D'
short ofinscriptional identity, so it is not simply a matter oflooking for a
relation a little stronger than intensional isomorphism.
But Carnap (1954) gets out of Mates's trap by a ploy he attributes to
Church, namely, by revoking his earlier stipulation that we take a person's
disposition to assent to a sentence as definitive evidence for his belief. If
that assumption is abandoned, and belief taken instead as a theoretical
construct for which many kinds of evidence may offer inductive support,
one can simply require (5) to be true for intensionally isomorphic D and
D', and discount any apparent evidence to the contrary. The question of
what constitutes evidence for belief is therefore of fundamental impor-
tance, and will be taken up in the next section.
3.2 . Evidence for Betief
3.2.1. Assent ta sentences. Carnap's assumption in Meaning and Necessity
(1947) that a person's beliefs can be inferred directly from his dispositions
to affirmative responses to presented sentences is not quite the same as
equating what a person believes with what he says he believes, since the
subject's understanding ofthe predicate 'believes that' is not brought into
the matter at all on Carnap's analysis. But in addition to letting in Mates's
problem, Carnap's· assumption has other undesirable consequences,
which also need to be avoided by an adequate theory of belief-sentences.
(a) Nonreductionism. One kind of objection to the assumption is voiced
by Carnap (1954), namely, that a term like 'believes' cannot be reduced to
any equivalent expres sion in the language of observables. This is, I
suppose, an aspect.of the general inadequacies of logical positivism, and
is related to Chomsky's insistence on linguistic competence (a theoretical
construct) as opposed to performance (directly observable data) as the
basis for linguistic theory.
One example of the possible misleadingness of disposition to assent
as criterial for belief would be the chronic liar. (lf his dispositions to assent
 THE SEMANTlCS OF BELIEF-SENTENCES 249

realIy reflected his beliefs, he should presumably be called not a liar, but a
fool.) For another example, consider the sentence "John believes that a
stalagmite is always larger than the corresponding stalactite." Assume that
John and his interlocutor, to alI appearances, speak the same language,
namely, standard English, and that John is disposed to assent to the
embedded sentence. It may nevertheless be false that John believes that a
stalagmite is ... , because of the (nonobservable) fact that John has
unwittingly mixed up the terms stalagmite and stalactite. Perhaps it could
be argued that such differences in language are always potentialIy observ-
able, but it seems clear that at the very last, a person's language would have
to be determina bIe on the basis of purely observational data in order for a
person's beliefs to be so determinable.
(b) Dogs. If "John believes that D" means that John is disposed to
assent to a certain sentence, then we would never be allowed to assert
of a dog, at least of a nontalking dog, "Alf believes that his mistress is in
this building." This might not be too consequential a limitation, since I for
one would be wilIing to be limited to assertions like "that dog is acting as
if he believed that .... " But even that sentence does not represent any-
thing I would want to assert under Carnap's explication, since it would
mean "that dog is acting as if he were disposed to an affirmative response
to ... ," which is not at all what I ha ve in mind if 1 say that that dog is
acting as if he believed such-and-such. The things that I take as evidence
for what a dog believes (or for thinking that he belitwes anything) include
only marginally his responses to presented linguistic stimuli - a much
more typical bit of evidence for the example cited would be his refusal to
be budged from the doorway until his mistress emerged. An adequate
theory of evidence for beliefs must at least alIow someone to be able to
claim that dogs have beliefs without thereby ascribing to them any linguistic
competence.
(c) Transparency. The assent criterion is manifestly inadequate for
examples like the foIIowing, which exhibit what Russell and Quine caII
the transparent reading of the verb.

(6) The students believe that the chancellor has more power than
he reaIIy does.
(7) Tom believes that you and 1 are sisters.
250 BARBARA HALL PARTEE

(8) Jones believes that that new mistress of Smith's is Smith's

wife.
In such cases the belief-sentence does not provide an extractable embedded
sentence appropriate for presentation to the alleged believer, nor does
there seem to be any uniformly effective procedure for reconstructing an
appropriate presentation sentence. Hence any account of belief-sentences
that attempts to include the transparent cases (which are in other respects
considered 10gicalIy simpler than the opaque cases) is doomed if it requires
belief to involve assent to the actual embedded sentence (or a sentence
intensionalIy isomorphic to it).
3.2.2. Belief as a theoretical construct. Carnap's later position (1954) is
that belief is a theoretical construct for which evidence such as disposition
to assent offers inductive support but not conclusive indication. He there
leaves open the question of what other evidence counts as relevant, and
appears to leave room for a reasonable account ofliars, dogs, and those of
us who tend to confuse stalagmite and stalactite. He also, and explicitly,
leaves open the question of whether a proposition or a sentence is the
more appropriate object of belief. The main condition he imposes on the
notion ofbeliefin the later article is that the problematical Mates sentence,
(5) Whoever believes that D, believes that D',
should count as 10gicalIy true when D and D' are synonymous. Any
psychological evidence to the contrary would ipso facto be untrustworthy
evidence.
Such a stipulation is quite appropriate for the philosopher engaged in
rational reconstruction. But the linguist, although he may agree whole-
heartedly that 'believes' is a term for whose correct application no single
kind of observational evidence is criterial, is not thereby free to discount
a priori whatever observational evidence happens to conflict with his
favorite hypothesis. Hence as a linguist I am simply unable to accept
Carnap's way out of Mates's problem, since I am unable to dismiss the
apparent counterevidence (Putnam, 1954) that not everyone who believes,
for instance, that alI Greeks are Greeks believes that alI Greeks are
Hellenes. (It may be possible to argue one's way out of such counter-
evidence, as Seiiars, 1955, attempts to do, but I cannot simply discount it
as Carnap is prepared to do.)
 THE SEMA"TTICS OF BELlEF-SENTENCES 251

3.3. Inconsistent Beliefs

Carnap's approach allows the possibility of a person's holding inconsistent

beliefs. This comes out in his discussion of the example of John believing
that D but not believing that D' for various sentences D' L-equivalent to
D, when he says, "This does not necessarily mean that he commits the
error of believing their negations."2 (That is, from not necessarily I am
(pragmatically) inferring possibly.)
Some logicians (e.g., Hintikka, 1962; Montague, 1970b) would prefer
to narrow the concept of belief by putting a requirement of consistency
on a person's beliefs. This is natural from a logician's viewpoint, since it
would be much more difficult, if not impossible, to construct a formal
system which allowed a person to hold inconsistent beliefs without thereby
believing everything. 3 But a formal system which disallows inconsistent
beliefs, no matter how elegant it may be, is of dubious value as an explica-
tion of the meaning of belief-sentences in ordinary language unless it can
be argued that ali purported attributions of inconsistent beliefs to a person
are necessarily in error.
There seem to be arguments for both sides of this last question. On the
one hand, we do not normally countenance flat contradictions. If someone
makes a statement like (9) or (l0), we would be inclined to question his
sincerity, his sobriety, or perhaps his understanding ofthe word believe; I
find it quite implausible that we could ever be led to accept (9) or (l0) as
true.
(9) John believes that Scott is the author of Waverley and that
Scott is not the author of Waverley.
(10) 1 believe that 2 is not equal to 2.
On the other hand, it is not so difficult to imagine a person sanely and
sincerely asserting something like (lI), and what is more important, it
seems to me that (lI) as normally understood could be true. 4
(lI) Smith believed that ali the women at the party were accom-
panied by their (monogamous) husbands, and that there were
more women at the party than men.
The most natural explanation of a case where (lI) is true is that Smith
simply failed to draw a connection between the two beliefs. Thus this
 252 BARBARA HALL PARTEE

situation is a result of the fact that the logical consequences of a person 's
beliefs are not automatically also beliefs of his - certainly at least not
consciously so. But are a person's beliefs only his conscious beliefs, and if
not, can unconscious beliefs be inconsistent? What are we to make of
(12) as a possible response to (11)?
(12) Smith couldn't rea/ly believe both of those things, because
they're incompatible.
We are back to the evidence question by another route. What would
constitute evidence for the truth of (11)? One piece of evidence might be
Smith's first-person declaration of belief, but such assertions, like dis-
positions to assent, ought not to be accorded the staţus of conclusive
evidence. I can imagine other more indirect sorts of evidence, most of
them verbal- that is, remarks made by him and responses to remarks made
by others. But there is possible nonverbal evidence as well, pointing to the
conclusion that beliefs can beinconsistent even at a nonverballevel, and
that (12) simply expresses an unjustifiable faith in human rationality. The
sorts of nonverbal evidence I can imagine include: (i) Smith is helping in
the kitchen and gets out more pink napkins than blue ones, (ii) Smith is
helping with coats and takes the women's coats to the larger of the two
closets, (iii) Smith doesn't ask any of the women to dance although at ali
other parties he has always asked ali women not accompanied by their
husbands to dance.
In a case where a person holds inconsistent beliefs, it may not be difficult
to get him to give up one of his beliefs, but that very phraseology supports
the contention that it was previously indeed a belief of his. Hence I think
we are forced on ali counts to conci ude that a person can hold inconsistent
beliefs without thereby believing everything, and that a sentence which
asserts that someone holds inconsistent beliefs (without believing every-
thing) must therefore not be counted as necessarily false.
This conclusion may be strengthened further by showing that even dogs
could conceivably hold inconsistent beliefs. (1 am stiH not arguing that
dogs do have beliefs, only that various kinds of sentences ascribing beliefs
to dogs are not necessarily false.) For example, (13) is perfectly analogous
to (Il) above, and could have similar (though probably much less clear)
kinds of evidence :
 THE SEMANTICS OF BELIEF-SENTENCES 253

(13) My dog believed (saw, noticed) that every dog who carne to
the dog show brought his own person with him, but he believed
that there were many more dogs there than people.
It seems to me that (13) could be made plausible (though this would
not be actual evidence for it) by considering a dog who is used to seeing
crowds of people but not used to seeing crowds of dogs and therefore the
number of dogs present seems larger to him than it actually is, even though
he noticed the one-to-one correlation of dogs and people as they entered.
The consequence of this requirement that at least some inconsistent
beliefs must be allowed for is that logical equivalence cannot be a sufficient
condition for interchangeability in belief-contexts. This principle was
already asserted in Carnap (1947), but it sti II needs to be argued for,
because the method Carnap (J 954) used to get out of Mates's trap can
be extended to the case of logical equivalence, as it is, for instance, in
Montague (1970b). If the only argument in favor of allowing inconsistent
beliefs were the argument that people sometimes c1aim that they (or
others) hold beliefs which in fact are inconsistent, then the insistence on
excluding inconsistent beliefs in principle would be a valid case of rational
reconstruction (much like Chomsky's competencejperformance distinc-
tion). The argument advanced here, like much ofthe argument ofHintikka
(1 970a), is intended to show that such a limitation would not be an idealiza-
tion but a distortion.
3.4. Sentences or Propositions
The question ofwhether beliefis a relation between a person and a sentence
or between a person and a proposition is fundamental for the substitution
problem and has ramifications far beyond it besides. A review of the
distinction is in order.
By a sentence we mean a Iinguistic object, a certa in form ofwords. There
is general agreement among linguists and philosophers on this poinL
The term proposition is used somewhat less uniformly. Intuitively it is
generally taken as that which is expressed by a sentence, that which can
be said to be true or false. F ormally it has been defined as the intension of a
sentence (Carnap, 1947) and as a function from possible worlds to truth-
values (Kripke, 1963). AII these definitions have in common the important
property that logically equivalent propositions count as identica!.
254 BARBARA HALL PARTEE

The relation of sentences to propositions is many-to-many. One sentence

corresponds to many propositions if the sentence is noneternal (Quine,
1960), i.e., ifit contains indexical terms like 1, here, yesterday, etc. On the
other hand one proposition always corresponds to many sentences, in
particular to a whole equivalence class of 10gicalIy equivalent sentences.
The relation of eternal sentences (in Quine's sense) to propositions is
many-to-one.
lf belief is construed as a relation between a person and a proposition,
then logical equivalence should guarantee interchangeability, and beliefs
would have to be consistent. Carnap (1954) appears in his discussion of
Church to be claiming otherwise, but I cannot understand the argument.
If belief is construed as a relation between a person and a sentence,
then problems arise with regard to noneternal sentences. For example,
in sentences like (14) and (15), as welI as (6)-(8) above, the question arises
as to what sentence, if any, is being asserted to be in the belief-relation
with some believer.
(14) John believed that he was already here.
(15) Mary believes that she saw you yesterday.
Linguistically, the problem can be described by saying that if belief is
a relation between a person and a sentence, then it ought to be expressed
by direct quotation, and yet in fact the only linguisticalIy possible form
is indirect quotation. And there is no general way to recover a unique
direct quotation form from a given indirect quotation.
Ifwe limit our attention to that-clauses which contain etern al sentences,
we can at least temporarily avoid the problemjust mentioned. In that case,
construing belief as a relation between a person and a sentence allows us
to make interchangeability conditions as strong as we like, since we can
detine them in terms of both the form of the sentence and of the unique
proposition the sentence expresses.
Note that such a view of sentence as object of belief does not commit
one to a view that believers must ha ve any linguistic capabilities. We
can assert that a dog stands in a certain relation to a certa in sentence of
English, and if we have abandoned the assent-definition of belief, we are
quite free to adduce alI sorts of nonverbal behavior as evidence for such an
assertion (as exemplitied above in the discussion of evidence for (II)).
 THE SEMANTICS OF BELIEF-SENTENCES 255

It is possible in the sentence-object view of belief to claim that no two

sentences are interchangeable in a belief context. It is also possible to
argue for a criterion stronger than logical equivalence but less strong than
total identity. Carnap's intensional isomorphism is such a criterion,
involving the linguistic form of a sentence plus the semantic interpretation
of its smallest units. Hintikka (1970a) discusses another proposal, one
which involves the logical form of a sentence rather than its purely linguis-
tic form. Hintikka's proposal looks very promising to my untutored eye;
Carnap's does not, and it may be worthwhile to examine why.
At first glance, the notion of intensional isomorphism would seem to
capture a relation of just about the right sort of strength: excluding
arbitrary logical equivalence, but allowing interchange of very close
paraphrases. But on closer inspection it can be seen to lead to a kind of
accidentallanguage-dependence in the ascription ofbelief. Take a sentence
like "Jones believes that Smith is hungry": this could be true if Jones is an
English-speaker, but not if Jones speaks, say, Spanish or French or
German, where the noun for 'hunger' is used with the verb for 'have'. The
same sorts of nonisomorphism crop up with many other expressions which
have no morpheme-by-morpheme translations between various pairs of
languages. In fact if 'be' and 'have' are taken as primitives in English, and
their nearest equivalents in other languages are also taken as primitives,
their designation rules would undoubtedly be different in every language.
Then no belief-sentence involving 'be' or 'have', including the classic
"Jones believes that the earth is round," would be true if the believer's
language was different from the language in which the belief-sentence was
stated.
At least a partial solution to this problem would be to say that terms
like 'be' and 'have' which have a wide range of use are actually multiply
ambiguous, with the number of distinct subterms to be determined by
inter-linguistic comparisons. Most prepositions and the most common
verbs would no doubt have to be split up into many subterms if there was
to be any hope of achieving real universality; and even then it is only cases
like "the earth is round" which would be handled, not those like "Smith
is hungry." The suggested partial solution is unsatisfactory on linguistic
grounds as well. It is artificial to determine ambiguity within a language
on the basis of cross-language comparisons, since a naive language-
256 BARBARA HALL PARTEE

learner (e.g., a child) has in general no access to data from outside his own
language, and his internally constructed system therefore takes no account
of such data (except for linguistic universals which can be regarded as in
some sense 'programmed' into his innate competence.)
An argument ofthis sort is a fortiori an argument against the c1aim that
no two sentences are interchangeable in a belief-context, for that would be
to say that belief-sentences can never be translated from one language to
another. What is needed is some semantic relation, one which hugs the
surface of syntactic form so as to disallow arbitrary logical operations, but
which is at the same time independent of particular languages (assuming
that it is correct that belief-sentences can in fact often be translated from
one language to another). Hintikka's proposal appears to be of just this
sort.
At this point we should return to the problem of how to analyze belief-
sentences whose that-c1auses contain noneternal sentences. Here it seems
to me is a place where formal logic can come to the linguist's rescue. Since
we have argued above that the relation governing interchangeability is a
semantic one, and since the semantics of naturallanguages must in some
sense include quantiticationallogic even though their syntax is very differ-
ent from the syntax oflogic, a deeper semantic analysis ofthese sentences is
in order. Take for example a case like that of Hintikka's cited in footnote
4, and add indexical terms:

(16) 10hn believes that my father was an only child, and that you
are my tirst cousin on my father's side.
The pronouns in the embedded sentence are c1early not to be taken as part
of 10hn's belief in their occurring form; the relevant semantic structure
can be shown by (17):

(17) (3x) (3y) (1 am x and you are y and 10hn believes: x's father
was an only child, and y is x's tirst cousin on x's father's side.)
We ha ve thus removed the indexical terms to a purely referential position,
but have not altered the structure of the sentence expressing 10hn's
belief - in particular, the contradiction in his beliefs is no harder and no
easier to spot in (\ 7) than in (16).
There are many problems remaining, but it seems to me that the way
 THE SEMANTlCS OF BELIEF-SENTENCES 257

toward a solution must be to take the object of belief as a sentence,

with appropriate quantificational extraction of alI indexical terms (and
in fact of any other terms that are to be taken in a transparent or de re
sense), and then to try to formulate strong principles ofinterchangeability
based on the semantic structure of the sentence. By semantic structure 1
mean something which is indeed rather close to the spirit of Carnap's in-
tensional isomorphism, and rather far removed from the model theoretic
"function from possible worlds to truth-values." But 1 am not sure that
there exists any independent definition of the sort of structure 1 have in
mind. lf the Katz-Postal (1964) hypothesis that transformations preserve
meaning had held up, syntactic 'deep structure' would have been a
reasonable candidate. Perhaps the rather different sort of tree structures
generated by a Montague-grammar will turn out to be appropriate.
4. OTHER PREDICATES

Verbs and other predicates which take sentences or propositions as one

of their arguments exhibit a wide range of behavior with respect to the
kinds of maUers examined above in connection with 'believes'. We
will examine here for a number of predicates whether the wording of the
embedded sentence is taken to reflect the actual wording of some prior
discourse; whether the predicate can be applied to dogs and other non-
linguistic species; what kinds of substitutions preserve truth; and whether
the predicate seems to have two Of more well-defined senses (or perhaps
uses).
4.1. Emotives

Certain verbs and adjectives express a relation between an animate being

and what certainly appears to be a proposition rather than a sentence.
The Kiparskys (1968) caII these "emotives" and point out that the
embedded sentence is always presupposed to be a fact. Examples of
emotive predicates, alI taking that-clauses, include regret, hate, be
surprised, be amazed, be sad, be glad, be upset, be delighted. (These predi-
cates also take infinitival complements, and some of them also take
gerundives, but we wilI be concerned only with that-clauses.) The emotives
alI seem to be clearly at the propositional-object end of the scaIe", since
logical equivalence seems to be quite sufficient for interchangeability.
 258 BARBARA HALL PARTEE

4.2. Verbs of lnference

At the other end ofthe scale, though not quite so clear-cut, are verbs like
deduce, prove, establish, show, discover, which we might caII 'verbs of
inference'.
(18) John proved in 3 lines that allleft-inverses in a group are right-
inverses.
Here the embedded sentence is virtually a direct quotation (although the
pronoun-shifts and other transformations associated with indirect dis-
course do apply), and the evidence for the truth of the whole sentence
crucially involves an overt occurrence of (something very close to) the
embedded sentence. 5 For truth-preserving substitution, logical equivalence
is blatantly too weak. This is a case where intensional isomorphism might
be the right strength, although the allowable differences due to 'merely
syntactic devices' would have to be restricted to those syntactic differences
which did not count as differences within any formal system.
Some verbs of inference can take as subject a noun phrase denoting
some kind of evidence as an alternative to a subject which, as in (18),
denotes the agent of the inference.
(19) The fact that the bullets pierced the door shows that the
police couldn't see who they were shooting.
In Fillmore's case grammar terms, we can say that such verbs take a
sentential object and either an agent or an instrument (perhaps stretching
that term slightly), and either ofthe latter two can be the subject. In some
sentences like (20) below, an animate subject may in fact seem more an
'instrument' than an agent.
(20) Sam proved that he was involved when he referred to Miss
Faust as 'Maizie'.
In (19) and (20) the sentential objects of the verbs of inference are not
understood as related to any overt occurrence of related sentences, as was
the case in (18). The uses of show and prove in (19) and (20) seem to appeal
to a commonsensical notion of what follows from what, rather than to a
formal system of deduction as in (18). It is hard to formulate reasonable
candidates for substitutional criteria for cases like (19) and (20). Perhaps
 THE SEMANTICS OF BELlEF-SENTENCES 259

logical equivalence might do, but it is hard to be certain, since common-

sensical analogues of formal notions like deducibility tend to break down
when confronted with nonstandard cases. For example, I am not sure
whether (21) should count as equivalent to (20) or not:
(21) Sam proved that he wasinvolved and (that) Bill either was or
was not at the scene the previous day, when he referred to Miss
Faust as 'Maizie'.
Perhaps the clearest verb of the class is deduce, which always requires an
agentive subject and for which logical equivalence is definitely too weak
to guarantee substitutivity. lnfer, surprisingly, cannot be included in this
class, because its object-clauses seem to be propositions rather than
sentences. Thus (22) is ambiguous, but if infer were replaced by deduce,
establish, or the like, only the self-contradictory sense of the embedded
clause would be possible.
(22) Jones inferred that your yacht was longer than it was.
It is interesting that infer is nonfactive, while all the verbs that I have
called 'verbs of inference' are factive; infer seems to denote a more subjec-
tive process of inference than the others. Compare the normalcy of (23a)
with the oddity of (23b).
(23) Because he considered only Mary's remarks, he incorrectly
(a) inferred } h S
{ (b) bl
established t at am was to ame.

An incorrect inference (in the subjective sense) is an inference; an incorrect

proof is not a proof.
In sum, then, the verbs of inference (excluding infer) in their strictest
use take sentences as object and have very strong substitutivity require-
ments; they thus occupy the opposite pole from the emotives.
4.3. Verbs of Communication

One would expect that a verb like say or tell would be even more literally
quotative than a verb like deduce, but that is not the case when the verbs
are used with that-clauses, or what is normally called indirect quotation.
All of the verbs say, assert, report, allege, tell, suggest, hint, imply can fit
260 BARBARA HALL PARTEE

naturally into the frame of example (22) above without forcing the con-
tradictory reading ofthe embedded clause. In fact, while for say and most
of the others the embedded clause can be a quotation except for manda-
tory shifts in indexicals, for hint and imply, such a near-quotative reading
seems to be impossible. For example, a sentence like (24) would be
regarded as false, or at least misleading, if what Nixon actually said was
"the new South Vietnamese government will include Communists."
(24) Nixon hinted that the new South Vietnamese government
would include Communists.
(The situation is complicated by the fact, pointed out to me by Larry
Horn, that if the embedded clause has might in it, the hint can be a near-
quotation.)
It appears that these verbs, when used with a that-clause, are used to
report the content of a communication, and not its verbatim form. But
there are other verbs of communication, which could be called 'manner-
verbs of communication', which take that-clauses that seem to be what I
have been caII ing 'near-quotes', i.e., quotations except for shifts in
indexicals. Sentences (25) and (26) exemplify this class, which includes
shout, whisper, scream, hiss, hoot, giggle, bark, etc.
(25) led hollered that them brown cows was back in the corn
patch again.
(26) She giggled that she would feeI just too, too liberated if she
drank another of those naughty martinis.
Since these verbs emphasize the manner of the communication, it is not
surprising that the form as well as the content of the embedded clause
is significant. Note that the verbs cannot be analyzed as 'communicate
by giggling', etc., since (27) and the like are quite odd, unless the dog is
assumed to have a bark-Ianguage:
(27) Fido barked that someone was in the front yard.
AII ofthe verbs of communication refer, by virtue of their central meaning,
to some overt utterance or other overt communication; but only for the
manner-verbs of communication is it the case that the embedded that-
clause must be a near-quotation of the overt utterance. Thus it would
 THE SEMANTICS OF BELIEF-SENTENCES 261

seem that the normal verbs of communication take propositions as

objects, but the manner-verbs of communication sentential ones.
4.4. Epistemic Predicates
Believe shares important semantic properties with know, realize, forget,
remember, be certain, think, suppose, doubt, be aware. Chomsky (1969)
cites examples with realize that fit Mates's schema for nonsubstitutivity
of synonyms, and realize also passes the dog-test:
(28) Fido finally realized that the children were nowhere around
the house.
And although (29) is odd, unlike (6}-(8), the oddness is due simply to the
factivity of realize. Hence (30), which avoids purporting as factive a
falsity, is perfectly accepta bIe.
(29) 1 realized that your yacht was longer than it is.
(30) 1 didn't realize that your yacht was as long as it is.
With this brief attempt to put believes into perspective among verbs
that take that-clauses, let us turn to the question of the linguistic 'deep
structure' for such constructions.

5. DEEP STRUCTURE

In Chomsky's earliest formulations of transformational grammar, e.g.,

Chomsky (1957), it was emphasized that in postulating underlying rep-
resentatians and transformations, the justification must always be purely
syntactic. Examples abound of resulting syntactic analyses that are quite
unwieldy as bases for semantic interpretation. For example: (a) Syn-
tactically, the contrast between definite and indefinite noun phrases was
always regarded as simply a minimal contrast in the article position.
(b) Quantifiers, demonstratives like this and that, the articles, and words
like only and other, were alI simply introduced by phrase structure rules
as components of the 'determiner' of a noun phrase. (c) AlI adjectives
were introduced in predicate position, so that the attributive use in (31)
would be derived from (32):
(31) Small elephants are big.
(32) Elephants which are small are big.
262 BARBARA HALL PARTEE

Furthermore, since the formation of relative pronouns was assumed to

involve deletion of a noun identical to the head noun, (32) would it self
be derived from the semantically inappropriate pair of sentences (33):
(33) Elephants are smal!.
Elephants are big.
(d) In the earliest treatments, negative sentences were optional transforms
of positive ones. Klima (l964) showed purely syntactic motivation for
postulating a deep structure NEG morpheme for negative sentences, but
his system included an optional some-any suppletion transformation
(limited primarily to negative contexts) which allowed (34) and (35) to be
derived from the same underlying representation:
(34) John couldn't solve some of the problems.
(35) John couldn't solve any of the problems.
5.1. Deep Structure Semantics
The arguments of Katz and Postal (1964) convinced Chomsky (1965) that
deep structure as established by "purely syntactic motivations" would
turn out to be the only le vei of syntactic structure relevant to semantic
interpretation. That hypothesis rapidly gained favor to the point of be-
coming widely regarded as criterial rather than empirical, so that trans-
formations like Klima's some-any rule carne to be regarded as untenable.
Although semantically inappropriate analyses like those mentioned
above were not immediately replaced, the general notion of deep structure
did and does look basically quite promising as a way of coming close in
many cases to the spirit of Carnap's "merely syntactic devices." Thus if
one considers the problem of defining intensional isomorphism for
natural languages, deep structure would be the appropriate level on
which to require sameness of structure. The following pairs are typical
cases of superficially distinct structures with identical underlying rep-
resentations:
(36) (a) That Mary wore a wig surprised Timothy.
(b) It surprised Timothy that Mary wore a wig.
(37) (a) Sam turned out the light.
(b) Sam turned the light out.
 THE SEMANTICS OF BELIEF-SENTENCES 263

But it has been widely disputed oflate whether the Katz-Postal hypothesis
really holds (for a survey of s.ome of the arguments see Partee 1971).
The following sets, which would classically be regarded as transforma-
tionally related, illustrate the problem:
(38) (a) Few rules are explicit and few rules are easy to read.
(b) Few rules are both explicit and easy to read.
(39) (a) It is particularly easy to get this baby into these overalls.
(b) This baby is particularly easy to get into these overalls.
(c) These overalls are particularly easy to get this baby into.
Sentences with believe in fact offer an interesting case in point. There is a
(disputed) transformation familiarly known as 'subject-raising' which
would transform (40a) into (40b).
(40) (a) Tom believes that Cicero denounced Catiline.
(b) Tom believes Cicero to have denounced Catiline.
But Quine (1960, pp. 145-50) argues that (40a) and (40b) are not synony-
mous, in particular that only in (40b) is Cicero in purely referential
position. Yet the syntactic evidence for subject-raising is strong; among
other things, it provides the only reasonable account for sentences like
(41) and (42).
(41) Tom believes there to have been an earthquake recently.
(42) Susan believes it to be likely that no one will show up.
Sentences like those two simultaneously suggest that Quine's semantic
intuition may be wrong, since there and it certainly cannot be taken as
referential. My own feelings about (40b) are not strong, though 1 am
inclined to regard it as ambiguous, with a slightly gre ater tendency for
Cicero to be regarded as referential there than in (40a). But (43) below
seems to me definitely ambiguous, which again argues against Quine's
interpretation.
(43) John believes a Communist to have been at the heart of the
ploL
5.2. Generative Semantics

The theoretical approach which is associated with the names of McCawley ,

264 BARBARA HALL PARTEE

Lakoff, Ross, Postal, and Bach, and which often goes under the name of
'generative semantics', can be thought of as a deep structure semantics
pushed to deeper structures. Within the framework of model theoretic
semantics it makes no sense to caII those deeper structures 'semantic';
that terminology is probably just a carryover from Katz-type semantics.
At any rate, the proponents ofthe generative semantics approach suggest,
among other things, having very different deep structures·for definite and
indefinite noun phrases, assigning the same deep structure to syntactically
disparate (putative) paraphrases such as "Seymour sliced the salami with a
knife" and "Seymour used a knife to slice the salami"; assigning appro-
priately different deep structures to sets like (38) and (39); etc. One of the
key differences between Chomskyan deep structure and Lakovian abstract
structure is that Chomsky regards deep structure as the level at which
virtually ali actual lexical items are inserted, but the corresponding
terminal elements in Lakoff's system are abstract semantic primitives,
with lexical insertion a complex transformational process. Actual lexical
items are thus part of relatively superficial structure in Lakoff's system.
Since the notion of 'intensional isomorphism' is relative to the smallest
units of a system, it would have quite different interpretations in the two
systems.
It is impossible to do justice in a short space to the prolific and stimula-
ting ftow ofideas that has resulted from the generative semantics approach.
Let me then overgeneralize and say that it looked most plausible when it
presented 'semantic-looking' abstract deep structures for classically
difficult cases, accompanied by arguments for independent syntactic
justification of those semantically appropriate structures. Some basic
problems arose in attempting to solve the problem of how the syntactic
transformations would be restricted to guarantee that a given abstract
deep structure would be mapped only onto the right surface structure.
For instance, sentences (38a-b) would have a single deep structure in the
'classical' theory; it was suggested in Partee (1970) that if a generative
semanticist assigned them different deep structures, he would sti II have to
separately prevent the usual conjunction-reduction transformation from
mapping (38a) onto (38b). Lakoff's response (1970a) was in part to add the
notion of 'global constraint' to his system, so that certain aspects of the
'semantic' deep structure could in effect control the subsequent syntactic
 THE SEMANTICS OF BELIEF-SENTENCES 265

processes. Whether and how the resulting system differs from alternative
theories is debatable and debated.
5.3. Interpretive Semantics
Again oversimplifying, one might describe the 'interpretive semantics'
approach of Jackendoff, recent Chomsky, and others as an attempt to
rectify the deficiencies of the Ka tz-Postal type of deep-structure semantics,
not by pushing the syntactic deep structure deeper, but by letting semantic
rules take account of surface structure as well as deep structure, with the
latter determined on the basis of 'purely syntactic arguments' as in the
earlier theory. Then sets like (38a-b) and (39a-c) could still have single
deep structures, but those structures would no longer be purported to be
the sole determinants ofmeaning. Interpretive semantics, or the 'surfacist'
approach, does not mean using only surface structure; and on some
accounts it may allow all levels of structure to be involved - e.g., some
semantic rules might operate in the transformational cycle. It is partly for
this reason that it is difficult to establish whether generative semantics
(with global constraints) and interpretive semantics might be different.
5.4. Application to Verbs of Propositional Attitude
The semantic problem to which linguists havc given the most attention
is the problem of how to show 'underlying' or 'deep' semantic similarities
among sentences which have 'superficially' dissimilar forms. 1 think the
reason for this has been that this is the kind of regularity transformational
grammar is particularly adept at capturing, as has been emphasized re-
peatedly with examples like "John is easy to please" vs. "John is eager to
please," etc.
Linguists have also accepted the tenet that the meaning of a sentence
should be a function ofthe meanings ofits parts, and have, 1 think, tended
to construe that tenet rather narrowly. In particular, the notion of seman-
tics as a 'deep level' has led to a conception of semantic structure as a
combinatorial function of 'deep' structures of component parts of a
sentence.
The interpretivists have been arguing against such a view, and 1 think
that my arguments to the effect that the object of believe is a sentence
rather than a proposition tend in the same direction. This alternative view
266 BARBARA HALL PARTEE

may perhaps be represented by the tenet that the meaning of a sentence is

a function of the form and the meanings of its parts. (1 believe that the
earlierformulation could be read this way, but sometimes is not.) Chomsky
(1969) illustrates this principle with the verb realize, and it is further
exemplified above by the indirect quotation construction with manner-
verbs like holler and giggle, and by the verbs of inference.
Perhaps the clearest cases of the relevance of ('surface') form to
meaning are cases with explicit quotation. Consider, for example, a
sentence like (44):
(44) 'Slurp' is an onomatopoeic word.
Such sentences are not novel to philosophers (cf. "Giorgione was so-called
because ofhis size," Quine, 1960, p. 153), but I do not believe that linguists
have given much thought to how to represent them. Clearly the meaning
of the subject of sentence (44) is not just a combination of the 'meaning'
of slurp with the 'meaning' of quotation marks, or the meaning would be
the same with any synonym substituted. The form, in this case the phonetic
form, of the word slurp must be included in the meaning of the result. In
some cases, such as (45) below, it might be argued that only the phonetic
form is involved, so that one could consider representing the quoted
material simply as a syntacticalIy unanalyzed phonological string.
(45) The little engine went, "Puff-puff, chug-chug, toot-too1. "6
But in a sentence like (44), the meaning of the whole clearly involves both
the phonetic form and the meaning of the word slurp; the predicate
onomatopoeic explicitly concerns a relation between the two.
If the meaning of a quoted word is to be constant, and independent of
the context, then it must always be taken to include alI aspects of its
form as well as its 'meaning' in the narrower sense. Different predicates
may ignore different aspects of this total meaning, i.e., may alIow inter-
changeability under different sorts of conditions. Consider the folIowing
contexts, alI of which can take quoted words:
(46) (a) "Burp" rhymes with --o

(b) "Myrrh" has as many letters as --o

(c) The Oxford Universal Dictionary does not contain words
like--.
(d) "Squuck" does not mean quite the same as --o
 THE SEMANTICS OF BELIEF-SENTENCES 267

Ifwe fit slurp into these contexts, its meaning does not change, but different
aspects of its meaning contribute to the meaning of the whole. Sentence
(46a) involves only the phonological structure of the part of the word
including and following the last stressed vowel (urp in the case of slurp);
(46b) only the spelling; (46c) is perhaps vague, but may be taken to involve
something like standards ofusage; (46d) involves 'meaning' in the narrow
sense. The fact that (44) involves both sound and meaning underscores the
impossibility oftrying to regard a quoted word as perhapsjust ambiguous
between pure form and pure meaning.
The same argument extends to quoted sentences. For instance, suppose
one were to try to represent the deep structure of (47a) as (47b), in order
to indicate that the meaning of the whole involves the surface form of the
quoted sentence.
(47) (a) Ray said, "Seymour sliced the salami with a knife."
(b) S

~
NP vp

I
N
~NP
V

I
Ray
I
sald
I
N?

ISlymor slayst 2b salamlY wid a nayf I

1 have no idea what category could be suggested for the quotation; the
essential feature is the hypothesis that there is no internal syntactic
structure. Such a proposal may be appropriate for sentences like (45),
but when the quotation is of an English sentence, the syntactic and
semantic content of the quoted sentence is also relevant to the meaning of
the total sentence. Consider, for instance, how we understand she in
(48a), one in (48b), the ellipsis in (48c), and the word opposite in (48d).
(48) (a) When the surgeon shouted, "1 need the nurse!", she carne
right in.
268 BARBARA HALL PARTEE

(b) When the surgeon said, "Give me the scalpel," she handed
him the wrong one.
(c) When he said, "Leave!", she wouldn't (--).
(d) When he said, "Turn right," she did the opposite.
Hence a structure in the style of (47b) is inadequate for lack of internal
structure; but clearly the phonological string is an indispensable part of
the total meaning, since substitution of a synonymous but nonhomoph-
onous sentence can lead to a change in the meaning of the whole.
1 ha ve gone on at length about quotation because 1 think it is simply
a clearer case ofthe sentential-object use of verbs like be/ieve, also exempli-
fied by indirect quotation with verbs like shout, and by the verbs of in fer-
ence. 1 think that the generative semantics approach has shed interesting
light on some aspects of the semantics of the propositional-object cases,
which seemed from the viewpoint of the classical linguistic theory much
harder to give semantically appropriate deep structures for. But it
appears that in the process the 'easier' sentential-object cases were made
impossible to account for.
The conclusion 1 am heading toward is that potentially, every aspect
of an em bedded sentence, from the most 'superficial' to the 'deepest', can
be relevant to the contribution that sentence makes to the meaning of the
whole of which it is a part. When a sentence is embedded as a relative
clause, it may be only its meaning in the narrow sense that contributes to
the meaning ofthe resulting nominal. But when it is embedded into a quote
context as in (47) or (48), the form ofthe sentence is a crucial component
of the total meaning. And the same holds for the verbs like holler even
without direct quotation. Furthermore, constructions which may not
ordinarily involve the superficial form of an embedded sentence in
determining the meaning of the whole may do so when a modifier like 'in
so many words' or 'in his usual verbose way' is added.
It follows that no two distinct sentences have exact1y the same semantic
potential or potential meaning, if by that we mean ali those aspects of a
sentence which could contribute to the meaning of a larger whole. Katz-
Postal or Lakoff-type 'semantic representations' should be viewed as
offering additional structural information about a sentence beyond its
superficial form, but not the whole semantic story.
 THE SEMANTICS OF BELIEF-SENTENCES 269

1 would therefore recast the distinction drawn above between verbs

which take propositions as objects and those which take sentences. 1 would
say instead that an embedded that-clause always contains a whole sentence,
but that different contexts require different aspects of the semantic
potential (or 'total meaning' or 'form plus meaning') ofthe sentence to be
taken into account. For some verbs, the only relevant part ofthe meaning
of an embedded sentence may be the proposition it expresses; for others,
various aspects of the form of the sentence may be relevant as well (or
instead). What we mean when we say that a certain verb 'takes a proposi-
tional object', then, is simply that the meaning of the total sentence is
invariant under substitutions of different sentences expressing the same
proposition.
A given sentence will participate in agreat many relations with in-
numerable other sentences; all of these relations are part of our under-
standing of a sentence, as are the truth-conditions for the sentence.
To look for 'the semantic representation' of a sentence seems to me mis-
g~ided; contexts like belief-sentences and quotation-sentences require a
much richer notion of semantics, one in which semantic structure includes
syntactic and phonological structure rather than existing as a parallel but
distinct level. 7

University of California, Los Angeles

NOTES

1 A linguist would, however, take issue with Montague's opinion that syntax is of no great
interest except as a preliminary to semantics (Montague, 1970a). In spite of the notorious
difficulty of making the notion of 'well-formed sentence' precise for naturallanguages, it is a
striking and well-confirmed fact that all natural languages share many highly specific
syntactic properties that lead the linguist toward the postulation of a notion of 'possible
naturallanguage' that is much narrower and more highly structured than the general notion
of 'possible language'. 1 would not as a linguist argue against the complaint voiced by
Montague and by Dana Scott among others that transformationallinguists have not made
much progress toward a rigorous formalization of their theories of gram mar ; there are stiH
too many unresolved problems at a preformal level. 1 would only argue against drawing an
inference therefrom that the aims and methods of linguistics are misguided and have been
unfruitful.
2 Note that Carnap is thereby rejecting the claim put forward by some linguists that sentences
of the form (2-1) and (2-2) are synonymous, a claim which is used as one argument for a rule
of 'NEG-raising':
270 BARBARA HALL PARTEE

(2-1) A believes that not-S

(2-2) A doesn't believe that S.

More recently the claim has been weakened to the following: that (2-2) is ambiguous, and
on one ofits readings is synonymous with (2-1). As a semantic claim, 1 believe it is indistin-
guishable from the claim thal (2-1) entails (2-2) and not conversely.
On the other hand, sentences like (2-2) do seem to be used to assert something stronger
than the absence of belief, at least in their most normal usage. Thus if I open a discourse with
(2-3 ),
(2-3) My three-year-old son doesn't beiieve that storks bring babies.
it is likely to be assumed by my hearers that my son has entertained the embedded proposition
and rejected it. Yet if I am asked (2-4) and answer (2-5), which is presumably simply an
elliptical form of (2-3), there will be no such assumption.
(2-4) Does your three-year-old son believe that storks bring babies?
(2-5) No, he doesn't.
The issues involved are not simple, since there are some very persuasive purely syntactic
arguments for NEG-raising. There is also the complication that first-person beiief sentences
seem to behave quite differently from third-person ones; i.e., first-person cases of (2-2) seem
much more nearly synonymous to (2-1) than third-person cases. Perhaps there is a prag-
matic explanation for the difference.
Kimball (1970) discusses beiief-sentences using mostly first-person examples, and comes
to rather different conclusions from those reached here. I am excluding what he calls the
'expressive' sense of belief from the ensuing discussion.
3 Hintikka (1 970a) describes a way of carrying out such a project, and has some very interest-
ing suggestions to make concerning possible factors involved in the failure to draw correct
inferences.
4 Hintikka (1 970a) offers a similar example of someone's believing that a's father was an only

child and Ihat a has first cousins on the father's side.

5 John Searle has pointed out to me that this claim is too strong as it stands, because of
sentences like (5-1), and that in any case it has as much to do with the phrase in 3 lines as
with the choice of verb.
(5-1) . John proved in 3 lines that arithmetic is consistent.
Nevertheless, it is worth noting that not ali verbs permit modification by phrases like in 3
Iines and the even more restrictive 'in so many words'. It seems to me that any verb which can
be modified by such a phrase cannot simply be taking a proposition as its object.
6 It is sentences like these that can make first -grade readers and the like very difficult to
analyze syntactically. Joyce Friedman and I noticed, when she was starting to write a trans-
formational grammar for Yngve's sample text Engineer Small, that the verb go as used in (45)
seems to allow as its complement any phonetic sequence that is not an ordinary sentence of
English; it is hard, ifnot impossible, to characterize that restriction in a grammar of English.
(Many teenagers now speak a dialect without that restriction on go; for them go can be used
just like say with direct quotation.)
7 l received many helpful comments and suggestions on an earlier draft of this paper, which
was presented at the conference at Stanford and also to linguistics colloquia at the University
of California at Berkeley and at San Diego. For helping me out of many philosophical
muddles 1 am grateful to Julius Moravcsik and David Kaplan. I am also indebted to Jaakko
Hintikka, who I think has already solved agreat many problems whose existence I am just
beginning to appreciate. Other useful comments carne from John Kimball, John Searle,
Chung-Ying Cheng, and Pieter Seuren.
 ASA KASHER

PRAGMATIC REPRESENTATIONS AND

LANGUAGE-GAMES: BEYOND INTENSIONS
AND EXTENSIONS*

1. THE CONCEPT OF LANGUAGE

"Throughout my life", says Rudolf Carnap in his intellectual auto-

biography, "1 have been fascinated by the phenomenon oflanguage. How
amazing and how gratifying it is that we are capable of communicating
with one another by spoken sounds or written marks, to describe facts
or express thoughts and feelings, to influence the actions of others".1
Neither the syntactic regularity nor the semantic variegation are, -in
this instance, the engaging properties of the naturallanguage, but rather
the opportunities for linguistic acts, the ability to express, to understand
and to influence. At the same time, Carnap's main contribution to the
understanding of the phenomenon of language is not in the realm of
theorization of the human linguistic activity - the sphere of prag-
maties - but actually in the realm of formal semanties and logical syntax,
an indirect contribution.
The tension between these intellectual curiosity and philosophical
activity is not an anecdotal matter only; it reflects an interesting twist
in Carnap's conception of language, this being a clearcut product of his
philosophical development. We shall attempt to clarify it by dwelling on
the double standard of pragmatics in Carnap's conception of language.
On the one hand, pragmatics is the point of departure for all the
linguistic, empirical abstractions. 2 The naturallinguistic units are human
acts in the performance of which there are intermingled, at one and the
same time, several systems of rules; among these are physiological,
psychological, and sociological systems. 3 In order to analyze and explain
meanings of sentences in a naturallanguage, a daring abstraction must
be carried out, systematically overlooking the contributions of the latter
systems to linguistic behavior: The transition from pragmatics to seman-
tics is achieved by abstraction. A step in the same category is taken in the
transition from semantics to syntax. 4
On the other hand, the language-systems whose construction Carnap

Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 271-292. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
272 ASA KASHER

had recommended, for the purpose of scientific explication and philo-

sophical analysis, do not have an explicit pragmatic component as part of
their definition. "The use of symbolic logic and of a constructed language
system with explicit syntactica/ and semantica/ ru/es is the most elaborate
and most efficient method". 5 Such language systems are defined by a
system of syntactical formation rules and a system of semantical rules of
interpretation. One can understand the reason for the absence of a
constitutive pragmatical component, when the reference is to language-
systems of science or mathematics. There is no need for pragmatical rules
oftype (1) about a community which uses such a language-system, since
there is nothing to them beyond the statement that the said community
does indeed use a certain rule of interpretation of the given language-
system. Nor is there any reason for pragmatical rules of type (2) re-
garding this community, since the occurrence of the described associa-
tions is not a constitutive part of a standard utterance or a correct
understanding.

(1) Whenever the people utter a sentence of the form ' ... ist kalt',
where ' ... ' is the name of a thing, they intend to assert that
the thing in question is c01d.
(2) When using the name 'titisee' the peop1e often think ofp1enty
of fish and good meals. 6

According to Camap's conception, both in his earlier writings as

well as in his 1ater writings on pragmatics, 7 the pragmatic rules are either
generalizations which note that in one community or another there is
prevalent use of the rules of a certain language-system or they are rules
which present the relationships between the linguistic behavior of the
members of that community according to the language-system and other
aspects of their existence - structures and behaviors, individual and
collective. In pure pragmatics the concept of assertion will be c1arified and
the relations between it and the concepts of belief and intention will be
explained, for example; 8 pure pragmatics does not supply us with a
general framework for presenting constitutive rules of use, but rather
with a segment of a meta-language.
This conception holds for the languages of science and mathematics
only. An explanation of the natural phenomenon of language demands
 PRAGMATIC REPRESENTATIONS 273

recognition of the existence of rules of use, alongside rules of formation

and interpretation, as elements of the definition of a language-system.
Since natural languages have constitutive rules of use and since "the
explicatum is intended to take the place of the explicandum",9 there is
reason to assume that there is no way of avoiding an extension of the
concept of a language-system - both for scientific, linguistic analysis
and for philosophical clarification. This extension will not surprise any-
one who has understood the transition from The Logical Syntax of
Language to Introduction to Semantics. Carnap himself, even when he
finally admitted the necessity for theoretical pragmatics, made no more
than a minor step in the direction of such extension; whereas the ties
between the semantic rules and the syntactical rules in the language-
systems are clear and tight, there is no more than a weak connection be-
tween the pragmatical analysis hinted at in Carnap (1955) and Carnap's
syntactical and semantical rules of language-systems. 1o
In the course of this paper we shall demonstrate a family of non-trivial
constitutive rules of use; the extension of the concept of a language-
system to include and explain such rules will be a substantial one, but the
connections between the proposed pragmatic framework and the
common semantic (and syntactic) systems will be clear and natural

2. BEYOND INTENSIONS: FIRST STEP

One of the deepest ideas on which Carnap's semantics is built is that

the meaning of a sentence may be adequately represented by its truth-
conditions. 11 When there is sufficient correspondence between what is
said in the sentence and the state of things in the world to which the sen-
tence refers, the sentence is true in that possible world. The semantic
value - the intension - of a sentence is a proposition, that is, a function
from an appropriate class of possible worlds to a class of truth-values.
For the sake of simplicity we shall assume that the said class of possible
worlds is that of alI possible worlds and that the class of truth-values
contains only the values 'true' and 'false', even though the two assump-
tions are not free of philosophical difficulties.
If any context is appropriate for the utterance of a sentence, then the
context will determine what is the possible world to which the uttered sen-
tence has reference. In this way a truth-value is paired with the said
274 ASA KASHER

sentence in the given context of utterance. Along with the semantic value
of the sentence, which is a function from possible worlds to truth-values,
we now find something which is seemingly a pragmatic value of a
sentence - a function from contexts of utterance to truth-values.
There are weighty reasons against the acceptance of functions of the
latter type as pragmatic values of sentences.
One reason was advanced by Stalnaker. 12 He argued that if the prag-
matic value of a sentence is a function from appropriate contexts of
utterance to truth-values, one cannot present the well-known difference
between the referential reading and the attributive reading of the definite
expression in such sentences as (3).
(3) The man in the purple turtleneck shirt is bald [uttered by
someone in a room containing exactly one man in a purple
turtleneck shirt].
If we assume, regarding any context of utterance of (3), that the only
man to whom the definite description in the sentence applies in that
context is indeed bald, then the truth-value of the sentence in its two
readings would be 'true' with reference to the context; if that man is not
bald, the truth-value of (3) in its two readings will be 'false' with respect
to that context of utterance. It would therefore be impossible to dis-
tinguish between the readings by means of the ascribed truth-values. On
the other hand as Stalnaker shows, if we first pair with the sentence cum
an appropriate context of utterance, a proposition which would express
what is said by means ofthe sentence in that context, and a possible world,
to which the sentence is related in the context, and only then would we
determine the truth-value, according to the proposition and the possible
world, then shall we be able to reflect the distinction between the two
readings. According to the referential reading, the definite expression
selects an individual in the context, and the appropriate proposition is
about this individual; according to the attributive reading, the definite
expression selects an individual in the selected possible world, as part of
the 'computation' of the value of the selected proposition for the selected
possible world.
The formal conclusion which may be derived from Stalnaker's argu-
ment and other similar arguments which he produces, is that the prag-
matic value of a sentence must be a function fromappropriate contexts of
 PRAGMA TIC REPRESENTATIONS 275

utterance to ordered pairs of propositions and possible worlds. Obviously

each such ordered pair will determine a truth-value.
The distinction between the referential reading and the attributive
reading of a definite expression becomes complicated when the latter
expres sion interacts with a modal or performative verb or with an in-
tensional verb of a propositional attitude. The definite expression in
sentence (4), for example, has three readings:
(4) Mary wants to marry the mayor of Majorca.
According to one reading, the definite expression selects an individual
in the context of utterance of (4); according to the second reading, the
definite expression selects an individual in the possible world to which
(4) is re1ated, that is, a possible world in which Mary has a desire toward
soemone in particular - to marry him; according to the third reading,
the definite expression selects an individual in each of the possible worlds
that are desire-altematives of the possible world to which (4) is related,
from Mary's standpoint. In the last instance Mary has no desire with
reference to a certain individual who exists in her actual world. 13
Example (4) indicates that not only the pragmatic value of a sentence
cannot be simply a function from contexts of utterance directly to truth-
values, but that the proposition mediating between contexts and truth-
values also cannot be simply a function from possible worlds directly to
truth-values. Here, too, there must be a mediating distinction between
various possible worlds.
In order to present systematically the distinctions required for a proper
mediation between the contexts of utterance of sentences and their
truth-values, we propose two formal changes in presenting the pragmatic
value of a sentence. First, since the context appropriate for the utterance
of any sentence determines the possible world to which the sentence is
related in that context, we may see in this possible world a constitutive
component of the context, on a par with its standard indices. 14 In
the transition from a context of utterance and a sentence to a truth-value,
the said possible world will appear as a defining part of the context and
will fill the same role intended for it previously - to determine, along with
a proposition, a truth-value.
The second change is more radical. Instead of the pragmatic value of
a sentence in an appropriate context of utterance being a proposition,
276 ASA KASHER

we propose that it be a representation of a proposition, in an appropriate

formallanguage. The mediating role of the representation would be the
former role of the proposition: to determine, together with the said ex-
ponent, a truth-value. The transition from the context to the truth-value
will parallel the previous transition: the sentence and the appropriate
context determine, firstly, a representation of a proposition; the rep-
resented proposition and the exponent of the context determine a truth-
value. However, the fact that the transition is performed not by means of
the proposition itself but by means of its representation allows us to
present a general framework into which all the required distinctions may
be set.
The pragmatic value of a sentence in an appropriate context of ut-
terance is only part of the pragmatic representation of the linguistic
unit in the elements ofwhich are refiected alI the components oflinguistic
competence - the speech act. A pragmatic representation of a speech-act
will contain an adequate representation of the context of ut1erance
(the standard indices, the exponent) and of the sentence itself. The
pragmatic value of the sentence in its context of utterance and the
representation of the context itself are the departure points of the
pragmatic theory for the description and the explication of the linguistic
act; out ofthese representations the theory is to determine the pragmatic
properties of the act - the performative role, the presuppositions of the
speaker, the implicatures. Obviously the pragmatic theory must answer,
on the basis of the pragmatic representation of the act under discussion,
the questions 'What is the proposition which the sentence expresses in a
given context ofits utterance?'; 'What are the individuals, the properties
and the possible worlds to which reference is made in the linguistic act,
directly or indirectly?'; 'What is the contribution of a certain expression,
which occurs in the sentence, to its pragmatic value?' We shall now see
how Stalnaker's distinction is to be refiected with the help of these
representations.

3. THE RELATIONAL THEORY OF BELIEF:

THE PRAGMATIC INTERPRETATION

In order to avoid quantification into a modal context, Quine 15 proposed

to distinguish between two readings of each of the verbs which form such
 PRAGMATIC REPRESENTATIONS 277

contexts - the re1ational reading and the notional reading. John Wallace
argued that the second reading is no more than a special case of the first
reading. 16 Be it as it may, there is a pragmatic interest in the relational
representation.
The belief relation is between a person, a finite sequence of individuals,
a time point and an attribute. Therefore an (approximate) representation
of (5) is (6), or, according to the notional reading - (7):
(5) Ralph believes that Ortcutt is a spy.
(6) BELIEVE (Ralph, (Ortcutt), t, x [x is a spy])
(7) BELIEVE (Ralph, (-), t, [Ortcutt is a spy]),
where the bracket notation stands for intensional abstraction, '( -)'
for an empty sequence, and 't' for a temporal argument that we ignore in
the sequel. The number of elements in the finite sequence of the second
argument ofthe beliefrelation is not fixed, nor is the number offreevaria-
bles in the attribute of the fourth argument. The only demand made of the
two is that they be equal: the belief of Ralph (he being the referent of the
first argument) at the time (which the third argument denotes) that (5)
is uttered by the speaker, is that a sequence of individuals (they being the
referents of the elements of the second argument sequence) satisfies an
attribute (denoted by the fourth argument).
The intensional abstractions are not problematic, in the present formal
framework. These are functions from possible worlds to functions from
the ranges of the attributive-variables to truth-values. In the case of an
attribute without variables - (7), for example - they are functions from
possible worlds to truth-values, that is, propositions. In representations
(6) and (7) these functions appear under some particular representations,
showing more than extensional relations between their arguments and
their values. In such a formal framework there is no particular difficulty
with intensional abstractions inside intensional abstractions, as is re-
quired in the case of Ralph's belief in his own or someone else's belief,
for example.
Let us now proceed to the representation of the various readings
of (4). According to the third, 'attributive' reading, the expression 'the
mayor of Majorca' does not refer, either in the context of utterance of
(4) or in the possible world to which the sentence is related, a possible
world in which Mary has a desire; the definite expression is part of the
278 ASA KASHER

attribute which Mary wants to apply to herself; in other words - the

attribute which applies to her in aH the desire-altematives of hers. A
natural representation of this reading would be, therefore, as foHows:
(8) WANT (Mary, <Mary), t, x [MARRY (x, (IY) (MAYOR
(y, Majorca))]).
Ifwe look at the possible world to which sentence (4) is related (in a con-
text), which is also the possible world in which (8) is said to be true, as a
possible context ofutterance for Mary, we see that her desire expression,
according to the attributive reading discussed here, involves only her
reference to herself. She wants the said attribute to apply to her - it
applies in each one of her desire-altematives. AH the elements of this
observation are shown in representation (8). We are talking about a
possible world in which Mary (as denoted by the fust argument) has a
certain desire (as denoted by the main predicate). The desire refers to an
individual - M ary herself, she being the referent of the single element of
the sequence (in the second argument). This is an individual existing in
the possible world under discussion, and had this world served as a
context of utterance for Mary in expressing her desire, she would have
performed a reference act to herself for this purpose. In each of the
desire-altematives of Mary's actual wOrld, a certain attribute applies to
the said individual - the individual is married to the one who is serving
as mayor of Majorca (as denoted by the fourth argument). The role ofthe
third argument is obvious.
We represent the second reading of (4) in a different manner:
(9) WANT (Mary, <Mary, (IY) (MAYOR (y, Majorca))),
t, xz [MARRY (x, z)]).
In this case Mary has a certain desire regarding a pair of individuals
existing in her actual world. Were this possible world to serve her as a
context of utterance for this desire-expression, she would perform an act
of reference to each member of this pair of individuals, they being she and
the one who is serving, in this possible world, as the mayor of Majorca.
Mary's desire is that a certain attribute should apply to this pair of
individuals - that one of the pair be married to the other; in other
words - in every possible world that is a desire-attemative of the actual
world (of Mary), it holds true for the members of the pair that they are
 PRAGMA TIC REPRESENTA TlONS 279

married to each other. This reading of (4) differs from the previous
reading in that the latter was attributive - the definite expres sion did not
serve for reference in any relevant context of utterance, while the second
reading has a relevant context ofutterance in which the expression which
denotes, in Mary's language, the function from possible worlds to in-
dividuals, which the expression 'the mayor of Majorca' denotes for us,
would serve her for reference to an individual in her actual world - an
individual who exists in alI of her desire-alternatives as welI. The second
reading of (4) is, therefore, a 'referential' reading.
However, the first reading of (4) is also referential, in a certain sense.
But while in the second reading the reference is performed in the possible
world to which the sentence is related (in a context); according to the
first reading, the reference is performed in the context of utterance. The
speaker is the one who performs the reference act, through the expression
'the mayor of Majorca', toward an individual in the speaker's actual
world. Another individual toward whom the speaker uttering (4) in an
appropriate context of utterance performs a reference act is, of course,
Mary. Regarding this pair of individuals the speaker c1aims that a specific
attribute applies to them in the exponent of the context - one of the
pair desires for herself and for the other member of the pair that they be
married to each other. In representing this reading of (4), along with
(8) and (9), we shall introduce the satisfaction-relation. This is a relation
between a person, a finite sequence of individuals in the person's actual
world, and an attribute. Again we request that the number of elements in
the sequence of the second argument be the same as the number of
variables in the expression of the third argument. This is, of course, a
simple extension ofthe standard satisfaction-relation offormal semantics.
The representation of the first reading of (4) will therefore be as follows:
(10) SATISFY (Speaker, (Mary, (IY) (MAVOR (y, Majorca))),
xz [WANT (x, (x, z), t, uv [MARRY (u, v)])]).
With the predicate SATISFY at our disposal we can represent the other
two readings of (4) with greater preciseness, in a manner which would
present the fact that in alI the readings, the speaker performs a reference-
act toward Mary, in the context of utterance:
(11=8*) SATISFY (Speaker, (Mary), x [WANT (x, (x), t, u
[MARRY (u, (IY) (MAVOR (y, Majorca)))])]).
280 ASA KASHER

(12=9*) SATISFY (Speaker, (Mary), x [WANT (x, (x, (IY)

(MAYOR (y, Majorca))), t, uv [MARRY (u, v)])]).

These three representations allow us to distinguish between the

various readings of (4) on the basis of the various pragmatical properties
of the readings; if we overlook temporal arguments, then each of the
main relations - WANT, SATISFY and, of course, BELIEVE, and each
modal, performative or propositional attitude verb determines a set of
possible worlds connected by a certain altemativeness-relation 17,
determined from the viewpoint of the referent of the first argument. The
second argument denotes a sequence of individuals in the actual world of
the referent of the first argument. Should this possible world serve this
referent as a context of utterance, it would perform therein a reference-
act for each of the elements in the sequence, by means of appropriate ex-
pressions. In each of the said altematives of that actual world, the at-
tribute of the last argument applies to this sequence.
The three readings of (4) show us that the distinction between the
'attributive' readings and the 'referential' readings is not absolute but is
relative to the predicates in arguments of which the definite expression
does Of does not occur. In (10) the expression 'mayor of Majorca' is
referential with regard to WANT and SATISFY; in (12) it is referential
with regard to WANT and is attributive for SATISFY; in (11) it is
attributive for both. 18
As in the representation of the pragmatical distinctions between the
various readings of (4), it is possible to represent the two readings of
(3), as follows:

(13) SATISFY (Speaker, ( - ), [BOLD ((IY) (MAN (y) &

PTS(y)))])
(14) SATISFY (Speaker, «(IY) (MAN (y) & PTS (y))),
x [BOLD (x)])
where rpTS (y)' stands for y wears a purple turtleneck shirt, and temporal
arguments are deleted.
The difference between the representations shows the difference in the
roles of the definite expression. In one case - (13) - its semantic value
is a function which is itself, and not one of its values, a component of the
proposition expres sed by the speaker, by means of (3) in an appropriate
 PRAGMATIC REPRESENTATIONS 281

context ofutterance. On the other hand, in the second instance-(14)-the

speaker performs, by means of the definite expres sion, a reference-act in
the context of utterance and only the value of the function in the context
is part of the proposition. In either case, the proposition is obtained by
applying the attribute of the last argument to the sequence of individuals
in the other argument.
These observations bring us to an interesting conclusion in the matter
ofthe objects ofbelief. Carnap felt that in the meantime there is no basis
for deciding between propositions and sentences, in this matter. 19
However, from a pragmatic viewpoint it is worth distinguishing between
various representations of propositions and there is, therefore, room
for an interim stand - representations of propositions which make it
possible to distinguish between various acts of reference and attribution
in various possible worlds. On the one hand, the interim stand is similar
to a stand which prefers propositions as belief objects, since an ap-
propriate pair of a reference-act and an attribution-act determines a
proposition. On the other hand, the interim stand is similar to a stand
which prefers sentences as belief objects, since the utterance of a sentence
in an appropriate context is a presentation of a proposition. The interim
stand somewhat alleviates the problem of individuation of belief-
objects, since the distinction between reference-acts in possible contexts
ofutterance can be reduced, for our purposes, to the distinction between
the referents, who are individuals. The main difficulty is in distinguishing
between various attributes; we see eye to eye with Hintikka (1970) on this
point, but shall not go into details here.
The transition from propositions to representations of propositions
through pairs of references and attributions seems to us to be a natural
generalization of a deep and famous idea of Frege:

It is natural ... to think of there being connected with a sign (name, combination of words,
letter), besides that to which the sign refers, which may be called the reference of the sign,
also what 1 should like to call the sense of the sign, wherein the mode of presentation is
contained. 20

From a pragmatical point of view, 'the mode of presentation' is likely

to include not only semantic elements, such as the function which, in a
context ofutterance, serves to indicate an individual or a truth-value, but
also the 'mode of presentation' of this function itself, by means of
282 ASA KASHER

reference and attribution acts in the utterance-contexts of sentences. Here

we are beyond intensions and extensions.

4. INDEFINITE EXPRESSIONS AND LANGUAGE-GAMES

In the foregoing section we presented referential and attributive readings

of definite expressions; we shall now attempt using the same general
framework for representing indefinite expressions. Referential and
attributive readings may be applied to indefinite expressions, and here
also we find the distinction between them relative to modal, performative
and propositional attitude verbs. 21 The various readings of (15) will be
represented by formulae similar to (IOHI2), with one change: where-
ever, in these representations, there appears the expression which
represents 'the mayor of Majorca', there will now appear an expression
representing 'a mayor'.

(15) Mary wants to marry a mayor.

How should the indefinite expression 'a mayor' be represented? Ob-

viously, there is no difficulty in introducing a representation of this ex-
pres sion or a mode of presenting similar expressions, but the ques-
tions are - what is the semantic value of such an expression, and what is
its pragmatic value?
It appears to us that the best way to obtain adequate answers to these
questions is to try to integrate with Hintikka'sjormallanguage-games 22
special rules which would be applied to indefinite expressions.
As a sentence is given, the game between two players has one of them
interested in showing that the sentence is true (in its context of utterance)
and the other interested in showing the opposite. The game may be re-
garded as being played by the speaker, who utters the sentence in an
appropriate context, and the listener; the former provides support for
his utterance, while the latter tries to subject it to extreme criticism, so as
to see whether the speaker's words contain flaws which may prevent the
listener from accepting them. 2 3
The game proceeds in line with the structure of the sentence, and with
every move the players go on from one structure to one of its partial
structures. If the given sentence is a negation of another sentence, the
 PRAGMATIC REPRESENTATIONS 283

game will proceed with the negated sentence, and with the players
changing their aims - the player who was interested in providing support
wiU now try to refute, and the player who was interested in refuting will
now try to provide support. If the given sentence is a disjunction, the
move will belong to the player attempting to provide support, and he will
select one of the disjuncts, according to his interests; since his aim is to
provide support to the given sentence, it would be in his interest to choose
the disjunct which, in his opinion, would provide this support. Ifthe given
sentence is a conjunction, the player trying to provide support will be
asked to provide it to each one of the conjuncts, and the move therefore
belongs to the player attempting to refute, and he thereupon selects,
according to his interests, one of the conjuncts; in line with his interests
he will choose a conjunct which, in his opinion, would refute, if there is
such.
If tlie sentence is existential - 'there are alphas' - it should be given
constructive support, that is, one has to point to such an alpha. The move
belongs to the player trying to provide support, and the act he performs is
the replacement of a proper name - standard and established or entirely
new - of an individual, instead of aU the variables bound by the exis-
tentional quantifier, and the deletion of the latter. The resulting sentence
is the data for the next move. A similar act is performed if the given sen-
tence is a universal one, except that the choice is with the player trying
to refute, acting in line with his own considerations.
If the given sentence is modal, such as 'It is necessary that such-and-
such' or 'LeRoy knows that so-and-so' or 'It is possible that so-and-so' or
'It is compatible with what LeRoy knows that such-and-such', the ap-
propriate choices are done by the players in the relevant class of possible
worlds. 'Necessary' and 'LeRoy knows' behave as universal quantifiers
over these classes while 'possible' and 'compatible with what LeRoy
knows' behave as existential quantifiers over the same classes of possible
worlds. These expressions introduce into the game an element which has
appeared in it - in the rules of the connectives and quantifiers - only in
disguised form. Each move is given not only a sentence but also a possible
world in which it has a truth-value; one player tries to show that the
truth-value of the given sentence in that possible world is 'true', while
the other player tries to prove the opposite. The possible world given at
the beginning of the game, along with the entire sentence, is the exponent
284 ASA KASHER

of the context of utterance, in which the sentence is said to be true, ac-

cording to what the speaker says.
If the given sentences in a game include also definite expressions, the
rules of the game should reflect this fact. Toward this end we shall first
define the depth of a definite expression within an expression (sentence,
formula, another definite expression). This is the number of predicates in
whose arguments the expression is to be found. The depth of the ex-
pres sion 'the mayor of Majorca' in formulae (10), (12) and (Il) is 1,2 and
3, respectively. The modal depth of a definite expression within an ex-
pression is the number of predicates which introduce altemativeness-
systems of possible worlds, such that the definite expression is to be found
in their attributive (last) arguments. Among these predicates are modal,
propositional attitude and performative verbs. For our convenience we
shall also include SATISFY among them. Therefore, the modal depth of
the expression 'the mayor of Majorca' in formulae (1O) and (Il) is O and
2, respectively. The modal depth of 'the capital of Palma' within 'the
mayor of the capital of Palma' is also O. We shall further define â pure
definite expression as a definite expression which does not contain free
variables. AlI the definite expressions in formulae (IOHI2) are pure,
but the definite expression in formula (16) is not pure:
(16) (Ex) (Ey) (X=(IZ) B(z, y}}.
And now, for the rule of definite expressions: At the start of each move,
with a sentence and a possible world given, an appropriate proper name,
established or new, is placed instead of every pure definite expression
whose modal depth in the sentence is O. The move is made according
to the rules of the game, with regard to the sentence emerging from these
substitutions 24 and the given possible world.
In order to reflect the presuppositional role of every definite expression,
the rule of definite expressions may be so formulated as to have the sub-
stitution performed by both players, that is, by mutual consent; ifthere is
no such agreement, the game ends with failure for both players. 25
The choice of the proper name for substitution in place of the definite
expression is not arbitrary, of course. The proper name must be ap-
propriate, that is, the name of the one and only individual in the popula-
tion of the given possible world to whom the definite description applies
in this world. In formula (10) the individual is chosen in the exponent of
 PRAGMATIC REPRESENTATIONS 285

the utterance-context, in formula (12) in the actual wor1d of Mary's

desire-alternative system, and in formula (11) in one of Mary's desire-
alternatives, chosen in a previous move of the game.
Heretofore the rules of the game have been simple, from the standpoint
of the information available to the players, in the course of the game: at
each stage of the game, each player knows precisely what has happened
in the game up to that stage, including the given sentence and possible
wor1d in that stage. In every instance that the rules of the game call on one
of the players to select an element - an individual, a possible wor1d - from
a class of such elements, the player possesses enough information about
the class to enable him to carry out the selection in an optimal way, in line
with his interests. 26 Such games are called 'games with perfect informa-
tion'. We shall now see that extending the rules ofthe game in such a way
as to be able to play with sentences which contain certain indefinite ex-
pressions will cause the loss of this important property of the previous
games.
Alongside the three different readings of (4), there are also three dif-
ferent readings for (15):
(4) Mary wants to marry the mayor of Majorca.
(15) Mary wants to marry a mayoT.
The three readings of (4) are distinguished one from the other by the
question of where does the function 'the mayor of Majorca' choose the
individual appropriate thereto - in the context of utterance, in the actual
wor1d of Mary's desire-alternative system, or in one of these alternatives.
The three readings of (15) are also distinct from each other with respect
to where the choice of one individual is made. However, in the case ofthe
definite expression in (4) the choice is made by virtue of this expression,
which is a description which applies, in the appropriate possible wor1d,
precisely to one element, so that it makes no difference who performs the
choice. On the other hand, in the case ofthe indefinite expression of(15),
the choice cannot be made by virtue of the expres sion itself only, so that
the question of who makes the choice is important. According to one
reading, the speaker is the one who selects a certain mayor in the context
of utterance and says something about him. According to another
reading, Mary selects such a mayor in the actual wor1d of her desire-
alternative system. According to the third reading, in the given desire-
286 ASA KASHER

alternative there is a mayor to whom Mary is married. If it is to be as-

sumed that in each of Mary's desire alternatives the marriage is mono-
gamous, then anyone who is fully familiar with any one of them is capable
of choosing the right individual. If this cannot be assumed, we shall
regard the indefinite expression (in the 'absolutely attributive' reading)
as an existential quantifier over the mayor population of the given
possible world, and we shall conduct the game accordingly.
According to the first reading, the speaker selects a certain mayor in the
context of utterance, gives him a proper name, replaces the name for the
indefinite expression, and thereby creates a new sentence for the continua-
tion of the game, but he does not perform this series of acts openly. The
identity of the selected mayor is not passed on to the listener, and the latter
goes on with the game without the proper name being passed on to him;
he keeps playing with the partial information, viz, that the selected in-
dividual is a mayor. Of course if the speaker selected a mayor, the latter
is known to him, to a certain extent 27 , and this information is known to
the listener. In order to facilitate the presentation ofthe game, we may put
the 'noise' into the given sentence, for the continuation of the game. The
indefinite expression will be replaced by a new proper name, previously
unknown to both players; for the speaker, this would be a new proper
name of an individual already known to him and whom he had chosen,
while for the listener it would be the proper name of an individual about
whom all he knows is that he is a mayor known in some way to the
speaker.
According to the second reading, Mary is the one who chooses the
mayor in the actual world of her desire-alternative system. The players
allow others - a certain portion of 'Nature' which is not subject to them -
to play, and they receive only partial information about the move which
has been made. The partial information is parallel to that which the
listener received in the foregoing reading - the chosen individual is a
mayor known to Mary in some way. Again it is possible to introduce a
new proper name which will denote the chosen individual, only to whom
Mary's move of selection is known; for the speaker and the listener this
will be the proper name of an individual who is a mayor known to Mary
in some way.
According to the third reading, a desire-alternative of Mary's is given;
in this, she is married to one of the mayors. In the beginning of the move,
 PRAGMATIC REPRESENTATIONS 287

the indefinite expression will be replaced by a variable, and an existential

quantifier will be added to the given sentence, in an appropriate position.
From here on, the game will proceed according to its standard rules. If
the quantifier is added to the sentence at its beginning, it will be the turn
of the player who is attempting to lend support to the given sentence to
play and to select a mayor, in line with his own interests.
We shall preface the indefinite expression rule with several definitions.
The modal depth of an indefinite expression within an expres sion is defined
along with the definition of the modal depth of a definite expression
within an expression, that is, according to the number of attributive
arguments of modal predicates (as well as SATISFY) in which the ex-
pres sion is to be found. According to this definition, there is no difference
in the modal depth of the expression 'the mayor of Majorca' in formulas
(17) and (18).
For the purpose of the rule of definite expressions, there is no dif-
ference between a referential argument of a modal predicate and some
argument of a non-modal predicate.

(17) WANT (Mary, «(IY) (MAYOR (y, Majorca)), Mary),

t, xz [MARRY (x, z)])
(18) MARRY (Mary, (IY) MAYOR (y, Majorca))).

Since, for the purpose of the indefinite expres sion rule, there is a dif-
ference between (17) and the similar thereto and (18) and the similar
thereto, we shall define the para-modal depth of an indefinite expression
within an expression as being the number of modal predicates (including
SATISFY) wherein the indefinite expression is found in their referential
(second) arguments. If we exchange the definite expressions· in (17) and
(18) for indefinite expressions, the para-modal depth of the expres sion
'a mayor' will be 1 in (17) and O in (18).
The definition of a pure indefinite expression will be fully parallel to
the definition of a pure definite expression.
The indefinite expression rule will not parallel the definite expression
rule, for now it is important to determine who performs each move of
selection and what is the information which each player has at every stage
ofthe game:
(a) At the beginning of each move, with a sentence and a possible world
288 ASA KASHER

given, a new proper name is substituted for every pure indefinite ex-
pression whose modal depth is Oand para-modal depth is 1. The referent
of this proper name is determined by the referent of the first argument
of the modal predicate in which the original indefinite expression ap-
peared. (Ifthis predicate is SATISFY, it would be the speaker.) We shall
call the referent of the first argument the chooser.
The chooser selects, in the appropriate population of the appropriate
possible world, one individual. The new proper name is the proper name
of this individual. The identity of the selected individual is not known to
the player, ifhe is not the chooser; what he knows, in this case, is that the
predicate which appears in the original indefinite expression applies to
the selected individual, which is known in a way to the chooser.
(b) At the beginning of each move, with a sentence and a possible
world given, a new variable is substituted for every pure indefinite ex-
pres sion whose modal depth is O and whose para-modal depth is o. We
shall add an existential quantifier of minimal scope possible binding the
new variable.
Several remarks about this rule are now in place. First, a comment on
the order of the rules. Even though the two paragraphs of this rule as
well as of the rule of definite expressions speak about what is done "at
the beginning of each move", there is no problem in the order of the rules.
The two rules may be regarded as definitions of simultaneous substitu-
tions, except for the inner order of various substitutions performed ac-
cording to the second paragraph of the indefinite expression rule; in this
case the order of the quantifiers should be determined if they have the
same matrix; however, since all of them are existential, the order may be
determined arbitrarily, let us say - according to the sequence, from left
to right, of the original indefinite expressions.
The fact that both rules talk about pure expressions actuaHy reflects
an inner order in the system of the rules of the game. First of all appropri-
ate substitutions are to be made of proper names in place of aH the free
variables in the expression, definite or indefinite, and only later will the
two rules of the last expressions be put into operation. The removal of
the free variables is done through the game rules of the quantifiers, or, if
these variables are free in the given sentence at the beginning of the game,
by means of a valuation, which is defined by the indices of the context of
utterance.
 PRAGMATIC REPRESENTATIONS 289

The indefinite expression rule may also be operated at the beginning of

each move several times, by iteration, as for the expression 'a book of a
Greek poet with interesting views'.
The indefinite expression rule (as well as the definite expres sion rule)
bears improvement in several directions. In its present form the rule does
not apply to some of these expressions - the generic, the plural. It may be
that the predication of the type 'Raymond is a member of Parliament'
may be based on one of the readings of the indefinite expression controlled
by the indefinite expression rule,28 but this point requires scrupulous
examination, which will not be made here.
We conclude this paper with two observations stemming from the
presentation of the indefinite expression rule.
First we can reply to questions about the semantic value and the
pragmatic value of indefinite expressions, these being the questions
which brought us to the framework oflanguage-games. It is important to
note that these values cannot be determined locally; unless we know
where in the sentence the indefinite expres sion appears we shall not be
able to ascribe a value to it, since the value varies with the modal depth
and the para-modal depth. This constitutes a significant strengthening of
Hintikka's thesis 29 that Frege's famous program, to the effect that the
semantic values of expressions are syntactic functions of the independent
semantic values of their components, is not feasible; there are expres-
sions to whom independent semantic values do not apply, even though
they are not ambiguous. The minimal revision which should be made in
this program is that the semantic values of expressions are syntactic
functions of the semantic values of their components, where the latter are
determined by semantic schemas, and by their positions in the syntactic
structures of the expressions which they form. Again we see that there
is reason for including, in the semantic values, meta-linguistic expres-
sions - not merely representations of functions but also schemas of func-
tions and of representations of functions.
We have not clarified, in this paper, the concept ofthe pragmatic value,
but from the discussion in this section and the foregoing section it is clear
that an adequate linguistic description is impossible without a deep
pragmatic component - not only for the representation of speech-acts of
reference but also for a presentation of the language-game which applies
also to sentences with indefinite expressions. This game, as we think we
290 ASA KASHER

have shown, has constitutive pragmatic rules, suchas the indefinite ex-
pression rule.
Finally, we shall mention a family of short and problematic sentences.
Elsewhere 30 we attempted to defend the claim that the distinction be-
tween specific and non-specific indefinite expressions appears also with-
out modal, performative or propositional attitude verbs:
(19) A picture is missing from the gallery.
In a semi-formal notation we can represent the two different readings
which we proposed for (19), as follows:
(20) SATISFY (Speaker, <a picture, the gallery), xy [MISS (x, y)J)
(21) SATISFY (Speaker, <the gallery), x [MISS (a picture, x)]).
In terms of the language-game the difference between (20) and (21) will
be in the degree of information-sharing by the speaker and listener. In
both cases the speaker would choose a picture in the context of ut-
terance, except that in the case of (21) the choice will be open and the
identity ofthe selected object will be passed on to the listener, while in the
case of (20) the choice will be hidden, and the only thing the listener will
leam is that the chosen object is a picture with which the speaker is
familiar, in a way. It s/eems to us that this is a difference which cannot be
ignored.

Te/-Aviv University

NOTES
* The author was partly supported by a grant from the Israel Academy of Sciences and
Humanities.
1 Carnap (1963, p. 67).
2 "Linguistics, in the widest sense, is that branch of science which contains ali empirical
investigation concerning language ... Pragmatics is the basis for alI of linguistics." Carnap
(1942, p. 13). See Lieb (1971, p. 101).
3 See Carnap (1939, p. 6).
4 See Carnap (1942, pp. 8-13).
5 Carnap (1963, p. 936); our italics.
6 (1) is Pragm. 1 of Carnap (1939, p. 5); (2) is based on its Pragm. 2a.
7 See examples of 'pragmatical investigations' in Carnap (1942, p. 10) and (1955).
8 Carnap (1963, p. 861).
9 Carnap (1963, p. 936).
10 This has been pointed out in Hintikka (1973, p. 393).
 PRAGMATIC REPRESENTATIONS 291

11 See Camap (1942, p. 22). In this paper we shalI overlook the question ofthe meaning of
non-indicative sentences; for a survey of common approaches and a new solution, see
Kasher (1974). Since we are interested in pragmatics, we shall overlook the main semantic
shortcoming of the Camapian semantics and proceed to a system of functions from possible
worlds to truth-values or to individuals, or to other values, as need be. For the decisive im-
portance of this step, see Hintikka (1973).
12 Stalnaker (1972).
13 The importance of such examples was pointed out in Karttunen (1969). We follow here
our related discussion in Gabbay and Kasher (1974) . .
14 To stress the important distinction between standard indices (speaker, addressee, time,
place) and that one, we dubbed it in Kasher (1974) 'exponent'. We shalI use this term in the
seque!.
15 Quine (1956).
16 Wallace (1972).
17 For the general conception involved, see Hintikka (1968, pp. 87-111).
18 The relativization of related distinctions has been suggested by Karttunen (1969). See
also Gabbay and Kasher (1974).
19 Camap (1953, p. 898). For a useful discussion ofthe Frege-Church method and Camap
methods, see Partee (1973) and Moravcsik's comments in his (1973).
20 Frege (1892; 1969, p. 57); our italics, except in the case of 'sense'.
21 Linguists distinguish between specific and non-specific indefinite expressions. There is
reason for the c1aim that these two distinctions are very similar, if not identica!. see Partee
(1972), Gabbay and Kasher (1974).
22 See Hintikka (1973a) and (1974).
23 On explicating some perplexing passages in Austin (1946), Mats Furberg formulates a
related principle of the speaker's trustworthiness, which is a 'principle of serious speech' -
"A serious utterance ... entitle[s] the audience to infer that the speaker thinks that he, when
asked to, can back it up in a way appropriate to it". (1971, p. 91)
24 As it is c1ear from such expressions as 'the mayor ofthe capital of Palma', an iteration of

substitutions may be required: first a suitable proper name will be substituted for 'the
capital of Palma', let us say 'Alpha', and then a suitable proper name will take the place of
'the mayor of Alpha'.
25 There is no sense in declaring the failure of one ofthem only or a draw, because this will
make possible a strategy which will promise achievements, which will be contrary to what
derives from the truth-values of the suitable sentences in the possible worlds under dis-
cussion.
26 What is 'enough information' here is a difficult philosophical question, with which we
shalI not deal in this paper. An optimal selection is certainly there, since in any situation
we are speaking about a sentence and a possible world in which the sentence has a truth-
value to begin with. (Ali this is on the assumption that the sentence does not contain empty
definite expressions.)
27 For an elaboration of this point, see Gabbay and Kasher (1974)
28 See Hintikka (1974a).
29 Hintikka (1975).
30 See Gabbay and Kasher (1974).

BIBLIOGRAPHY

Austin, J. L.: 1946, 'Other Minds', in Philosophical Papers, Oxford University Press, 1961.
292 ASA KASHER

Camap, R.: 1939, Foundations of Logic and Mathematics, The University of Chicago Press,
Chicago, I11.
Camap, R.: 1942, Introduction to Semantics, Harvard University Press, Cambridge, Mass.
Camap, R.: 1955, 'On Some Concepts ofPragmatics', in Meaning and Necessity (2nd ed.),
The University of Chicago Press, Chicago, 111., 1956.
Camap, R.: 1963, 'Intellectual Autobiography' and 'Replies and Systematic Expositions',
in P. A. Schilpp (ed.), The Philosophy of Rudolf Carnap, Open Court, La Salle, 111.
Frege, G.: 1892, 'Uber Sinn and Bedeutung', English translation in P. Geach and M. Black
(eds.), Translations from the Philosophical Writings of Gottlob Frege, Basil Blackwell,
Oxford, 1960.
Furberg, M.: 1971, Saying and Meaning, Basil BlackwelI, Oxford.
Gabbay, D. and Kasher, A.: 1974, 'On the Semantics and the Pragmatics of Specific and
Non-Specific Indefinite Expressions, 1', in B. Partee (ed.), Montague Grammar, Academic
Press, New York, forthcoming.
Hintikka, K. J. J.: 1969, Modelsfor Modalities, D. Reidel Publishing Company, Dordrecht,
HolIand.
Hintikka, K. J.1.: 1970, 'Knowledge, Belief, and Logica1 Consequence', Ajatus 32,32-47.
Hintikka, K. J. J.: 1973, 'Camap's Semantics in Retrospect', Synthese 25; also in this
volume.
Hintikka, K. J. J.: 1973a, Logic, Language-Games, and Information, D. Reidel Publishing
Company, Dordrecht and Boston.
Hintikka, K. J. J.: 1974, 'Quantifiers vs. Quantification Theory', Linguistic Inquiry 5, 153-
177.
Hintikka, K. J. J. : 1974a, 'On the Proper Treatment of Quantifiers in Montague Semantics',
in S. Stenlund (ed.), Logical Theory and Semantic Analysis, D. Reidel Publishing Com-
pany, Dordrecht and Boston.
Hintikka, K. J. J.: 1975, 'A Counter-Example to Tarski-type Truth Definition as Applied
to Natural Language', in A. Kasher (ed.), Yehoshua Bar-Hillel's Festschrift, D. Reidel
Publishing Company, Dordrecht and Boston, forthcoming.
Karttunen, L.: 1969, Problems of Reference in Syntax, unpublished Ph.D. thesis.
Kasher, A.: 1974, 'Mood Implicatures: A Logica1 Way of Doing Generative Pragmatics',
Theoretical Linguistics 1, 6-38.
Lieb, H.: 1971, 'On Subdividing Semiotics', in Y. Bar-Hillel (ed.), Pragmatics of Natural
Languages, D. Reidel Publishing Company, Dordrecht.
Moravcsik, J. M. E.: 1973, 'Comments on Partee's Paper', in K. J. J. Hintikka et al. (eds.),
Approaches to Natural Longuage, D. Reidel Publishing Company, Dordrecht and Boston.
Partee, B. H.: 1972, 'Opacity, Coreference, and Pronouns', in D. Davidson and G. Harman
(eds.), Semantics ofNatural Longuage, D. Reidel Publishing Company, Dordrecht.
Partee, B. H.: 1973, 'The Semantics of Belief-Sentences', in K. J. J. Hintikka et al. (eds.),
Approaches to Natural Longuage, D. Reidel Publishing Company, Dordrecht and
Boston.
Quine, W. V. : 1956, 'Quantifiers and Propositiona1 Attitudes', also in The Ways ofParadox,
Random House, New York, 1966.
Stalnaker, R. C.: 1972, 'Pragmatics', in D. Davidson and G. Harman (eds.), Semantics of
Natural Language, D. Reidel Publishing Company, Dordrecht.
Wallace, J.: 1972, 'Be1ief and Satisfaction', Nous 6, 85-95.
 RUDOLF CARNAP

NOTES ON PROBABILITY AND INDUCTION*

TABLE OF CONTENTS
Introductory Remarks
1. The Three Main Conceptions of Probability
II. The Two Explicanda
III. Preliminary Remarks on Inductive Logic
IV. Some Concepts of Deductive Logic
The Theory of Degree of COllfirmatioll
V. Fundamental Axioms (Al-A5)
VI. Regular m-Functions and c-Functions (A6)
VII. Coherence
VIII. Symmetrical c-Functions (A7)
IX. Estimation
X. The Functions c t and c*
XI. Further Axioms of Invariance (A8-AlI)
XII. Learning from Experience (AI2)
XIII. The Language fi' F with One Family F (AI3)
XIV. The Axiom of Predictive Irrelevance (Al4)
XV. The A-System (AI5)
XVI. Various c-Functions in the A-System
XVII. A Language with Two Families (AI6)
XVIII. An Infinite Domain of Individuals (Al7)
Bibliography

INTRODUCTOR Y REMARKS

I. The Three Main Conceptions of Probability

(1) The classical conception (Bernoulli, Bayes, Laplace).
(2) The frequency conception (Mises, Reichenbach; mathematical
statistics).
(3). The logical conception (Keynes, Jeffreys).
Read: Nagel, 1939, Ch. 1.

* University of California, Los Angeles, graduate course Philosophy 242, 'Probability

and Induction', fali semester 1955 (and spring semester 1960). Corrected and slightly
revised from the version published in Synthese 25 (1973) 269-298, with enlarged Bibli-
ography. Edited for publicat ion by Arthur J. Benson, California State University,
Los Angeles.

Jaakko Hilltikka (ed.), Rudolf Carnap, Logical Empiricist, 293-324. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
294 RUDOLF CARNAP

II. The Two Explicanda

There are two explicanda, both called 'probability':
(1) logical or inductive probability (probabilitYl)'
(2) statistical probability (probabilitY2)'
Read: Carnap, [Prob.] Ch. II, esp. §§ 9 and 10.
The logical concept of probability appears in three forms ([Prob.] § 8):
(a) the classificatory concept (confirming evidence),
(b) the comparative concept (higher confirmation),
(c) the quantitative concept (degree of confirmation).

III. Preliminary Remarks on lnductive Logic

Read: [Prob.] Ch. IV, in particular:
(1) Logical probability (as explicandum) is explained as a fair betting
quotient, and as an estimate of relative frequency ([Prob.] § 41).
(2) If logical probability is used, no synthetic assumption (e.g.,
uniformity of the world) is needed as presupposition for the validity
of the inductive method ([Prob.] § 41 F).
(3) Comparison of inductive and deductive logic ([Prob.] § 43).
(4) The main kinds of inductive inference ([Prob.] § 44 B):
(a) direct inference (from the population to a sample),
(b) predictive inference (from one sample to another),
(c) inference by analogy,
(d) inverse inference (from a sample to the population),
(e) universal inference (from a sample to a universal law).
(5) The use of inductive logic for the choice of a practical decision
([Prob.] §§ 50, 51). The rule of maximizing the estimate of utility. Daniel
BernoulIi's law of utility.

IV. Some Concepts of Deductive Logic

Read: [Prob.] §§ 14-20, esp. 18-20.
(1) State-descriptions (Z, § 18A; comp. individual distributions,
D26-6a). A state-description describes a (possible) state or model.
(2) The requirement of the logical independence of the primitive
predicates (§ 18B) can be abandoned if the dependences are expressed
by meaning postulates (see Carnap, [postulatesD.
(3) Families of related primitive predicates (§ 18C).
 NOTES ON PROBABILITY AND INDUCTION 295

(4) The range of a sentence (§§ 18D, 19).

(5) L-concepts (§ 20).1 write 'f- i' for 'i is L-true', and hence 'f- i=>j'
for 'i L-implies1', and 'f- i=.j' for 'i is L-equivalent to1'.
In simple languages (e.g., those used in [Prob.]) every model is describ-
able by a state-description. In richer languages this is not possible; here
the definitions of L-concepts and degree of confirmation are to be based
on models rather than state-descriptions.

THE THEORY OF DEGREE OF CONFIRMATION

V. Fundamental Axioms (Al-AS)

The axioms apply to any sentences e and h in a given language !l' (finite
or infinite). We presuppose throughout that the second argument of
c (usually e) is not L-false (see [Prob.] pp. 29Sf.).

Al. Range of values. O:;;;; c(h, e) :;;;; 1.

A2. L-implication. If f- e => h, then c(h, e) = 1.
A3. Special addition principle. If e . h . h' is L- false, then
c(h v h', e) = c(h, e) + c(h', e).
A4. General multiplication principle.
c(h.h', e) = c(h, e) >< c(h', e.h).
AS. L-equivalent arguments. If f- e =. e' and f- h =. h', then c(h, e) =
= c(h', e').

(These axioms, except for AS, are those of Shimony (19SS). They are
together equivalent to the Conventions C53-l and 2 in [Prob.] § 53. Most
axiom systems of other authors are essentially equivalent to this one;
see [Prob.] § 62.) The usual theorems of the probability calculus are
provable on the basis of these axioms. Among them are the theorems
[Prob.] TS3-la to f. ('t' is the tautology.)
VI. Regular m-Functions and c-Functions (A6)

For degrees of confirmation (d. of c.) we need a measure function for

the ranges of sentences. For this purpose we define regular m-functions
([Prob.] 55A).

Dt. m is a regular m-function for !l'N= Df

(a) for every Zi in !l'N' m(Zi) > O;
296 RUDOLF CARNAP

h
e e

e.h..
Deductive logic Inductive logic
'e L-implies h' means that the range of 'c(h, e) = 3/4' means that three-fourths
e is entirely cOntained in that of h. of the range of e is contained in that of
h. ([Prob.] § SSB.)

(b) Lim(Zi) = 1;
(c) if j is L-false, m(j) = O;
(d) if j is not L-false, m(j) = Lm(Zi) for aU Zi in the range
ofj.

D2. c is a regular c-function for fi' N = Of there is a regular m-func-

tion m s'uch that c is based upon m, i.e.

m(e.h)
c(h,e) = m(e
).
A6. RegulaTity. In a finite domain of individuals, c(h, e) = 1 only
if 1- e => h.
(This axiom corresponds to [Prob.] C53-3.)
Nul! confirmation is the d. of c. on the tautological evidence t ([Prob.]
D57-1, where the symbol 'co' is used):

D3. ct(j) = Of c(j, t).

Tl. A c-function c for .fi'N satisfies the axioms AI-A6 if and

only if c is a regular c-function.

Proo! 1. Let c be a regular c-function for fi' N' Then c satisfies A l-A6
according to [Prob.} T59-la, Ib, 11, In, lh and i, T59-5a, respectively.
2. Let c satisfy AI-A6. Then Ct is a regular m-function (by [Prob.] C53-3
and T53-1). c is based upon Ct (comp. [Prob.] § 54B, (3». Therefore c is
a regular c-function.
According to TI, the theorems stated in [Prob.) §§ 55, 57A and B, 59,
 NOTES ON PROBABILITY AND INDUCTION 297

60, and 61 for regular c-functions in finite systems .2N are provable on
the basis of Al to A6.

°
If c satisfies Al to AS, but not A6, we shall call it a quasi-regular
c-function (not in [Prob.]). In this case, C t is for some Zi; therefore,
even in .2N , c(h, e) cannot always be represented as ct(e·h)/ct(e).
(Example: the straight rule, [Prob.] p. 227.)
The following theorems are provable on axioms Al to AS; hence
they hold for all regular or quasi-regular c-functions.

T2. c(h, e) x c(i, e. h) = c(i, e) x c(h, e. i). (From A4, AS.).

T3. General division theorem, in two forms.

. . c(h,e) x c(i,e.h)
a. If c (1, e) > 0, then c (h, e. 1) = .) .
C(I, e
(From T2.)
°
b. If c(i,e) > and c(h,e) > 0, then
c (h, e. i) c (i, e. h) ) )
--- = . (From (a .
c(h, e) c(i, e)

The fraction on the left-hand side of the equation is known _as the
relevance quotient; the numerator is the posterior confirmation of hand
the denominator is the prior confirmation of h.
T4. Special division theorem. Suppose that c(i, e) > 0, c(h, e) > 0,
and c(i, e. h) = 1 (i is predictable or explainable by h). Then
c (h, e. i) 1
---,------,-- = - - (From T3b).
c(h, e) c(i, e)
See the explanations and examples for these theorems in [Prob.] §§ 60
and 61.
VII. Coherence
Informal explanation. Let X be willing to accept any system of bets in
which the betting quotients are equal to the values of a function c. If
there were a betting system such that X would suffer a loss in every
logically possible case, c would obviously be unsuitable. If there is no
such betting system, we shall call c coherent (Ramsey, De Finetti). If,
298 RUDOLF CARNAP

moreover. there is no betting system such that X would Iose in at least

one possible case and would not gain in auy possible case, we shall caII
c strictly coherent (Shimony).
We assume for the following definitions that 2 is an interpreted
language. that e and h are sentences of 2. that e is not L-false, that c is
a function whose value for any h, e is a real number, and that q and S
are real numbers (and likewise for ei> hjo qjo Si).
We represent a bet (of the person X) on h, given e, in language 2.
with the betting quotient q and the total stake Sas the ordered quintuple
(2, h. e, q, S) (without reference to X):

Dt. B is a bet = Df for some 2. h, e, q, and S, B = (2, h, e, q. S).

We represent a betting system BS ba sed on the assumption k and com-
prising the bets BI' B2' ... , Bn in language 2. in accordance with c
(i.e .• the betting quotients are determined by the values of c) as the ordered
quadruple ({Bl •... ,Bn}. 2. k. c):
D2. BS is a betting system = Df BS = (K, 2, k, c). where K = {Bi}
(i = 1•...• n). Bi = (2, hjo ei> qjo Si). k is a non-L-false
sentence in 2, each ei is either k or a conjunction containing
k as a component and is not L-false, and qi = c(h i , e;).
If X regards a bet on h. given e, with betting quotient q as fair. then he
is willing to make a corresponding bet on either side. i.e .• either for h
or against h. If e is true, the gaius are as follows (with S>O):

Gain

h for h against h
- --
(a) true (l-q) S -(1-q)S
- --
(b) false -qS qS

Thus X's bet against h can be regarded as a bet for h with negative S.
Therefore we admit S ~ O; then D3 covers both bets, for hand against
h. g (B, j) is the gain which X would obtain from his bet B if j were true.

D3. Let B be a bet (2, h, e, q. S). Letj be a non-L-false sentence

 NOTES ON PROBABILITY AND INDUCTION 299

in 2 which L-implies either e or '" e and L-implies either

h or '" h. g (B, j) = Of the value u such that
either (a) 1- j:::> e.h, and u = (1- q) S,
or (b)l-j:::>e.",h,andu=-qS,
or (c) 1- j:::> "'e, and u=O.
We define G(BS,j) as the total gainjrom the betting system BS which X
would obtain if j were true:

D4. <
Let BS be {BJ, 2, k, c) (i = 1, ... , n). Letj be a non-L-false
sentence in 2 such that, for every i, j L-implies either ei or
'" ei' and j L-implies either h i or '" h i • Then G(BS,j) = Of
2:7= 1 g (Bi,j)·
<
Let BS be {BJ, 2, k, c). Let CBS be the c1ass of the conjunctions j
such that (1) j contains as components, for each of the sentences el' ... ,
en , h1 , ... , hn , either the sentence it self or its negation but not both, and
no other components, and (2) j is compatible with k. These conjunctions
represent the possible cases on the basis of the assumption k. We shall
say that for a given BS loss is necessary if, for every conjunction j in
CBS , G (BS, j) < O; that loss is possible if, for at least one j in CBS ,
G(BS,j) < O; and that positive gain is impossible if, for every j in CBS ,
G(BS,j) ~ O. We shall say that BS is vacuous if, for every j, G(BS,j) = O.
D5. c is a coherent c-jullctioll for 2 = Df there is no betting
system in 2 in accordance with c for which 10ss is necessary
(in other words, for every betting system there is a possible
outcome without 10ss).
D6. c is a strictly coherent c-junction for 2 = Of there is no
betting system in 2 in accordance with c for which 10ss is
possible and positive gain is impossible (in other words, for
every non-vacuous betting system there is a possible outcome
with positive gain).
TI. If c is strictly coherent, it is also coherent.

T2. (Ramsey, De Finetti.) Every coherent c-function satisfies the

axioms Al to A5. In other words, if c violates at least one of
the axioms Al to AS, then there is betting system in accor-
dance with c for which 10ss is necessary.
300 RUDOLF CARNAP

Example for A4. Suppose that c violates A4 in .!Z'. Then there are
sentences e, h, and h' in .!Z' such that
c(h, e) x c(h', e.h) - c(h.h', e) =F O.
Let Cl =c(h, e), c2=c(h', e.h), c3=c(h.h', e), and let D=C1C2-C3'
We choose the betting system BS = ({BJ, .!Z', e, c) (i = 1, 2, 3),
TABLEI
Example for A4

B, g(BI,j) for the four conjunctions in CBS

i hi ei qi Si e.h.h' e.h. ~h' e. ~h.h'

e. ~h. ~h'
- ------
1 h e CI C2 (1 - CI) C2 (1 - CI) C2 - CIC2
- --
D D D D
- ------
2 h' e.h C2 1 1 - C2 - C2 O
- -- --
D D D
- ------
3 h.h' e C3 -1 - (1 - ca) C3 C3
- -
D D D D

G(BS,j) = -1 -1 -1

with e as k, and with ei' h i, and Si as specified in Table 1. By D2, q;=

= C(hi' eJ The values of g are determined by D3, and those of G by D4.
We find that, for every j, G = - 1. Thus for the chosen BS, loss is necessary.
T3. (Shimony.) If c violates A6, then there is a betting system in
accordance with c for which loss is possible and positive
gain is impossible. Therefore every strictly coherent c-function
satisfies the axioms Al to A6.
ProoJ. Suppose that c violates A6 in .!Z'N' Then there are sentences
e, h in .fl/N such that c(h, e) = 1 but e does not L-imply h, hence e. "" h is
not L-false. We take a system of one bet (.!Z'N, h, e, c(h, e), 1), and again
e as k. The two possible cases j are e. hand e. "" h. The gain is O in the
first case, and - 1 in the second. Thus 10ss is possible and positive
gain is impossible. This applies to any quasi-regular c-function, e.g. to
the straight rule (VI).
 NOTES ON PROBABILITY AND INDUCTION 301

T2 gives a validat ion for the axioms Al to A5, T3 for A6. The following
theorem shows that an analogous validation is not possible for any
further axioms. (The proof for T4 is given in Kemeny, 1955.)

T4. (Kemeny) a. Every c-function in 2 which satisfies the axioms

Al to A5, is coherent in 2.
b. Every c-function in 2 which satisfies the axioms
Al to A6, is strictly coherent in 2.
T5. a. A c-function is coherent if and only if it is regular or
quasi-regular.
b. A c-function is strictly coherent if and only if it is regular.
(From T2, T3, T4, and VI-TI.)
The classification of c-functions defined by T4 and T5 can be presented
in the form of the following table:

Axioms satisfied Type of c-function

A6 regular
Al to AS strictly coherent coherent
not A6 quasi-regular

VIII. Symmetrical c-Functions (A7)

The system Al to A6 is very weak. It determines no value of c(h, e)
except O or 1 in special cases. For any pair of factual sentences e, h such
that e L-implies neither h nor '" h, the system does not exclude any
number between O and 1 as a value of c(h, e) ([Prob.] T59-5f, see remark
on p. 323). Thus additional axioms are needed. A7 is the first of several
axioms of invariance of c(h, e) with respect to certain transformations of
e and h. These axioms represent the valid core of the classical principle
of indifference. Axiom of symmetry (with respect to individuals):

A7. c(h, e) is invariant with respect to any permutation of the

individuals.

DI. m-functions and c-functions which satisfy A 7 are said to be

symmetrical (with respect to individuals). (See [Prob.] §§ 90,
91.)
302 RUDOLF CARNAP

Read the definitions and explanations of the following concepts in

[Prob.]: Ch. III: division (D25-4), isomorphic sentences (D26-3) and
isomorphic state-descriptions (§ 27), individual and statistical distributions
(D26-6), structures (§ 27) and structure-descriptions (Str, D27-1), Q-
predicates (§ 31) and Q-numbers (§ 34).
Henceforth it is assumed, unless the contrary is stated, that c satisfies
Al to A7 and hence is regular and symmetrical. m is CI; hence c is based
onm.

Tl. Let e be isomorphic to e', and h to h'.

a. c(h, e) = c(h', e'). (From A7).
b. m(h) = m(h'). (From (a».

T2. Let i be an individual distribution for n given individuals

with respect to the division MI' ... , M k , with the cardinal
numbers ni' ... , nk •
a. The number of the individual distributions for the same
n individuals which are isomorphic to i is
n!
(i = . ([Prob.] T40-32b.)
ni!'" nk!

b. Let j be the statistical distribution corresponding to i. Then

m(j) = (i x m(i). (From Tlb).
T3 is a special case of T2.
T3. Let fi' N be a language with N individual constants and k
Q-predicates. Let Zi be a state-description in !l'N with the
Q-numbers N j (j = 1, ... , k).
a. The number of those state-descriptions in fi' N which are
isomorphic to Zi is
N!
(i = N 1 !N 2 ! ... N k !· (FromT2a.)

b. Let Stri be the structure-description corresponding to

Zi' Then m(Str;) = (i x m(Z;). (From T2b.)

Therefore a regular and symmetrical m-function for fi' N is uniquely

determined if we choose as its values for the structure-descriptions in
 NOTES ON PROBABILITY AND INDUCTION 303

.fi'N arbitrary positive numbers whose sum is 1. Then, for any Zi' m(Zi)
is determined by T3b and hence the other values by VI-Dlc and d.
The subsequent theorems T4 to T6 on the direct inductive inference
refer to the following situation. e is a statistical distribution for n given
individuals (the 'population') in .fi'N with respect to the division Mi> M 2
'i
(which is non-M1) with the cardinal numbers n 1 , n 2 • = n;/n (i = 1, 2).
h is an individual distribution for s of the n individuals (the 'sample') with
the cardinal numbers S1' S2 (Si ~ nJ h st is the statistical distribution
corresponding to h.

T4. a. c(h, e) =

(For [:], see [Prob.] D40-3.)

b. c(h.t , e) =
C:)C)-.
(:~)

(For (;), see D40-2.)

c. For given e and s, c(h. t , e) has its maximum if S1/S is equal,

or as near as possible, to '1.
d. For fixed s, let hp(p = O, ... , s) be the statistical distribution
h. t with S1 = P and S2 = S - p. Then
•
L [p x c(hp , e)] = S'1·
p=o
e. Letj be a fuII sentence of 'M1 ' with one of the n individual
constants in e. Then c(j, e)= '1.
(For proofs see [Prob.] T94-1.)
We see from T4d that, for given s, the estimate of S1 on e is S'1. Hence
the estimate of sds is r1 • T4e shows that c for a singular prediction with
'Mt' is r1. Thus for the direct inference something analogous to the
304 RUDOLF CARNAP

straight rule holds for all symmetrical regular (or quasi-regular)

c-functions.
T5. The folIowing holds app,oximately for sufficient1y large n, ni>
and n2' It holds exact1y for Iim c (n ~ (0) if Iim (nJn) = 'i'
a. c (h, e) = ,~1 X r~2.

b. Binomiallaw. c(h st , e) = (:J r~lr~2.

For proofs and explanations, see [Prob.] § 95.

We shall use the following notations in T6: (F = Jsr 1r2 ('standard

deviation'); D=Sl-sr1 (deviation of Sl from its estimate); cf>(u)=
= (l/J2n) e- u2 / 2 (thenormal function; [Prob.] D40-4a); hp as in T4d; h' is
the disjunction of sentences h p with p running from S'l - D' (or the
integer nearest to it) to S'l + D' (= Sl'); thus h' says that Sl deviates
from its estimate sr 1 to either si de by not more than D', in other words,
that Sl/S (the relative frequency of M 1 in the sample) does not deviate
from '1
by more than D' /s.
T6. The folIowing holds approximately for sufficiently large s
and n/s.

a. The normallaw.

_
c (h st, e) - J-
1
e
-6 2 /2(12 _ 1 (D)
- - cf> - •
(F 2n (F (F

b. Bernoulli-Laplace theorem

f
+0'/(1

c(h',e) = cf>(U) du.

-0'/(1

c. Bernoul!i's /imit theo,em. For fixed '1 and fixed q = D'JS,

Iim c(h', e) = 1.

T6c says the folIowing. If the sample size s increases but a fixed interval
'1 ±q around the given '1 is chosen, then c(h', e) (Le., the probability
that the relative frequency of M 1 in the sample lies within the chosen
 NOTES ON PROBABILITY AND INDUCTION 305

interval) can be brought as near to 1 as desired by making the sample

sufficiently large. For explanations and numerical examples, see [prob.]
§ 96.

IX. Estimation
Read: [Prob.] § 98 about the present situation of the problem of estima-
tion.
Definition of the general estimate-function.
Suppose that, on the basis of e, the magnitude u has n possible values:
Uh ••• , U". Let h i say that u has the value Ui (i = 1, ... , n). The c-mean
estimate of u is the weighted mean of the possible values, with their
c-values as weights:
n
Dt. est(u, e)=Df L
i= 1
CUi x c(h i , e)].

TI. A and Bare arbitrary fixed constants.

a. est (Au, e) = A x est(u, e).
b. est(u + B, e) = est(u, e) + B .
.c. est(Au + B, e) = A x est(u, e) + B.
([Prob.] TIOO-3, 4, and 5).
Analogous results do not generally hold for a non-linear function of u.
For example, in general est(u 2 , e) =F est 2 (u, e). This leads to a paradox
in the practical application of estimates ([Prob.] § 100C). The paradox
is eliminated if the rule for the determination of a decision refers to the
estimate of only one magnitude, e.g., the gain or the utility resulting from
an action.
Truth-frequency. Let K be a c1ass of s sentences il' ... ' i•. Let tf(K) be
the truth-frequency in K, i.e., the number of true sentences in K. Let
rtf(K) be the relative truth-frequency in K, i.e., tf(K)/s .

•
T2. a. est (tj, K, e) = L c (i", e). (For this proof, see [Prob.]
11=1
Tl04-2a.)

b. est(rtj, K, e) = ~ L
11=1
c(in> e). (From (a), Tia.)
306 RUDOLF CARNAP

c. If aII sentences in K have the same c-value on e, then the

estimate of rtf(K) is equal to this c-value. (From (b).)

The frequency of a property of individuals. Let K be a class of n indivi-

duals defined by enumeration. Let af(M, K) be the absolute frequency of
M in K, and rf(M, K) the relative frequency, i.e., af(M, K)/n. Let K' be
the class ofthe full sentences of'M' with those individual constants which
designate the individuals in K. Then

af(M, K) = tf(K') and rf(M, K) = rtf(K').

Therefore the resuIts T2 on estimates of tf and rtf can now be applied

to estimates of af and rf.
Direct estimation of frequency. This is based on the direct inference
(see VIII-T4). Let e, n, M lo nI' r lo s, and SI be as before (VIII-T4). Thus
e says that the rf of M 1 in the population is r l' Let K be the class of the
s individuals of the sample.

T3. a. est(af, MI' K, e)=sr 1 • (From VIII-T4d.)

b. est(rf, Mi> K, e)= r1 • (From (a), TIa.)

Predictive estimation of frequency. Here the estimate depends on the

chosen c-function. Let e be any non-L-false sentence, h a fuII sentence
of 'M' for a new individual, and K any finite, non-empty class of new
individuals.

T4. est(rf, M, K, e)=c(h, e). (From T2c.)

Thus the confirmation of a singular prediction with 'M' is equal to the

estimate of rf of M. This relation was used earlier for an informal
explanation of inductive probability ([Prob.] § 41D).

X. The Functions c t and c*

In discussions on the principle of indifference, some authors have pro-
posed to give equal a priori probabilities to alI individual distributions
(for a given domain of individuals and a given division of properties).
Other authors have proposed the same for alI statistical distributions. In
our terminology, the controversy concerns the choice of one of the folIow-
 NOTES ON PROBABILITY AND INDUCTION 307

ing two rules:

(A) AlI individual distributions have equal m-values.
(B) AlI statistical distributions have equal m-values.
However, each of these rules leads to contradictions if applied to
different divisions (see the examples in Carnap [Continuum] p. 39).
Each of the rules becomes consistent if it is restricted to one division
(for a given finite language), viz. the division of the Q-predicates, as
follows:
(A') All state-descriptions have equal m-values.
(B /) AII structure-descriptions have equal m-values.
The junc/ion ct . There is exactly one symmetrical, regular m-function
which fulfills (A'), viz. mt defined by D 1.
Let 2 N be a language with N individual constants and k Q-predicat~s.
TI. a. The number of state-descriptions in 2 N is (N = k N.
([Prob.] T40-31c.)
b. The number of structure-descriptions in 2 N is
N + k - 1) (N + k - 1)!
TN = ( k 1 = ) . ([Prob.] T40-33b.)
- N!(k - 1 !

Let ZN be any state-description in 2 N with the Q-numbers NI' ... , N k.

We define:
1
DI.
t
m (ZN) =Df e'
t mt(e.h)
D2. c (h, e) = Df --:;:-t--
m (e)

Let eN be an individual distribution for any N individuals for the

division of the k Q-predicates with the same Q-numbers NI' ... , N k (the
same as in ZN)' Let hj be a full sentence of 'Q/ for a new individual.
T2. a. m t is regular and symmetrical. (From D1.)
t 1
b. m (eN) = k N' (From Dl, since eN is isomorphic to ZN')

c. ct (h j , eN ) = l/k.
308 RUDOLF CARNAP

Proof. eN' h j is isomorphic to a state-description in fL'N+ 1; hence

m t = 1/0+ 1 (from Dl). The result is obtained by D2 and (b).
T2c shows that ct (h i' eN) is independent of eN0 It violates the principle
of learning from experience and hence is unacceptable ([Prob.) p. 565).
However, this function was proposed by C. S. Peirce, Keynes, and
Wittgenstein.
The function c*. There is exactly one symmetrical, regular m-function
which fulfills (B'), viz. m* defined by D3.
1
D3. m*(ZN.;) =Df-
7: N (i

N 1 ! ... Nk!(k - 1)!

= (From Tlb, VIII-T3a.)
(N +k - 1)!
1
T3. a. For any structure-description in fL'N, m* = -. Thus m*
7: N

fulfills (B'). (From VIII-T3b.)

b. m* is regular and symmetrical. (From D3.)
* N 1 ! ... Nk!(k - 1)!
c. m (e N ) = (N + k _ 1)! . (From D3.)

c* is based on m* :

* m*(e.h)
D4. c (h, e) = Df *
m (e)
.

T4.
*
c (h j , eN) = N j + 1.
N +k
Proof. eN • h j is isomorphic to a state-description in 2 N+1

with the Q-numbers N 1 , ... , N j + 1, ... , N k. Therefore its

m*-value is like that of eN in T3c, but with N j + 1 instead of
N j and N + 1 instead of N. Hence the result by D4.

Let 'M' be a disjunction of w Q-predicates (O<w<k) and N M be the

sum of the Q-numbers of these Q-predicates in eN0 Hence w is the logical
width of 'M' ([prob.] § 32). Let hM be a full sentence of 'M' for a new
individual.
 NOTES ON PROBABILITY AND INDUCTION 309

NM +W
T5. c*(hM , eN) = . (From T4 and A3.)
N+k
Consider a sequence of samples of increasing size N but such that
r = NM/N remains constant. Then the value of c*(hM, eN) moves from
w/k (for N = 0, i.e., tautological evidence) towards r (which is the limit
forN- (0).
For further explanations and theorems on c* see [Prob.] § 110.

XI. Further Axioms of lnvariance (A8-AU)

AS. c(h, e) is invariant with respect to any permutation of the
predicates of any family.

TI. Let F be a family of k primitive predicates 'P l ', ••• , 'Pk '. Let
h l , ... , hk be fuU sentences of these predicates with the same
individual constant, and h be the disjunction of these sentences.
a. (Lemma.) For any e, c(h, e) = 1. (From A2, since h is
L-true.)
b. Suppose that e' does not contain any predicate of F. Then
for any i( = 1, ... , k), c(h;, e') = l/k.
Proof. The k values c(h;, e') are equal (by A8). Their
sum = c(h, e') (by A3) = 1 (by (a»). Hence the assertion.
c. m(h i ) = l/k. (From (b).)

A9. c(h, e) is invariant with respect to any permutation ofJamilies

of the same size.

AlO. For non-general hand e, c(h, e) is independent of the total

number of individuals.
(AlO corresponds to the requirement of a fitting c-sequence,
[Prob.] § 57C.)
AU. c(h, e) is independent of the existence of other families than
those occurring in h or e.
XII. Learning Iram Experience (AI2)
The intuitive principle of learning from experience says that, other
things being equal, the more frequently a kind of event has been observed,
the more probable is its occurrence in the future. This is expres sed more
310 RUDOLF CARNAP

exactly in the axiom of instantial relevance (first proposed in Carnap,

[comparative]) :
A12. Suppose that e is non-L-false and non-general, and i and h
are fun sentences of the same factual, molecular predicate
'M' with distinct individual constants not occurring in e.
a •. c(h, e. i) {: c(h, e).
b. c(h, e. i) "# c(h, e).
Both ct (X) and the straight rule (VI, p. 296) fulfill part (a) of A12, but
violate part (b). With ct , i is always irrelevant for h. With the straight rule,
i is irrelevant for h if e is a conjunction of full sentences of 'M'; in this
case both c-values are 1.
TI. Let e, i, h, and 'M' be as in A12.
a. c(h, e.i) > c(h, e); i is positiveIy relevant for h on e.
b. Let j be a conjunction of n fun sentencesof 'M' (n ~ 2)
with n distinct individual constants which do not occur in e
or h. Then
c(h, e .j) > c(h, e). (From (a).)
c. c(h, e. ,..., i) < c(h, e); ,..., i is negatively relevant for h on e.
(From (a) and [Prob.] T6S-6e.)
d. c(h, e.i»c(h, e.,...,i). (From (a), (c).)

XIII. The Language 21' with One Family F (A 13)

This and the subsequent sections refer to a language 2 F whose primitive
predicates are k predicates 'P 1', ... , 'Pk ' of a family F (k ~ 2). A sentence
in 2 F may contain any number of individual constants but no variables.
eF is an individual distribution for s individuals with respect to F with

the cardinal numbers Si (i = 1, ... , k). h l , ... , hk are fun sentences of

'P l ', ... , 'Pk ', respectively, with the same individual constant, which does
not occur in eFo

A13. Meaning postulates for F:

a. 1- hl v h2 V ... V hk •
b. If i "# j, h i • h j is L-false.

m(eF ) is independent of other individuals (AlO) and other families

(A11). It depends not on the particular individuals in eF but only on their
 NOTES ON PROBABILITY AND INDUCTION 311

numbers Sio Therefore:

TI. For any m-function m fulfilling the axioms, there is, for any
k, a representative mathematical junction M k of k arguments
such that, for any eF ,
m(eF) = Mk(st> S2, Sk)o000'

T2. M k is invariant with respect to any permutation of the k

argumentso (From A8o)
eFoh l is an individual distribution for S + 1 individuals with the
cardinal numbers Sl + 1, S2, Sko We define:
000'

Mk(Sl + 1, S2, Sk)

DI.
000'

Ck (Sl;S2,ooo,Sk)=Of o

Mk(Sl, S2, 000' Sk)

T3. a. For any c-function c and any k, there is a representative
mathematical junction Ck of k arguments such that, for
any eF, C(hl' eF) = Ck(Sl; S2, Sk)o Analogously for h 2, etc.
000'

b. Ck is invariant with respect to any permutation of the

k-l arguments following the first.
1 shall sometimes write 'M' and 'C' without subscriptso
T4. For any k numbers n, p, S3, Sk whose sum is s, the following
000,

holdso ('---' stands for 'S3, Sk'; this expression drops out
000'

if k = 2; in this case n + p = so)

C(n;p + 1,---) C(n;p,---)
C(p; n + 1, ---) C(p; n, ---)

(Here n+(p+ 1)+ --- = p+(n+ 1)+ --- =S+ 1 and

n+ p+ --- = p+n+ --- =S

Proojo The following holds identically:

M(n + 1, p + 1, ---) M(n, p + 1, ---)
---------------- x =
M(n,p + 1,---) M(n,p,---)
M(n + 1,p + 1,---) M(n + 1,p,---)
x ----------,-
M(n + 1,p,---) M(n,p,---)
According to Dl, the first quotient is C(n; p + 1, ---); the
312 RUDOLF CAR NAP

second is (by T2) equal to

M(p + 1, n, ---)
- - - - - - = C (p; n, - - -) ;
M(p, n, ---) .
the third becomes (again with reordering of arguments)
C(p; n + 1, ---), and the fourth C(n; p, ---). Hence the
theorem.
k
T5. a. L c (h i , eF) = 1. (From A13a.)
i= 1
k
b. L C (Si; Sl' ... , Si-l' Si+
i=l
l ' •.. , Sk) = 1. (From (a).)

XIV. The Axiom of Predictive lrrelevance (AI4)

Let el be formed from eF by replacing each predicate except 'Pt' with
, '" P 1'. Hence el is an individual distribution for the S individuals
with respect to the division P l , ' " P l , with the cardinal numbers Sl
and s- SI. e2 , ••• , ek are formed analogously.
For given k, c(h 1 , el) depends only on Sl and s. It can therefore be
represented by a function Gk(Sl; s)o Analogously for i = 2, 000' k(by A8)o
TI. For any c-function c and any k, there is a representative
mathematical function Gk such that, for i = 1, .00' k,
C(hi' ei) = Gk(Si; S)o

T2. Suppose that Sl < So Let e' 1 be like el but with the cardinal
numbers SI + 1 and S- SI - 1.
a. c(hl, e' 1) > C(hl' el). (From XII-Tldo)
b. Gk (Sl + 1; s) > Gk(Sl; s) (From (a)o)

The axiom of predictive irrelevance says that of the k cardinal numbers

in eF alI except Sl are irrelevant for h l :

This axiom is not a necessary condition for the adequacy of c. But it is

a customary (usually tacit) assumption, and it leads to agreat simplifica-
tion ofthe system. If k = 2, then el is the same as eF and therefore A14 is
fulfilled trivially.
 NOTES ON PROBABILITY AND INDUCTION 313

T3. For any k (~2) and any i:

a. c(h;, eF) = c(h;, ej). (From A14, A8.)
b. For any numbers S2' ... , Sk whose sum is s- si>
Ck(Sl; S2' ... , Sk) = Gk(Sl; s). (From (a).)

1 shall often write 'G' for 'G k '.

T4. For any sequence of k numbers Sl' ... , Sk whose sum is s,
k
L G(Sj; s) = 1. (From XIII-T5.)
j=l

Special cases of T4:

T5. a. G(s; s) + (k- 1) G(O; s) = 1. (From T4 for the sequence
s, O, ... , O.)
b. G(s + 1; S + 1) = 1- (k- 1) G(O; S + 1). (From (a).)
c. G(l; 1) = 1- (k- 1) G(O; 1). (From (a).)
d. G(s; s+ l)+G(l; s+ 1)+(k-2) G(O; s+ 1)= 1.
(Sequence s, 1, O, ... , O.)
The folIowing development has the aim to show (1) that, if alI values
of G for sare given, the values for s + 1 are uniquely determined, and (2)
if G(O; 1) is given, alI values of G are uniquely determined. For these
results it is presupposed that k > 2.
T6. For k>2, for any n, p, s such that n+ p~s,
G(n;s+1) G(n;s)
a. = --. (From XIII-T4.)
G(p; s + 1) G(p; s)
G (n; s)
b. G(n;s + 1) = G(O;s + 1) --o
G (O; s)
(From (a) with p = O.)
T7. For k>2,

G(S;S) G(1;s) ]
G(O;s + 1) [ - - + - - + k - 2 = 1.
G(O;s) G(O;s)

(From T5d, by transforming the first two of its G-terms accord-

ing to T6b.)
Now aim (1) has been reached. Ifall G-values for sare given, G(O; s + 1)
314 RUDOLF CARNAP

is determined by T7, then the values G(n; s + 1) for n = 1, ... , sare

determined by T6b, and G(s + 1; s + 1) by TSb. Thus all values for
s + 1 are determined.
We have also attained aim (2). If G(O;I) is given, G(1; 1) is determined
by TSc. These are all the G-values for s = 1. They determine the values
for s = 2, and so on. Thus all G-values are determined by G(O; 1). The
following theorem gives the explicit form.
T8. For k > 2, for any s and n (O ~ n ~ s),
n-(kn-l)G(O;I)
G(n;s ) = .
s - (s - l)kG(O; 1)
(This can be proved by mathematical induction with respect
to s. (1) The theorem holds for s = 1 (for n = O it holds
identically, for n = 1 by TSc). (2) If the theorem holds for
a given s, it holds likewise for s + 1; this can be shown with
the help of the theorems T7, T6b, and TSb, which determine
the G-values for s + 1 on the basis of those for s. Hence the
theorem holds for every s.)
Suppose that the value of G(O; 1) has been chosen. Then all values of
G can be determined. The following theorem T9 shows that the m-value
of any state-description is determined by the value of G. Thereby the
m-values for alI sentences and the c-values for alI pairs of sentences are
determined (see VI).
T9. Let ZF be a state-description for N individuals and for the
k predicates of the family F, with the cardinal numbers NI
(i = 1, ... , k). Then
N,-l
m(ZF) = TI n=O
i
TI G(n; Si + n),

where TIi runs through those values of i for which Ni > O;

i-1
Si = LN
h=l
h, Sl = O. (For the proofsee [Continuum]§S.)

XV. The Â-System (A1S)

We shall construct a system of all c-functions fulfilling our axioms. We
 NOTES ON PROBABILITY AND INDUCTION 315

call it the A-system, because the c-functions will be characterized by the

values of a parameter A.
We have seen that, for a given k(> 2), all values of G are determined
by G (O; 1). The Iatter value can be freely chosen within certain boundaries.
We shall now determine these boundaries.
From XI-TIc:
(1) c(h i , t) = l/k.
Hence with XII-TIc (based on A12):
(2) G(O; 1) < I/k.
If we were to choose G(O; 1) > l/k, then c would violate not onIy A12b
but aiso A12a and therefore be unacceptabIe. If we choose G(O;l) = l/k,
then on1y A12b is violated. This c does not belong to the A-system,
but will nevertheless be discussed as a boundary case (we shall find
that it is the same as ct in X).
The following is obvious (from Al):
(3) G(O; 1) ~ O.
If we choose G(O; 1) = O, then the resulting c fulfills Al to AS, but
violates A6. Hence it is quasi-regular. Therefore it does not belong to
the A-system. It will, however, be discussed as a boundary case; we shall
see that it corresponds to the straight rule.
TI. O < G(O; 1) < 1Jk.
It can also be shown that, if any value between O and lJk is chosen
for G(O; 1), then the resulting c-function fulfills our axioms.
We define first an auxiliary parameter:
D1. A~' =Df kG k (O; 1).
From TI:
T2. O < A;' < 1.
1 shall usually write 'A"" for 'Ak"". From XIV-T8:

T3. For k > 2, for any s and n (O ~ n ~ s),

G (n- s) = _n_-_(_n_-_l/_k_H_'"
, s - (s - 1) A'"
316 RUDOLF CARNAP

We shall mostly use, not A'II, but A = A"'J(l- A"'). The use of A leads to
a simpler formula for G(n;s) (T4c). However, in the case of G(O; 1) = lJk,
A'II = 1, while A is infinite. Therefore in this case A is less convenient than
AII/. But this case is not inc1uded in our system.
kGk(O; 1)
D2. A -
k -Df 1 - kGk(O; 1)
T4. a. A = AJ(A
III
+ 1).
A
b. G(O; 1) = k(A + 1)·
n + AJk
c. For k > 2, G(n; s) = . (From T3, (a).)
S+A
The case k = 2. The important results at the end of XIV can be proved
only if k > 2. (This seems surprising, since A14, on which the results are
based, holds also for k = 2.) A new axiom must be added for k = 2. T4c
shows that, for given k(> 2), s, and G(O; 1), G(n; s) is a linear function
of n. Weassume as an axiom that the same holds for k = 2:
A15. For given s and G2 (O; 1), G 2 (n; s) is a linear function ofn.
Note that G 2 (n; s) = e2 (n; s- n) (see XIV-T3b). Therefore we have
(without use of AI5):
(4) G2 (n; s) + G 2(s - n; s) = 1. (From XIII-TSb.)
G2 (n; s + 1) G2 (n; s)
(5) (From XIII-T4.)
G2 (s-n;s+1) G2 (s - n; s)

From (4) and (5) we can derive with A15:

G2 (n; s)
T5. G2 (n; s + 1) = G2 (O; s + 1) .
G2 (O; s)

This corresponds to XIV-T6b. Then, in analogy to the earlier proofs,

we can now prove the analogues of T3 and of T4c for k = 2; the latter is

n + AJ2
T6. G2 (n; s) = .
S+A
 NOTES ON PROBABILITY AND INDUCTION 317

For any A, let c). be the c-function characterized by A, and m). be the
corresponding m-function. Using our result for G(n; s) (T4c, T6), we
obtain T7a from XIV-T9 (for the proofs of T7b and c and the notation
[:], see [Continuum] § 10).
n + A/k
U TI
N,-I
T7. a. m).(ZF) = =
I
(N,>O)
n=O Si +n+A

b.

c.

d. Let StrF be the structure-description corresponding to ZF.

Then
Q(Ni + ~ik -1)
m).(StrF) = ( N + A-1 ) .
N
(From (c) with VIII-T3.)

XVI. Various c-Functions in the A-System

Let the predicate 'M' be defined as a disjunction of w predicates of
F: 'P I v P 2 V .. V Pw'o Hence its logical width is Wo Let eF be as before,
0

and hM a full sentence of 'M' for a new individual. Let SM be SI + S2 +

+ ... + sW. Then we have (from XV):
SM + wAJk
(1) C). (h M , eF) = .
S+A
This is
SM· w
-S+ -A
S k
S+A
318 RUDOLF CARNAP

thus it is the weighted mean of the observed relative frequency SM/S and
the relative width w/k, with weights S and A, respectively. SM/S is an
empirical factor in the situation, and w/k is a logical factor. A is thus the
weight of the logical factor. The greater the chosen A, the closer to w/k is
the above c-value.
Example. For an even k, we take a predicate 'M' with w = k/2. In eF,
let S = 10, SM = 1. Then the c;.-value in (1), for various choices of A, is as
follows (see [Continuum] (12-19)):
A. = O 1 2 4 8 16 32
C;.(hM, eF) = 0.1 0.136 0.167 0.214 0.278 0.346 0.405 0.5

For A = O, c = SM/S, This is the straight rule, which violates A6.

([Continuum] § 14.)
For A = 00, c = w/k =c(hM , t). This is ct , which violates A12b.
([Continuum] § 13.)
These are the two extreme methods, not included in the A-system. In
this system, we take O< A < 00; hence the above c is between SM/S and
w/k (if these are unequal).
For families of different sizes (each in a separate language) we distin-
guish two kinds of inductive methods.
lnductive methods of the first kind: a fixed va1ue is chosen for Â,
independent of k. ([Continuum] § 11.)
lnductive methods of the second kind: Ak is dependent upon k. The
simplest form is: Ak = Ck, with a constant C. The simplest method of this
form takes C = 1, hence Ak = k; thus from (1):

(2)

This is the function c* (see X).

XVII. A Language with Two Families (AI6)

The language .fi> contains two families: FI consists of k l predicates:
'pf', 'P~', etc; and F 2 of k 2 predicates: 'pi', 'P~', etc. There are
k=k l k 2 Q-predicates; Qij is the conjunction 'pl.p]' (i= 1, ... , k l :
j= 1, ... , k 2 ).
Let el be an individual distribution for FI, and e2 for F 2 , both for
the same S individuals. Let e be el . e 2 • This is an individual distribution
for the k Q-predicates; let sij be the number of individuals with Q;j'
 NOTES ON PROBABILITY ANO INOUCTION 319

We take the same A. for both families. Then we can determine m A(e i )
and mA(e2 ) (by XV-T7).
Problem: What is to be taken as value of mA (e)? This is not determined
by the previous axioms. We shall now consider two attempts at a solution,
and then take a combination of them.
First tentative solution. We take the cIass of the k Q-predicates as the
pseudo-family Fi. 2. Then we define m 1 • 2 for Fi. 2, as if the Iatter were a
real family; hence, in analogy to XV-T7c:

,[Lv, [3U+ ~ - 'J

DI. mi" (el ~'" [3 + : _3,:]
m1' 2(e) depends only on the Q-numbers Sij in e, not on the P-numbers in
el or e 2.
Second tentative solution. We define m 112 (e) as the product of the
m-values for the two families separately:
D2. mA112 ( e) =or
m 112 (e) depends only on the P-numbers, not on the Q-numoers.
We shall examine the two solutions with the help of the foIlowing
three examples A, B, C, of individual distributions for s = 20 individuals,
with k 1 =k2 =2 (the numerals in the four ceIls indicate the Q-numbers;
the marginal numerals indicate the P-numbers for the two families).
A F2 B F2 c F2

I ~~21
110 I 10I o 20 I 10 I 10
1
10 I O I 10 O I O I O

The foIIowing two requirements (or desi de rata ) 1 and II seem plausible.
(1) We should have: m(A) < m(B), because B is more uniform
than A.
This requirement is satisfied by m 1 • 2 (because the Q-numbers are equal
320 RUDOLF CARNAP

in A, unequal in B), but not by m 112 (this has equal values for A and
for B, because the P-numbers are the same).
(II) We should have: m(B) < m(C), because the distribution for
F 1 is more uniform in C than in B, while that for F 2 is the
same in C as in B.
This requirement is in accord with the customary analogy inference
('horse-donkey inference'). However, it is not satisfied by m 1 •2 (this has
equal values for B and for C, because the Q-numbers are the same). It is
satisfied by m 112 •
Thus both solutions are unsatisfactory. Generally, any solution that
uses only the P-numbers cannot satisfy 1, and any solution that uses
only the Q-numbers cannot satisfy II. An adequate solution must use
both the P-numbers and the Q-numbers. This is done in the third
solution, which satisfies both requirements.
Third solution. We define m),je) as a weighted mean of the first
two solutions, with the weights 1] and 1-1], where 1] is a new parameter:
D3. mÂ..q(e)=Df 1]m1 12 (e)+(I-1]) ml· 2 (e).
The parameter 1] may be chosen, independent1y of l, such that
0<11 < 1. The greater 1] is, the stronger is the influence by analogy (i.e.,
the greateris the difference between the two c-values in A16 below). The
method can easily be extended to more than two families; no new para-
meter is needed. (The method was worked out in collaboration with
John Kemeny.)
The requirement II can be represented in a generalized form as follows:
A16. Axiom of analogy. Let e be an individual distribution for
two families (with any k 1 and k 2 ). Let i andj be full sentences
of Qll and Q21' respectively, with the same individual
constant not occurring in e. Let h be a full sentence of Q12
with another individual constant not occurring in e. Then
c(h, e. i) > c(h, e .j).
The generalization for other Q-predicates follows by A8.

XVIII. An Infinite Domain of Individuals (AI7)

Let the domain of 2 N contain N individuals, and that of 200 be denumer-
ably infinite. According to AlO, the values of c for non-general sentences
 NOTES ON PROBABILITY AND INDUCTION 321

are in .!&' "" the same as in .!&'N' If either e or h or both contain variables,
a new axiom is needed. We take the value of c in .!&'"" as the Iimit of its
values in finite languages (see [Prob.] § 56):
A17. Axiom of the infinite domain. Let NC be a c-function for.!&' N'
Then the corresponding c-function ""c for.!&'"" is determined
as follows:
""c(h, e) = Iim Nc(h, e).
N-+""

BIBLIOGRAPHY

A list of selected publications, almost aII since 1950. For earlier publications, see the
bibliography in Carnap [Prob.]. (Items marked with an asterisk were added, and those
marked with a dagger were omitted, when the course was given in the spring of 1960. -
Ed.)

Abbreviations

BJPS The Brilish !ournal for the P R The Phi/osophical Review

Philosophy of Science PSc Philosophy of Science
JSL The Journal of SYl11bolic Logic PSt ,Philosophical Studies

*Ayer, A. J., 'The Conception of Probability as a Logica) Relation', in: K6rner, 1957,
pp. 12-17. Repr. in: Madden, 1960, pp. 279-284.
Bar-Hillel, Yehoshua,
'A Note on State-Descriptions', PSt 2 (1951) 72-75.
'A Note on Comparative Inductive Logic', BJPS 3 (1952/53) 308-310.
t'Comments on "Degree of Confirmat ion" by Professor K. R. Popper', BJPS 6
(1955/56) 155-157.
t'Further Comments on Probabilityand Confirmation: A Rejoinder to Professor
Popper', BJPS 7 (1956/57) 245-248.
Bar-Hillel, Yehoshua, and Carnap, Rudolf, 'Semantic Information', BJPS 4 (1953/54)
147-157. AIso in: Willis Jackson (ed.), COl11l11unication Theory, Butterworths
Scientific Publications, London, 1953, pp. 503-511.
Black, Max,
Language and Philosophy, Cornell Univ. Press, Ithaca, 1949. (Essay III, 'The Justi-
fication of Induction'.)
*Problel11s of Analysis, Cornell Univ. Press, Ithaca, 1954. (Part 3, 'Induction'.)
Braithwaite, Richard Bevan,
Scientific Explanation, Cambridge Univ. Press, Cambridge. 1953.
*'On Unknown Probabilities', in: K6rner, 1957, pp. 3-11.
Burks, Arthur W.,
'Reichenbach's Theory of Probability and Induction', The Review of Metaphysics 4
(1950/51) 377-393.
'The Presupposition Theory of Induction', PSC 20 (1953) 177-197.
 322 RUDOLF CARNAP

'On the Significance of Carnap's System of Inductive Logic for the Philosophy of
Induction', in: Schilpp, 1963, pp. 739-759.
Carnap, RlIdolf,
[Prob.] Logical Foundations of Probability, Univ. of Chicago Press, Chicago, 1950;
2nd ed., 1962.
The Nature and Application of Inductive Logic, Consisting of Six Sections from
Logical Foundations of Probability, Univ. of Chicago Press, Chicago, 1951.
'The Problem of Relations in Inductive Logic', PSt 2 (1951) 75-80.
[Continuum] The Continuum of Inductive Methods, Univ. of Chicago Press, Chicago,
1952.
[Postulates] 'Meaning Postlllates', PSt 3 (1952) 65-73. Repr. in: Carnap, Meaning
and Necessity, 2nd ed., Univ. of Chicago Press, Chicago, 1956, pp. 222-229.
[Comparative] 'On the Comparative Concept of Confirmation', BJPS 3 (1952/53)
311-318.
'Remarks to Kemeny's Paper', Philosophy and Phenomenological Research 13 (1952/
53) 375-376.
'Inductive Logic and Science', Proceedings of the American Academy oI Arts and
Sciences 80 (1953) 189-197.
'What Is Probability?', Scientijic American 189:3 (Sept. 1953) 128-130, 132, 134, 136,
138.
1. Statistical and Inductive Probability, II. lnductive Logic and Science, Galois Institute
of Mathematics and Art, Brooklyn, N.Y., 1955. I is repr. in Madden, 1960, pp.
269-279; II is repr. from Proc. Amer. Acad. Arts & Se. above.
t'Remarks on Popper's Note on Content and Degree of Confirmation', BJPS 7 (1956/
57) 243-244.
*'Replies and Systematic Expositions, V. Probability and Induction', in: Schilpp,
1963, pp. 966-998.
Carnap, RlIdolf, and Bar-HiIlel, Yehoshua, An Outline of a Theory of Semantic lnforma-
tion (Technical Report No. 247), Research Laboratory of Electronics, Massachusetts
Institute of Technology, Cambridge, Mass., 1952 [i.e. 1953]. Repr. in: Bar-HiIlel,
Language and lnformation, Addison-Wesley Publishing Co., Reading, Mass., 1964,
pp. 221-274.
tDavidson, Donald, McKinsey, J. C. c., and Suppes, Patrick, 'Outlines of a Formal
Theory of Value, 1', PSc 22 (1955) 140-160.
*Davidson, Donald, and SlIppes, Patrick, Decision Making, Stanford Univ. Press,
Stanford, Calif., 1957.
*Edwards, Paul, 'Russell's Doubts about Indllction', Mind 58 (1949) 141-163.
Feigl, Herbert, 'Scientific Method without Metaphysical Presuppositions', PSt 5
(1954) 17-29.
Feys, Robert (ed.), Theorie des probabilites (Collection de logique matbematique,
Serie B, 1), E. Nauwelaerts, Louvain; Gauthier-Villars, Paris, 1952.
Finetti, Bruno de
'Recent Suggestions for the Reconciliation of Theories of Probability', in: Jerzy
Neyman (ed.), Proceedings ofthe Second Berkeley Symposium on Mathematical
Statistics and Probability, Univ. of California Press, Berkeley & Los Angeles,
1951, pp. 217-225.
'Le Vrai et le probable', Dialectica 3 (1949) 78-92.
Goodman, Nelson, Fact, Fiction, and Forecast, Harvard Univ. Press, Cambridge, Mass.
1955; 2nd ed., Bobbs-Merrill Co., Indianapolis, 1965.
 NOTES ON PROBABILITY AND INDUCTION 323

*Halmos, Paul R., Measure Theory, D. Van Nostrand Co., New York, 1950.
Hutten, ErnestH., 'Probability-Sentences', Mind 61 (1952) 39-56.
Kemeny, John G.,
'Extension of the Methods of Inductive Logic', PSt 3 (1952) 38-42.
'A Contribution to Inductive Logic'. Philosophy and Phenomenological Research 13
(1952/53) 371-374.
'The Use of Simplicity in Induction', PR 62 (1953) 391-408.
'A Logical Measure Function', JSL 18 (1953) 289-308. .
'Fair Bets and Inductive Probabilities', JSL 20 (1955) 263-273.
• A Philosopher Looks at Science, D. Van Nostrand Co., Princeton, 1959. (Ch. 4, 'Prob-
ability'; Ch. 6, 'Credibility and Induction'.)
·'Carnap's Theory of ProbabiIity and Induction', in: Schilpp, 1963, pp. 711-738.
Kemeny, John G., and Oppenheim, Paul, 'Degree of Factual Support', PSC 19 (1952)
307-324.
*Kolmogorov, A. N., FOllndations of the Theory of Probability, 2nd EngIish ed., Chelsea
PubIishing Co., New York, 1956.
*Kărner, Stephan (ed.), Observation and Interpretation, Butterworths Scientific Publica-
tions, London, 1957.
*Leblanc, Hugues,
'On LogicaIly False Evidence Statements', JSL 22 (1957) 345-349.
'On Chances and Estimated Chances of Being True', Revue philosophiqlle de LOl/vain
57 (1959) 225-239.
Statistical and Indllctive Probabilities, Prentice-HaIl, Englewood Cliffs, N.J., 1962.
Lehman, R. Sherman, 'On Confirmat ion and Rational Betting', JSL 20 (1955) 251-262.
*Lenz, John W.,
'Carnap on Defining "Degree of Confirmation''', PSC 23 (1956) 230-236.
'The Frequency Theory of Probability', in: Madden, 1960, pp. 263-269.
'The Pragmatic Justification of Induction', in: Madden, 1960, pp. 299-303.
*Loeve, Michel, Probability Theory, D. Van Nostrand Co., New York, 1955; 3rd ed.,
D. Van Nostrand Co., Princeton, 1963.
*Madden, Edward H. (ed.), The Strl/cture of Scientific Thought, Houghton Mifflin Co.,
Boston, 1960. (Ch. 5, 'Probability Notions'; Ch. 6, 'The Riddle of Induction'.)
*Martin, R. M., 'A Formalization of Inductive Logic', JSL 23 (1958) 251-256.
*Mises, Richard von, Probability, Statistics and Truth, 2nd rev. English ed., George
Allen & Unwin, London; MacmiIIan Co., New York, 1957.
Nagel, Ernest, Principles of the Theory of Probability (International EncycIopedia of
Unified Science, VoI. 1, No. 6), Univ. of Chicago Press, Chicago, 1939.
tNeyman, Jerzy, 'The Problem of Inductive Inference', Communications on Pure and
Applied Mathematics 8 (1955) 13-45.
tPap, Arthur, Analytische Erkenntnisthearie, Springer-Verlag, Wien, 1955. (Kap. III,
Abschn. B, 'Kernfragen der Wahrscheinlichkeitstheorie'; Abschn. C, 'Das Induk-
tionsproblem'.)
Popper, Karl R.
t'Degree of Confirmation', BJPS 5 (1954/55) 143-149.
t'Two Autonomous Axiom Systems for the Calculus of Probabilities', BJPS 6 (1955/
56) 51-57.
t'''Content'' and "Degree of Confirmation": A Reply to Dr. Bar-HiIIel', BJPS 6
(1955/56) 157-163.
t'The Demarcation between Science and Metaphysics', in: SchiIpp, 1963, pp. 183-226.
 324 RUDOLF CARNAP

Also in: Popper, Conjectures and Refutations, Basic Books, New York; Routledge
& Kegan Paul, London, 1962; pp. 253-292.
*'The Propensity Interpretation of the Calculus of Probability, and the Quantum
Theory', in: Kărner, 1957, pp. 65-70.
*'The Propensity Interpretation of Probability', BJPS 10 (1959(60) 25--42.
*The Logic of Scienti/ic Discovery, Basic Books, New York, 1959.
tRubin, Herman, and Suppes, Patrick, 'A Note on Two-Place Predicates and Fitting
Sequences of Measure Functions', JSL 20 (1955) 121-122.
*Salmon, Wesley c.,
'Regular Rules of Induction', PR 65 (1956) 385-388.
'Should We Attempt to Justify Induction?', PSt 8 (1957) 33--48.
'The Predictive Inference', PSC 24 (1957) 180-190.
Savage, Leonard J., The FOlllldations of Statistics, John Wiley & Sons, New York;
Chapman & Hali, London, 1954. 2nd rev. ed., Dover Publications, New York, 1972.
Schilpp, Paul Arthur (ed.), The Philosophy of Rudolf Carnap (The Library of Living
Philosophers, VoI. XI), Open Court, La Salle, III.; Cambridge Univ. Press, London,
© 1963 [pub. 1964].
Shimony, Abner, 'Coherence and the Axioms of Confirmation', JSL 20 (1955) 1-28.
tSimon, Herbert A., 'Prediction and Hindsight as Confirmatory Evidence', PSC 22
(1955) 227-230.
tStegmUller, Wolfgang, 'Bemerkungen zum Wahrscheinlichkeitsproblem', Stlldium
Generale 6 (1953) 563-593.
*Strawson, P. F., lntroduction to Logical Theory, Methuen & Co., London, 1952. (Ch.
9, 'Inductive Reasoning and Probability'.)
tVietoris, L., 'Zur Axiomatik der Wahrscheinlichkeitsrechnung', Dialectica 8 (1954)
37--47.
Will, Frederick L., 'Kneale's Theories of Probability and Induction', PR 63 (1954)
19--42.
Wisdom, John Oulton, Foulldatiolls of lnference in Natural Science, Methuen & Co.,
London, 1952.
Wright, Georg Henrik von,
A Treatise on lnduction and Probability, Routledge & Kegan Paul, London; Harcourt,
Brace & Co., New York, 1951. Repr. Littlefield, Adams & Co., Paterson, N.J.,
1960.
'Carnap's Theory of Probability', PR 60 (1951) 362-374.
*The Logical Problem of lnduction, 2nd rev. ed., Basil Blackwell, Oxford; Macmillan
Co., New York, 1957. Repr. Barnes & Noble, New York, 1966.
 RICHARD C. JEFFREY

CARNAP'S INDUCTIVE LOGIC

Carnap pretty much agreed with Keynes:

Part of our knowledge we obtain direct; and part by argument. The theory of probabi-
lity is concemed with that part which we obtain by argument, and it treats of the
different degrees in which the results so obtained are conclusive or inconclusive.
(J. M. Keynes, A Treatise on Probability, Ch. 1)

Of course, Carnap would speak of the theory of inductive probability

where Keynes speaks simply of the theory of probability. But Carnap's
/
main deviation from Keynes at this point concerns the assertion that
probabilities give degrees of conclusiveness of the results of inductive
arguments. There, Carnap adopted Ramsey's view, that
the kind of measurement of belief with which probability is concemed ... is a measure-
ment of belief qua basis of action.
(F. P. Ramsey, 'Truth and Probability,' §3.)

Carnap thought it essential that we describe people's beliefs, not in an

all-or-none fashion, but in terms of degrees of credence. Jones's beliefs at
time tare to be given by a function crJones,t which obeys the laws of the
elementary probability calculus. (Either or both of the subscripts may be
dropped, where the context makes clear what is intended.) With Ramsey,
Carnap interpreted statements of form 'cr(h)=x' as dispositional (or
perhaps theoretical) statements about Jones's behavior. In a rough and
ready way, both men took cr(h) to be the largest part of a dollar that
Jones would be willing to pay, at time t, in order to receive a dollar if and
only if h is true. Carnap followed Ramsey in taking this to be a rough
indication of the full account, in which Jones's degrees of belief are
deducible from certain characteristics of his preference ranking.
So much for actual credence. Carnap took the task of applied inductive
logic to be that of telling people like Jones what their credence functions
ought to be, given their experience. Carnap's paradigm was the case in
which we consider a sequence of moments, t=O, 1,2, ... , and a corre-
sponding sequence of sentences el' e2 , ... , where for each positive t, et

Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 325-332. AII rights reserved.
Copyright ©1975 by D. Reidel Publishing Company, Dordrecht-Holland.
326 RICHARD C. JEFFREY

expresses the information which Jones got directly from experience be-
tween times t-l and t. Then Jones's total evidence at time t is e(t)=DF
el 'e2' ... ·e,. In such a case, Carnap thought, Jones's degree of credence in
a statement h at time t ought to depend only on hand e(t), i.e., cr,(h) ought
to be afunction of hand e(t). Carnap called that function 'c': cr,(h) = c(h,
e(t)) if Jones's beliefs at time tare determined in a logically satisfactory
manner by his experience up to time t. The function c should have the
characteristic that for any sentence h in Jones's language, the number
c(h, e(t)) will seem (after sufficient reflection) to be the degree of credence
that anyone ought to have in h at time t, provided his total directly
experimential evidence at time t is expres sed by the sentence e(t). To
carry out that task, Carnap thought it important that the function c be
defined even when the second argument is not the sort of sentence that
could represent someone's total direct observational experience. Thus, for
any sentences hand e (with certain exceptions, e.g. where e is self-contra-
dictory), Carnap wanted c(h, e) to be a definite number. He took the task
of pure in3uctive logic to be that of providing a mathematical definition
of the function c.

1. IN WHAT SENSE IS c 'LOGICAL'?

Carnap had two different, Pickwickian senses in mind for the word
'logical' in this question. Briefly,
logica/! = nonfactual
/ogical 2 = in agreement with inductive intuition

Let me explain and evaluate the situation as I see it.

'c' ought to be a logical! functor, just as 'sin' and 'IOglO' are logical l
functors. This is a true description of Carnap's program, which is to
identify a satisfactory c-function in such a way that, once hand e are
identified either ostensively (as in quotation-mark names) or by structural
descriptions, it is a matter of ca1culation to determine the value which c
assigns to the pair (h, e): no empirical investigation is needed. Thus, in
his paper 'On Inductive Logic' (1945) Carnap defined a functor 'c*' in
such a way that it is a matter of calculation to discover that C*('Pa2"
'Pal')=j-. Note that it is the functor, 'c', not the function, c, which it
denotes, which is logical l . It is entirely possible for one and the same
 CARNAP'S INDUCTIVE LOGIC 327

function to be denoted by two functors, one ofwhich (say, 'c') is logical 1 ,

while the other (say, 'crJones. V-E da;' thought of as a function of two ar-
guments, where cr(h, e)=cr(h·e)/cr(e)) is factual. (To discover Jones's
degrees of credence you must study Jones, but the values of 'c' can be
determined with pencil and paper.)
The function c i$ logica/z: Its values will be in agreement with our
inductive intuitions as they will exist at the time when the program has been
carried out. At the moment, we are often at a loss to say what the values
c(h, e) ought to be. Carnap's method for carrying out his program was to
consider cases where we are fairly clear about what numbers we want, and
then ask how various unclear cases might be decided so as to get the
required results in the clear cases. If the program ever succeeds, it will be
because we have been able to apply that method on a broad front, thus
developing and sharpening our inductive intuitions - creating them, one
ffiight every say - while at the same time narrowing the set of candidates
for the role of c. Then the processes of finding c and of developing our
inductive intuitions are interwoven in such a way that the chosen c func-
tion must agree with inductive intuition - if the program ever succeeds.
Then 1 take it to be a true description of Carnap's program, that if it
succeeds, the functor 'c' will be logica!l' and the function it denotes will be
/ogica!z. Note that as 1 have been using and mentioning it, the symbol 'c'
has no referent unless Carnap's program is eventuaIly carried to a success-
fuI conclusion. Then my claim about the two senses of logicality can be
put more precisely and less technically as foIlows:
If Carnap's program ever succeeds, we shall then be in posession of a method for
computing values c(h,e) when hand e are specified either ostensively or by structural
descriptions; and these values will agree with our logical intuitions (as they shall then
be) about the degree of credence in h that would be appropriate for someone whose
total observational experience is summarized by the sentence e.

This is what 1 take to be the content of Carnap's claim that c is logica/o

Interpreted in this way, the cIaim is cIearly true - but uninteresting, un-
less Carnap's program is in fact destined to succeed. Carnap hoped and
expected that others would continue his work on the program, and would
eventually bring it off. Therefore he took the claim to be interestingly
explicative.
It seems cIear to me that Carnap explicitly distinguished two senses of
'logical', essentially as 1 have done here - except (a propos logical z) for
328 RICHARD C. JEFFREY

the latitudinarianism which 1 shalI now discuss, and except for a tendency
on Carnap's part to speak of the inductive intuitions which will emerge
as the program proceeds as already existing within us (if only covertly).

II. LA TITUDINARIANISM

Carnap was prepared to admit the possibility that different people might
have somewhat different inductive intuitions, e.g. when, ca. 1951, he
thought the right c-function might be found somewhere in the continuum
of inductive methods, he thought that different people might discover that
they had somewhat different values of  and hence that their inductive
intuitions were describable by somewhat different functions c.... Re thought
it possible that these differences were irreducible, so that his program, as
described above, might fail in the mild sense that there might be no such
thing as the c-function which represents our inductive intuitions. Rather,
there might prove to be a range of inductive intuitions among us, which
might be represented by different c-functions. The failure would be mild
- so mild as not to deserve the name, failure - if the various inductive
intuitions are sufficiently similar so that their differences are swamped
out by experience. Thus, suppose that the functions c and c' represent your
inductive intuitions and mine, respectively, and that at time t our ex-
periential evidence is represented by sentences e(t) and e' (t), for you and
me, respectively. Then the difference between c and c' is swamped out by
experience if my rational credence function at time t would have been
pretty much the same if 1 had used c instead of c', and similarly yours
would have been pretty much the same if you had used c' instead of c.
Thus, the numbers c(h, e(t» and c' (h, e(t» may be very close, for alI h,
and so may be the numbers c(h, e' (t» and c' (h, e' (/». If so, the differ-
ences between our inductive logical intuitions are alI but irrelevant at
time t, having been 'swamped out' by our experience.

III. SUBJECTIVISM

Carnap's latitudinarianism is suggestive: perhaps he is describable as

a special sort of subjectivist. Differences among the c functions which
represent our various inductive intuitions are matters of individual psy-
chology, while the respects in which alI of our c functions agree represent
 CARNAP'S INDUCTIVE LOGIC 329

human nature - general psychology. Perhaps Carnap would accept this

subjectivist reading, but would think it appropriate to dignify with the
name 'logical' the common features of aH c functions, since the agreement
concerns norms for credence. Note that de Finetti's subjectivist program
seeks (Foresight, Ch. 6),

to show that there are rather profound psychological reasons which make the exact
or approximate agreement that is observed between the opinions of different indivi-
duals very natural,

and here 1 take his posi tion to be compatible with Carnap's. But de
Finetti then urges
that there are no reasons, rational, positive, or metaphysical, that can give this fact
any meaning beyond that of a simple agreement of subjective opinions.

Presumably the disjunction of rational, positive, and metaphysical is

philosophical: de Finetti declines to try to derive the similarities from
more basic, philosophical considerations. But then, so does Carnap.
lnductive logic is a description of the norms we shaH find ourselves accep-
ting, once we have thought the matter through. It is no part of Carnap's
program, to base those norms on stiH deeper philosophical considera-
tions.
But if Carnap is a subjectivist, he is a very exigent one: he thinks it
possible and worthwhile to try to describe someone's inductive tempera-
ment quite explicitly and generally, via a c-function. lf your experience
differs from mine, that difference will appear in the second argument
places of our respective c-functions - it will not affect the functions them-
selves. At times he sounds Kantian: one's inductive temperament is
evoked but not determined by one's experience. But subjectivists like de
Finetti despair of such neat separation of experience from temperament;
anyway, they think it hopeless, and profitless, to attempt the separation
along Carnapian lines.

IV. RESPONSIBILITY FOR EXPERIENCE

Let me now turn to a problem which Carnap never treated explicitly (as
far as 1 know), but which can be treated readily enough within the frame-
work of his system.
330 RICHARD C. JEFFREY

Suppose that cr is Jones's actual credence function at time t, and that

e(t) represents his total observational knowledge at time t. Suppose
further that Jones's inductive temperament is represented by the function
c. Then if cr(h) fails to be the same number as c(h, e(t», Jones's belief
about h is not fuIly rational, for it is not as it ought to be, given his ex-
perience and temperament. Thus, inductive logic can be applied to
impugn the rationaIity of someone's beliefs.
Can inductive logic also be applied to endorse the rationaIity of some-
one's beIiefs? This may not be so dear. Thus, suppose that cr(h) and
c(h, e(t» are one and the same number. Does it foIlow that Jones's
belief about h is rational? Surely, that belief is as it ought to be, given his
temperament and his experience. But are his temperament and experience
as they ought to be?
Let us suppose that the answer to this question about his temperament
is 'yes', and examine his experience. Can Jones's experience be faulted as
inadequate to suppori the thesis that cr(h) is a rational degree of credence
in h for him, at time t? The question is not simply, whether cr(h) is rational
for him at time t, given what his experiences have been, up to that time. We
are also asking, whether his experience at time t should not, perhaps, be a
bit richer than it is.
No doubt, there are situations in which Jones's experience can be so
fauIted, e.g. where h is the statement that it wilI be raining, at times during
the day at which Jones wiII want to be outdoors; and where at time t
Jones is in his bedroom dressing for bad weather, partly on the basis of a
weather forecast he heard the night before, and partly through a passive,
sleepy reluctance to undertake a trip to the window, look out, and see
what sort of day it looks Iike being. Perhaps, in fact, e(t) is such that
crt(h)=c(h, e(t»=.7, but by drawing the blinds, Jones could have
provided himself with a bit more evidence, which might have driven his
credence in h nearly up to 1, or nearly down to O. We (or Jones himself,
Iater in the day), might well fault him for not having bestirred himself to
augment e(t).
It is of no great moment, whether we put the maUer as above - as a
criticism of the sIightness of e(t) - or as a recommendation concern ing
e(t+ 1), viz., that it be the conjunction of e(t) with a report ofwhat Jones
sees when he looks out the window. In either case, the conditions
crt(h) = c(h, e(t» - or, as it may be, crt + 1 (h) = c(h, e(t + 1» - is seen to be
 CARNAP'S INDUCTIVE LOGIC 331

necessary but not sufficient for reasonableness of Jones's credence in h

at the time in question.
MORAL: Experience is only sometimes the Given. It is always the Taken.
(Taking can be rather passive, as when one accepts a gift; but it may also
be rather active - more of a seizing.)
Experience can be had fortuitously, or by design. 1 don't mean that
Jones can decide to look out the window and see that it looks like being a
fine day. But he can decide to look out the window and see what he can
find out that way. In the stylized examples studied in statistical decision the-
ory, one thinks about the maUer in advance and sees that there are a num-
ber of mutually exclusive, collectively exhaustive possible outcomes, 01'
... , an' ofthe act of observation which Jones is contemplating. At the time
when he is considering whether to make the observation (to perform the
act), Jones has definite degrees of credence Pl' ... , Pn in the possible out-
comes, and associates definite 'payoffs' U1, ... , Un with them. Making the
observation has a certain cost, say U; and the observation is worth making
if its expected payoff, Pl U1 + ... + Pnun' exceeds its cost. (In other words
Jones ought to perform the act ( = make the observation) if its expected
utility, P1Ul + ... +Pnun-u, is positive.)
The utility of a possible outcome tends to be greater, the less expected
that outcome is. Thus, since Jones rather expected rain anyway, seeing that
it looks like rain would be a less useful outcome than seeing that it looks
like being fine - not because Jones prefers fine weather to rain, but be-
cause the unexpected outcome, unlike the expected one, would lead him
to take a course of action (dress for fine weather) which he would not
have undertaken without the observation. Thus, the utilities of outcomes
of observations depend on what acts are contemplated. But if one wishes
to express the general tendency of outcomes to be more useful, the less
likely they are, one can take -log Pi to be a measure of the utility of
outcome Oi (insofar as that can be measured without knowing what actions
are in view), and then take the entropy of the observation, -(pllogPl +
+ ... +Pn log Pn), as an action-independent estimate of its expected
utility.
In this way, the element of responsibility for experience can be ac--
comodated within the framework of Carnapian inductive logic, which
then seems adequate 'in principle' to the task of endorsing, as well as
impugning, the rationality of people's credence functions.
332 RICHARD C. JEFFREY

v. IDEALIZATION OR FALSIFICATION?

1 see no knock-down argument (no glory, in Humpty Dumpty's sense)

either for or against the thesis that Carnap's program is feasible and
worthwhile. The program aims at a certain sort of 'rational reconstruc-
tion' of OUT notions of how we ought to proceed, in acquiring what we
like to think of as knowledge. Any such enterprise abounds with elements
which its proponents call 'idealizations' and its opponents call 'falsifica-
tions'. And the opponents are generally right, to begin with. 1 mean (in
the case of Carnap's program) that the business of discovering what our
inductive intuitions are is generally not so much a matter of uncovering
preexistent, covert intuitions, as of creating intuitions: forging an in-
ductive temperament out of materials which were not inductive intuitions
before they pas sed through the Carnapian fire, and rang between that
hammer and that anvil. For my part, 1 find it plausible that there is a
human inductive temperament in the making - less plausible, of course,
that it will emerge with the definite features which Carnap designed for it
over the past twenty-five years or so. But then, Carnap's design kept
changing, as the work went ono It would be no service to him, or to
philosophy, to try to stop the process at the point it happened to have
reached when he had to leave it.

UniversUy of Pennsylvania
 RISTO HILPINEN

CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC*

During his life Carnap published two extensive, systematic works on

inductive logic: Logical Foundations of Probability [4] and The Continuum
of lnductive Methods [5]. These works report Carnap's research in induc-
tion and probability in the 1940's and early 1950's. After the publication
of the Continuum in 1952 Carnap continued his research in inductive
logic, and in the course of this research both his theory of inductive
probabiIity and his philosophy of induction underwent development.
Most of Carnap's publications since 1952 have dealt with the philosophy
and methodology ofinduction rather than inductive logic proper.1 A part
of his recent research in inductive logic will be published in an extensive
work 'The Basic System of Inductive Logic', forthcoming in Studies in
lnductive Logic and Probability, I-II (edited by Rudolf Carnap and
Richard Jeffrey).2
This paper discusses the development of Carnap's system of inductive
logic and his philosophy of induction during the past twenty years. Our
exposition and discussion of Carnap's inductive logic is based mainly on
'The Basic System of Inductive Logic' [13B-C] and partly on an earlier
(unpublished) version of the same work, 'An Axiom System of Inductive
Logic' [13A]. The theory presented in the 'Basic System' will be termed
'Carnap's New System' or the 'Basic System', and that presented in Log-
ica/ Foundations of Probability and the Continuum will be called 'Carnap's
Old System'. Sections II-III of this paper discuss Carnap's philosophy of
induction, Sections IV-VII contain an exposition of the main features of
the 'Basic System', and the concluding Sections VIII-IX discuss the
philosophical implications of Carnap's New System.

* An earlier version of this paper was read in a Symposium on Camap's Philosophy at

Stanford University, Stanford, 1970. This work has been supported by a U.S. Depart-
ment of State Grant No. 70-066-A.

Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 333-359. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
334 RISTO HILPINEN

II

In Logical Foundations of Probability Carnap explains the notion of log-

ical probability from three different points of view. 3 First, a conditional
logical probability c(h, e) can be understood as a measure ofthe degree to
which the hypothesis h is confirmed or supported by the data e. This
interpretation may be termed the degree of cOlifirmation-explication or
briefly the inductivist conception of logical probability and inductive logic.
Secondly, logical probabilities can be regarded as (rational) degrees of
belief or as 'fair betting quotients': according to this explication,
c(h, e) = r means that a bet on h with a betting quotient r is a fair bet if
e describes the total knowledge concerning h available to the bettors. 4
This viewpoint will be termed the belief-explication of logical probability.
Thirdly, logical probabilities can in certain cases be used as estimates of
statistical probabilities (relative frequencies in the long run). 5
The inductivist conception of logical probability is associated with the
classical empiricist philosophy of science, in particular, with the problem
of explicating how general theories or hypotheses are justified (or sup-
ported) by observational data. The belief-explication (or betting quotient-
explication), on the other hand, is c10sely related to utility theory and the
foundations of decision-making under uncertainty. (We might also call
it the 'decision-theoretic conception oflogical probability' .)6 These expli-
cations are not mutually exclusive; they merely illustrate different aspects
and fj.elds of application of logical probability and inductive logic. 7
In the early 1940's Carnap was mainly interested in the c1assical induc-
tivist application of inductive logic; in fact, originally he used the expres-
sion 'degree of confirmation' as a technical term for logical probability
(in order to distinguish it from the statistical concept of probability).8
Carnap assumed that the degree to which an evidential statement e
confirms a hypothesis h is expressible by a conditionallogical probability
c(h, e) = r, where the value r depends only on certain logical (or seman-
tical) relations between hand e.
This conception has been criticized by several philosophers, especially
by Ernest Nagel and Karl R. Popper. 9 Nagel and Popper have argued
that no adequate explicatum of the presystematic notion of confirmation
can have the same formal properties as the concept of conditional proba-
bility. In many cases 'confirmation' seems to mean in crease ofprobability
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 335

rather than just conditional probability; this is indicated, for instance, by

the fact that we speak of disconfirmation as well as confirmation. This
ambiguity in the intuitive notion of confirmation was pointed out by
Carnap in the Preface to the Second Edition of Logical Foundations
of Probability; here Carnap noted that 'confirmation' can mean either the
firmness or the in crease of firmness of a hypothesis (relative to evidential
data).10 Popper's criticism of Carnap's theory seems to be based partly on
a confusion between these two notions. l l In this later work Carnap
preferred the belief-explication and emphasized the application of induc-
tive logic to decision problems.1 2
This change of interest was perhaps not motivated merely by the ambi-
guity in the presystematic notion of confirmation. Even if the degree of
confirmation cannot be expressed as a conditional probability (of a
hypothesis relative to data), it is plausible to as sume that it can be defined
in probabilistic terms, and the theory of logical probability can thus serve
as a foundation for the theory of confirmation. Most formal definitions
of support proposed in recent literature are in fact based on the concept
of probability.1 3 In the case of Carnap's inductive logic, the emphasis on
the decision-theoretic viewpoint may also be motivated by considerations
of a different type. The inductivist conception is primarily relevant to
sÎtuations in which we are interested in whether - and to what degree -
some general hypothesis or theory is supported by empirical data. How-
ever, it has turned out to be very difficult to evaluate the degree of support
of general theories in terms of conditional probabilities; for instance, in
Carnap's Old System all genuinely universal hypotheses with factual
content have zero probability relative to any type of observational evi-
dence.1 4 This difficulty is not specific to Carnap's inductive logic; the
customary statistical methodology is subject to similar problems. 15 The
belief-explication is related to decision-problems, and in this context the
problem of assigning probabilities to general theories does not automat-
ically arise. In practical decision problems it is normally not necessary to
consider the probabilities of universal generalizations. Thus, quite apart
from the ambiguity involved in the inductivist interpretation, Carnap's
inductive logic seems to be more appropriate to applications related to the
belief-explication than to discussion of the classical philosophical prob-
lems of inductive inference, among which questions related to general
theories loom large. 16
336 RISTO HILPINEN

III

Carnap's conception of inductive probability is a variant of the person-

alist or Bayesian conception ofprobability. However, there are important
differences between Carnap' s theory and the customary Bayesian approach.
First, Carnap defines probabilities as relative to a formalized language or
a conceptual system with a fully specified logical structure. In Carnap's
Old System probabilities are assigned to the sentences of a given formalized
language .P. In the New System they are assigned to propositions, and
propositions are identified with sets of models. However, the fields of sets
(or propositions) considered in the New System are defined in terms ofthe
atomic sentences of a formalized language; thus probabilities are relative
to languages in Carnap's New System, too. Preoccupation with formalized
languages or conceptual systems is characteristic of Carnap's philosophy
of science in general; this feature of his probability theory can perhaps be
seen as a consequence of his general philosophical methodology. Inductive
logic is an extension of semantics and can best be studied by similar
methods. However, this interest in conceptual systems is also related
to another aspect of Carnap's theory. Carnap's probability theory is
stronger than the standard Bayesian theory: he discusses and accepts
axioms which are not among the axioms of standard probability theory.
His theory of inductive logic has always been in an unfinished state; even
in his last works he often says that his inductive logic is still "in the initial
phase of the whole construction".17 Carnap did not present a completed
axiom system, but he was constantly attempting to find new axioms and
conditions for logical probabilities. In the 'Basic System' the standard
axioms are termed basic axioms or axioms of coherence; the c-functions
satisfying these axioms are termed regular or coherent c-functions.
Carnap points out that for any pair of contingent and logically independent
sentences h, e and any real number r (O < r < 1), there exists a coherent
c-function c such that c(h, e) = r. This shows, according to Carnap, that
quite unacceptable and unreasonable c-functions are consistent with the
basic axioms. In his paper 'On Inductive Logic' Carnap says that the
systems of standard axioms

restrict themselves to the fust part of inductive logic, which, although fundamental and
important, constitutes only a very small and weak section of the whole of inductive
logic. The weakness of this part shows itself in the fact that it does not determine the
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 337

value of c for any pair h, e except in some special cases where the value is O or 1. The
theorems of this part tell us merely how to calculate further values of c if some values
are given. Thus it is c1ear that this part alone is quite useless for application and must
be supplemented by additional axioms.18

The main objective of Carnap's research in inductive logic from the early
1940's to recent years has been the formulation of such additional axioms
and specification of a more restricted c1ass of probability measures which
would correspond to the intuitive notion of reasonable belief better than
the c1ass defined by the basic axioms alone.
Originally, in the 1940's, Carnap seemed to believe that one can find,
for a given formalized language 2, a unique c-function c which can be
regarded as the explicatum of 'rational degree of belief' or 'degree of
confirmation' (this can be seen, for instance, from the paragraph cited
above). Certain passages in 'On Inductive Logic' and in Logical Founda-
tions of Probability indicate that at a certain time Carnap assumed that he
had in fact found this special measure of confirmation; this confirmation
function was the function c*.1 9 However, in the Continuum of lnductive
Methods he defined a comprehensive c1ass of c-functions, alI of which
satisfy the basic axioms and certain plausible additional requirements. In
this c1ass c* does not seem to be the only reasonable or acceptable c-func-
tion; the main distinguishing feature of c * is only the relative simplicity of
its definition. Carnap did not c1aim any more that there exists a unique
'correct' c-function. In the Continuum the adequacy of an inductive meth-
od in a given universe is characterized in terms of its 'success' in this
universe; this success is measured by the mean square error of the corre-
sponding c-estimate function in the universe in question. Different c-
functions are appropriate in different circumstances, and the choice of an
inductive method (i.e., the choice of a c-function) should thus be depen-
dent on empirical considerations. 20
Later Carnap rejected this viewpoint, and assumed that the choice of a
c-function is, at least partly, a subjective and personal decision which
depends on certain personality traits of the individual concerned, and no
objectively founded choice is possible. One of the relevant personality
traÎts was termed by Carnap "inductive inertia"; this trait is reflected by
the parameter A (the characteristic parameter of the A-system). The larger
A is, the stronger is the inductive inertia of the person in question, and the
slower he modifies his a priori beliefs on the basis of observational evi-
338 RISTO HILPINEN

dence. Carnap calIs this the "personalist point of view", but the term
"psychologism" may be more appropriate. (This view must be distin-
guished from Savage's "personalism".)21
In the 'Basic System' Carnap returned to a more 'objectivist' and ratio-
nalist position and argued again that the choice of an inductive method
should depend, not on observed facts and subjective factors, but only on
a priori considerations. 22 However, he did not maintain that there exists
only one accepta bIe c-function. Now he expressed the folIowing view of the
.nature of inductive logic:
The person X wishes to assign rational credence values to unknown propositions on
the basis of the observations he has made. It is the purpose of inductive logic to help
him to do this in a rational way; or, more precisely, to give him some general rules,
each of which warns him against certain unreasonable steps. The rules do not in general
lead him to specific values; they leave some freedom of choice within certain limits.
What he chooses is not a credence value for a given proposition but rather certain fea-
tures of a general policy for determining credence values. 23

Carnap's aim was to find methods of determining, for each knowledge

situation, the 'correct' or 'acceptable' probability values (or intervals of
probability values) of various propositions on the basis of the conceptual
structure of the situation. The standard probability theory is not sufficient
for this purpose: it says that if the probabilities of certain propositions
are given, then the probabilities of certain other propositions are deter-
mined in accordance with the axioms of probability, but it cannot justify
the total set of credence-values used in a given situation. According to
Carnap's original program, the theory of logical probability is a "strong
extension" of the standard theory in which the structure of the probability
space (or sample space) determines a unique measure or a restricted class
of probability measures,24 and the structure of the probability space is
determined by the structure of language. The latter assumption was
rejected in Carnap's later work. 25 The theory presented in the 'Basic
System' did not lead to a unique measure, but even in his last works
Carnap was optimistic about the possibility of obtaining - at least in
certain special cases - positive results in this direction and restricting the
range of accepta bIe measures on the basis of a priori considerations.
However, Carnap wanted to keep an "open mind" in this respect:
Suppose that at some point in the context of a given problem, say. the choice of a para-
meter value, we find that we have a free choice within certain boundaries, and that at
the moment we cannot think of any additional rationality requirement which would
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 339

constrict the boundaries. Then it would certainIy be imprudent to assert that the?,resent
range of choice wilI always remain open. How could we deny the possibility that we
shaII find tomorrow an additionaI requirement? But it would aIso be unwise to regard
it as certain that such an additionaI requirement wiII be found, or even to predict that
by the discovery of further requirements the range wiII shrink: down to one point. 26

In the 'Basic System' Carnap distinguishes the 'subjectivist' and the 'ob-
jectivist' approach to inductive logic as follows: Subjectivism emphasizes
the existing freedom of choice, whereas objectivism tends to stress the
existence of limitations. Carnap was interested in finding new rationality
requirements, and this led to an objectivist tendency in his inductive
logic. 27 His success (or lack of success) in actually restricting the range of
acceptable probability measures by new axioms is not the philosophically
most significant aspect of his work, however. Carnap developed an inter-
esting theory about certain conceptual factors that influence rational
probability assignments, and this aspect of his work is of fundamental
importance to the philosophy of induction. It has inspired philosophers
who have studied other dimensions of inductive reasoning by similar meth-
ods, for instance the work on inductive generalization by Jaakko Hintik-
ka 28 and Juhani Pietarinen. 29
IV

In Carnap's Old System the functions m (measure function, absolute

probability) and c (conditional probability) are defined on the sentences
of a formalized language 2. In the 'Basic System' probabilities are defined
on events or propositions; propositions are defined as sets of models (of a
formalized language 2). Instead of the functions m and c applied to
sentences, Carnap now uses the corresponding functions Jt and re applied
to sets of models. Following Carnap's notation, we shall denote proposi-
tions by E, H, A, R, etc.; the absolute probability of Hand the probability
of H relative to E are expressed by Jt(H) and re(H lE), respectively.
In Carnap's New System probabilities are defined as measures of sets,
but they are nevertheless relative to a formalized language 2, in partic-
ular, to the descriptive constants of 2. For each type of language discussed
in the 'Basic System', Carnap defines the set of atomic propositions on
!l', ,ffat. Let Zz (briefly Z) be the space ofmodels of!l'. The class ofmolec-
ular propositions on!l', ,ff'J/" is the field of sets generated by,ff'j) on Z, and
the class of propositions on !l', ,ffz, is the u-field generated by ,ff'j) on Z.
The functions Jt and f(f are defined for subfields of ,ffz.30
 340 RISTO HILPINEN

For the most part Carnap discusses only languages with a relatively
simple structure which are termed monadic predicate languages. The New
System is developed in detail only for such simple languages; in this
respect it is similar to the Old System. The descriptive symbols of these
languages include individual constants (names of individuals) al' a2, ... ,
and monadic predicates which denote attributes of individuals. The attri-
butes denoted by the primitive predicates (primitive attributes) are clas-
sified into families; the attributes of a single family belong to the same
general kind or modality. For instance, colors and shapes are modalities;
these are qualitative modalities, but a monadic language 2 can, according
to Carnap, also contain predicates for quantitative modalities or numer-
ical magnitudes such as age, weight, length, etc. It is assumed that the
class of families ~ = {FI, F 2 , ••• , F"} in 2 is always finite, and each
family F m E ~ contains a finite or denumerable number of primitive
attributes P7, P~, ... , Pj, ... 31
The attributes denoted by the predicates of a family form an attribute
space. The points of an attribute space represent the most elementary or
specific properties of the modality in question, but the attributes in F m
are less specific; they correspond to finite regions of the attribute space
u m • These regions form a finite or countable partition of U m , X';', X~, ... ,
where each Xj corresponds to Pj. The distance between any two points
in u m reflects the similarity between the corresponding qualities: the more
two qualities P an P ' resemble each other, the nearer each other their
representative points are in um • A quantitative property of individuals is
based on a measurable magnitude Gm, the points of the corresponding
attribute space represent the possible numerical values of Gm • Such an
attribute space um can thus be termed a va/ue space. In this case different
attributes of F m correspond to parts of a countable partition of the value
space; for instance, if U m is an interval of the set of real numbers, the
attributes of F m correspond to a partition of subintervals X7, X~ . ... ,
Xj, .... The magnitude Gm itself is not denoted by any sign of a monadic
predicate language 2, the language 2 contains only the predicate sym-
bols Pj.33
In addition to the basic probability axioms, Carnap accepts a number
ofaxioms of invariance for..ll and C(j'. According to the axiom of symmetry
for individuals, the ..11- and C(j'-values are invariant with respect to per-
mutations of individuals,33 and according to the axiom of invariance for
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 341

families of predicates, the Jt- and ~-values are invariant with respect to
the introduction of new families of predicates into the language 2. If we
wish to determine the numerical value ofCC(H lE) for given propositions
Hand E, then, by virtue of the axiom of invariance for families, it suffices
to consider only a sublanguage 2' of 2 containing only those families
ofwhich some attributes are involved in H or E. Conversely, ifCC(H lE)
is ba sed on 2', this language can be extended to a new language 2 with
new families of predicates without revising the CC-values based on 2'.34
In the 'Basic System' Carnap discusses mainly a case in which 2 con-
tains only one finite family F with k attributes P 1, P 2 , ••• , Pk • According
to the axiom of invariance for families, the results concerning this lan-
guage hold for any family of attributes in languages containing any num-
ber of families. In the next sections 1 shall describe the main aspects of
Carnap's theory for a single family of predicates.
Carnap's Old System (as presented in Logical Foundations of Probabi-
lity and The Continuum of Inductive Methods) does not concern a single
family of predicates, but a language 2 with several families F m , each of
which contains two attributes, p m and "" pm. However, the Old System
can easily be reinterpreted in such a way that it becomes comparable to
the New System: the Q-predicates defined in terms of the predicates pm
form a partition; let us as sume that the number of Q-predicates in 2 is
k. 35 We shall reinterpret the Q-predicates as the primitive predicates PJ
of a one-family language and ignore the original two-attribute families
F m ; thus alI results and formulae of the Old System become applicable
to a single family with k primitive attributes. This reinterpretation will be
presupposed in the comparisons made below. 36

v
Let Ind be a denumerable set of individuals {al' a 2 , ••• } with an index set
D and let D' be the index set of Ind' cInd. The class of aII atomic propo-
sitions with an individual index i is termed tlft, and tI~, is the class of
atomic propositions with individual indices iED'; thus
~at
0D' = U ~at
0; •
ieD'

E is termed a sample proposition for Ind' if and only if there is a class

.r;1 c tI~, such that for every iED', .r;1 contains exactly one atomic propo-
342 RISTO HILPINEN

sition from tf~t, and E = n.s;1 (the intersection of aII propositions in d).
Thus a sample proposition for Ind' assigns an elementary attribute to
every member of Ind'. Let E. be a sample proposition for a set Ind.; the
index set D. of Ind. consists of the first S indices in D. Thus E. specifies
the primitive attributes ofthe individuals al' a 2 , ••• , a•. For each attribute
indexj, Sj members of Ind. havethe attributePj. Thek-tuples = (SI' ... , Sk)
is termed the k-tuple of E•. 37
In the sequel I shalI employ Carnap's notation: sis any k-tuple (SI' ... ,

Sj + 1, ... , Sk) with the sum S + 1. The k-tuple (O, ... , O) with S =
termed so.38
°
Sj, ... , Sk), where LjSi = s. If s = (SI,·.·, Si' •.. , Sk), sj is the k-tuple (SI' ... ,
is

If two sample propositions have the same k-tuple, they can be transfor-
med into each other by permutations of individuals, and by virtue of the
axiom of symmetry for individuals, they have the same Jt -value. Thus the
measure function Jt can be represented by its representative junction MI
which determines the prior probability of each sample proposition as a
function of its k-tuple. In a similar way, a symmetric confirmation func-
tion can be represented by functions which take numbers (i.e., k-tuples
of numbers) as arguments: let s be the k-tuple of E. and let Hj be the
atomic proposition P ia, + 1 ; the functions CI' ... , C k are termed the repre-
sentative Crjunctions for <ţj' if and only if

C/s) = <ţj'(Hj lE.)

for every j and s. C is the representative C-function for <ţj' if and only if
C(s) = (CI (s), ... , Ck(s), where each Ci is a representative Crfunction
for <ţj'. The subfunctions of Cj and C restricted to k-tuples with the sum S
are denoted by Cj and C'.39
In Carnap's Old System, the probability of a hypothesis H depends on
the logical width of the predicates involved in H. If Pj is an elementary
attribute in a family of k attributes, the (relative) logical width of Pj is,
according to the Old System, always l/k, and the a priori probability of
a hypothesis Pja l is also l/k, i.e.
(1) C/s o) = Jt(Pja l ) = l/k.
In the New System (1) does not always hold. Jt(PPI) is dependent on the
width of P i , but the relative width of an elementary attribute is not always
l/k. The number k indicates the size of a family (the number of different
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 343

basic attributes); in the New System width is not measured in terms of

size, but by a separate width function.
The Jt- and ~-values of hypotheses involving attributes of a family F
depend on the structure of the corresponding attribute space U. The pri-
mitive attributes of F correspond to disjoint and connected regions of the
space U. In the construction of his inductive logic, Carnap accepts the
following methodological conjecture:
(C) Only those properties of Pj (relations between Pj and P h ) are
essential for inductive logic, i.e., for the determination of val-
ues of ~ for propositions involving Pj (and Ph ), which are re-
flected in the topological and metric properties of the corre-
sponding region X j (in the topological and metric relations
between X j and X h).40
More specifically, Carnap suggests that it suffices to take as basic mag-
nitudes relevant to the determination of Jt- and ~-values the width of a
ba sic region X j and the distance between two disjoint regions X j and X h .41
The concept of distance between points in an attribute space is ba sed
on judgments of similarity. If Xl' X 2 and X 3 are regions in an attribute
space U, the distance between Xl and X 2 is shorter than that between Xl
and X 3 if and only if Xl is judged to be more similar to X 2 than to X 3 •
In some cases it is possible to define a quantitative distance function over
the attribute space, that is, a function d such that if u, v and w are points
in U, d(u, v) = O if and only if u = v; d(u, v) = d(v, u); and d(u, v) +
+ d( v, w) ~ d(u, W).42 If the attribute space is the value space of a mea-
surable magnitude, the distance function can be defined in terms of the
accepted scale for the magnitude in question (for instance, spatiallength
and temperature are such magnitudes).
The width of an attribute Pj is, roughly speaking, the size or volume
ofthe region Xj' If a distance function dfor U is available, a widthjunction
w can be defined in terms of d. For instance, if U is al-dimensional con-
tinuous space, it can be mapped in a l-to-l correspondence onto a sub set
of real numbers by means a coordinate x such that if the coordinates of u
and VE U are Xl and x 2 respectively, d(u, v) = X 2 - Xl' In this case we can
take as the width function w the Lebesgue measure on the set of real num-
bers. Each connected region Xc U corresponds to an interval in the set
of real numbers; w(X) is simply the length of this interval. If U is an
344 RISTO HILPINEN

n-dimensional space, the width function can be defined in a similar way

in terms of an n-dimensional coordinate system as an n-dimensional Le-
besgue measure function. 43
Given a width function w, a normalized width function is defined by
(2) w'(Xj ) = w(Xj)jw(U).
The corresponding normalized distance function can be defined by
(3) d'(u, v) = d(u, v)jw(U)
if U is a one-dimensional space; in the case of an n-dimensional space
, d(u,v)
d (u, v) = 11'"
w(U)
If F is a family of k qualitative attributes for which no natural metric is
available, but only a comparative serial order, Carnap suggests the foIIow-
ing choice of w and d: (i) For each of the k primitive attributes, w(Pj) =
= Ijk; (ii) for any two adjacent attributes in the series, d(Pj , P h) = Ijk. 44

VI

In the 'Basic System' the dependence of Jt- and ~-values on the widths
of attributes and distances between attributes is expressed in terms of two
new parameters (or types of parameters), Y and 11. Ifj is an attribute index,
(4) Yj = CJ(so) = Jt(P j a 1 )·
According to Carnap, Yj is equal to the relative width of the attribtite Pj'
If no quantitative concept of width is available, but there is no (a priori)
reason for expecting the occurrence of Pj more than that of P h , we should,
according to Carnap, take Yj = Yh' and if Pj has a greater width than P h ,
we should take Yj > Yh' If w is a quantitative width function,
(5) Yj = aw(Pj) ,
where a is the normalizing constant Ij L~= 1W(Ph)' If w is a normalized
widthfunction, Yj= W(P).45 The y-values for molecular attributes are
determined in an obvious way on the basis of y-values for primitive
attributes.
Carnap's early work, for instance, the system of the Continuum, con-
cerns only a case in which Yj = ljk for every attribute P i . In this case the
Jt- and ~-values satisfy the condition of attribute symmetry. In his
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 345

early work Carnap took this condition as an axio_m for alI 'G'-functions. 46
In the 'Basic System' the condition of attribute symmetry is satisfied only
by certain special families F.
According to (5), the a priori probability of Pjat is equal to the relative
width of Pj; thus Carnap accepts an even a priori distribution over the
attribute space. In the case of very simple attribute spaces, for instance,
in those considered in Carnap's Old System, an even distribution seems
natural, but in more complex cases it is often implausible or even impos-
sible. For instance, if U is the value space of a real-valued quantitative
magnitude G, and the set of possible values of G is not bounded, the
'7-values cannot be determined in terms of a normalized distance function
(3).47 Carnap discusses mainly finite families with bounded value spaces,
but in many cases (e.g. if the value space can be mapped onto a subset of
real numbers) the choice of the boundaries seems partly arbitrary, and
this very arbitrariness indicates that an even a priori distribution over the
whole space is not acceptable. Moreover, in many cases it seems natural
to regard countable and bounded 'observational' families as approxima-
tions to continuous, unbounded 'theoretical' families; this requires that
the corresponding probability distributions be mutually consistent.
If j and h are two attribute indices,

(6) '7jh = CJ (si)/cJ (so)

= 'G'(p j a2 1 Pha 1 )/J1(Pj a2);
hence

(7) 'G'(Pja21 Pha1) = '7jhJl(Pja2).48

Thus the parameter '7jh indicates how the knowledge that some individual
has the property P h affects the credibility of the proposition that some
other (as yet unobserved) individual has the property Pj.
It is natural to regard this influence as dependent on the similarity - that
is, the distance - between the attributes Pj and Ph. The parameter '7 can
be termed an analogy parameter. (7) implies that observation of an in-
dividual with the attribute Ph is positively relevant to the proposition Pja2
if '7jh > 1; if '7jh < 1, it is negatively relevant to Pja2. In the 'Basic System'
Carnap accepts an axiom of instantial relevance according to which

(8) I
'G'(Hj Es (") H) > 'G'(Hj IEs),
346 RISTO HILPINEN

where
Hj=P j a.+ 1 , Hj=Pja.+2' and 0<~(HjIE.)<1.49
This axiom implies
(9) f1jj> 1
for each attribute indexj. Moreover, the similarity between PJ and P" is
maximal when j = h; hence
(10) f1jj > f1jll forevery j#=h.
(10) is termed the principle of self-similarity: Any basic attribute is more
similar to itself than to any other attribute. 50 Since
(11) I I
~(Pja2 Pha1)/.4(Pp2) = ~(Pha1 Pja2)/.4(P"a 1)
(according to the basic axioms), (6) implies
(12) f1"j = fii'"
and according to (5),
(13) ~(Pja21 P"a1) = f1j"Yi·
Since ~(Pja21 P"a 1) < 1, (11) implies that fii" < I/Yj and f1"i < l/y". fi is
always positive; hence (12) implies
(14) 0< f1j" < ljmax(Yj' y,,).
(14) expresses the possible range ofvariation of f1j" in terms ofYi and y".51
Carnap lays down e.g. the following rules for the choice of f1-values:
If no quantitative distance function for U is available, but we see no reason
to regard Pj as more (or less) similar to Pm than to P", we should take
f1im = f1j". If comparative judgments of relative similarity are possible, we
should choose f1jm > f1j" if and only if Pj is more similar to Pm than to
P".52 If a normalized distance function d fpr U is available, the l1-values
may be determined in terms of a function f such that
(15) l1im = f (djm).
This function is termed the l1-function for F. (15) is always positive, and
according to the principle of self-similarity, it assumes its maximum when
djm = O; this maximum is always greater than 1 (cf. [9]). Carnap discusses
the shape of the l1-function in detail; here it suffices to note that it seems
plausible to assume that for sufficiently long distances the similarity in-
fluence has no noticeable effect. This means that for long distances d,
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 347

say d> d*, the '1-function has a constant value '1*.53 The choice of d*
seems to be partly subjective; the problem of the choice of '1* will be discus-
sed below in Section VIII. It is also important to observe that '1jh may
exceed 1 not only at d = O, but also for small positive values of d, that is,
for pairs P j , Ph withj '# h. Observation of an attribute Pj may be positively
relevant to a hypothesis about the occurrence of some other attribute suf-
ficiently similar to Pj. This is genuine (Positive) 'similarity influence'.

VII

Carnap discusses in detail certain special kinds of families, namely

(i) families with Yj = Yh (for every j, h), and
(ii) families with '1jh = '1mn (for alI pairs j '# hand m '# n).
Pamilies of type (i) are termed families with y-equality; those of type (ii)
are termedfamilies with '1-equality.54 Carnap's early work is restricted to
families with y-equality, for instance, in the system ofthe Continuum Yj =
= lJk for every j; in this system '1-equality holds as weB. The principle
of self-similarity (10) can be regarded as an axiom of inductive logic, thus
is it also satisfied by families with '1-equality. In these families we thus
have k + 1 '1-parameters, '1jj for each attributePj and '1jhV '# h). The para-
meter '1jh will be abbreviated '1.
According to (I 1),

(16) L rt'(Pma21 Pjat) = L '1mjYm = 1,

m~ m
where

If F is a family with '1-equality and '1 = '1mj' (I6) and (17) imply

(18) '1jjYj + '1 L Ym = '1jj'1j + '1(1 - Yj) =1;

m*j
hence

(19)

The axiom of instantial relevance implies that '1 Jj > 1 ; consequently '1 < 1.
348 RISTO HILPINEN

In the present case the axiom of instantial relevance implies the principle
of self-similarity. (19) implies
1-'1
(20) '1jj = '1 + --o

"Ij
(19) and (20) show that 0<'1 < 1 and 1 < '1jj < l/"Ij, and '1 and '1jj are
inversely related to each other. "Ij and '1 determine the value of '1jj.
For families with '1-equality it is possible to define ~-functions which
satisfy the following Â-principle:
For any S = (S1' S2, ••• , Sk), Cis)(i.e., rl(PJ.(J&+ 1 IEs» depends
only on Sj and s, but is independent of the other k - 1 mem-
bers of S.55
This principle is inapplicable if the ~-values depend on similarity influ-
ence. However, if we accept '1-equality, the Â-principle may be accepted
too. rl-functions satisfying the Â-principle are termed  - ~-functions. In
Carnap's earlier systems a principle equivalent to the Â-principle was
taken as an axiom. 56 However, it excludes similarity influence and is thus
not generally acceptable; in Carnap's New System it is merely a principle
satisfied by certain special ~-functions.
If C is a C-function for a family of k attributes with '1-equality, and if
the k y-values and '1 are given, alI values of C for S = Oand S = 1 are deter-
mined in accordance with (4) and (6). However, if C satisfies the Â-prin-
ciple, the values of C and ~ are determined for any s, and consequently
alI values of ~ are determined. If C is a representative function for a
 - ~-function,

(21) C .(s) = Sj + "1/1/(1 - '1)

J S + '1/(1 - '1)
for an arbitrary k-tuple s. The representative function of the Continuum
is obtained as a special case of (21): if we define a new parameter Â:

(22) Â = '1/(1- '1),

and assume that the family of attributes under discussion satisfies attri-
bute symmetry (y-equality), (21) simplifies to
Sj + Â/k
(23) Cj(s) = rl(P j as+l IEs) = ,
s+Â
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 349

which is the representative function of the A-system. 57 The A-continuum

is a special case of Carnap's Basic System (for one family of predicates)
in which
(i) The family of attributes under discussion has 17-equality and
~ satisfies the A-principle, and
(ii) alI primitive attributes of the family have the same relative
width lJk, and consequently the family has 'l'-equality.
It is of interest to observe here the dependence of A on 17. The range of
admissible values of 17 is O< 17 < 1 ; according to (22), 17 = O implies A = O
or the 'straight rule' of induction; this value is excluded by the coherence
requirements on ~.17 = 1 implies A = 00 and 17jj = 1; this value is excluded
by the axiom of instantial relevance and the principle of self-similarity.
The difference between 17 and 17jj' and hence the value of 17 (the larger
rt is, the smaller is the difference between rt and 17Jj), indicates the differen-
tiai relevance of Probservations to a hypothesis PJ.fls + 1 in comparison to
other observations. '1-equality implies that on/y Probservations have po-
sitive differential relevance to Pps+ 1. If 17 = 1, Probservations possess no
differential relevance to PJ.fls + 1, and consequently alI observations are ir-
relevant. On the other hand, very small values of '1 overemphasize the
differential relevance of Probservations.

VIII

The definition (22) puts the choice of A into a new prespective. In the
Continuum of lnductive M ethods Carnap distinguished two types of c-func-
tions and inductive methods. In methods of the first kind, the value of A.
is constant, independent ofthe size ofthe family. In methods ofthe second
kind, the value of A for a given family is a function of its size, that is, a
function of k. The simplest function of this type is obviously A(k) = k;
this yields the c-function c*. 58
In Carnap's Old System width is defined in terms of size; the size of a
family is thus an important parameter in inductive logic, and it is perhaps
not entirely implausible to regard A as dependent on k. However, in the
New System the widths of attributes and distances between attributes are
not based on the size of a family, but on separate width- and distance-
functions for the underlying attribute space. The size of a family has little
350 RISTO HILPINEN

theoretical significance; it indicates merely how the attribute space has

been partitlOned into basic attributes. In this case it is not reasonable to
define A as a function of k. In the Basic System inductive methods of the
second kind (including c*) do not seem plausible.
A - '6'-functions are applicable to families with '1-equality, that is, to
families in which the distances between different attributes are equal, and
to 'long-distance families' in which the distances exceed d* and '1jh = '1 = '1*.
'1 is a decreasing function of djh ; consequent1y '1 and A should be the larger,
the smaller is the distance between different attributes. On the basis of
this relationship, it is possible to make in certain special cases objective
comparisons between the A-values appropriate to different families. If the
average distance between different attributes in a family F" is larger than
that in F', we should use a larger '1-value (and hence a larger A-value) for
F" than for F'. An objective comparison between the average distances in
the two families is possible e.g. if the attribute spaces of F' and F" are
subspaces of a common attribute space U.59 However, such comparisons
are possible only in exceptional cases, and they cannot determine the ab-
solute (numerical) value of '1 and A. According to Carnap's assumptions
concerning the shape of the '1-function, all long-distance families with
'1-equality should have the same '1-value '1*, but the choice of the numeri-
cal value of '1* seems partly subjective. There seem to be no compelling
a priori grounds for choosing any particular numerical value for 17*.
In the 'Basic System' Carnap argues, however, that the range of accept-
able A-values for long-distance families can be narrowed down to A = 1
or A = 2 on the basis of a priori considerations. (A = 1 corresponds to
'1 = t, A = 2 to '1 = i.) Let El and E 2 be two sample propositions and let
S1 and S2 be the corresponding structure propositions. (The structure
proposition corresponding to a given sample proposition specifies the
k-tuple of the sample proposition.) 60 According to the A-system, .A (El) >
> .A(E2) if the degree of order or uniformity of El exceeds that of E2'
that is, if the distribution of individuals among different attributes is ac-
cording to E 2 more even than according to El' Carnap suggests that this
inequality can be extended to the structure propositions S 1 and S2' that is,
(24) (i) .A(Sl) ~ .A(S2) or
(ii) .A(Sl) > .A (S2) ,
if S1 exhibits a higher degree of uniformity than S2' If the same A-value
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 351

is used for alI long-distance families, including those with k = 2, (24.i)

holds for alI families only if A = 2, and the stronger condition (24.ii) holds
only if A < 2. If (for reasons of simplicity) we consider only integers as pos-
sible values of A, (24.ii) implies A = 1. According to (24.i), ..1.= 2 is also an
admissible value of ..1.. 61
Carnap shows that A-values A < llead to intuitively unacceptable '(?-val-
I »
ues, for instance, the values of C/s6) (Le., '(?(Pja2 P j a 1 seem too high
or 'overoptimistic'.6{l However, the values Cj (s6) may seem too high even
for A = 1 and A = 2; for instance, if Fis a family of three basic attributes
with ')'1 = ')'2 = ')'3' ..1.= 1 implies Cj (s6) = 0.67 and ..1.=2 implies Cis6) =
= 0.56. Even the latter value may seem too high. Carnap's argument for
..1.= 1 is not very convincing. The postulate (24) expresses a strong a priori
belief in a high degree of uniformity in our universe of individuals. It may
be justified in special cases (for instance, on the basis of suitable back-
ground information), but is hardly accepta bie as a general principle of
inductive logic. In Carnap's New System an argument of this type (Le.,
an argument based on the degree of order of an attribute partition) is
also weakened by the fact that the degree of order of structure proposi-
tions depends on the choice of the attribute partition, but A should not
depend on it.
In the 'Basic System' Carnap says that the definability of A in terms of
the similarity parameter rr supports an 'objectivist' viewpoint in the choice
of A. The value of rr does not depend only on idiosyncratic, personal fac-
tors, but on the structure of the attribute space, viz. on the distance be-
tween different attributes. The concept of distance is based on judgments
of similarity; these judgments are, in a certain sense, 'subjective', but they
are not arbitrary. Moreover, the similarity judgments of different persons
show a high degree of interpersonal consistency.
As was mentioned above, the rr-parameter indicates the differential re-
levance of Probservations to a hypothesis Pja s + 1 in comparison with
other observations. In normal circumstances our judgments of differential
relevance are ba sed on extensive information about past inductions and
on knowledge about the world (including theoretical knowledge). From
this standpoint, the question of the a priori differential relevance of ob-
servations seems impossible to answer or even meaningless. Carnap's pro-
posal to use innate perceptual spaces as a basis of a priori judgments of
differential relevance is an ingenious and highly plausible idea. In the ab-
352 RISTO HILPINEN

sence of factual information, innate similarity judgments form a natural

basis for judgments of relevance.
How are relevance judgments ba sed on perceptual similarity? The
concept of similarity is notoriously elusive; Nelson Goodman has
remarked that "as it occurs inphilosophy, similarity tends under analysis
either to vanish entirely or to require for its explanation just what it
purports to explain".63 If we think of similarity from the behavioral
viewpoint, the 'inductive behavior' associated with relevance judgments
provides a convenient and natural method of measuring similarity.64 If
the structure ofthe perceptual space is determined in this way, it is almost
analytic (true by·definition) that the value of A is determined by similarity
judgments, and alI 'long distance families' have the same A-value. The
value of A cannot be determined by rational a priori considerations, in
fact, there can be no question of 'choosing' (or 'justifying') the value of
A. 1'/ and A reflect an innate, natural standard of 'similarity' which is given,
not chosen. On this interpretation of 'similarity', there is an analytic
connection between similarity and relevance, and Carnap's proposal to
take similarity as the basis of relevance judgments (and his conjecture
(C) as welI) becomes a truism. On the other hand, if perceptual similarity
is measured by psychophysical methods which require explicit verbal (or
'introspective') similarity judgments on the part of the subject, it is
not obvious that these alone should determine relevance judgments. In
this case relevance judgments may depend, not only on perceptual similar-
ity, but on other factors as welI. This viewpoint leaves room for 'person-
alism' (in Carnap's sense) in Carnap's New System, and makes the conjec-
ture (C)0 a problematic hypothesis. .
Similarity is a context-dependent and theory-dependent concept; simi-
larity judgments depend on one's theoretical standpoint. The interest of
Carnap's proposal depends on the existence of non-theoretical, 'natural'
similarity relations. However, quite apart from this question, the general
theory-dependence of similarity relations has interesting consequences
for Carnap's inductive logic. In the 'Basic System' Carnap accepts the
'two-Ievel-conception' of the language of science. According to Carnap,
the Basic System is constructed only for an observationallanguage. How-
ever, if an inductive logic is constructed for the theoreticallanguage, the
transition from the observational level to the theoretical language may
involve a total restructuring of the similarity relations, and consequently
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 353

abrupt changes in the Jt- and 'G'-values. As an example of this possibility

Carnap mentions the family of colors: in the observational language,
similarity relations between colors are based on perceived similarity; on
the theoretical level they are defined in terms of the wavelength of
Iight. The similarity relations (i.e., the distances) between the same
colors are entirely different in the two cases. 65 This result can be gener-
alized: in the same way as Carnap's 'observational' and 'theoretical'
attribute spaces may involve different similarity relations, different theo-
ries may involve different concepts of similarity. This implies that strict
probabilistic comparisons between theories are in some cases in principle
impossible. Thus Carnap's New System vindicates the skepticism expressed
by some philosophers about the probabilistic comparisons between differ-
ent theories. 66
IX

In Carnap's Old System the measure function and confirmation function

are defined for the sentences of a formalized language !l'; in the New
System they are defined for the propositions (sets of models) of a language.
This formal difference reflects a deeper philosophical difference between
the two systems. In the New System Carnap discusses mainly monadic
predicate languages, but the confirmation functions defined for these
languages depend on parameters that cannot be expressed in these
Ianguages, for instance, on the widths of attributes and distances between
attributes. The width function and the distance function are not reflected
in the structure of the language. In the Old System the width of a predicate
can be shown in the language itself by transforming the predicate into
normal form (into a disjunction of Q-predicates), but in the New System
such procedure is not possible.
As was mentioned above, Carnap's aim was to define a 'strong exten-
sion' of probability theory in which the structure of the sample space
determines a unique probability measure or a restricted class of 'accept-
able measures'.67 In the Old System the structure of the sample space is
determined by the structure of the language !l'; in the New System it is
determined by the structure ofthe attribute space U, that is, by the topolog-
ical and metric properties of U (in the case of one family of attributes; in
languages with several families, the structure of the probability space is
determined by the spaces Ut, ... , un). This structure is, in general, much
354 RISTO HILPINEN

richer than the structure ofthe language!l'. According to the Old System,
the conceptual factors important for inductive logic are properties of
language; in the New System they are rather properties of the reality to
which the language is applied. It is perhaps appropriate to say that in the
New System the focus of Carnap's interest has shifted from methodolog-
ical questions to the epistemological foundations of probabilistic reason-
ing. In this respect Carnap's use of the '1-parameter is of great interest.
His discussion of similarity influence and the structure of perceptual space
brings his inductive logic into contact with his own early epistemological
work in which the structure of the world is constructed on the basis of a
single undefined relation of similarity between 'elementary experiences', 68
and also with recent discussions of the epistemological importance of
innate standards of similarity.69
In his methodological writings Carnap often says that the purpose of
inductive logic is to "help people make decisions in a rational way," and
emphasizes the application of inductive logic to practical decision prob-
lems. These statements should perhaps not be understood too literally.
Carnap's work has mainly foundational interest; it is an analysis of the
foundations of probabilistic reasoning. Many ofhis axioms and conditions
are not applicable to actual knowledge situations: they do not concern
the credence- junctions used in actual knowledge-situations, but only
credibility- junctions which are independent of factual information and
from which the credence functions are derived. 70
As was observed above, Carnap was even in his latest works optimistic
about finding new axioms and conditions for rational probability mea-
sures and thus narrowing down the range of admissible ~-functions -
even though he, in his characteristically cautious spirit, wanted to 'keep
an open mind' in this respect. In the 'Basic System' he writes:
1 have repeatedly experienced a development of the following kind, similar to that de-
scribed in this section in connection with the problem of choosing a A.-value. At the
beginning of the investigation there is agreat number, sometimes even an embarrassing
abundance of possibilities to choose from. But then, with the gain of deeper insight
into the situation, the range of choice is gradually narrowed down. 71

However, in many respects the development of Carnap's inductive logic

exhibits also the opposite pattern: not only did he find new axioms and
conditions that re strict the range of acceptable probability functions, but
he also had to reject conditions previously accepted. For instance, the
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 355

condition of attribute symmetry and the Â-principle were first tentatively

accepted as axioms, but on the basis of new insights and developments it
was realized that these conditions are satisfied only by certain special -
though very important - probability measures. Thus, new insights into
the foundations of inductive inference not only restricted the possibility
choice, but also opened up completely new possibilities. This is especially
clear in the case of Carnap's discovery of 1] and Hintikka's discovery of
inductive methods for general hypotheses. 72 These developments do not
detract from the value and importance of Carnap's research. Carnap,
perhaps more than anyone el se, has contributed to our understanding of
the complex factors underlying rational probability assignments - and
this, of course, is the principal task of the philosophy of induction.

University of Turku
NOTES

1 These publications include Camap [6], [8], [9]. [10]. and [11]. Certain developments
ofCamap's theory of inductive probability are described in Carnap and Stegmiiller [14],
Appendix B. and in Camap [7].
2 Part 1 (the tirst 13 sections) of this work has appeared in 1971 in VoI. 1 of Studies in
lnductive Logic and Probability [13B]; Part II will be published in the forthcoming
VoI. II of Studies in lnductive Logic and Probability [13C].
3 Camap [4). pp. 164-175.
4 [4). pp. 165-167.
5 [4). pp. 168-175.
6 This conception of the application of inductive logic has been presented in detaiJ by
Camap in [6) and [12]. ([12) is a revised and expanded version of [6).)
7 Some philosophers have assumed that betting on general hypotheses and theories
does not make sense. and the betting interpretation (or belief-interpretation) and the
contirmation-interpretation of inductive logic are therefore somehow imcompatible. See
Hacking [17). pp. 215-216. and Lakatos [29]. p. 361. This prejudice against betting on
general hypotheses appears unfounded. however; see Hintikka [25]. pp. 339-340 and
Pietarinen [32). p. 26. See also note 16 below.
8 Cf. Camap [4). pp. 19-25.
9 Cf. Nagel [31), and Popper [33). pp. 251-276 and [34]. pp. 213-216.
10 Camap [4B). pp. xv-xix. See also Camap [7). p. 967.
11 The confusions involved in the 'Popper-Camap-controversy' are discussed in detail
in Michalos [30); see especially Chapter III. Cf. also Bar-Hillel [1).
12 Cf. Carnap [7). pp. 967-969.
13 Several measures of support proposed in recent litera ture are discussed and com-
pared in Kyburg [28]. See also Hintikka [22), pp. 328-329.
14 Cf. Camap [4]. pp. 570-571.
15 Cf. Hintikka [24).
16 However. the inductivist or conflrmation-theoretic applications of inductive logic
356 RISTO HILPINEN

can also be studied from the decision-theoretic viewpoint; see Hilpinen [19]. The con-
firmation-theoretic and the decision-theoretic conception of inductive logic are not in-
compatible.
17 Carnap [13A], section 18 ('Lambda-families').
18 Carnap [3], p. 76. Cf also [12], p. 15.
19 See Carnap [3], section 6, and [4], p. ix and pp. 562-563.
20 Cf. Carnap [5], pp. 56--79.
21 Carnap [13A], section 18 ('Lambda-families').
22 Cf. Carnap [11], pp. 313-314, and [l3A], section 18.
23 [13A], section 18 ('Lambda-families').
24 Cf. Suppes [37], chapter 3, pp. 100-101.
25 Cf. section IX of this paper.
26 Carnap [13A], section 18 ('Lambda-families').
27 [13A], 'Lambda-families'. 'Objectivism' heredoes not mean 'empiricism', but 'ratio-
nalism', according to which it is possible to restrict the c1ass of 'G'-functions by rational
a priori considerations.
28 Cf. Hintikka [20], [21], and [23].
29 Pietarinen [32] has studied the problems of analogy and lawlikeness by the methods
developed by Carnap and Hintikka.
30 Carnap [l3B], pp. 40-41.
31 [13B], p. 43.
32 [13B], pp. 43-44.
33 [13B], pp. 117-118.
34 [13B], p. 121.
35 For Q-predicates, see Camap [4], pp. 122-126.
38 This reinterpretation of the Old System is described in Camap [7], p. 973-974.
37 [13B], pp. 121-122.
38 [13B], p. 62.
39 Representative MI-, Cj- and C-funetions are diseussed in detail in [13B], sections
11-12 (pp. 131-160).
40 Carnap [13C], section 15.
41 [13C], sectionl5.
42 [13C], section 14B ('Distance and Width').
43 [13C], section 14B.
44 [13C], section 14B.
45 [13C], section 16A ('The Analogy Influence: Gamma- and Eta-parameters').
46 Cf. Carnap [7], p. 975, and Carnap and Stegmiiller [14], p. 244.
47 Carnap [13C], section 14B.
48 [13C], section 16A. Cf. also note 72.
49 [13B], p. 161 This principle is not independent of the other axioms of inductive
logic; it can be proved from the basic axioms, the axiom of symmetry (for individuals),
and the axiom of eonvergence ('The Reichenbach Axiom'); ef. Humburg [26].
50 [13C], section 16C (The Search for Principles of Analogy by Similarity').
51 [13C], section 16A.
52 [13C], section 16B ('Rules for the Eta-parameter').
53 [13C], section 17A ('The Eta function: General Considerations').
54 [13C], section 18 ('Some Special Kinds of Families').
55 [13A], section 18 ('Lambda-families'). 1. J. Good ([15], p. 26) has termed this prin-
eiple 'W. E. Johnson's sufficieney postulate'; it has been accepted by Johnson in [27].
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 357

56 The axiom (C9) on p. 14 of [5] is equivalent to the J.-principle.

57 [13A], 'Lambda-families'.
58 Carnap [5], sections 11 and 15.
59 In [13A], section 18 ('Lambda-families') Carnap discusses an example of this type
in which the families F' and FU are subfamilies of the family of colors.
60 For structures and structure-propositions, see Carnap [13B], p. 123.
61 This argument is presented in Carnap [13A], section 18 ('Lambda-families').
62 [13A], section 18 ('Lambda-families').
63 Goodman [16], p. 29.
64 Cf. Quine [35], p. 123.
65 See Carnap [13B], pp. 51-52.
66 Cf. e.g. Bar-Hillel [1].
67 Cf. Section III of this paper.
68 Cf. Carnap [2].
69 Cf. Quine [35] and Stemmer [36].
70 These concepts are explained in Carnap [12].
71 Carnap [13A], 'Lambda-families'.
72 Additional new possibilities of this kind are created by the extension of inductive
methods to relationallanguages; cf. Hilpinen [18]. (Added in proof) Notethatthel1-para-
meter discussed here must be distinguished from that defined in [14], p. 251, and in
Carnap's articIe in the present volume. The latter parameter concerns analogy between
different families (inter-family analogy); the l1-parameter described here is related to
intra-family analogy.

BIBLIOGRAPHY

[1] Yehoshua Bar-Hillel, 'Popper's TheofY of Corroboration', in The Philosophy of

Karl R. Popper (ed. by P. A. Schilpp), Open Court Publ. Co., La Salle, III.,
1974, pp. 332-348.
[2] Rudolf Carnap, Der logische Aufbau der Welt, Felix Meiner Verlag, Leipzig
1928. (English translation The Logical Structure ofthe Worldby Rolf A. George,
University of California Press, Berkeley and Los Angeles, 1967)
[3] Rudolf Carnap, 'On Inductive Logic', Philosophy of Science 12 (1945), 72-97.
Reprinted in Probability, Confirmation, and Simplicity (ed. by M. H. Foster and
M. L. Martin), The Odyssey Press, New York, 1966, pp. 35-60.
[4A] Rudolf Carnap, Logical Foundations of Probability, The University of Chicago
Press, Chicago, 1950.
[4B] Rudolf Carnap, Logical Foundations of Probability, 2nd Edition, The University
of Chicago Press, Chicago, 1962.
[5] Rudolf Carnap, The Continuum of Inductive Methods, The University of Chicago
Press, Chicago, 1952.
[6] Rudolf Carnap, 'The Aim of Inductive Logic', in Logic, Methodology and Philo-
sophy of Science (ed. by E. Nagel, P. Suppes and A. Tarski), Stanford University
Press, Stanford, Calif, 1962, pp. 303-318.
[7] Rudolf Carnap, 'Replies and Systematic Expositions V. Probability and Induc-
tion', in The Philosophy of RudolfCarnap (ed. by P. A. Schilpp), Open Court
Publ. Co., La Salle, 111 , 1963, pp. 966-998.
[8] Rudolf Carnap, 'Probability and Content Measure', in Mind, Matter and Method
358 RISTO HILPINEN

(ed. by P. K. Feyerabend and G. Maxwell), University of Minnesota Press,

Minneapolis, 1966, pp. 248-260
[9] RudolfCarnap, 'Inductive Logic and Inductive Intuition', in The Problem of Jn-
ductive Logic (ed. by 1. Lakatos), North-Holland Publ. Co., Amsterdam, 1968,
pp.258-267.
[10] Rudolf Carnap, 'On Rules of Acceptance', in The Problem of Jnductive Logic
(ed. by 1. Lakatos), North-Holland Publ. Co., Amsterdam, 1968, pp. 146-150.
[11] RudolfCarnap, 'Reply to J. Hintikka', in The Problem of Jnductive Logic (ed.
by 1. Lakatos), North-Holland Publ. Co., Amsterdam, 1968, pp 312-314.
[12] Rudolf Carnap, 'Inductive Logic and Rational Decisions', in Studies in lnductive
Logic and Probability, VoI. I (ed. by R. Carnap and R. Jeffrey), University of
California Press, Berkeleyand Los Angeles, 1971, pp. 5-31.
[BA] RudolfCarnap, 'An Axiom System of Inductive Logic' (unpublished; a pre-
Iiminary version of 13B-C)
[13B] Rudolf Carnap, 'The Basic System of Inductive Logic, Part 1', in Studies in
Jnductive Logic and Probability, VoI. 1 (ed. by R. Carnap and R. Jeffrey), Univer-
sity of California Press, Berkeley and Los Angeles, 1971, pp. 33-165.
[13C] Rudolf Carnap, 'The Basic System of Inductive Logic Part II', in Studies in
lnductive Logic and Probability, VoI II (ed by R Jeffrey), University of Califor-
nia Press, Berkeley and Los Angeles (forthcoming).
[14] Rudolf Carnap and Wolfgang Stegmilller, Jnduktive Logik und Wahrscheinlich-
keit, Springer Verlag, Wien, 1959.
[15] Irving John Good, The Estimation of Probabilities, Research Monograph No 30,
The MIT Press, Cambridge, Mass, 1965.
[16] Nelson Goodman, 'Seven Strictures on Similarity', in Experience and Theory
(ed. by L. Foster and J. W. Swanson), The University of Massachusetts Press,
Amherst, 1970, pp. 19-29.
[17] lan Hacking, The Logic of Statistical lnference, Cambridge University Press,
Cambridge, 1965.
[18] Risto Hilpinen, 'Relational Hypotheses and Inductive Inference', Synthese 23
(1971),266-286.
[19] Risto Hilpinen, 'Decision-Theoretic Approaches to Rules of Acceptance', in Con-
temporary Philosophy in Scandinavia (ed. by R. E. Olson and A. Paul), The Johns
Hopkins Press, Baltimore, 1972, pp. 147-168.
[20] Jaakko Hintikka, 'A Two-Dimensional Continuum of Inductive Methods', in
Aspects of Jnductive Logic (ed. by J. Hintikka and P. Suppes), North-Holland
Publ. Co., Amsterdam, 1966, pp. 113-132.
[21] Jaakko Hintikka, 'Induction by Enumeration and Induction by Elimination', in
The Problem of Jnductive Logic (ed. by 1. Lakatos), North-Holland Publ. Co,
Amsterdam, 1968, pp. 191-216.
[22] Jaakko Hintikka, 'The Varieties of Information and Scientific Explanation', in
Logic, Methodology and Philosophy of Science III (ed. by B. van Rootselaar and
J. F. Staal), North-Holland Publ. Co., Amsterdam, 1968, pp. 311-332.
[23] Jaakko Hintikka, 'Inductive Independence and the Paradoxes of Confirmation',
in Essays in Honor of Cari G. Hempel (ed. by N. Rescher et al.), D. Reidel Publ.
Comp., Dordrecht, 1969, pp. 24-46.
[24] Jaakko Hintikka, 'Statistics, Induction and Lawlikeness: Comments on Dr.
Vetter's Paper', Synthese 20 (1969), 72-83.
 CARNAP'S NEW SYSTEM OF INDUCTIVE LOGIC 359

(25] Jaakko Hintikka, 'Unknown Probabilities, Bayesianism, and de Finetti's Repre-

sentation Theorem', in Boston Studies in the Philosophy of Science, VoI. VIII
(ed. by R. C. Buck and R. S. Cohen), D. Reidel Publ. Comp., Dordrecht, 1971,
pp. 325-341.
[26] Jiirgen Humburg, 'The Plinciple of Instantial Relevance', in Studies in lnductive
Logic and Probability, VoI. I (ed. by R. Carnap and R. Jeffrey), University of
California Press, Berkeley and Los Angeles, 1971, pp. 225-233.
[27] W. E. Johnson, Appendix (ed by R. B. Braithwaite), to 'Probability: Deductive
and Inductive Problems', Mind 41 (1932), 421-423.
[28] Henry E. Kyburg, Jr., 'Recent Work in Inductive Logic', American Philosophical
Quarterly 1 (1964), 249-287. '
[29] Imre Lakatos, 'Changes in the Problem of Inductive Logic', in The Problem of
lnductive Logic (ed. by 1. Lakatos), North-Holland Publ. Co., Amsterdam, 1968,
pp.315-417.
[30] Alex C. Michalos, The Popper-Carnap Controversy, Martinus Nijhoff. The Hague,
1971.
[31] Ernest Nagel, 'Carnap's Theory of Induction', in The Philosophy of Rudolf Car-
nap (ed. by P. A. Schilpp), Open Court Publ. Co, La Salle, 111.,1963, pp. 785-825.
[32] Juhani Pietarinen, Lawlikeness, Analogy, and lnductive Logic, Acta Philosophica
Fennica 26, North-Holland Publ. Co., Amsterdam, 1972.
[33] Karl R. Popper, The Logic of Scientific Discovery, Hutchinson, London, 1959.
[34] Karl R. Popper, 'The Demarcation Between Science and Metaphysics', in The
Philosophy of RudolfCarnap (ed. by P. A. Schilpp), Open Court Publ. Co., La
Salle, 111., 1963, pp. 213-226.
[35] W. V. Quine, 'Natural Kinds', in Ontological Relativity and Other Essays (by
W. V. Quine), Columbia University Press, New York and London, 1969, pp.
114--138.
[36] Nathan Stemmer, 'Three Problems of Induction', Synthese 23 (1971), 287-308.
[37] Patrick Suppes, Set-Theoretical Structures in Science, Institute for Mathematical
Studies in the Social Sciences, Stanford University, Stanford, 1967.
 THEO A. F. KUIPERS

A GENERALIZATION OF CARNAP'S
INDUCTIVE LOGIC

1. Carnap's continuum of inductive methods for a family of monadic

predicates, the so-called A-continuum or A-system, is restricted by the as-
sumption that the size of the family is known at the start of the learning
process. In the proper inductive situation we do not have this information.
In this article a two-dimensional system will be constructed that is
appropriate to deal with the proper inductive situation. This system
describes a learning process both for relative frequencies of properties
and for the size of the family; it may therefore be called a generalization
of Carnap's continuum.

2. The A-system can be formulated in an interpreted language as folIows.

The individuals (tI' t 2, ... ), called trials, are successive throws with a baU.
The ball has at most k colours (P1 , P2' ... , Pk ) on its surface. The result of
a trial is supposed to be uniquely determinable as the colour of the rest-
point.
Let eS state the result of the first S trials, such that it includes the infor-
mation, for every Pj, how many trials Sj have resulted in Pj. Let hj be the
hypothesis that the next trial will result in Pj.
According to the A-system the 'special value' inp (h j le"), the inductive
probability of hi on the evidence e', is: (Si + A.jk)j(s + A). A is a parameter
for a positive real number.

3. The fundamental assumption of the A-system is that the size of the

family of colours is known to be k. Let us now assume that we do not
have this information, but that alI other things remain the same.
In order to obtain a suitable terminology we will give each new ob-
served colour a new name in a systematic way. Ifthere have been observed
n different colours in the first S trials and if at the (s + 1)th trial a new
colour is instantiated then we will give that colour the new name Pn +1 •
Let e S state the result of the fust S trials, such that it includes the infor-
mation how many different colours n are instantiated, and Sj(1 ~ i ~ n),

Jaakko Hintikka (ed.). Rudo/f Carnap. Logical Empiricist. 361-363. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company. Dordrecht-Holland.
362 THEO A. F. KUIPERS

the number of trials that have resulted in P i • The hypothesis that the next
trial will result in PiCI :::;:; i:::;:; n) is symbolized by hi . H is the hypothesis
that a new colour will be instantiated at the next tria!.

4. Because h i logically implies '" H we can prove easily the following

basic theorem:

I I
inp(h i ej = inp( '" H ej'inp(h i Ies & '" H); 1:::;:; i:::;:; n:::;:; s.

The second factor at the right si de is obviously equal to the corre-

sponding special value of the A-system with k = n, viz. (Si + A/n)/(s + A).
But this function can also be obtained independent1y along the same li nes
as Carnap has done for the 2-system.
In order to determine the first factor· in the above equation we have to
analyse the learning process in detai!. There are essentially two sequences
of events: the sth trial results in a certain colour resp. that colour is, or is
not, already instantiated before the sth tria!. The second kind of event
is a second order event for it is a relation between the sth event of the
first sequence and all the preceding events in that sequcnce. A con se-
quence ofthis relational aspect is that Hand'" H start to be interesting after
the first trial. The relative frequency ofthe event 'a new colour' remains the
ratio ofthe number ofinstantiated colours to the number oftrials.
Because there are in the second sequence only two possible events, viz.
'a new colour' or 'a previously instantiated colour', and because these
events are mutually exclusive, the second sequence is a proper example
to be treated by the A-system with k = 2. There is no reason to identify
the parameter for this sequence with that of the first; therefore we intro-
duce the new parameter 6, again for a positive real number.

l
The two special values of the second sequence that we obtain in this
wayare:

I
inp (H ej = n_+_o_/2
s+o 1~ ~
s-n+u1:/2 "" n "" s.
I
inp( '" H ej = 1 - inp(H ej = I -s-+-o-

Combining both results we obtain as the special values of the first

sequence for instantiated colours:
 A GENERALIZATION OF CARNAP'S INDUCTIVE LOGIC 363

. I
mp(h i ej =
s - n + D/2 Si + Ain
. ---;
s+D s+A
Using the laws ofthe probability calculus we can extrapolate the system
to other than special values. The complete system will be called the
Â.-D-system. The formulation of the whole system, however, will be much
less transparent than the formulation that could be used to present the
special values.
It remains to be investigated whether the Â.-D-system has any short-
coming relative to the inductive situation under discussion. But in the
last section we will only try to clarify some aspects of the system.

5. lf n does not exceed a certain finite N and if s goes to infinity, then

the special value for an instantiated colour goes to sJs, the same as in the
Â.-system. Under the mentioned conditions inp(HI ej goes to O; this
value is the relative frequency of the event 'a new colour' in an infinite
domain if the size of the family of colours is finite.
If s is very large and n is small relative to s then it is very probable but
never certain that n is the actual size of the family. In this situation it may
be better to introduce, for computational reasons, the new property Pn+l
defined as the complementary property of the first n properties, and we
obtain a Carnapian problem situation with n + I properties and Sn+l = O.
This procedure can certainly also be followed at an earlier stage, but then
we close our eyes for new and perhaps valuable distinctions.

Centrale lnterfaculteit,
Universiteit van Amsterdam
 WILHELM K. ESSLER

HINTIKKA VERSUS CARNAP

Let D be a denumerably infinite domain of elements and L a language of

first order logic whose sentences formulate propositions over this
domain. Carnap's inductive methods may then be characterized as
strictly coherent methods which are symmetric both in relation to
individual terms and predicates, and which make it possible to learn
from experience, whereby limes accounts for the value in the infinite case,
and the only relevant factor for a positive singular hypothesis (say
'alSeF3') is whether or not the things named in the experiential data fall
under the notion in question ('F3 '). The totality of methods meeting
these conditions can be characterized by a parameter Â. It can then be
shown that on the basis of finite experiential data (concerning this infinite
domain) every inductive method c always assigns the value O to univers-
aIly quantified sentences. 1 This contradicts our intuitions about adequate
inductive methods, which, already according to Aristotle, should lead
from the particular to the general. Hence, we are faced with two possible
consequences:
(1) hold on to our intuitions and reject Carnap's methods;
(II) hold on to Carnap's methods and reject our intuitions.
Hintikka drew the first of these two consequences 2 and weakened the
condition of irrelevance to the effect that for the value of a positive
hypothesis it is also relevant which notions have an empty and which a
non-empty extension in that part of the domain covered by the experien-
tiaI data. Universal quantifications also receive then positive values on
the basis of finite experiential data for infinite domains - indeed the
simplest receives the largest. With the accumulation of new experiential
data compatible with this simplest of universal quantifications its con-
firmation value approaches ever closer to 1. Thus it appears that Hin-
tikka's methods are superior to Carnap's.
In the foIlowing I should like to show, however, (1) that as far as
generalization processes are concerned Carnap's methods are not inferior
to Hintikka's confirmation methods, (2) that Hintikka's methods do not

Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 365-369. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Hol/and.
366 WILHELM K. ESSLER

conform to certain requirements that must be made of confirmation

methods, and (3) that Hintikka's methods do have the edge on Carnap's
in one area where Carnap's, however, were never intended to be used.
In this sense then consequence (II) must be drawn, i.e., our intuitions
concerning the use of the expression 'induction' have since Aristotle been
partially in error.
(1) Let's take a look at sciences with laws that are not represented by
mathematical functions, e.g., biology or everyday physics. The pre-
sumption that the domains over which such disciplines make claims are
denumerably infinite is wholly unconvincing. We mortals have always
encountered but finite numbers of such things and will in the future
continue to be confronted with mere finite numbers of them. The assump-
tion of denumerably infinite domains is an idealization enabling mathe-
maticians to pass from certain sums to integrals and utilize then the
apparatus of analysis when applying inductive methods. Such idealiza-
tions are often useful, but not, however, always. In particular, they prove
to be of little use where the application of Carnap's confirmation methods
are concerned. This is because Carnap's methods do assign universal
quantifications positive values which (with consistantly positive experien-
tiaI data) increase as Â. decreases, when the domain (be it ever so large) is
finite.
The same holds for disciplines utilizing metric concepts and mathemat-
ical functions to exhibit laws when the domain is identified with the class
of realizations of a certain experimental instruction. In this case too the
domain will always be de facto finite. To assume a denumerably infinite
class of such realizations is to make a mistake that can lead to others.
When in such a case D is supposed to be denumerably infinite, one
cannot determine with Carnap's methods that each, or no thing has a
certain property, i.e., that the objective probability in question (in the
infinite case and in finite parts) is 1 or O, but only that the actual value
(of the objective probability) has the degree of confirmation (the logical
probability) l-r. lying within a certain interval (e.g., [0.99; 1] or
[0.01; O]; in other cases, e.g., [0.495; 0.515]). Thus, with these methods
one can, in terms of statistics, only estimate intervals, not points. Selecting
a value from such an interval is then no longer a matter of applying a
method of confirmation, but rather a matter of applying criteria such
as simplicity, or simplicity tempered by other results and accepted laws.
 HINTIKKA VERSUS CARNAP 367

(For example, from the interval [1; 0.99] the value 1 might be selected
as being the simplest. But when, by virtue of other results, one knows
that counter-examples are to be expected, a somewhat smaller value
might be selected instead; a possibility prec1uded by Hintikka's methods
which in this case dictates the value 1.)
(2) In theories of subjective as weB as logical probability estimations of
a quantity are introduced as mean values. For example, Kemeny and
others identify the estimation of winning in a betting system with the sum
of the products of the single possible winnings and the probabilities of
their realizations (the quotient disappears, because the sum of the probab-
ilities is 1).
In statistics the expectation value is defined as the mean value of ob-
jective probabilities weighed with their (objective, subjective or logical)
probabilities. With respect to finite domains in inductive logic we have
therefore the equation: with respect to a datum <jJ the c-estimation of
relative frequency of a property in some part of the domain is the sum of
products of the possible relative frequencies (of the property in that
domain) with the c-values ofthese relative frequencies relative to <jJ. Using
this definition we get the result that for some strict1y coherent methods
singular predictions are identical with estimations of relative frequencies
resp. expectation values, namely those being symmetric to individual
constants. Therefore this equation holds especially for Carnap's A-meth-
ods and for Hintikka's ex-A-methods.
Now take a Hintikka-method being different from alI Carnap-methods.
Let k be the number of observations being positive instances of a gener-
alization and let k ~ ex. Then the estimation of relative frequency with
respect to this datum differs from 1 with a small value, say 0.000001, and
the c-value for the generalization regarding that datum is near to 1, say
0.999. Let x be an imaginary person with human powers of cognition,
whose span of life has no outer bounds, e.g. the personification of a
temporal infinite sequence of scientists, and suppose that x is using that
Hintikka-method at betting with someone else on that generalization
regarding the given datum. If the domain is infinite, x cannot win (since
the generalization then induces an infinite betting system) but may Iose -
and x knows that he may Iose since the expectation value is smaller
than 1, i.e. the method does not rule out the possibility of counter-
examples. This method, therefore, does not give us adequate probability
368 WILHELM K. ESSLER

values at applications like betting in infinite cases, and, since this method
was arbitrarily chosen, the same holds for all Hintikka-methods different
from Carnap-methods.
But Hintikka-methods don't give us adequate probability values at
applications in finite domains neither. For let the cardinal number of
the domains be large but finite and let k<t,n, being the normal case in
applying induction methods. Let u be the degree of uniformity of the
domain, where u is the objective mean value (i.e. the sum of the products
of the relative frequencies of the property in the domain with their ob-
jective probabilities); then there exists a Carnap-method CA being optimal
in applications, e.g. in betting systems (and vice versa). Therefore no
Hintikka-method being different from all Carnap-measures is optimal in
a domain with a certain degree of uniformity, and therefore no method
of this kind gives us adequate probability values in finite cases of applica-
tion.
(3) Hintikka's inductive methods are standardized additive measures
and in this sense, but only in this sense, probability functions or con-
firmation methods. As far as the question of estimations is concerned,
they deviate considerably from what one expects of probability functions.
This does not, of course mean that they must be discarded. On the
contrary, applied to universal quantifications as premises they offer the
only explication to date for the notion "degree of simplicity of a universal
quantification with respect to finite experiential data", showing, thus,
how the simplicity notion may be applied where empirical laws are
concerned. (Where concepts or rule systems describing the use of ex-
pressions are concerned, a different notion of simplicity seems to be
involved. 3) They indicate which of those universal quantifications that
are neither falsified nor confirmed by the experiential data are the sim-
plest in respect of this data. Hence they offer the scientist an exact proce-
dure for selecting simple laws from the totality of those compatible with
experience in place of his hitherto ad hoc intuitive practice.
This notion of simplicity for empirical laws must still be generalized
in two respects. First, it must be widened to include statistical hypotheses
in its range of applicability, not just universal and existential quantifica-
tions. Second, it must establish a relationship not only between hypo-
thesis and data, but also between hypothesis, data, and other accepted
theories (cf. the end of(1».
 HINTIKKA VERSUS CARNAP 369

NOTES

1 Cf. [3], p. 217ff.

2 Cf. [5], [6], [7] and [8].
a Cf. [4], p. 44ff.

BIBLIOGRAPHY

[1] Carnap, Rudolf: 1962, Logical Foundations ofProbability, Chicagol , 19502 •

[2] Carnap, Rudolf: 1952, The Continuum 01 Inductive Methods, Chicago.
(3] Essler, Wilhelm K.: 1970, Induktive Logik - Grundlagen und Voraussetzungen,
Freiburg(Miinchen.
[4] Essler, Wilhelm K.: 1973, Wissenschaltstheorie III - Wahrschein/ichkeit und
Induktion, Freiburg(Miinchen.
[5] Hintikka, Jaakko: 1964, 'Towards a Theory of Inductive Generalizations', in:
Proceedings olthe 1964 International Congress lor Logic MethodologyandPhilosophy
01 Science, Jerusalem, pp. 274-288.
[6] Hintikka, Jaakko: 'On a Combined System of Inductive Logic', in: Acta Philo-
sophica Fennica 18 (1965), 21-30.
[7] Hintikka, Jaakko: 'A Two-Dimensional Continuum of Inductive Methods', in:
Hintikka, J. and Suppes, P. (eds.), Aspects 01 Inductive Logic, Amsterdam, 1966,
pp.113-132.
[8] Hintikka, Jaakko: 'Induction by Enumeration and Induction by Elimination', in:
1. Lakatos (ed.), The Problem of Inductive Logic, Amsterdam, 1968, pp. 191-216.
 JAAKKO HINTIKKA

CARNAP AND ESSLER VERSUS

INDUCTIVE GENERALIZATION

In his interesting and clarifying note,! Essler sets up a contrast between

Carnap's inductive logic,2 in which alI nontrivial generalizations obtain
the value O as their probability in an infinite domain, and mine,3 where
suitable generalizations can get confirmed by increasing evidence and can
even have asymptotically the degree of confirmation one. I suspect that
this contrast is not a completely accurate one historicalIy, in the sense that
Carnap's inductive logic was probably shaped less by philosophical rea-
sons for assigning zero probabilities to genuine generalizations in an in-
finite domain than by the logical and mathematical difficulties one en-
counters in mastering such probability distributions as enable us to discuss
also inductive generalization. 4 (Cf. below my discussion of (3).) Be this as
it may, the three arguments outlined by Essler in his note in favor of
Carnap do not seem to me to get to the root of the matter.
(1) Essler's first line of thought follows c10sely Carnap's defense of his
notion of instance-confirmation of laws. BasicalIy it amounts to arguing
that generalizations are only idealizations which do not concern the con-
crete situations we actually encounter in science or in everyday life. For
these practical purposes, on1y inferences from finite evidence to finite
events are involved. Hence (1 take Essler's point to be that) the prior
probabilities of generalizations in an infinite domain do not matter for the
intended applications ofCarnap's inductive logic.
This is simply an incomplete picture of the logical situation, as Essler
is undoubtedly well aware. For there is (as shown by the results of de
Finetti, Gaifman, etc. 5) a strong interdependence between the degrees of
confirmation ofjinite events onjinite evidence and the prior probabilities
of generalizations. When someone has made up his mind how to bet on
any finite event on finite evidence, he has ipso facto made up his mind
what probabilities to associate with generalizations, we may say, assuming
on1y the symmetry (exchangeability) ofthe underlying probability distri-
bution. It folIows as a corolIary that the traditional dilemma as to
whether induction is inference from particu1ars to particulars or from

Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 371-380. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
372 JAAKKO HINTlKKA

particulars to generalizations is empty if one believes that inductive in-

ference can be handled by means of symmetrical probabilities.
Thus the question whether nontrivial generalizations receive zero
probabilities a priori has palpable consequences for the finite betting
situations to which Essler apparent1y wants to restrict his, or rather
Carnap's, attention.
For instance, suppose the question is whether alI members of a finite
uni verse satisfy a generalization whose counter-examples have together
the width (01. In Carnap's 1950 system6 (which corresponds to the choice
A=the number of different Q-predicates), the degree of confirmation is
approximately (for large evidence) (n/N)ro1, where n is the number of
(completely known) individuals in the sample and N the number of
individuals in the domain (universe). From this it is seen that the degree
of confirmation in question is non-negligible only if nIN is non-negligible,
that is, only if our sample already exhausts an appreciable part of the
whole universe. For instance, a generalization concerning alI electrons
is not to be trusted before we have verified it for an appreciable portion
of the electrons in the universe, assuming that it is finite.
Of course different choices of A give us different degrees of confirma-
tion. Ifwe choose a small (but still finite A), it may admittedly happen that
a strict generalization receives an appreciable degree of confirmation on
the basis of relatively limited evidence in a large but finite universe.
However, other sorts of absurdities will then ensue. Given any fixed A, the
degree of confirmation of a generalization will depend on the size of the
domain, which has to be known in order for us to be able to assign a
definite degree of confirmation to the generalization. Thus a Carnapian
phycisist who sets up a generalization, say, concerning aU eIectrons must
make the reliability of his generalization conditional on cosmologists'
conclusions concerning the finiteness (and the size) ofthe universe. Ifthis
size turns out to be much greater than anticipated, the physicist's generali-
zation must (on Carnap's principles) be allotted a much lower degree of
confirmation than earlier.
If you find results of this kind unacceptable, you have in your hands
evidence against Camap's handIing of generalizations, taken from the
very kinds of situations to which Essler wants to restrict the applicability
of Carnap's methods. Further evidence to the same effect is easily
forthcoming.
 CARNAP VERSUS INDUCTIVE GENERALIZA TION 373

Thus the problem of inductive generalization just cannot be disen-

tangled from the evaluation of the performance of Carnap's methods in
the finite realm (i.e., from the evaluation of the way they assign prob-
abilities to singular events on finite evidence). Furthermore, Essler's way
of posing the general problem of his note is now seen to be misleading.
In 80 far as the confirmation procedures prescribed by my inductive
logic are superior to Carnap's, this superiority will be in evidence already
in the finite realm. My reasons for preferring the former are therefore
not due to any unarticulated 'intuition' concerning inductive generaliza-
tion, but on an evaluation of the performance of different inductive
logics in finite as well as infinite domains and on their performance
vis-a-vis singular inductive inference and not just vis-a-vis inductive
generalization.
Although the point is not very central, it also seems to me that Essler
is oversimplifying when he implies that practical men do not care what
always happens (Cf. "We mortals have always encountered but finite
numbers of such things and will in future continue to be confronted with
mere finite numbers of them.") It may be that a Detroit executive only
cares whether most of the automobiles he produces keep running for a
certain limited period of time. But it is more questionable whether an
engineering company is satisfied when told that most of their bridges are
likely to last so many years. Inductive logic ought not to be tailored to the
idea of planned obsolesence, it seems to me. And in any case one of the
main concerns of insurance companies is, not the probability that they
will go broke next year, but the probability of their ever going broke.
Highly sophisticated mathematics is in fact being employed to that
undoubtedly practical purpose.
(2) Essler's main specific criticism against my inductive logic is an
alleged inconsistency (in some weak sense) between the degrees of
confirmation it assigns to generalizations and the degrees of confirmation
(posterior probabiIities on finite evidence) it assigns to singular events.
Essler (folIowing Carnap) takes the Iatter probabilities to be estimates of
relative frequency in the whole universe. He argues that on the basis of
such estimates the posterior probability of alI nontrivial generalizations in
an infinite universe must be zero, thus apparent1y contradicting the assump-
tion that they have nonzero probabilities.
It is in fact true that degrees of confirmation for singular predictions will
374 JAAKKO HINTIKKA

on quite general assumptions equal estimates (in the sense of expected

values or probabilistic averages) of the relative frequency of individuals of
the corresponding kind in the whole universe. Carnap already proved
results to this effect in his Logical Foundations ofProbability, pp. 540-557,
and they can be generalized further.
However, what Essler's observations in reality iIlustrate is not a flaw in
my system but a general feature of sequences of experiments which (like
successive draws of an individual from a Carnapian domain) are not
(probabilistically) independent of each other. In such cases, estimates (in
the Carnapian sense of probabilistic averages) of a quantity just cannot
be interpreted as one's beliefs as to what will happen to this quantity in the
long run, just because subsequent experiments will inevitably change the
estimate ofthis quantity.
Yet this is precisely Essler's line of thought. He interprets the fact that
one's so-called estimate of relative frequency of individuals of a certain
kind is 8>0 to imply that in a sufficient1y large sample the probability of
there being at least one individual of this kind to be close to one. But to
draw this conc1usion is to treat subsequent observations of individuals
as being independent of each other, which in turn means ceasing to learn
from experience after the finite number of 'experiments' comprised
in one's evidence e.
In general, we can thus see that for probabilisticaIly dependent se-
quences of experiments the terms 'estimate' and 'expected value' for a
probabilistic average are misnomers. Because of the dependence, such an
'estimate' cannot reflect one's expectations as to what happens in the
long run. Neither Carnap nor Essler emphasizes this fact sufficiently.
These observations may be illustrated in different ways. As a simple
example, take a coin which is (on the evidence we are assumed to have)
as likely to be a fair normal coin (probability of tails t) as to be a two-
tailed one (probability of tails 1). (We know that it is of the one or the
other sort.) It is tossed once, and shows tails. What is the rational betting
ratio (degree of confirmation) of its showing taiIs at the next toss? Clearly
i. 7 But this betting ratio cannot by any stretch of imagination be under-
stood as a rational expectation or an educated guess concerning the
relative frequency of tails in an infinite sequence of tosses. This frequency
just cannot be i or in the vicinity of i, for this frequency is known to be
either t or 1. (The rational guess in the circumstances envisaged might
 CARNAP VERSUS INDUCTIVE GENERALIZA TION 375

perhaps be 1.) Note also that, on the assumptions we have made, interval
estimation has no sense here.
More generally, assume a certain finite body of evidence e consisting
of n fully known individuals. Then the probability P (Cto(a)/e) concerns
only the next individual a to be observed. After it has been observed, the
probability P (Cto(b)/e & Cto(a))) that a further unknown individual b
satisfies Cto(x) need no longer equal P (Cto(a)/e), though it is trivially
true that P (Cto(b)/e)=P (Cto(a)/e), for we now have an additional item
of evidence Cto(a) at our disposal. And this discrepancy between the
situation on the evidence e and on the evidence (e & Cto(a») (or on the
evidence (e & '" Cto (a», for the matter) shows that the degree of confir-
mation P(Cto(a)/e) just cannot be taken to reflect a rational inductivist's
beliefs about the relative frequency of individuals satisfying Cto(x) in the
whole universe. This can be done only if P(Cto(a)/e)=P(Cto(b)/(e &
Cto(a»)=P(Cto(c)/(e & Cto(a) & Cto(b)) = ... But this would mean that
aU the further individuals are independent of each other probabilistical1y.
This in turn would mean that we stop heeding experience when we have
reached the evidence e. And this is in so many words said to be ruled out
in Carnap's own approach. On these general grounds, then, degrees of
confirmation of singular predictions, even though they equal estimates of
the relative frequencies of the corresponding kinds of individuals in the
whole universe in Carnap's technical sense of 'estimate', they cannot be
understood as a rational agent's actual expectations concerning these
relative frequencies on the basis of any nontrivial inductive logic.
The line of thought I have criticized does not originate with Essler, but
goes back to Carnap (1950), p. 168, where we read: "Since the probabilitYl
[degree of confirmation] of h on e is intended to represent a fair betting
quotient, it will not seem implausible to require that the probabilitYl of h
on e determine an estimate of the relative frequency of M in K" (Carnap's
italics). Here we can perhaps have a glimpse of Carnap's motivation. In
the technical sense of the word, it is nonsense to ask whether degrees of
conflrmation can plausibly be identified with estimates of relative fre-
quencies. It follows from Carnap's other assumptions that they must be
so identified. What Carnap is in reality saying is therefore that these other
assumptions have to be chosen in such a way that this conc1usion fol1ows.
But why? Why did Carnap require these other assumptions to be such as
to entail the identity of degrees of confirmation of singular predictions
376 JAAKKO HINTIKKA

and expected values (estimates) of relative frequencies? Clearly because he

was worried about the concrete 'empirical' or 'operational' meaning of
his probabilitYl (degree of confirmation). He wanted to tie it somehow to
actual relative frequencies. It was this purpose that seemed to be served
so very well by the identification of degrees of confirmation with esti-
mates of relative frequency.
However, what we have seen shows that this idea does not serve the
purpose it was supposed to serve. Carnap's 'estimates' of relative frequen-
cy cannot be equated with one's expectations ofwhat happens inthe long
run, and in the absence of such an equation the idea does not contribute
to the operationalization of probabilitYl' Notice the jump in Carnap's
own formulation from a betting ratio to a relative frequency. From the
former you can get to the latter only by means of supplementary assump-
tions, such as independence.
Of course what Carnap was trying to do is indeed reasonable and even
important. But the partial operationalization he wanted to achieve is
more complicated than he thought. The true connection between induc-
tive probability and relative frequency is spelled out in de Finetti's
famous representation theorem. 8 This theorem shows that any probability
distribution, including the one defined by Carnap's prior probabilities
PC -) as weII as his degrees of confirmation on finite evidence P( - le), in
effect define, and are defined by, a second-Ievel probability distribution,
i.e., a probability distribution on first-IeveI probabilities. Now one natural
way of viewing these second-Ievel probabilitities for the purposes of de
Finetti's theorem is to interpret them as an assignment of probabilities to
relative frequencies of individuals of different kinds in the whole universe.
The confirmation (epistemic) aspect then comes into play through the
probabilities we associate with them. This is in fact the precise generaliza-
tion of Carnap's own procedure of assigning prior probabilities to struc-
ture-descriptions and then dividing each of these probabilities evenly
among the corresponding state-descriptions. (For what a structure-
description specifies is precisely the frequencies of individuals having the
different Q-predicates in the whole universe.) This is what happens in his
1950 system in a special case, and this is what happens in his A-continuum
in all finite domains.
What the evidence e effects is then a transition from one such prob-
ability distribution on limiting relative frequencies to another. Hence
 CARNAP VERSUS INDUCTIVE GENERALIZATION 377

Carnap's suggestion that certain degrees of confirmation be interpreted

as estimates of relative frequencies is in an ironical contrast to his
attitude elsewhere to objective hypotheses. Re insists that we should not
accept definite hypotheses on evidence, but rather let the evidence change
our probability distribution on the various alternative available hypo-
theses. Yet the Carnapian suggestion under discussion means that he
proposed to deal with relative frequencies by means of (tentative) accep-
tance (estimates) rather than in terms of changing probability-distribu-
tions, as he ought to have done. (This 'ought to' refers back to our ob-
servations above.) If there is an inconsistency to be found in this area, it
therefore !ies in Carnap's approach, not in mine.
It does not help to try to drive a wedge between acceptance and estima-
tion, either. The hypothesis one accepts is simply one's estimate of the
true hypothesis, and conversely the main purpose of estimates apud
Carnap is to supply us with guides to action, i.e., guidelines for acceptance.
The only way out here would be to view estimation as a purely formal
operation. This, however, would deprive it of all value for the purpose
for which we saw Carnap to use it in the case at hand.
Notice how well the idea of reshuffiing one's probability distribution
on different relative frequencies (in the long run) fits our !ittle example
above. The effect of obtaining tails in the first toss of our coin changed the
probability of t frequeney from t to t and likewise changed the probabili-
ty of 1 frequency to t, as one ean verify by means of Bayes' formula.
The unsatisfaetory characterofCarnap's suggested operationalization of
degrees of confirmation is likewise indicated already by the fact that on it
relative frequencies compatible with strict generalizations in an infinite
universe virtually never oecur as Carnapian estimates even when ac-
cording to the underlying probability distribution on all relative frequen-
cies (as determined by de Finetti's theorem) they do receive nonzero
probabilities.
Rence Essler's second line of criticism is inconclusive.
(3) Essler formulates his third line of criticism as follows: "Rintikka's
inductive methods are standardized additive quantities and in this sense,
but only in this sense, probability functions or confirmation methods.
As far as the question of estimation is concerned, they deviate consider-
ably from what one expects of probability functions" (Essler's italics).
In so far as this is not just another way of putting Essler's other
378 JAAKKO HINTIKKA

criticisms, it apparently amounts to saying that while the probability

distributions on which my inductive logics are based are arbitrary,
Carnap's inductive logic somehow has a deeper justification. Unfortu-
nately Essler does not indicate what this deeper justification might be.
The well-known derivation of Carnap's Â-continuum from a simple
assumption concerning the form (the arguments) of the representative
function is of course the most plausible candidate for this role.
Be this as it may, probably the most effective way of rebutting Essler's
charges would be to show that Carnap's Â-continuum arises through an
arbitrary choice of parameters from a wider continuum of inductive
methods which can be motivated through simple general arguments and
which in most cases allow for inductive generalization also in an infinite
universe. Such a wider class of inductive methods have recent1y been
studied in a joint paper by Ilkka Niiniluoto and myself. 9 This class of
inductive policies is defined by the corresponding representative func-
tions. They are now allowed to depend, over and above the two arguments
of these functions that are considered in Carnap's Â-continuum, also on
the number c ofnonempty Q-predicates in the sample.
A couple of explanations may be in order here. The representative
function specifies the probability, on the basis of a sample of n completely
observed individuals, that the next one has a given Q-predicate. The
Â-continuum comes about when it is required that the representative
function depends only on n and the number n' of individuals in the
sample which have this Q-predicate. The new class of inductive policies
arises when the representative function is ofthe formj(n, n', c), where c
is the number of Q-predicates instantiated in the sample. Of course it is
also required that the representative function has to create a coherent
probability distribution.
Taking c into account is of course inevitable un1ess one refuses to
countenance inductive generalization at all. For how can one hope to
take into account the possibility of generalization if one does not even
allow one's inductive strategies to depend on the number of generaliza-
tions compatible with one's sample? Thus taking c into account is surely
"what one expects of probability functions", Essler notwithstanding.
An examination of the new class of inductive methods shows that all
of them will assign nonzero prior probabilities to nontrivial generaliza-
tions, except Carnap's Â-continuum, which thus emerges as the most
 CARNAP VERSUS INDUCTIVE GENERALIZATION 379

pessimistic choice of an inductive method compatible with Niiniluoto's

and Hintikka's basic assumption concerning the form of the representa-
tive function. It consists of precisely those members of the new wider
class of inductive methods in which c is disregarded. From the point
of view of this basic assumption, to opt for Carnapian methods thus
comes close to a deliberate a priori refusal to recognize the possibility
of inductive generalization. Accordingly it seems to me that the less
pessimistic (non-Carnapian) choices of inductive methods are less
arbitrary than the Carnapian one.
These choices result in a class of inductive methods which overlaps
with my two-dimensional oc-A-continuum but does not coincide with it.
Several important methods, among them my 'Jerusalem system' and the
whole of 'generalized combined system' belong to both.
Thus tables can be turned completely on charges of arbitrariness
against inductive methods allowing for the confirmation of inductive
generalizations.
NOTES

1 Wilhelm K. Essler, 'Hintikka versus Carnap', this issue, p. 229. Cf. also Wilhelm K.
Essler, Induktive Logik: Grundlagen und Voraussetzungen (Verlag Karl Alber, Freiburg
and Munich, 1970), especially pp. 342-351.
2 Rudolf Carnap, Logical Foundations of Probability (University of Chicago Press,
Chicago, 1950); The Continuum of Inductive Methods (University of Chicago Press,
Chicago, 1952).
3 See especially 'A Two-Dimensional Continuum of Inductive Methods', in Jaakko
Hintikka and Patrick Suppes, editors, Aspects of Inductive Logic (North-Holland,
Amsterdam, 1966), pp. 113-132, 'Towards a Theory of Inductive Generalization',
in Yehoshua Bar-Hillel, editor, Logic, Methodology and Philosophy of Science: Pro-
ceedings of the 1964 International Congress (North-Holland, Amsterdam, 1965), pp.
274-288.
4 See p. 977 of Carnap's 'Replies' in P. A. Schilpp (ed.), The Philosophy of Rudolf
Carnap (The Library of Living Philosophers, Open Court, La Salle, Illinois, 1963).
Carnap says there that he has "constructed c-functions of this kind [viz. assigning
nonzero probabilities to generalizations even in an infinite domain], but they are
considerably more complicated than those of the Â-system", and for this reason un-
satisfactory in Carnap's view.
5 Bruno de Finetti, 'Foresight: Its Logical Laws, Its Subjective Sources', in H. E.
Kyburg and H. E. Smokler, editors, Studies in Subjective Probability (Wiley, New
York, 1964), pp. 93-158 (translation of 'La prevision', Ann. de l'Inst. Henri Poincare7
(1937), 1-68); Haim Gaifman, 'Concerning Measures on First-Order Calculi', Israel
Journal of Mathematics 2 (1964),1-18. Cf. also Dana Scott and Peter Krauss, 'Assign-
ing Probabilities to Logical Formulas', in Hintikka and Suppes, editors (note 3 above),
pp. 219-264, and Jaakko Hintikka, 'Unknown Probabilities, Bayesianism, and de
380 J AAKKO HINTIKKA

Finetti's Representation Theorem', in Roger C. Buck and Robert S. Cohen, editors,

PSA 1970: In Memory of Rudolf Carnap (Boston Studies in the PhiIosophy of Science,
VoI. 8, D. Reidel, Dordrecht, 1971), pp. 325-341.
6 See Logical Foundations (note 2 above), pp. 570-571.
7 This is seen by a straightforward application of Bayes' Theorem.
8 See de Finetti, 'Foresight' (note 5 above).
9 Jaakko Hintikka and Ilkka NiiniIuoto, 'An Axiomatic Foundation for the Logic of
Inductive Generalization', forthcoming in the proceedings of the 1974 Warsaw Con-
ference on Formal Methods in the Methodology of Empirical Sciences.
 ABNER SHIMONY

CARNAP ON ENTROPY

Introduction to 'Two Essays on Entropy' by Rudolf Carnap

I. HISTORY

Carnap's two essays on entropy were written during his tenure of a fellow-
ship at the Institute for Advanced Study in Princeton between 1952 and
1954. The 'Autobiography' in the Library of Living Philosophers volume
tells about conversations concerning entropy with mathematicians and
physicists in Princeton
I certainly leamed very much from these conversations; but for my problems in the logica!
and methodologica! analysis of physics, I gained less help than I had hoped for. At that time
I was trying to construct an abstract mathematica! concept of entropy, analogous to the
customary physica! concept of entropy. My main object was not the physica1 concept, but
the use of the abstract concept for the purposes of inductive logic. N evertheless, I also
examined the nature of the physica! concept of entropy in its c1assica! statistica! form, as
developed by Boltzmann and Gibbs, and I arrived at certain objections against the custom-
arydefinitions, not from a factual-experimental, but from a logica! point ofview. It seemed
to me that the customary way in which the statistica! concept of entropy is defined or inter-
preted makes it, perhaps against the intention of the physicists, a purely logica! instead of
physica! concept; if so, it can no longer be, as it was intended to be, a counterpart to the
classica!'macro-concept of entropy introduced by Clausius, which is obviously a physica!
and not a logica! concept. The same objection holds in my opinion against the recent view
that entropy may be regarded as identica! with the negative amount of information. 1 had
expected that in the conversations with the physicists on these problems, we would reach,
if not an agreement, then at least a clear mutual understanding. In this, however, we did not
succeed, in spite of our serious efforts, chiefly, it seemed, because of great differences in
point ofview and in language. [1963, pp. 36-7.]

Because of the discouraging reception of his ideas, Carnap abandoned

his original plan to publish the two essays together, and then decided
against the immediate publication of the two separate essays. Theyare
listed separately under 'To Appear: New Works' in Carnap's Biblio-
graphy [Benson 1963, p. 1054J, but according to Professor Arthur
Benson he had kept open the possibility of making revisions. They are
now being published as Two Essays on Entropy by Rudolf Carnap,
edited with an introduction by Abner Shimony, The University of
California Press, Berkeley and Los Angeles, 1976. The present essay also
serves as an introduction to that volume.

Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 381-395. Ali rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
382 ABNER SHIMONY

II. CRITICAL EXAMINATION OF THE STA TISTICAL

CONCEPT OF ENTROPY

Any simple classification of conceptions of entropy is certain to be inade-

quate because of the many considerations involved: various levels of
description, the large number of components in a thermodynamic system,
permutability of identical components, dynamics of an isolated system,
interaction of a system with different types of environment, uncertainty
and information, and experimental control. One rough way of classifying
conceptions, however, which is of great philosophical interest, is along a
spectrum from entropy as an objective physical property to entropy as a
non-physical (perhaps subjective, perhaps logical or epistemological)
concept. Of the two great classical mas ters of statistical mechanics,
Boltzmann and Gibbs, the former was somewhat closer to the objective
end of the spectrum and the latter to the non-physical end, but in fact the
conceptual differences between them were probably much less sharp
than some of their respective followers have made out. Carnap unequiv-
ocally locates himself at the extreme physical end of the spectrum, and
by giving a lucid formulation and defense of this conception of entropy
he has made a significant contribution to the literature on the subject
Unfortunately, a fuH exposition of the other extreme of the spectrum did
not exist at the time that Carnap wrote his essay, even though the non-
physical conception was 'in the air'. Had Jaynes' information-theoretical
formulation of statistical mechanics [1957] been available to Carnap,
his essay might have gained both in sharpness of focus and in historical
accuracy, by taking the formulation of Jaynes rather than of Gibbs as the
target in his analysis of the relation between entropy and uncertainty.
The core of Carnap's argument is given in § 6. Entropy in thermody-
namics is asserted to have the same general character as temperature,
pressure, heat, etc., aH of which serve "for the quantitative characteriza-
tion of some objective property of a state of a physical system". The
thermodynamic description of a system is presumably correct, but it is
very coarse in comparison with the precise microscopic description
Dprec, which in c1assical mechanics consists in the specification of the
positions and momenta of aH the constituent partic1es. In order to relate
the thermodynamic description to finer descriptions, inc1uding the most
precise one, the thermodynamic concepts must have counterparts
 CARNAP ON ENTROPY 383

defined in terms of the microscopic theory. For Carnap, a controlling

principle in the search for such counterparts is the principle of physical
magnitudes, essentially asserting that if a physical quantity is ascribed a
definite value (with a certain precision) at a given time by a coarse de-
scription, then a finer description of the system at the same time must
ascribe a compatible value to the quantity. To use the terminology which
Carnap has recommended elsewhere [1950, p. 3J, his principle is a partial
explication of 'physical quantity', and indeed a very plausible one. Both
the constructive and the critical parts of Carnap's essay are almost
corollaries of the foregoing statements.
Carnap's main constructive result (§5-§6) concerns a monatomic dilute
gas in an insulating container, the diluteness guaranteeing that the energy
of interaction per particle is negligible compared to the average kinetic
energy of individual particles. Ris treatment is modelled upon that of
Boltzmann, who supposed that the 6-dimensional ţl-space ('ţl: for
'molecule', 3 dimensions for position, 3 for momentum) available to a
particle was divided into a number K of cells, where K is much greater
than 1 but much less than the number N of particles, so that the average
number of particles per ceH is not too small. A specification of the numbers
Nj(j= 1, ... , K) of particles in each of the ceHs is called by Carnap 'a
statistical description', D S!. Boltzmann's own concept of entropy can be
expressed in terms of the statistical description as
SB(D S' )= -k L N j InNj+C,
where C is a constant depending on N but otherwise independent of the
N j' and the factor k (Boltzmann's constant) has only the significance of
relating a combinatorial expression to standard physical units. Carnap
then proposes an extension of Boltzmann's concept,

for any description Di which implies DSt, i.e., consistent with Ds! and at
least as fine. The extended concept evidently satisfies the principle of
physical magnitudes. What Carnap proposes is in the spirit of Boltz-
mann's objective and microphysical approach to the foundations of
thermodynamics, and it is even more evidently in the spirit of the mono-
graph of Boltzmann's disciples, P. and T. Ehrenfest [1959, especially
Sections 14c, 24c, and 28]. (Indeed, Carnap's essay can be profitably read
384 ABNER SHIMONY

together with this monograph, for he presents fully some ideas which are
highly condensed by them, while they present physical considerations,
such as ergodicity, which are lacking in Carnap's exposition. Mutual
illumination is especially provided by the Ehrenfests' Sections 14c and
14d, and Carnap's discussion of the H-theorem in §9.)
Carnap restates Boltzmann's results on the agreement of statistical
entropy SB and thermodynamic entropy sth as follows. Let L1t be a time
interval of duration sufficient for the measurement of a macroscopic
quantity like temperature or pressure, and let D~~p (not Carnap's notation)
be a typical DSt, in the sense that its ceH numbers equal approximately the
respective averages of the time-varying ceH numbers N j during L1t.
Boltzmann gave arguments which are powerful, though - because of
difficulties in ergodic theory - not entirely definitive, that if the gas has
been kept in isolation sufficiently long then D~~p will with overwhelming
probability be such as to make SB(D~~p)very c10se to the maximum value
of SB compatible with the constraints offixed energy, volume, and number
of partic1es. (Lebowitz and Penrose [1973] present an excellent sum-
mary of the problems and results of ergodic theory.) He also showed that
this maximum is equal to the value of sth when the gas is in equilibirum
under the given constraints. Carnap's general position suggests at this
point an option, which he does not mention, of making thermodynamic
as weH as statistical entropy satisfy the principle of physical magnitudes
by simply identifying sth with SB(D~~p). The agreement of statistical and
thermodynamic entropy would thereby become a tautology, and Boltz-
mann's fundamental result would be restated, with no loss in depth, as
asserting that sth (interpreted as a time-averaged microscopic quantity)
has with overwhelming probability the value predicted by thermo-
dynamic theory and also the value obtained by standard operations for
measuring entropy.
Another way (caHed 'Method II' by Carnap) of extending Boltzmann's
concept to descriptions other than DS! is to define the following concept
(or variant of it in which the individual descriptions are unequally
weighted):
Sg(D;)=dfk Inm(D;) + C.
Here Di is equivalent to the disjunction of m individual descriptions (i.e.,
descriptions specifying the ceH to which each of the N individual partic1es
 CARNAP ON ENTROPY 385

is assigned, rather than merely giving the occupation numbers of the

cells} By proper choice of the constant C' the equality of S~(DSI) to
Sk(DSI) is easily established Unlike Sk(D;), however, S~(Di) does not
satisfy the principle of physical magnitudes, as can be seen by considering
a D SI which does not assign all particles to the same celL Because of the
permutability of the individual particles
m(D SI ) N!
N 1 !N 2 !. .. N k !
which is greater than 1. But for any individual description Dind, m(D ind ) = 1.
Hence, even if Dind implies Dsl one has the inequality
S~(DSI)"# S~(Dind),

contrary to the principle of physical magnitudes. Since the entropy as

defined by Method II depends upon the specificity of the description Di'
it is a logical or epistemological rather than a physical concept Those
statistical mechanicians who conceive of entropy as a measure of lack of
information are committed to something like Method II.
Gibbs' innovations were motivated by an aim at generality (whereas
Boltzmann's methods were most appropriate to systems with a large
number of identical particles, preferably weakly interacting) and also
freedom from dubious assumptions about the microscopic constitution
of matter [1948, pp. ix-x]. Accordingly, he represents the state of a physi-
cal system with f degrees of freedom (where f is very large in the case of a
thermodynamic system) not by occupation numbers in the cells of Jl-space,
but by a point U in a 2fdimensional phase space, with f dimensions for
the generalized coordinates and f for the conjugate momenta lnciden-
tally, Carnap is incorrect in crediting Gibbs with introducing this rep-
resentation into statistical mechanics, for Boltzmann had previously
considered ensembles of systems described by a density p(U) over the
phase space [1871]. What Gibbs did that was entirely new was to identify
thermodynamic quantities with parameters or functionals of p(U} He
examines a number of possibilities regarding entropy [1948, ch. XIV,
'Discussion of Thermodynamic Analogies']; see also the commentary
by Epstein [1936, pp. 491-509]. The concept which Carnap attributes to
Gibbs is
Sg(Di )= -k L Pi(Q~) InPi(Q~)+const.
n
386 ABNER SHIMONY

(8-3, 8-5, slightly rewritten} Here the phase space has been divided into
cells Q~ of finite volume, and Pi(Q~) is the probability that an arbitrary
member of the ensemble is in Q~ if the ensemble is described by the density
Pi(U) associated with the description Di' i.e.,

Pi(Q~)= f
QY n
Pi(U) dU

(8-1, rewritten). Sg(D i) is thus the concept of entropy which the Ehren-
fests call 'coarse-grained' [1959, p. 54], in contrast with the fine-grained
J
entropy - k P In P dU, which is not helpful in considerations of irre-
versibility since it is constant in time. Gibbs himself did not write down a
formula which can be regarded as a transcription of the above expression
for sg (D;). The ascription to him of the concept of coarse-grained entropy
is not on this account erroneous, for it is surely implicit in the discussion
of irreversibility in the famous Ch. XII, where no formulae are written,
and Tolman's Gibbsian exposition [1938, pp. 165f1] makes essential use
of it. Nevertheless, the fact that in the discussion of thermodynamic
analogies Gibbs prefers to use either the fine-grained entropy, or else
special expressions appropriate for particular ensembles, may indicate a
different philosophical viewpoint than the one attributed to him by
Carnap, a point which will be discussed in comment 4 below.
Carnap's objection to sg (D i) as a statistical counterpart of thermo-
dynamic entropy is evident, if Di is permitted to be a description of arbi-
trary specificity. The more specific Di is, presumably the more concen-
trated in fewer cells. Consequent1y, Sg(D i) depends upon the specificity
of Di and does not conform to the principle of physical magnitudes.
lndeed, as Carnap explains in the paragraph containing 8-6, Sg(D i) has
the characteristics of a concept constructed in accordance with Method II.
A number of comments will now be offered about the cogency of
Carnap's critique of classical statistical mechanics. Some of them can be
made with confidence, either in favor of Carnap's position or against it.
Some, however, raise difficult problems, concerning which only brief
summaries of arguments and references to recent literature can be given.
(1) There are many thermodynamic entropies, corresponding to
different degrees of experimental discrimination and different choices of
parameters (Grad, [1961] and [1967]; Jaynes [1965]; but see cautions in
 CARNAP ON ENTROPY 387

Tisza and Quay [1963, footnote 22]~ For example, there will be an in-
crease of entropy in certain senses, but not in others, as a result of mixing
samples Of0 16 and 018; and one ofthe former senses would be appro-
priate if (and, it might be argued, only ii) isotopes are experimentally
distinguished Jaynes therefore claims that "Even at the purely phe-
nomenological level, entropy is an anthropomorphic concept. For it is a
property, not of the physical system, but of the particular experiment
you or 1 choose to perform on it" [1965, p. 398]. Carnap's point of view
can clearly accommodate a multiplicity of thermodynamic entropies, to
each of which the principle of physical magnitudes would be applied when
statistical counterparts are sought That human considerations dictate
the use of one or another of this family of concepts does not detract from
the objective character of each member of the family. (See Griinbaum
[1973, Ch. 19], for a criticism of Jaynes' statement.)
(2) An objection may be raised against applying the principle of physi-
cal magnitudes below the phenomenologicallevel Entropy, generically,
is associated with disorder, but it may not be possible to choose in a non-
arbitrary way a quantitative measure of the disorder of a system without
reference to the level of description. Suppose, for example, that a system
is corn posed of Kn identical particles and the DSI describes the system
as having n particles in each of the K cells of the Jl-space. The disorder, at
this level of description, is usually considered to be the maximum possible.
However, a particular precise description Dprec may exhibit a highly
ordered structure - such as locating the particles exactly at lattice points,
or clustering of particles in pairs - even though Dprec is compatible with
the maximally disordered D SI• Even if Dprec does not manifestly exhibit
an ordered structure, it may contain a hidden order which will become
apparent as the system evolves in accordance with the laws of dynamics:
for example, the particles uniformly distributed at time O may become
concentrated in one ceH at a time t (as in the 'reversibility paradox'~ To be
sure, Carnap's principle of physical magnitudes could be satisfied pro
forma by defining statistical entropy at a chosen level and simply dis-
regarding order at finer levels of description; Carnap has done exactly
this in defining Sk(D;), the chosen level being that of the statistical
description DSI • Such a strategy, however, is open to the charge of arbi-
trariness.
It might be suggested that the arbitrariness be resolved by finding out
388 ABNER SHIMONY

what is the deepest level of description relevant to the operational

quantities involved in actual measurements of entropy, quantities like
flow of heatinto a calorimeter and length of a column of mercury in a
thermometer. The appropriate level could hardly be that of DSt, since some
precise descriptions compatible with DS! imply that sequences of collisions
between particles of the system and those of the thermometer and cal ori-
meter will occur which are extremely improbable from the standpoint of
Dst and which will yield anomalous values of the measured quantities.
For the same reason the level of description could not be chosen to be one
using celIs in the 12-dimensional space ofpairs ofparticles, or even celIs in
the 6n-dimensional space of n-tuples of particles for any small n; the
reason is the large number of encounters with other particles that any one
particle makes during the time Llt needed for a measurement, the mean
time between 2 successive encounters of a hydrogen molecule with other
hydrogen molecules at normal pressure and temperature, for example,
being approximately 6.6 x 10- 11 s. (Chapman and Cowling [1961, p.
92]). If one goes alI the way to Dprec in choosing the level of description
there are overwhelming technical problems in computing the degree of
order and some conceptual problems as welI.
(3) The difficulty of formulating a non-arbitrary objective concept of
the order of a system suggests that it may be simpler and more appro-
priate to define the microscopic counterpart of thermodynamic entropy
in terms of randomness rather than of disorder. In ordinary discourse and
in much of the literature of statistical mechanics the two words are used
almost interchangeably. A clear and useful distinction is made between
them, however, in an unpublished paper, 'The Concept of Degree of
Order', which is referred to in §2 of Carnap's essay:
The term 'randomness', as defined in statistics, refers to a physical characteristic of the
proeedure of se/eetion of a sample (c1ass or sequence) from a given population. A procedure
is defined to be random if ali possible samples of the same size (or certain specified kinds of
samples) have equal probability (in the statistical sense) to be selected. Hence randomness
is not a characteristic of the mathematical structure of a sample .... Any sample may be
selected by a random procedure. (Therefore the term 'random sample' is misleading.) On
the other hand, the degree of disorder is a characteristic of the mathematical structure.

Immediately folIowing this distinction are remarks which are very sug-
gestive for the problem of entropy:
The concepts of randomness and degree of disorder, although different, are inductively
connected. If we know that a procedure of selection is random, w~ can determine the
 CAR NAP ON ENTROPY 389

probability of the occurrence of certain kinds of samples and the expectation value of
certain parameters with respect to samples. On the other hand, the inductive inference from
the evidence describing a given sample S to the hypothesis h that S was produced by a
random procedure is very weak and chieflynegative .... In no case can the mere inspection
of a sample lead to the acceptance of h. This holds even if the sample is very large and, in
addition, the values ofthe relevant parameters in the popu1ation are given.

Though Carnap's remarks are quite genera~ they surely allow the possi-
bility of a randomizing procedure which is such that with overwhelming
probability the degree of disorder at many levels of description is near the
maximum Knowledge of the randomizing procedure may also permit the
calculat ion of the probabilities of non-negligible discrepancies between
the degree of disorder at one level and at another. By contrast, if one is told
only that at the level of D S! th~disorder is maximum, the probability
that the state was produced by a random procedure may be low, as
Carnap points out, or hard to determine, and therefore one cannot be
confident about the degree of disorder at deeper levels of description N
particles may have been partitioned equally into K cells for the purpose of
equitable distribution of booty, or for artistic simulation of randomiza-
tion, or because the dynamics is such as to destroy correlations at that
level but not (in the time allowed) at a deeper level; in all these cases one
might expect high order at some deeper level than D S!. If follows that if
entropy is defined in terms of the randomizing, it may provide more in-
formation about disorder at various levels than does an entropy concept
defined in terms of the disorder at a chosen level. The mechanism of
randomization, which may involve both dynamics and considerations
of large numbers, can establish a reliable, though probabilistic, linkage
among the interesting statistical features at various levels of description,
whereas the principle of physical magnitudes establishes a rigid and non-
probabilistic link that may entirely miss the most interesting features at
deep levels.
(4) Gibbs' treatment of entropy can now be reconsidered with the fore-
going remarks on random procedures in mind Entropy, in most ofGibbs'
formulations, is a functional of a density p(U) on the phase space. Now
suppose that p(U) is established by some random physical procedure for
selecting the state of the system, and is not merely.a compendium of the
observer's knowledge. Then an entropy defined in terms of p is directly
related to randomness (although it is related indirectly and probabilisti-
cally to disorder, by the argument of the preceding paragraph). Now
390 ABNER SHIMONY

Gibbs' Chapter XIV is most naturally read as suggesting that p(U) is

specified by a physical procedure - a suggestion that is lost, perhaps, in
earlier chapters because of the abstractness of the treatment (Note the
remark by Klein [1970, p. 130J that Ch. XIV is the climax of the devel-
opment of Gibb's ideas.) In particular the canonical distribution is appro-
priate for a system which has reached equilibrium with a heat reservoir,
for contact with the reservoir is a mixing or randomizing procedure, the
result of which is the selection of states according to the canonical
distribution Carnap's critique of Gibbs in §8 pays no attention to the
physical circumstances relevant to Gibb's own choices of p(U~ Instead,
Carnap constructs a series of examples in which p(U) is determined only
by rather artificial knowledge of the observer. It is not surprising that the
entropy calculated as a functional of such a distribution is unrelated to
thermodynamic entropy.
(5) In the reading of Gibbs suggested here, p(U) may be understood to
be a probability distribution in the 'propensity' sense. The propensity
interpretation of probability seems to have been formulated explicitly
first by Peirce, who considered the probability that a tossed die will turn
up a certain face to be "a certain 'would-be', and to say that adie has a
'would-be' is to say that it has a property, quite analogous to any habit
that a man might have" [1932, p. 409]. The name 'propensity' is due to
Popper, whose interest was mainly in the sense ofprobability appropriate
to quantum mechanics [1957]. The probability distribution has an
objective status in nature, according to the propensity interpretation, not,
however, as the constitution of any actual ensemble, but rather as the
weighted set of potentialities of outcomes in a type of physical situation
Both Peirce and Popper arrived at a propensity interpretation of proba-
bility after first espousing a frequency interpretation and then realizing
that in most cases of interest the population or collective in which the
relative frequencies are to be found does not actually exist Since Gibbs
considers the ensemble described by p(U) to be virtual [1948, p. 17J, it is
much more natural to attribute to him a propensity than a relative fre-
quency sense of probability.
Attractive as a propensity interpretation of probability may be, it is
undoubtedly very obscure. Popper is inclined to take it as a primitive
concept, but if one wishes to use the concept in the context of classical
physics, where the fundamentallaws are deterministic, then it appears to
 CARNAP ON ENTROPY 391

be illegitimate merely to postulate the existence of propensities without

grounding them in dynamics. There is at least one way in which the con-
cept of propensity can be explicated dynamicalIy, namely via ergodic
theory. (For an explanation of the terminology used here see Lebowitz
and Penrose [1973]; and for a more extended discussion of the relation
between propensity and ergodicity see Friedman and Shimony [1975*].)
Suppose that for a given physica1 system T an energy surface is metricalIy
indecomposable. Then the microcanonical distribution is justified by the
fact that for almost alI trajectories on the surface the time spent by the
phase point in any subset of the surface is asymptoticalIy proportional
to the measure of the subset Suppose further that T is an isolated system
consisting of a very large part R (reservoir) and a relatively small part (1,
which interact with each other. Then for almost alI trajectories on the
energy surface of T, the time spent by (1 in a subset of its phase space is
proportional to the integral of an appropriate canonical distribution over
that subset. In other words, the randomizing procedure of interaction
with R almost certainly has the effect of making (1 'visit' alI states accessible
to it with frequencies prescribed by the canonical distribution The proof
by Sinai of the metrical indecomposability of energy surfaces of a gas of
hard spheres can be considered to legitimate the propensity interpretation
in one very important class of cases though the transition from 'almost alI
trajectories' to 'almost certainly' requires more analysis than one finds in
the current literature on ergodic theory. It is still a matter of research how
far Sinai's result can be extended, and also whether weaker properties than
metrical indecomposability will suffice to provide a dynamical justifica-
tion for the distributions used in statistical mechanics (Grad [1967,
pp. 69ff]; Khinchin [1949, Chs. 5 and 6]).
(6) A strong dissent to the foregoing reading of Gibbs is expressed by
Jaynes, who considers him to be usirig probability in the epistemic sense
of reasonable degree of belief [1967, p. 97]; note that Jaynes sometimes
characterizes this sense of probability as 'subjective', but always in
quotation marks, and indeed many of his remarks in this paper and
elsewhere indicate a commitment to something like a logical concept of
probability. Whether or not Jaynes is historically accurate regarding
Gibbs, it is important to assess his claim that the appropriate probability
concept for statistical mechanics is epistemic probability, and that the
concept of entropy is accordingly a measure of 'amount of uncertainty'
392 ABNER SHIMONY

[ibid, p. 97]. Carnap's exhibition of the discrepancy between s~ and sth in

cases 1 through 6 of §8 constitutes a serious objection to Jaynes' program.
These cases are revealing because in them the information upon which the
computation of Pi(U) is to be based is quite different from the usual kinds
of information about the preparation of a system, e.g., that it is in contact
with certain reservoirs. Consequently, in these cases it is easy to distin-
guish epistemic probability from probability in the sense of propensity,
and it is difficult to borrow from the latter in evaluating the former. An
entirely different objection to Jaynes' program is that his maximum-
entropy prescription for evaluating probabilities can be saved from in-
consistency with the calculus of probability only by a highly implausible
assumption about prior probabilities (Friedman and Shimony [1971]).
Finally, the incorrectness of conflating thermodynamic entropy with
uncertainty in an information-theoretic sense can be illustrated by noting
that in molecular biology the latter depends upon the number of isomers
or well-defined variants of a type of molecule, a number which is inde-
pendent of the thermodynamic entropy (Tisza and Quay [1963, VI]).
Maintaining a clear conceptual distinction between thermodynamic
entropy and information -theoretical uncertainty is, however, consistent
with acknowledging that the price of acquiring information is always an
increase of entropy, as Szilard discovered and Carnap lucidly explains
in § 10. A wonderful example is the increase of entropy due to the chemical
reaction required for the 'recognition' of an isomer (Monod [1971, p. 39]).
(7) An exclusive commitment to taking statistical entropy to be a func-
tional of p(U) is by no means entailed by the above defense of Gibbs.
Since an ensemble described by a canonica! or microcanonical distribu-
tion will contain some (though relatively very few) members with anoma-
lously high order at the level of D st (or any other level of interest), it cer-
tainly is useful to have in one's arsenal of concepts a measure of disorder
at this level, especially in treating the fluctuations that occur at thermo-
dynamic equilibrium. The following passage is a judicious statement of
the value of conceptual pluralism:
The Boltzmann entropy of a particular system at a particular time t depends only on its
observational state at that time; its statistica! entropy, on the other hand, depends on what
ensemble the system is to be regarded as belonging to, that is, on the experimental procedure
to which the system was subjected before time t and which, if replicated many times, would
generate this ensemble .... Thus the two definitions of entropy refer to difIerent physical
situations and are, in a sense, complementary. (Penrose [1970, p.215].)
 CARNAP ON ENTROPY 393

It may be added that both Boltzmann's and Gibbs' concepts of entropy

are physical and objective, though in different ways.

III. AN ABSTRACT CONCEPT OF ENTROPY

Although Carnap prefers an extended vers ion of Boltzmann's concept of

statistical entropy to that of Gibbs, he has some misgivings about Boltz-
mann's reliance upon a ceH system in J.L-space. The value of Sk(Dprec)
depends on the size of the ceHs and the exact locations of their boundaries,
both of which are arbitrary. Furthermore, as the system evolves in time,
with Dprec changing continuously in accordance with the laws of motion,
sk changes discontinuously as molecules pass from one ceH to another.
The main achievement of Carnap's second essay is the formulation of a
concept of entropy S** which is free from both of these defects. Re also
indicates how S** could be applied in inductive logic.
The definition of S** presupposes the concept of the environment of the
point bi occupied by the ith molecule, which is the set of aH points in the
J.L-space c10ser to bi than to any point occupied by another molecule. S** is
then defined (Equation (11-10)) in terms of the volumes of the environ-
ments of the system, and it is indeed independent of the choice of a ceH
system and continuous in its temporal evolution. Furthermore, S** is a
natural and aestheticaHy pleasing concept
It may be questioned, however, whether s** is free from arbitrariness.
Some of the earlier objections concerning the order of a system at various
levels of description can be readily modified so as to apply to S**. Suppose,
for example, that aH of the N environments have equal volumes, so that
S**, and therefore the disorder which it measures, are maximized. The
system would nevertheless be highly ordered, in an intuitive sense, if the
equality of volumes of the environments were achieved by a perfect1y
regular lattice arrangement of the N molecules. Another source of arbi-
trariness is the expression 'c1oser to' in the definition of 'environment'. If
the point bi in the J.L-space is specified by a sextuple xL xt, x~, pf, pt, p~,
which are respectively position and momentum coordinates, tl1en the
distance between bi and bj depends not only upon the 'position distance'
lXi-X) and the 'momentum distance' Ipi-pjl, for both ofwhich there are
natural metrics, but it also depends upon the relative weights placed
upon these two 'distances'; but neither kinematics nor dynamics supplies
394 ABNER SHIMONY

a non-arbitrary method of assigning relative weights. By contrast, the

dynamical theorem of Liouville does provide a natural volume measure
in the phase space of the N molecules, from which follows a natural
volume measure in the Jl-space of a single molecule; and in Boltzmann's
procedure, using a ceH system, the volume measure suffices to compute
the entropy.
It is hard to assess the potential value of S** in inductive logic, espe-
cially since Carnap himself did not seem to make much use of it in his
work on induction between 1954 and 1970 (cf. Carnap and Jeffrey [1971 J).
Possibly the difficulty in computing S**, which has a geometrical charac-
ter in contrast to the combinatorial concept SB> discouraged him from
applying it The complications in § 14 and § 15, which treat relatively
simple cases, show that a strong motivation would be needed to justify
the effort of calculating S** in practical situations.
Apart from the relative merits of S** and SB> Carnap's second essay is
interesting as a continuation of his pioneering work on the relations
among the concepts of confirmation, order, and entropy ([1952J and
Carnap and Bar-Hillel [1952J). Recent developments ofthis active line of
investigation in inductive logic are reported in two collections edited by
Hintikka and Suppes, [1966] and [1970].

ACKNOWLEDGEMENTS

The author is grateful to Prof. Arthur J. Benson, to Prof. Howard Stein,

and to Prof. Charles R Willis for their very helpful suggestions and their
encouragement The research was supported in part by the National
Science Foundation.
The University of California Press has kindly given permis sion to
print the introduction to Carnap's Two Essays on Entropy in the present
volume.
BIBLIOGRAPHY

Bar-Hillel, Y. and Carnap, R., 'Semantic Information', British Joumal for the Philosophy
of Science 4 (1952), 147-157.
Benson, Arthur J., 'Bibliography of the Writings of Rudolf Carnap', in P. A. Schilpp (ed.),
The Philosophy of Rudolf Camap (Library of Living Philosophers, VoI. XI), Open House,
La Salle, I1l., 1963.
Boltzmann, Ludwig, 'Zusammenhang zwischen den Siitzen iiber das Verhalten mehr-
atomiger Gasmolekiile mit Jacobi's Princip des letzten Multiplicators', Wiener Berichte
 CARNAP ON ENTROPY 395

63 (1871),397. (Reprinted in Fr. Haseniihrl (ed.), Wissenschaftliche Abhandlungen, VoI. 1,

Leipzig 1909.)
Carnap, Rudolf, Logical Foundations of Probability, Univ. of Chicago Press, Chicago, 1950.
Carnap, Rudolf, The Continuum of lnductive Methods, U. of Chicago Press, Chicago, 1952.
Carnap, Rudolf, 'Intellectual Autobiography', in P. A. Schilpp (ed.), The Philosophy of
Rudolf Carnap, Open Court, La Salle, Ill., 1963.
Carnap, Rudolf, 'The Concept of Degree of Order', unpublished.
Carnap, R. and Jeffrey, R., Studies in lnductive Logic and Probability, VoI. 1, Univ. of
California Press, Berkeley and Los Angeles 1971.
Chapman, S. and Cowling, T. G., The Mathematical Theory of Non-Uniform Gases,
University Press, Cambridge, England 1961.
Ehrenfest, P. and Ehrenfest, T., The Conceptual Foundations of the Statistical Approach in
Mechanics, Cornell Univ. Press, Ithaca, 1959. (Translated by M. J. Moravcsik from
Encyclopiidie der mathematischen Wissenschaften, VoI. IV, 2, ii, Teubner, Leipzig 1912.)
Epstein, Paul S., 'Critical Appreciation of Gibbs' Statistical Mechanics', in A. Haas (ed.),
A Commentary on the Scientific Writings of J. Willard Gibbs, VoI. II, Yale Univ. Press,
New Haven 1936.
Friedman, K. and Shimony, A., 'Jaynes's Maximum Entropy Prescription and Probability
Theory', Joumal of Statistical Physics 3 (1971),381-384.
Friedman, K. and Shimony, A., 'The Propensity Interpretation of Probability' (1975).
Gibbs, J. Willard, Elementary Principles in Statistical Mechanics, in Collected Works of
J. Willard Gibbs, VoI. II, Yale Univ. Press, New Haven 1948. (Reprinted from edition
of 1902.)
Grad, Harold, 'The Many Faces of Entropy', Communications on Pure and Applied Mathe-
matics 14 (1961), 323-354.
Grad, Harold, 'Levels of Description in Thermodynamics and Statistical Mechanics', in
M. Bunge (ed.), Delaware Seminar in the Foundations of Physics, Springer, N.Y., 1967.
Griinbaum, Adolf, Philosophical Problems of Space and Time, 2nd. ed., Reidel, Dordrecht,
1973.
Hintikka, J. and Suppes, P., Aspects of lnductive Logic, North-Holland, Amsterdam 1966.
Hintikka, J. and Suppes, P., lIiformation and lriference, Reide1, Dordrecht 1970.
Jaynes, E. T., 'Information Theory and Statistical Mechanics', Physical Review 106 (1957),
620-630; 108,171-190,
Jaynes, E. T., 'Gibbs vs Boltzmann Entropies', American Joumal of Physics 33 (1965),
391-398.
Jaynes, E. T., 'Foundations of Probability Theory and Statistical Mechanics', in M. Bunge
(ed.), Delaware Seminar in the Foundations of Physics, Springer, New York 1967.
Khinchin, A. 1., Mathematical Foundations of Statistical Mechanics, Dover, New York
1949. (Translated from the Russian by G. Gamow.)
Klein, Martin, Paul Ehrenfest, VoI. 1, North-Holland, New York 1970.
Lebowitz, J. and Penrose, O., 'Modern Ergodic Theory', Physics Today 26 No. 2 (1973),
23-9.
Monod, Jacques, Chance and Necessity, Knopf, New York 1971. (Translated from the
French by A. Wainhouse.)
Peirce, Charles S., in C. Hartshorne and P. Weiss (eds.), Collected Papers, VoI. II, Harvard
Univ. Press, Cambridge 1932.
Penrose, Oliver, Foundations of Statistical Mechanics, Pergamon, Oxford 1969.
Popper, Karl R., 'The Propensity Interpretation of the Ca1culus of Probability, and the
Quantum Theory', in S. Kiirner (ed.), Observation and lnterpretation, Butterworths,
London 1957.
Tisza, L. and Quay, P., 'The Statistical Thermodynamics of Equilibrium', Annals of Physics
25 (1963), 48-90. (Reprinted in L. Tisza, Generalized Thermodynamics, M.I.T. Press,
Cambridge 1966.)
Tolman, Richard, The Principles of Statistical Mechanics, University Press, Oxford 1938.
 INDEX OF NAMES

Achinstein, P. 94, 100 Cooley, J. XXV

Ajdukiewicz, K. XVI, 98,124 Copi,l. 158
Alembert d' 49 Cowling, T. G. 388
Aristotle LIX, 365
Austin, J. L. 291 Dalla Chiara Scabia, M. L. 100, 106, 117,
Avenarius, R. XIV 118, 120
Ayer, A. J. XVI, XLIII, XLVlf, 88, 94, 321 Davidson, Donald 214, 241, 292, 322
Davis, J. W. 242
Bach 264 Dorpfeld, W. LI
Bar-Hillel, Y. XXX, LVI, LXV, 214, 243, Duhem214
32lf, 355, 357, 379, 394
Bayes 293, 336, 377
Eberle, R. A. IX, 55, 73
Benacerraf, P. 215
Edwards, P. 322
Benson,A.J. 293, 381, 394
Ehrenfest, P. and T. 383f
Berlin, J. 88, 94
Einstein, A. XVII
Bemays, P. 75, 77, 84
Epstein, P. 385
Bemoulli, D. 293, 294, 304
Essler, W. K. 365, 369ff
Beth, E. W. 120, 213
Euc1id 10
Black, M. 321
Bohnert, H. G. X, XLIV, 84, 94,183
Boing, E. G. XVII Feigl, H. XIII, XIV, XVII, XXV, XL,
Boltzmann, L. 38lf, 392f LXV, 322
Bolzano, B. 37 Feyerabend, P. XVII, 358
Braithwaite, R. B. 321, 359 Feys, R. 322
Bridgman, P. W. 106, 120 Finetti, B. de XXII, XXIII, 297, 299, 322,
Brightman XXI 329, 359, 371, 376f
Buck, R. C. 359, 380 Foster, L. 358
Bunge, M. 100 Foster, M. H. 357
Burks, A. XXIX, 321 Fraassen, B. C. van 120
Frank, P. XVI
Cantor 204,215 Frege, G. XIII, XXXII, LVIII, 3, 15, 37,
Care, N. S. 241 169,170, 183f, 196f, 203, 213, 214, 217f,
Camap, Ina XXIVf, XLVII, LII, LIV 222, 227f, 239, 240, 281, 289, 291
Chapman, S. 388 Friedman, Joyce 270
Chomsky, Noam LVI, 243, 248, 253, 261f Friedman, K. 391, 392
Chung-Ying Cheng 270 Furberg, M. 291f
Church, A. 88, 94, 201, 213, 215, 248, 254,
291 Gabbay, D. 291f
Clausius 381 Gaifman, Haim XXIIf, 371, 379
Cohen, R. S. XLV, 359, 380 Glymour, C. 96, 120
Coffa, A. 158 Gibbs, J. W. 381f
398 INDEX OF NAMES

GOdel, K. XVI, XLII, XLIII, 78, 84, 185f, Kanger,S.121,24O

198, 204f, 215 Kant, E. 11, 189, 191
Good,I.J.356,358 Kaplan, D. IX, XLIX, 87, 94, 218, 270
Goodman, N. XV, XXVII, LXIII, LXV, Karttunen, L. 29lf
10, 16, 22, 30, 31, 44, 50, 52, 56, 322, Kasher, A. IX, 271, 291
352,358 Katz 257, 262f
Gotari, de XX Kemeny, J. XXIf, XXVIII, 141, 214, 215,
Grad, H. 386, 391 301, 32Of, 367
Grelling, K. XVI Keynes, J. M. 263, 308, 325
Grice 215 Khinchin, A. J. 391
Grimm, R. H. 241 Kimball, J. 270
Gross, M. XXVII Kiparsky 257
Griinbaum, A. XIII, 387 Kleene, S. 188, 190, 213, 214
Klein, M. 390
Hacking, J. 355, 358 Klima262
Hahn, H. XIV, XV, 184 Kneale 324
Halmos, P. R. 323 Kohler, W. XVI
Harman, G. 292 Komer, S. 323
Hartshome, C. XLVII KokoszyDska,M.98
Hayek, F. von XVII Kolmogorov, A. N. 323
Heidegger, M. 174 Krauss, P. 379
Heijenoort, J. van 216 Krauth, L. LXI
Heidsenberg, W. 115 Kripke, S. 240, 241, 253
Hempel, K. G. X, XIII, XVI, XXIf, XLIX, Kiilpe, O. XIV
LXV, 1,84,95, 145, 147, 152,358 Kuhn, L. LXI
Hendry, H. E. 214 Kuhn, T. S. 121
Henken 215 Kuipers, T. 361
Hertz, P. XVI Kyburg, H. E. 355, 359, 379
Heyting 183
Hilbert, D. XLIX, LV, 75, 77, 78, 79, 185f Lakatos, 1. 355, 358, 359, 369
Hillman, D. J. 53 Lakoff 264, 268
Hilpinen, R. X, 242, 333, 356, 357f Laplace 293
Hintikka, J. 112, 120, 213, 218, 240, 251f, Leblanc, H. 323
270, 28lf, 289f, 339, 355f, 365f, 379, 394 Lebowitz, J. 384, 391
Hitler, A. LIII Lehman, R. S. 323
Hockney, D. J. 242 Leibniz, G. W. 49, 220, 224
Hogben, Lancelot 8 Lenz, J. 323
Hom,L. 260 Lewis, D. XLV, 215
Humburg, J. 356, 359 Lieb, H. 29lf
Hume, D. XLV, LVIII Loeve, M. 323
Husserl, E. 49 Lorenzen, P. XLV
Hutten, E. H. 323 Lovejoy, A. XXVII

Jackson, W. E. 321 Mach, E. XIV, 47, 68

Jaynes, E. T. 382, 386f Madden, E. H. 323
Jeffrey, R. X, XIII, XXIV, 263, 325,358, Martin, M. L. 357
394 Martin, R. M. 215, 240, 323
Johnson, W.E. 356,359 Mates 247f, 253, 261
Jordan, P. XVI Matthews, G. XXIV
 INDEX OF NAMES 399

Maxwell, G. 158,358 Quay, P. 387, 392

McCawley 263 Quine, W. XIII, XVII, XXX, XLIX, LV,
McKeon, R. XLIII LXV, 3, 31, 52, 84,118,145,148,195,197,
McKinsey, J. C. C. 322 199, 200, 203, 205, 206, 208, 210, 212,
Michalos, A. C. 359 214, 218, 219, 223, 23lf, 249, 254, 263,
Mill, 1. S. 5, 169, 170 266, 276, 291, 357, 359
Mises, R. von XVII, 293, 323
MolinaXX Ramsey, F. P. XXII, 82f, 97, 146f, 153, 159,
Monod, J. 392 184,208,297
Montague, R. XLVIII, 105, 120, 222f, 240, Reichenbach, H. XV, XVI, XVII, L, LI,
251, 253, 257, 269 LIII, LXI, II, 204, 293, 321, 356
Moore, G. E. XLV Reichenbach, M. LIV
Moravcsik, J. 241, 270, 291f Rembrandt 206
Morris, C. XLV, L Rescher, N. 358
Richards, 1. A. XVII
Naess, A. 37 Robinson, A. 159
Nage1, E. XVI, XVII, LXV, 263, 323, 334, Rootselaar, B. van 358
355, 357, 359 Ross 264
Neumann, von 183 Rozeboom, W. W. 94,159
Neurath, O. XV, XVI, XVII, 4, 6, 7,12 Rubin, H. 324
Neyman, J. 322, 323 Russell, B. XIII, XIV, XV, XVII, XXIV,
Niiniluoto, I. 378, 380 XXVII, XL V, LIII, LV, LXVIII, 3, 7,
Nixon, R. 260 15,17,22,23,26,50,169,170, 183f, 189,
Nohl, H. XXXII 196, 199, 213, 249, 322
Ryle 8, 169, 171

Olson, R. E. 358 Salmon, Wesley C. 324

Oppenheim, P. XVI Savage, L. J. XXII, 324, 338
Ostwald, W. XIV Schilpp, P. A. XIII, LIII, LX
Schlick, M. XIV, XV, XVI, II, 184, 214
Pap, A. 323 Schroder, E. 203
Parsons, C. 212 Schrodinger, E. XVI
Partee, B. Hali IX, 241, 243, 263, 264, 291, Scott, Dana 122, 269, 379
292 Searle, J. 270
Paul, A. 358 Sellars, W. 84,250
Peano, G. 187, 192f, 208, 209 Seuren, P. 270
Peirce, C. S. XLVI, 308, 390 Shimony, A. X, XIII, XXII, XXIX, 295,
Penrose, O. 384, 39lf 298,300,324,381,391,392
Perry, C. XL VI Shoenfield 215
Pietarinen, J. 339, 355f, 359 Simon, H. A. 324
Plato 18, 49 Smokler, H. E. 379
Poincare, H. 212, 214, 379 Sneed, J. D. 96, 97, 105, 120
Popper, K. XVI, XXVII, LVIII, LIX, LXV, Staal, J. F. 358
4, 100, 110, 121, 137, 32lf, 334, 355, 357, Stalnaker, R. C. 274, 276, 291, 292
359,390 Stegmiiller, W. XXII, XXIV, LXVIII, 324,
Postal 257, 262f 355,356,358
Prall, D. XXVI Stein, H. 394
Przelecki, M. 95f, 124, 129 Stemmer, N. 357, 359
Putman, H. XXI, 100, 122,215,250 Stevens, S. S. XVII
400 INDEX OF NAMES

Stine, Gail 241 Waismann, F. XIV, XV, XVI, 11

Strauss, M. XVI Wallace, J. 277, 291, 292
Strawson, P. F. 13, 215, 324 Wedberg, A. X, 15, 161
Suppe, F. 120 Weyl, H. 78, 84, 85
Suppes, P. 122,241,322, 356f, 369, 379, 394 Whitehead, A. N. XLV
Swanson, J. W. 358 Wiener, N. XVII
WiII, F. L. 324
Tarski, A. XVIf, XLII, XLVI, LVII, 7,10, Williams, P. M. IX, 97, 100, 120, 124
105,174,165,188,198,200,205,207,214, WiIIis, C. R. 394
243,357 Wilson, W. K. 242
Thost, Hanna XXIV Winnie, J. A. IX, 99, 100, 120, 143
Tisza, L. 387, 392 Wisdom, J. O. 324
To1man, R. 386 Wittgenstein, L, XV, XXIV, XXXIX, 9, 15,
Tora1do di Francia, F. 100, 106, 120 37, 175, 184, 189, 197, 198,202,208,308
Tuomela, R. 95, 96 Wojcicki, R. IX, 94f, 124, 129
Works, C. 94
Veblen, O. 205, 213 Wright, G. H. von XLVI, 324
Vietoris, L. 324
Vetter 358 Y ourgrau, W. 94
 SYNTHESE LIBRAR Y

Monographs on Epistemology, Logic, Methodology,

Philosophy of Science, Sociology of Science and of Knowledge, and on the
Mathematical Methods of Social and Behavioral Sciences

M anaging Editor,'
JAAKKO HINTIKKA (Academy of Finland and Stanford University)

Editors,'
ROBERT S. CoHEN (Boston University)
DONALD DAvmsoN (The Rockefeller University and Princeton University)
GABRIEL NUCHELMANS (University of Leyden)
WESLEY C. SALMON (University of Arizona)

1. J. M. BOCHENSKI, A Precis of Mathematical Logic. 1959, X+ 100 pp.

2. P. L. GUIRAUD, Problemes et methodes de la statistique linguistique. 1960, VI + 146 pp.
3. HANS FREUDENTHAL (ed.), The Concept and the Role of the Model in Mathematics and
Natural and Social Sciences, Proceedings of a Colloquium held at Utrecht, The Nether-
lands, January 1960. 1961, VI+ 194 pp.
4. EVERT W. BETH, Formal Methods. An Introduction to Symbolic Logic and the Study
of Effective Operations in Arithmetic and Logic. 1962, XIV + 170 pp.
5. B. H. KAZEMIER and D. VUYSJE (eds.), Logic and Language. Studies dedicated to Profes-
sor Rudolf Carnap on the Occasion of his Seventieth Birthday. 1962, VI + 256 pp.
6. MARX W. WARTOFSKY (ed.), Proceedings ofthe Boston Colloquiumfor the Philosophy
of Science, 1961-1962, Boston Studies in the Philosophy of Science (ed. by Robert
S. Cohen and Marx W. Wartofsky), Volume 1. 1973, VIII +212 pp.
7. A. A. ZINOV'EV, Philosophical Problems of Many-Valued Logic. 1963, XIV+155 pp.
8. GEORGES GURVITCH, The Spectrum of Social Time. 1964, XXVI + 152 pp.
9. PAUL LORENZEN, Formal Logic. 1965, VIII + 123 pp.
10. ROBERT S. COHEN and MARX W. WARTOFSKY (eds.), In Honor of Philipp Frank, Boston
Studies in the Philosophy ofScience (ed. by Robert S. Cohen and Marx W. Wartofsky),
Volume II. 1965, XXXIV +475 pp.
Il. EVERT W. BETH, Mathematical Thought. An Introduction to the Philosophy of Mathe-
matics. 1965, XII + 208 pp.
12. EVERT W. BETH and JEAN PrAGET, Mathematical Epistemology and Psychology. 1966,
XII+326pp.
13. Gumo KONG, Ontology and the Logistic Analysis of Language. An Enquiry into the
Contemporary Views on Universals. 1967, XI+21O pp.
14. ROBERT S. COHEN and MARX W. WARTOFSKY (eds.), Proceedings ofthe Boston Col/o-
quiumfor the Philosophy of Science 1964-1966, in Memory of Norwood Russell Hanson,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W.
Wartofsky), Volume III. 1967, XLIX +489 pp.
15. C. D. BRoAD, Induction, Probability, and Causation. Selected Papers. 1968, XI + 296 pp.
16. GONTHER PATZIG, Aristotle's Theory of the Syllogism. A Logical-Philosophical Study
of Book A of the Prior Analytics. 1968, XVII + 215 pp.
17. NICHOLAS REsCHER, Topics in Philosophical Logic. 1968, XIV +347 pp.
18. ROBERT S. CoHEN and MARX W. WARTOFSKY (eds.), Proceedings ofthe Boston Collo-
quium for the Philosophy of Science 1966-1968, Boston Studies in the Philosophy of
Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume IV. 1969,
VIII + 537 pp.
19. ROBERT S. CoHEN and MARX W. WARTOFSKY (eds.), Proceedings ofthe Boston Collo-
quium for the Philosophy of Science 1966-1968, Boston Studies in the Philosophy of
Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume V. 1969,
VIII +482 pp.
20. J. W. DAVIS, D. J. HOCKNEY, and W. K. WILSON (eds.), Philosophical Logic. 1969,
VIII + 277 pp.
21. D. DAVIDSON and J. HINTIKKA (eds.), Words and Objections: Essays on the Work of
W. V. Quine. 1969, VIII+366 pp.
22. PATRICK SUPPES, Studies in the Methodology and Foundations of Science. Selected
Papers from 1911 to 1969, XII +473 pp.
23. JAAKKO HINTIKKA, Models for Modalities. Selected Essays. 1969, IX + 220 pp.
24. NICHOLAS REsCHER et al. (eds.), Essays in Honor of Cari G. Hempel. ATribute on the
Occasion of his Sixty-Fi/th Birthday. 1969, VII + 272 pp.
25. P. V. TAVANEC (ed.), Problems ofthe Logic ofScientijic Knowledge. 1969, XII +429 pp.
26. MARSHALL SWAIN (ed.), Induction, Acceptance, and Rational Belie! 1970, VII+232 pp.
27. ROBERT S. CoHEN and RAYMOND J. SEEGER (eds.), Ernst Mach; Physicist and Philos-
opher, Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and
Marx W. Wartofsky), Volume VI. 1970, VIII + 295 pp.
28. JAAKKO HINTIKKA and PATRICK SUPPES, Information and Inference. 1970, X+336 pp.
29. KAREL LAMBERT, Philosophical Problems in Logic. Some Recent Developments. 1970,
VII + 176 pp.
30. ROLF A. EBERLE, Nominalistic Systems. 1970, IX+217 pp.
31. PAUL WEINGARTNER and GERHARD ZECHA (eds.), Induction, Physics, and Ethics, Pro-
ceedings and Discussions of the 1968 Salzburg Colloquium in the Philosophy of Science.
1970, X+382 pp.
32. EVERT W. BETH, Aspects of Modern Logic. 1970, XI + 176 pp.
33. RISTO HILPlNEN (ed.), Deontic Logic: Introductory and Systematic Readings. 1971,
VII + 182 pp.
34. JEAN-LoUIS KRIVINE, Introduction to Axiomatic Set Theory. 1971, VII+98 pp.
35. JOSEPH D. SNEED, The Logical Structure of Mathematical Physics. 1971, XV + 311 pp.
36. CARL R. KORDIG, The Justijication of Scientijic Change. 1971, XIV + 119 pp.
37. MILIC CAPEK, Bergson and Modern Physics, Boston Studies in the Philosophy of
Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume VII. 1971, XV +414
pp.
38. NORWOOD RUSSELL HANSON, What I do not Believe, and other Essays, (ed. by Stephen
TouImin and Harry Woolf), 1971, XII+390 pp.
39. ROOER C. BucK and ROBERT S. COHEN (eds.), PSA 1970. In Memory of RudolfCarnap,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W.
Wartofsky), Volume VIII. 1971, LXVI +615 pp. Also available as a paperback.
40. DONALD DAVIDSON and GILBERT HARMAN (eds.), Semantics ofNatural Longuage. 1972,
X + 769 pp. Also available as a paperback.
41. YEHOSHUA BAR-HILLEL (ed.), Pragmatics of Natural Languages. 1971, VII+231 pp.
42. SOREN STENLUND, Combinators, A-Terms and Proof Theory. 1972, 184 pp.
43. MARTIN STRAUSS, Modern Physics and lts Philosophy. Selected Papers in the Lagic,
History, and Philosophy of Science. 1972, X+297 pp.
44. MARIO BUNGE, Method, Model and Matter. 1973, VII + 196 pp.
45. MARIO BUNGE, Philosophy of Physics. 1973, IX + 248 pp.
46. A. A. ZINOV'EV, Foundations of the Logical Theory of Scientific Knowledge (Complex
Logic), Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and
Marx W. Wartofsky), Volume IX. Revised and enlarged English edition with an
appendix, by G. A. Smirnov, E. A. Sidorenka, A. M. Fedina, and L. A. Bobrova
1973, XXII + 301 pp. Aiso available as a paperback.
47. LADISLAV TONDL, Scientific Procedures, Boston Studies in the Philosophy of Science
(ed. by Robert S. Cohen and Marx W. Wartofsky), Volume X. 1973, XII+268 pp.
Aiso available as a paperback.
48. NORWOOD RussELL HANSON, Constellations and Conjectures, (ed. by Willard C.
Humphreys, Jr.), 1973, X+282 pp.
49. K. J. J. HINTIKKA, J. M. E. MORAVCSIK, and P. SUPPES (eds.), Approaches to Natural
Language. Proceedings of the 1970 Stanford Workshop on Grammar and Semantics.
1973, VIII + 526 pp. Aiso available as a paperback.
50. MARIO BUNGE (ed.), Exact Philosophy - Problems, Tools, and Goals. 1973, X + 214 pp.
51. RADU J. BOGDAN and ILKKA NIINILUOTO (eds.), Lagic, Language, and Probability. A
selection 'of papers contributed to Sections IV, VI, and XI of the Fourth International
Congress for Logic, Methodology, and Philosophy of Science, Bucharest, September
1971. 1973, X+323 pp.
52. GLENN PEARCE and PATRICK MAYNARD (eds.), Conceptual Chance. 1973, XII +282 pp.
53. ILKKA NIINILUOTO and RAIMO TUOMELA, Theoretical Concepts and Hypothetico-Induc-
tive lnference. 1973, VII + 264 pp.
54. ROLAND FRAiSSE, Course of Mathematical Logic - Volume 1: Relation and Logica!
Formula. 1973, XVI + 186 pp. Aiso available as a paperback.
55. AooLF GRUNBAUM, Philosophical Problems of Space and Time. Second, enlarged edi-
tion, Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx
W. Wartofsky), Volume XII. 1973, XXIII+884 pp. Also available as a paperback.
56. PATRICK SUPPES (ed.), Space, Time, and Geometry. 1973, XI +424 pp.
57. HANS KELSEN, Essays in Legal and Moral Philosophy, se1ected and introduced by Ota
Weinberger. 1973, XXVIII + 300 pp.
58. R. J. SEEGER and ROBERT S. COHEN (eds.), Philosophical Foundations of Science. Pro-
ceedings of an AAAS Program, 1969. Boston Studies in the Philosophy of Science (ed.
by Robert S. Cohen and Marx W. Wartofsky), Volume XI. 1974, X+545 pp. Also
available as paperback.
59. ROBERT S. COHEN and MARX W. W ARTOFSKY (eds.), Logical and Epistemological Studies
in Contemporary Physics, Boston Studies in the Philosophy of Science (ed. by Robert
S. Cohen and Marx W. Wartofsky), Volume XIII. 1973, VIII +462 pp. Aiso available
as paperback.
60. ROBERT S. COHEN and Marx W. WARTOFSKY (eds.), Methodological and Historical
Essays in the Natural and Social Sciences. Proceedings of the Boston Colloquium for
the Philosophy of Science, 1969-1972, Boston Studies in the Philosophy of Science (ed.
by Robert S. Cohen and Marx W. Wartofsky), Volume XIV. 1974, VIII+405 pp. Aiso
available as paperback.
61. ROBERT S. COHEN, J. J. STACHEL and MARX W. W ARTOFSKY (eds.), For Dirk Struik.
 Scientific, Historical and Political Essays in Honor of Dirk J. Struik, Boston Studies
in the Philosophy ofScience (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume
XV. 1974, XXVII+652 pp. Aiso available as paperback.
62. KAZIMIERZ AJDUKIEWICZ, Pragmatic Logic, transl. from the Polish by Olgierd Wojta-
siewicz. 1974, XV +460 pp.
63. SOREN STENLUND (ed.), Logical Theory and Semantic Analysis. Essays Dedicated to Stig
Kanger on His Fi/tieth Birthday. 1974, V +217 pp.
64. KENNETII F. SCHAFFNER and ROBERT S. COHEN (eds.), Proceedings ofthe 1972 Biennial
Meeting, Philosophy of Science Association, Boston Studies in the Philosophy of
Science(ed. by Robert S. Cohen and Marx W. Wartofsky), Volume XX. 1974, IX + 444
pp. Aiso available as paperback.
65. HENRY E. KYBURG, JR., The Logical Foundations of Statistical Inference. 1974, IX + 421
pp.
66. MARJORIE GRENll, The Understanding of Nature: Essays in the Philosophy of Biology,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W.
Wartofsky), Volume XXIII. 1974, XII + 360 pp. Aiso available as paperback.
67. JAN M. BROEKMAN, Structuralism: Moscow, Prague, Paris. 1974, IX + 117 pp.
68. NORMAN GESCHWIND, Selected Papers on Language and the Brain, Boston Studies in
the Philosophy of Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume
XVI. 1974, XII + 549 pp. Aiso available as paperback.
69. ROLAND FRAiSSE, Course of Mathematical Logic - Volume II: Model Theory. 1974,
XIX + 192 pp.
70. ANDRZEl GRZEGORCZYK, An Outline of Mathematical Logic. Fundamental Results and
Notions Explained with AII Details. 1974, X+ 596 pp.
71. FRANZ VON KUTSCHERA, Philosophy of Language. 1975, VII + 305 pp.
72. JUHA MANNlNEN and RAIMO TUOMELA, Essays on Explanation and Understanding.
Studies in the Foundations of Humanities and Social Sciences. VII+435 pp.
75. JAAKKO HINTIKKA and UNTO REMES, The Method of Analysis. Its Geometrical Origin
and Its General Significance. Boston Studies in the Philosophy of Science (ed. by
Robert S. Cohen and Marx W. Wartofsky), Volume XXV. 1974, XVIII + 144 pp. Aiso
available as paperback.
76. JOHN EMERY MURDocH and EDITII DUDLEY SYLLA, The Cultural Context of Medieval
Learning. Proceedings of the First International Colloquium on Philosophy, Science,
and Theology in the Middle Ages - September 1973. Boston Studies in the Philosophy
of Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume XXVI. 1975,
X + 566 pp. Aiso available as paperback.
77. STEFAN AMSTERDAMSKI, Between Experience and Metaphysics. Philosophical Problems
of the Evolution of Science. Boston Studies in the Philosophy of Science (ed. by Robert
S. Cohen and Marx W. Wartofsky), Volume XXXV. 1975, XVIII+193 pp. Aiso
available as paperback.
80. JOSEPH AGASSI, Science in Flux. Boston Studies in the Philosophy of Science (ed. by
Robert S. Cohen and Marx W. Wartofsky), Volume XXVIII. 1975, XXVI + 553 pp.
Aiso available as paperback.
90. JAAKKO HINTIKKA, The Intentions of Intentionality and Other New Models for
Modalities. XVIII + 254 pp.
 SYNTHESE HISTORICAL LIBRARY

Texts and Studies

in the History of Logic and Phi1osophy

Editors:

N. KRETZMANN (Cornell University)

G. NUCHELMANS (University of Leyden)
L. M. DE RUK (University of Leyden)

1. M. T. BEONIo-BROCCIllERI FUMAGALLI, The Logic of Abelard. Trans1ated from the

Italian. 1969, IX+ 101 pp.

2. GOTTFRIED WILHELM LEIBNITZ, Philosophical Papers and Letters. A se1ection trans1ated

and edited, with an introduction, by Leroy E. Loemker. 1969, XII + 736 pp.

3. ERNST MALLY, Logische Schriften, ed. by Karl Wo1f and Paul Weingartner. 1971,
X+340 pp.

4. LEWIS WIllTE BECK (ed.), Proceedings of the Third International Kant Congress. 1972,
XI+718pp.

5. BERNARD BOLZANO, Theory of Science, ed. by Jan Berg. 1973, XV + 398 pp.

6. J. M. E. MORAVCSIK (ed.), Patterns in Plato's Thought. Papers arising out ofthe 1971
West Coast Greek Philosophy Conference.1973, VIII + 212pp.

7. NABIL SHEHABY, The Propositional Logic of Avicenna: A Translationfrom al-Shifii: al-

Qiyiis, with Introduction, Commentary and G1ossary. 1973, XIII + 296 pp.

8. DESMOND PAUL HENRY, Commentary on De Grammatico: The Historical-Logical

Dimensions of a Dialogue of St. Anselm 's. 1974, IX + 345 pp.

9. JOHN CORCORAN, Ancient Logic and Its Modern 1nterpretations. 1974, X+208 pp.

10. E. M. BARTH, The Logic ofthe Articles in Traditional Philosophy. 1974, XXVII+533
pp.

11. JAAKKO HINTIKKA, Knowledge and the Known. Historical Perspectives in Epistemology.
1974, XII + 243 pp.

12. E. J. ASHWORTH, Language and Logic in the Post-Medieval Period. 1974, XIII + 304 pp.

13. ARISTOTLE, The Nicomachean Ethics. Trans1ated with Commentaries and G10ssary by
Hypocrates G. Apostle. 1975, XXI + 372 pp.
14. R. M. DANCY, Sense and Contradiction: A Study in Aristotle. 1975, XII + 184 pp.

15. WILBUR RICHARD KNORR, The Evolution of the Euclidean Elements. A Study of the
Theory of lncommensurable Magnitudes and lts Significancefor Early Greek Geometry.
1975, IX + 374 pp.

16. AUGUSTINE, De Dialectica. Translated with the Introduction and Notes by B. Darrell
Jackson. 1975, XI+ 151 pp.