Professional Documents
Culture Documents
SYNTHESE LIBRAR Y
MONOGRAPHS ON EPISTEMOLOGY,
M anaging Editor:
Editors:
VOLUME 73
Photograph by Adya, 1962
RUDOLF CARNAP,
LOGICAL EMPIRICIST
M aterials and Perspectives
Edited by
JAAKKO HINTIKKA
PREFACE IX
ACKNOWLEDGMENTS XI
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
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
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.
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
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.
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
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.
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
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.
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.
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
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.
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
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
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
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
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.
10
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.
11
12
13
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
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
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
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
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
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
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
Princeton University
NOTES
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
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
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
(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
L 3 The language Ll is not used in the Aujbau, but the Aujbau contains
0
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
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
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
(a) Translating from science into the 'Aujbau'. I shall use the phrase
38 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.
(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.
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.
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,
VII. CONCLUSION
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.
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:
46 From § 180.
47 When Carnap looked back on the Aujbau in 'Testability and Meaning', he described
52 ANDERS WEDBERG
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.
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
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
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
(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.
(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
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
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
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:
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
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
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 :
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
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
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
Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 75-85. AII rights reserved.
Copyright Dialectica and the Rudolf Carnap Trust.
76 RUDOLF CARNAP
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
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
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
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
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
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.
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
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
(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
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.
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
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
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
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
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.
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.
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
DEFINITION 6. Thefactual content ofrt., F cont (rt.), is the set ofali struc-
tures in sms(P~) in which rt. is not true.
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'.
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
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.
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
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
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
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 ) ,
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
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
(**) 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,
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
THEORETICAL ANALYTICITY
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
where the first conjunct represents its theoretical postulates T, and the
146 JOHN A. WINNIE
(3) FTC'A 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
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
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
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.
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
::;. (---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.
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
BIBLIOGRAPHY
[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
I. INTRODUCTION
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
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
form:
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
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.
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
(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
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
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
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
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
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
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
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
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
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
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
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
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
[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
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:
Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 217-242. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
218 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
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
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
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.
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
NOTES
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
1. INTRODUCTION
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
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
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
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
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
5. DEEP STRUCTURE
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
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
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?
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
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
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
Jaakko Hintikka (ed.) , Rudolf Carnap, Logical Empiricist, 271-292. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
272 ASA KASHER
(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
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
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
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
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
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
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
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
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
Jaakko Hilltikka (ed.), Rudolf Carnap, Logical Empiricist, 293-324. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
294 RUDOLF CARNAP
(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)
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.
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):
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.
. . 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
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.
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.
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
G(BS,j) = -1 -1 -1
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.)
.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) =
b. c(h.t , e) =
C:)C)-.
(:~)
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
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
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)].
•
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. ct (h j , eN ) = l/k.
308 RUDOLF CARNAP
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
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.
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).)
Ck (Sl;S2,ooo,Sk)=Of o
holdso ('---' stands for 'S3, Sk'; this expression drops out
000'
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)
G(S;S) G(1;s) ]
G(O;s + 1) [ - - + - - + k - 2 = 1.
G(O;s) G(O;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)
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.
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
(2)
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:
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.
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
*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
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.
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
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
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.
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
v. IDEALIZATION OR FALSIFICATION?
UniversUy of Pennsylvania
RISTO HILPINEN
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
III
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
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
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'
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
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,
(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
If F is a family with '1-equality and '1 = '1mj' (I6) and (17) imply
(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,
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
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
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
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
BIBLIOGRAPHY
A GENERALIZATION OF CARNAP'S
INDUCTIVE LOGIC
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!.
I I
inp(h i ej = inp( '" H ej'inp(h i Ies & '" H); 1:::;:; i:::;:; n:::;:; s.
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-
. 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.
Centrale lnterfaculteit,
Universiteit van Amsterdam
WILHELM K. ESSLER
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
(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
BIBLIOGRAPHY
Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 371-380. AII rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
372 JAAKKO HINTlKKA
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
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
CARNAP ON ENTROPY
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.]
Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist, 381-395. Ali rights reserved.
Copyright © 1975 by D. Reidel Publishing Company, Dordrecht-Holland.
382 ABNER SHIMONY
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
(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
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
ACKNOWLEDGEMENTS
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
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)
Editors:
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.
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.