Mario Bunge - Semantics (I) - Sense and Reference PDF
Mario Bunge - Semantics (I) - Sense and Reference PDF
Mario Bunge - Semantics (I) - Sense and Reference PDF
Volume I
Semantics I:
move on to the next, namely the building of further systems. And this for
three reasons: because the world itself is systemic, because no idea can
become fully clear unless it is embedded in some system or other, and
because sawdust philosophy is rather boring.
The author dedicates this work to his philosophy teacher
Kanenas T. Pota
in gratitude for his advice: "Do your own thing. Your reward will be
doing it, your punishment having done it".
CONTENTS OF SEMANTICS I
PREFACE XI
ACKNOWLEDGEMENTS XIII
SPECIAL SYMBOLS XV
INTRODUCTION 1
1. Goal 1
2. Method 4
1. DESIGNATION 8
1. Symbol and Idea 8
1.1. Language 8
1.2. Construct 13
1.3. Predicate 15
1.4. Theory and Language 18
2. Designation 21
2.1. Name 21
2.2. The Designation Function 23
3. Metaphysical Concomitants 26
3.1. Basic Ontology 26
3.2. Beyond Platonism and Nominalism 27
2. REFERENCE 32
1. Motivation 32
2. The Reference Relation 34
21. An Unruly Relation 34
2.2. Immediate and Mediate Reference 36
2.3. Reference Class 37
2.4. Factual Reference and Object Variable 39
2.5. Denotation 42
2.6. Reference and Evidence 43
2.7. Misleading Cues in the Search for Referents 46
3. The Reference Functions 48
3.1. Desiderata 48
3.2. Principles and Definitions 50
VIII CONTENTS OF 'SEMANTICS I'
3. REPRESENT A TION 83
1. Conceptual Representation 83
2. The Representation Relation 87
2.1. A Characterization 87
2.2. The Multiplicity of Representations 93
2.3. Transformation Formulas and Equivalent Theories 97
3. Modeling 99
3.1. From Schema to Theory 99
3.2. Problems of Modeling 101
4. Semantic Components of a Scientific Theory 104
4.1. Denotation Rules and Semantic Assumptions 104
4.2. Philosophical Commitment of the SA's 108
4.3. Application to Quantum Mechanics III
5. Conclusion ll3
4. INTENSION 115
l. Form is not Everything 115
1.1. Concepts of Sense 115
1.2. Extension Insufficient 118
1.3. 'Intensional': Neither Pragmatic nor Modal 120
basic concept has been the object of a theory, and the various theories
have been articulated into a single framework. Some use has been made
of certain elementary mathematical ideas, such as those of set, function,
lattice, Boolean algebra, ideal, filter, topological space, and metric space.
However, these tools are handled in a rather informal way and have been
made to serve philosophical research instead of replacing it. (Beware of
hollow exactness, for it is the same as exact emptiness.) Moreover the
technical slices of the book have been sandwiched between examples
and spiced with comments. This layout should make for leisurely reading.
The reader will undoubtedly apply his readermanship to skim and skip
as he sees fit. However, unless he wishes to skid he will be well advised to
keep in mind the general plan of the book as exhibited by the Table of
Contents. In particular he should not become impatient if truth and
extension show up late and if analyticity and definite description are
found in the periphery. Reasons will be given for such departures from
tradition.
This work has been conceived both for independent study and as a
textbook for courses and seminars in semantics. It should also be helpful
as collateral reading in courses on the foundations, methodology and
philosophy of science.
This study is an outcome of seminars taught at the Universidad de
Buenos Aires (1958), University of Pennsylvania (1960-61), Universidad
Nacional de Mexico (1968), McGill University (1968-69 and 1970-71),
and ETH Zurich (1973). The program of the investigation and a preview
of some of its results were given at the first conference of the Society for
Exact Philosophy (see Bunge, 1972a) and at the XVth World Congress of
Philosophy (see Bunge, 1973d).
ACKNOWLEDGEMENTS
It is a pleasure to thank all those who made useful comments and criti-
cisms, whether constructive or destructive, in the classroom or in writing.
I thank, in particular, my former students Professors Roger Angel and
Charles Castonguay, as well as Messrs Glenn Kessler and Sonmez Soran,
and my former research associates Professors Peter Kirschenmann,
Hiroshi Kurosaki, Carlos Alberto Lungarzo, Franz Oppacher, and
Raimo Tuomela, and my former research assistants Drs David Probst
and David Salt. I have also benefited from remarks by Professors Harry
Beatty, John Corcoran, Walter Felscher, Joachim Lambek, Scott A.
Kleiner, Stelios Negrepontis, Juan A. Nuiio, Roberto Torretti, Ilmar
Tammelo, and Paul Weingartner. But, since my critics saw only frag-
ments of early drafts, they should not be accused of complicity.
I am also happy to record my deep gratitude to the Canada Council for
the Killam grant it awarded this research project and to the John Simon
Guggenheim Memorial Foundation for a fellowship during the tenure of
which this work was given its final shape. Finally I am grateful to the
Aarhus Universitet and the ETH Ziirich for their generous hospitality
during my sabbatical year 1972-73.
MARIO BUNGE
l. GOAL
<
a closely knit and autonomous discipline, this kind of semantics is the
union of two fields: a chapter of linguistics and one of psychology:
Linguistic semantics: the semantics of natural
Empirical Semantics languages
measurement to test its conjectures and models - nor does it need to,
b~cause this kind of semantics does not describe and predict facts. In
other words, nonempirical semantics is concerned not only with linguistic
items but also, and primarily, with the constructs which some such items
stand for as well as with their eventual relation to the real world. (More
on constructs in Ch. 1, Sec. 1.2.) This branch of semantics is then closer
to the theory of knowledge than the theory of language. (We shall take a
look at this point in Ch. 10, Sec. 3.) Moreover, under penalty of being
useless, nonempirical semantics should account for our experience with
conceptual objects - thus being vicariously empirical. In particular, it
should be concerned with our experience in interpreting conceptual
symbols, elucidating the sense of constructs, uncovering their referents,
and estimating truth values. Furthermore, the way it performs this task
should constitute the supreme test of this branch of semantics: there is no
justification for a semantic theory that fits neither mathematics nor sci-
ence nor ordinary knowledge. Conceived in this way, nonempirical
semantics may be split up as follows:
Applied or special
<S.OfmathematiCS
(model theory)
S. of science
S. of ordinary
knowledge
2. METHOD
find out in each case which logical theory justifies them would be going
beyond the scope of this work. We take it for granted that, given a useful
mathematical or scientific concept, there exists a branch of logic on which
it can roost.
As stated above, we assume that logic and mathematics are sufficient
to uncover the form and structure of every construct. We assume also
that they are necessary to build theories aiming at elucidating and system-
atizing the semantic concepts of factual reference, factual sense, and
factual truth. But, of course, we do not claim that logic, mathematics and
the semantical theories contrived with their help will suffice to reveal the
syntax and the semantics of every particular scientific construct - any
more than geometry suffices to triangulate the universe. However, a
combination oflogic, mathematics, semantics and substantive knowledge
can do the trick of uncovering the formalism and the semantics of a
scientific theory. At any rate nothing else has done it, so it is worth
trying.
The preceding assumptions concerning the role of logic and mathem-
atics in constructing philosophical theories, such as semantics, character-
ize what may be called exact philosophy (see Bunge, ed., 1973a). They are
likely to be challenged by philosophers of a traditional cast of mind. But
they may also be doubted by progressives. A case in point is the notion of
sense (or intension or connotation or content) - a favorite with conser-
vatives as well as a bete noire of progressives for being allegedly impreg-
nable to mathematization. We shall meet the challenge by proposing a
mathematical theory of sense - or rather three such theories, one for each
component of the sense of a construct (Chs. 4 and 5). Should these parti-
cular theories fail, others could take their place: approaches and programs
die only if nobody works on them. After all, simpler ideas, like that of
many, or class, were regarded as typically nonmathematical, hence
unclear, until about one century ago. In any event, a bill of difficulties
and failures does not constitute a philosophy.
Ordinary language philosophers and hermeneutic philosophers will
complain that our method is misplaced, because logic and mathematics
are incapable of discerning the structural subtleties of ordinary language
(which one pray?). And if smart they will score a point of two, either
because the locution in question has not yet been tamed or because the
mathematicians have failed to pay proper attention to it. Given chance
6 INTRODUCTION
DESIGNATION
The goal of this chapter is to characterize the most basic of all semantic
concepts: that of designation. This concept occurs in statements such as
rSign x designates concept (or proposition) yl. Since a significant sign
is a member of some language or other, we must start by defining the
notion of a language and, in particular, that of a conceptual language, ie.
one capable of expressing propositions. But this will necessitate a clari-
fication of the very nature and status of concepts and propositions. Which
will in turn lead us to discussing some of the philosophical underpinnings
and ramifications of our enterprise - for philosophical semantics, though
a distinct discipline, is not an isolated one.
1.1. Language
An artificial sign, whether written, uttered, or in any other guise, is a
physical object - a thing or a process that a thing undergoes. But of course
it is a very special object, namely one that
CONCEPTUAL:
Designates constructs
instead of, or in addi-
tion to, facts, feelings,
etc. Ex.: English.
SYMBOLIC:
Represents objects
immediately relevant
to state and drives
of animal.
1.2. Construct
From now on we shall restrict our attention to conceptual languages,
which are the ones employed in mathematics, science, and philosophy.
In anticipation ofthe formal definition to be given in Sec. 2.2 we may say
that what characterizes a conceptual language is that some of its ex-
pressions symbolize ideas. If we abstract from ideation, which is a con-
crete brain process, as well as from communication, which is a concrete
physical process, we get constructs: concepts (in particular predicates),
propositions, and bodies of such - for example, theories. Unlike cognitive
psychology, psycholinguistics, and pragmatics, all of which are concerned
with real people engaged in thinking and communicating, philosophical
semantics abstracts from people, and hence does not handle communica-
tion. (Nor does mathematicallingui~tics for that matter.) Philosophical
semantics handles constructs as if they were autonomous - ie., Platonic
Ideas - without however assuming that there are such.
The existence of constructs may be regarded as a pretense on a par
with the infinite plane wave and the self sufficient steppenwolf: all three
are fictions. (See Vaihinger, 1920; Henkin, 1953.) In each case· the real
thing is far more complex but, if we want to theorize, we must start by
building more or less sketchy models: once the modeling is under way we
may contemplate complication and articulation. In any event we shall
agree that a conceptual language is a language suitable for expressing
constructs of some kind, e.g., biological theories. Moreover we assume
that the chief linguistic categories correspond to (but are not identical
with) the conceptual categories: (some) terms correspond to concepts,
(some) sentences to propositions, certain fragments of languages to
theories. (Cf. Kneale, 1972.) See Table 1.1., which summarizes the pre-
ceding informal characterization of the sign-construct relations.
(Note that we .make no use of the notion of a semantic category
[Bedeutungskategorie] introduced by E. Hussed and worked out by S.
14 CHAPTER 1
TABLE 1.1
Linguistic categories and conceptual categories
1.3. Predicate
We shall analyze the concept of a predicate with the help of the mathe-
matical concept of a function. A function f is a correspondence between
two sets A and B such that, to every member x of A, there is a single
element y of B. The correspondence is written 'f: A --+ B', where A is
called the domain and B the range of f. The value f takes at XE A is desig-
nated by f (x), in tum an element y of B. That is, f (x) = y. This is the gist
of the general concept of a function.
We are interested in a particular kind offunctions, namely propositional
functions. A propositional function P is a function whose values are
propositions. That is, a propositional function, or predicate, is a function
from individuals to statements. Thus "lives" ("is alive") may be regarded
as a certain mapping L from a set D of objects such that, for an individual
c in D, L(c) is the proposition "c is alive". Briefly, L:D--+S, where the
domain D is in this case the set of organisms and S a certain set of state-
ments, namely the class of propositions in which the concept L occurs.
Likewise "dissolves" may be analyzed as a function from the set of
ordered pairs <solvent, solute) to a set of statements. In general, a predi-
cate (or propositional function) of rank (or order) n, where n is some integer
greater than zero, will be analyzed as a function
P:A l x A2 x ... x An--+S
where each Ai' for I ~ i ~ n, is a set of objects, S is a class of statements
(propositions), and the cross represents the cartesian product of the sets
of objects concerned In simpler terms: a predicate of rank n is that which
combines n objects, say Xl' X2, •.. , X n, not necessarily real nor necessarily
distinct from one another, to produce something else, namely PX l X 2 .•• Xn>
16 CHAPTER 1
"&" or "<:>". The advantages of this construal are multiple. First, it holds
for all predicates. Second, it exhibits the referents of a statement. Third, it
does not require the concept of truth - whence it is independent of any
particular theory of truth. Fourth, it writes off without further ado, as ill
formed, compounds such as 'black thought' and 'cleverly melting at
1()() oK' because their components, though bonafide predicates, are defined
on disjoint domains. Such pseudopredicates can be assigned neither
sense nor reference.
Note that our construal of a predicate differs from Frege's analysis as
a function from individuals to truth values. (Recall Frege (1891), in
Angelelli (1967) p. 133: "ein Begriff ist eine Funktion, deren Jtert immer ein
Wahrheitswert ist".) To put it in contemporary mathematical jargon,
Frege identifies a predicate F with the characteristic function XD of the
domain D of F. In short he sets F = XD:D -+ {O, I}. But then he is unable
to distinguish among the various predicates with the same domain,
because there is only one characteristic function for each set. Frege's
DESIGNATION 17
Theory 1 Theory 2
Language of Tl = !l' = Language of T2
Logic of Tl = First order predicate calculus = Logic of T2
Axiom 1 I Pa v Pb. Axiom I (Pa v Pb)
Axiom 2 Pa Theorem 1 I Pa 1\ I Pb.
Theorem 1 Pb Theorem 2 IPa
Theorem 2 (3x) Px Theorem 3 I Pb
Theorem 4 (3x) I Px.
2. DESIGNATION
2.1. Name
Names are those terms in a language that designate objects of some kind.
Thus the numerals '3' and 'III' name the number three. In a conceptual
language all names designate constructs. But the converse is false: not-
withstanding nominalism, most constructs go unnamed. Thus, only a
few irrational numbers and a few functions have got standard names.
Still, just as anonymous persons can be identified by their features and
actions, so constructs can be identified by their properties even if they are
assigned no individual names. For example, the function "two fifths of
the cube plus seven" has got no special name but it is unambiguously
characterized by the preceding description, which amounts to saying what
the function does - namely sending x into !X3 + 7, where x is a real
number. In short, there are unnamed constructs but not ineffable ones.
There are names of (some) individuals, names of (some) classes (or
common names), of (some) relations, and so on. In either case names
may designate a fixed or an arbitrary member of a collection, i.e. an un-
specified individual. Thus we may agree to call 'x' an arbitrary real
number. (Note that a variable is not quite the same as a blank. While
all blanks or voids are the same, namely nothing, variables can differ
from one another and can be manipulated Thus 'x + y = z' is not the
same as' + = '. Nor is a variable something that varies.)
Names are symbols and as such they act as proxies for their nominata
We should not forget what they stand for - unless we happen to be com-
puters. For example, strictly speaking we should not say 'Let R be the
real line' but rather 'Let R name (deSignate) the real line'. However, as
long as we are alert to the distinction between signs and their designata
we can afford to confuse them in our speech and writing for the sake of
brevity. In short, we can indulge in self-designation provided we keep in
mind that symbols are just that - conventional signs proxying for some-
thing else.
The refusal to distinguish symbols from what they symbolize can be
made into a philosophy. It is, indeed, the core of the nominalist or vulgar
materialist philosophy of mathematics. This philosophy appeals to the
haters of intangibles and the lovers of simplicity: instead of having sym-
bols on the one hand and constructs on the other it offers a single bag of
22 CHAPTER I
tangible entities, some natural and others (the signs) artificial. Thus a
member of this persuasion will claim, for example, that "The expression
or string consisting of '(x) ('followed by 'P' followed by 'x:::.' followed by
'P' followed by 'x)' is an analytic sentence of the language r:. And he
will be happy with the nursery school identification of the number one
with a vertical stroke. A simple creed indeed - hence one inadequate to
tackle the complex facts of life. For example the creed makes no room
for the principle that symbols, in particular names, are replaceable be-
cause conventional: that "the real thing" is the designatum not its name:
that no particular sign is indispensable. (For a vigorous defense of this
thesis see Frege (1895). Even Bourbaki (1970, Ch. I), despite its nominalist
flare-ups, warns against confusing symbol with designatum.)
The preceding semantic principle, that although names may be nec-
essary (or convenient) none of them is indispensable, has been credited
to Shakespeare. In fact, in Romeo and Juliet he held that the scent of a
rose is name-invariant This principle, sometimes called the "principle of
reference" (Linsky, 1967), can be given a more exact though far less poetic
formulation, such as the following. Let x be part of an expression e(x)
in a language !l' and let e(y) be the expression resulting from trading x
for y in e(x), where y is yet another sign in !l'. If x and y have the same
designatum, then so have e(x) and e(y~ Consequently the two can be
mutually substituted salva signijicatione et salva veritate. One may adopt
this necessary triviality as part of the definition of the designation rela-
tion.
Thus we accept Shakespeare's thesis that nothing is in a name: that
what really matters is the nominatum. If the latter happens to be a con-
struct, not a physical object, then we may say, following Frege and
Church, that the name points to the sense or to the referent of the con-
struct. But we may as well substitute 'and' for 'or', as the concept of
ambiguity has a pragmatic ring about it. In our theory of meaning a sign
that stands for a concept performs both functions: it signifies the sense
as well as the referent of the construct it designates. (See Ch. 7, Sec. 1.)
Thus the class name Homo sapiens symbolizes the technical concept of
man. The sense of this concept is given by some of the hypotheses oc-
curring in the sciences of man, while its reference class is of course the
set of humans. Whether or not this latter set is empty (as some of us are
beginning to suspect) is no business of the semanticist: the determination
DESIGNATION 23
I ~efi~rentd
I
!?tenote
Nominata:
Location
extralinguistic
relation
items
3. METAPHYSICAL CONCOMITANTS
dependent upon man - they were and are being created by mankind and
will follow the fate of mankind. Surely signs, e.g. inscriptions, are phys-
ical objects: but they depend upon man for their coming into being as
well as for their functioning as signs (proxies) for whatever they stand
for. As to constructs, they are total fictions: what is real is the brain
<
process that consists in thinking of some object. Let us elaborate on this
point.
Concrete thing
Extralinguistic
~ Complex
Simple
Factual
~ <
Term
Phrase
Linguistic Expression
Sentence
Language
Object
Conceptual
Icons truc t)
. Contextle.g."(N.O.')")
Conceptual~
body ~ Set of formulas
Hypothetico-deductive system
and their funds of experience are similar, so will be their thought pro-
cesses, whence their various utterances, though possibly different, will
stand for the same statement or proposition. (See Fig. 1.1.)
I
.....---Natural science-----,,~~Linguistics ~ Semantics--_~
Sentence
IT1
CT2
0""3
Proposition p
Fig. 1.1. A physical stimulus cp, the brain process p (thinking) it elicits, and its linguistic
outputs l1i. sentences expressing the proposition p.
Propositions are not physical objects: they have no reality aside from
brain processes - just as there is no motion independent of moving
things. It is a fiction, albeit an indispensable not an idle one, to assume
that, in addition to the factual items and independently of them, there
is such a thing as a proposition. For Platonists like Bolzano there are
propositions in themselves, that need not have been thought by anyone,
hence that may never be "discovered" (Bolzano, 1837). We make no such
metaphysical assumption: we make instead the methodological pretense,
or useful fiction, that things happen as though there were autonomously
existing propositions (statements) as the designata of some sentences.
(Cf. Sec. 1.2.) The concept of existence involved herein is that of concep-
tual existence not physical existence (Ch. 10, Sec. 4.1). We can turn the
Platonist and the fictionalist loose in the field of constructs precisely be-
cause these are fictions. It is only where real (concrete, material) entities
are concerned that the restrain advocated by the nominalist is in place.
Let us take another look at the situation depicted by Figure 1.1. The
real events are the lightning, its perception, the thought process triggered
in the subject by that perception, and his linguistic utterances. Each of
these real events can be studied by some factual science. In particular,
linguistics may take care of the sentences - and literary criticism may
decree which sentences are "felicitous". Philosophical semantics starts
thereafter, where linguistics leaves off. The former is not concerned with
speech acts or even with linguistic items for their own sake or as con-
stituents of human behavior but only insofar as they represent constructs.
DESIGNATION 29
TABLE 1.2
Sign and construct: main views
REFERENCE
1. MOTIVATION
which is not the case with the latter statement, whence the two are dif-
ferent. He even suggested the possibility of iterating this procedure, thus
coming up with an infinite ladder of statements, everyone of which has
the previous one as its subject or referent (Bolzano, 1851, p. 85).
Case 2 The same philosopher had held previously that rSome people
are literate'" concerns only literate persons while rSome people are not
literate'" is about illiterates only (Bolzano, 1837, III, Sec. 305). True or
false? And how about their equivalents ~ot everyone is illiterate'" and
~ot everyone is literate'" respectively?
Case 3 Aristotle taught that "A single science is one whose domain
is a single genus" (Post. Anal., Bk. I, Ch. 28). Right or wrong? How about
ecology?
Case 4 Evolutionary biologists are divided on the question of the
referents ("units") of population genetics and of the theory of evolution
(see Williams, 1966, Ch. 4~ Is it individual organisms, species, or popu-
lations? And does the theory state that selection acts on genotypes (in-
dividuals) or rather on phenotypes (populations)?
Case 5 Some proponents of the so-called identity theory argue that,
although neurophysiology and psychology employ concepts with dif-
ferent senses, they have exactly the same referent, namely the person,
whence those sciences constitute different ways of viewing the same facts,
namely mental (=neurophysiological) events. This particular defense of
the identity theory rests then on the semantic hypothesis that Referent =
=Fact. How about the ethology and the physiology of birds, which share
their referents?
Case 6 Eminent physicists have claimed that the special theory of
relativity is about the behavior of clocks and yardsticks. Others have
held that it concerns observers in relative motion. Still others, that the
referents of the theory are point masses inhabited by competent and well
equipped experimenters communicating with one another through light
signals. Finally, there are those who contend the theory to be about any
systems connectible by electromagnetic signals. Take your pick.
Case 7 Most physicists act on the assumption that the quantum theory
refers to autonomously existing microsystems such as neutrons and
photons. But when it comes to "philosophizing" many claim that the
theory is about sealed (unanalyzable) blocks constituted, in arbitrary
proportions, by microsystems, measuring instruments, and observers.
34 CHAPTER 2
Still others hold that the theory is concerned with the knowledge of
nature rather than the latter (Heisenberg, 1958, p. 1(0).
Case 8 Consider the formula rFor all integers x: x+ 1 = 1 +x-'. Is it
>
just about integers or rather about the whole system (Z, 1, + (Rosen-
bloom, 1950, p. 110)? In any case, by which criterion is it either?
Case 9 The theory of control systems has been developed by engineers
but has applications in biology and other fields of inquiry concerned with
things, whether inanimate or alive, that have control devices built into
them. Does the theory have a definite reference class at all?
Case 10 What is the reference class of elementary logic? Combinations
of letters? Sentences? Propositions? Argumentative people? The world?
There should remain little doubt that identifying the referents of a
statement or of a theory can be a thorny problem and that we need a
semantic theory capable of helping us to perform that task. We shall
presently expound one such theory. More precisely, we shall first study
the rather unruly relation of reference and shall later on introduce a
couple of law-abiding reference functions - one for predicates, the other
for statements. The outcome may be regarded as a calculus of reference
allowing one to compute the reference class of any composite statement
as a function of the reference classes of its constituents. This calculus
should help us to solve problems of referential ambiguity. By the same
token it should free us from the recourse to authority as the "method"
for ascertaining what scientific theories are about. On the other hand our
calculus will not presume to tell us when a given predicate is applicable,
i.e. what its correct reference or field of validity is: this is a business of
science. In other words, we distinguish reference from extension - a sub-
ject to be studied in Ch. 9, Sees. 1 and 2.
where 'C' stands for the class of constructs and 'Q' for the class of objects.
9l has no simple formal properties. In particular 9l is not reflexive
throughout its graph. For example, the concept of a star concerns stars
rather than itself. On the other hand the number 7 refers to nothing. Nor
is 9l either symmetric or anti symmetric. Finally 9l is not transitive either,
as shown by the following counterexample:
r = 'The statement below is false"l (1)
s (2)
where s refers to a third statement t. Clearly, although 9lrs and 9lst, it
is not the case that 9lrt.
This last result has an important application in the foundations and
philosophy of science, namely in relation to the semantic status of meta-
nomological statements, or laws of laws (Bunge, 1961a; Angel, 1970).
Consider the following propositions:
A fundamental physical statement (e.g. equation) should in-
volve no constants other than universal constants. (3)
Newton's laws of motion are invariant under Galilei trans-
formations. (4)
Every quantum-mechanical formula should correspond to
some classical formula. (5)
If a local field theory is relativistically invariant, then it is
also invariant under the combined conceptual inversion of
charge, time, and parity. (6)
These statements and many others - some descriptive, some prescrip-
tive - are often treated on a par with the object statements of a theory.
However, it is apparent that they are metastatements, i.e. that they concern
further statements. (Caution: not every metastatement belongs to some
meta theory. None of the above does. Moreover some metastatements
belong to object theories - as is the case with (4) and (6) above. For a
typical confusion of 'metastatement' and 'metatheoretical statement' see
Freudenthal (1971).) Further, since the reference relation is not transitive,
the above propositions fail to refer to the same object their referents are
36 CHAPTER 2
about. Thus statement (4) is not a law of motion and (6) is not a field
law - hence neither can be tested by observations on physical objects.
Which suggests that, in spite of operationism, semantics should precede
methodology: before we pose the problem of testing a statement we
should know what it refers to. Enough of this for the moment.
In conclusion, fJI is neither reflexive nor symmetrical nor transitive -
nor, indeed, seems to have any other definite formal trait. It is a wishy-
washy relation and as such not amenable to theory. All our remarks in
this section will therefore be intuitive. To obtain regularity we shall have
to introduce a reference function. This will be done in Sec. 3; but before
that we must take a closer look at fJI.
TABLE 2.1
Examples of immediate and mediate reference
In every such case (a) the formulas of the theoretical model (specific
theory) concern directly the model object itself, and mediately the thing
modeled: for example, a theorem in enterprise theory may be about the
degree of a vertex in the hierarchical tree of the enterprise; (b) the for-
mulas are true of the model object (e.g. the unperturbed rabbit-fox econ-
REFERENCE 37
omy) but only approximately true of the real thing - e.g. the rabbit-fox
system in a variable environment subject to droughts, viruses, etc.; (c)
the reference relation holds between every pair: 9ttm&9tms&9tts even
though it is not transitive.
2.3. Reference Class
The set of referents of a given construct c is called its reference class. More
explicitly, we introduce
Factually
~ umts
Proportionality constants
nonreferential Dimensional constants (e.g.
Boltzmann's k)
Scientific
concepts
Spatiotemporal: coordinates,
metric tensors, etc.
Factually
Property variables: mass, utility,
referential
etc.
Object variables: field, cell,
ecosystem, etc.
of light or the electric charge of the electron, Boltzmann's constant k is
not the value of some property of a physical system.
The variables that are factually referential can in turn be classed into:
(a) object variables, referring to things such as photons or persons; (b)
property variables, representing properties of concrete things and rela-
tions among them, and (c) spatiotemporal variables, concerning the basic
world framework. Variables of all three kinds are likely to occur in one
and the same statement. Examples:
rThe mass of (rocket) r at (time) t, reckonned in tons, equals m.'
i i i i i
Property Object Coordinate Scale / Number
population of the ith country'. However, of all the variables the object
variables are the most important. Without them the property variables
would be baseless and the spatiotemporal variables would be purely
mathematical constructs. Indeed, a property other than a purely formal
(mathematical) property is a property of some concrete individual or
other, whether actual or conjectural. And the spatiotemporal variables
occurring in factual science have a similar though less apparent concrete
support. Thus in physics every distance value is a distance between two
points in some concrete system. And a time interval is the time separation
between two events, one of which may be taken as the start of a process.
No things, neither properties nor events - nor space or time. However, in
the case of spatiotemporal variables it is usually possible and convenient
to feign that the spatiotemporal framework is given prior to, and in-
dependent of, things and events. This pretense (the hypothesis of absolute
space and time), if taken literally, ensues in an illusion of thinglessness,
hence of absence of factual reference. The illusion is dispelled by a
foundational analysis of spatiotemporal concepts and by the associated
metaphysical construction of a relational theory of spacetime, i.e. one
according to which space and time are defined on the class of facts.
Without such a metaphysical groundwork the semantic analysis of
statements containing spatiotemporal variables would be incomplete
and misleading. Thus semantics and metaphysics, far from being
mutually exclusive, are complementary. (More in Ch. 10, Sec. 4.)
2.5. Denotation
Consider the following table.
Objects
Testable statements
Reference Evidence
And only some of the statements in a theory, even when adjoined specific
assumptions and data, are susceptible to empirical test: all others must
remain content with vicarious evidence, if any. (For details see ,Bunge,
1967a, Ch. 5 Sec. 5.6; Ch. 8 Sec. 8.4; Ch. 12; 1973a, Ch. 2; and 1973b,
Ch. 10.) In sum Reference # Evidence. Consequently Reference c1ass#
# Evidence class.
The differences between reference and evidence are best understood in
relation ~o any theory concerning entities that are not accessible to direct
REFERENCE 45
the latter unanalyzed.) Positivist semantics has had its chance and has
failed. Give now realist semantics a chance.
(ii) Psychologism,' "Search for the intentional object, i.e. for the object
of consciousness associated with the given construct". This recipe of
Brentano's is definitely misleading: one and the same formula can be
"read" in different ways by different persons. In other words, the intended
referent of a construct depends not only on the construct but also on the
person who thinks of it and the circumstances under which he or she is
thinking. This is of interest to psychology not to semantics: the latter is
concerned with the hypothetical though possibly real referents of con-
structs.
REFERENCE 47
Reference class
Truth value
3.1. Desiderata
Consider the table on the next page.
A first point to note is that we are tacitly distinguishing reference from
extension. Thus in the 2nd row we state that "Clairvoyance" concerns
mind readers and the like without claiming that there are any such
REFERENCE 49
Construct Referent(s)
1 Human Mankind
2 Clairvoyance Mind readers, soothsayers, etc.
3 Conducting Bodies
4 The moon is round Moon
5 The moon is round or the Moon and cat
cat is fat.
6 The moon is round and the Moon and cat
cat is fat.
7 All quarks are charged. Quarks
8 Nobody has so far observed Quarks and people
any quark.
9 Some people loathe animals. People and animals
10 Nobody loathes animals. People and animals
from predicates to the power set of the union of the cartesian factors
of the domains of the former is called the predicate reference function
iff it is defined for every P in iP> and
alp(P}= U Ai'
l~i~n
relation < on some set A. Then the reference class of < is A. Example
3. Let F:A-'>B be a function. Since F is a particular case of a relation
on A x B, Blp(F)=AuB.
Bls:S-,>~( U Ai)
l~i~n
fltp
Predicate P -----!:.---+
1 ~i:E;n
1~.
Domain At x A2 X ••• X An--+ Statement family S
Remark 1 Constructs other than predicates and statements, such as
the individual x and the set y occurring in the statement rxey"', are
assigned no referents in our theory. The reason is that individuals and
collections thereof are to function as referents themselves. Remark 2
According to Definition 9i, r Pa'" is taken to concern the object named
a, not the name 'a'. One might be tempted to introduce a new symbol,
say 'a', for the nominatum of a. But this would start an infinite regress.
Remark 3 Our analysis of reference is applicable in all contexts, whether
or not "extensional" (truth functional). For example,
fIt(Smith knows that p) = {Smith, p} .
A binary epistemic predicate E, such as "knowing", "believing", or
"doubting", will be analyzed as a function
E: People x Statements-+Statements,
whence
fit (E) = Peopleu Statements.
There is nothing opaque about "oblique" reference if analyzed as
multiple rather than single. More on "intensional" contexts in Ch. 4,
Sec. 1.3. Remark 4 Our definitions of the reference functions are nec-
essary but not sufficient for a correct identification of the referents of
a construct. A precise identification requires some fund of knowledge
and it depends upon one's metaphysics. Consider, indeed, the statement
that the weather is fine. The ostensive (or manifest or superficial) referent
of this statement is the weather. But the weather is not a thing: it is a
state of a thing. The deep (or hidden or genuine) referents of the given
statement are the terrestrial atmosphere and mankind. In fact the given
proposition may be construed as short for ~he atmosphere is in a
state that suits humans"'. The metaphysical principle behind this analysis
REFERENCE 53
COROLLARY 2.3a The reference class of a set relation equals the union
of the sets involved: If p is a binary relation between the sets A and
B, then
Blp(p)=AuB.
In particular
Bl.(A~B)={A, B}
COROLLARY 2.6a The reference class of a predicate and its negate are
the same: If P is in Ifll, then
COROLLARY 2.6b The reference class of a statement and of its denial are
the same: if p is in S, then
~S(P) = ~s( -,p).
4. FACTUAL REFERENCE
DEFINITION 2.19 Let al(c) =A 1 uA 2 u ... uA f ... uAn be the reference class
of a construct c, where the Ai for i between 1 and f < n are sets of non-
conceptual items, i.e: such that Ai ¢ C. Then the factual reference class
of c is the union of the classes of factual items:
alp(C)=Al UA2U ... uAf~al(c).
Example 1 alp (Light rays are represented by straight lines)=Light
rays. The total reference class includes also the set of straight lines.
Example 2 Let f: A - B be a function from a set A of concrete systems
to a set B of numbers. For example,! could represent the relative dilation
of bodies. Then alp (f)=A, whereas alp (f)=AuB. Example 3 Let
g: A x B- R, where A is a set of concrete systems, B the set of conceivable
scale-cum-unit systems associated to the magnitude g, and R the set of
real numbers. Since B is conventional, it is included in the class of
constructs, so that we are left with alp(g)=A.
Not all the constructs occurring in factual science have a factual
reference. For one thing the logical concepts, such as "not" and "all",
have no such reference. Nor do scale concepts, proportionality constants,
and dimensional constants have referents (recall Sec. 2.4). It will therefore
be convenient to coin a name for such constructs. The following con-
ventions will do.
DEFINITION 2.22 The triple CF =(S, iP>, D) is called a factual context iff
(i) it is a context and (ii) iP> includes a nonempty subset of factual pred-
icates.
referent. It would seem then that logic is a factual context after all,
contrary to our previous contention. Not so, because logic is concerned
with the whole of the conceptual universe C, in particular with statements
- not with their referents. (E.g. the reference class of a propositional
connective is the class of statements on which it is defined.) And, as
we saw in Sec. 2.1, the reference relation is not generally transitive.
Of all the factual contexts the most interesting ones, because the
richest, are the factual sciences. A factual science is a science with a
factual reference, i.e. it is a factual context capable of being tested in
accordance with the scientific method. This quick characterization of a
factual science is of a methodological character. A strictly semantic
definition of science seems to be out of the question: a precise sense
and a factual reference are necessary but insufficient to have a science,
and a high degree of truth, though desirable, is neither necessary nor
sufficient. For this reason we do not give here a formal definition of
factual science: this is a matter for methodology (cf. Bunge, 1967a). But
once we have settled for some characterization of factual science we can
use our semantics to shed some light on certain aspects of the former.
First of all the concept of subject matter of a scientific subject:
DEFINITION 2.26 T is a factual theory iff (i) T is a theory and (ii) T contains
factual predicates.
We leave metamathematics, in particular the theory of theories, the
elucidation of the concept of a theory. Our present concern is with the
objects a factual theory is about:
62 CHAPTER 2
al(S)=al(~n(S)).
screen out all those that violate semantic closure, i.e. that are not refer-
entially germane to the premises. (One of the rare occasions on which
such a precaution is taken explicitly, is in stating Craig's interpolation
theorem. A standard formulation of this theorem is as follows. Let p and q
be formulas such that I- p ~ q. Then there exists a third formula r, con-
taining only predicates occurring in p and q, such that I- p ~ r and I- r => q.)
The very choice of premises in an argument should be regulated by the
unspoken yet time honored principle of relevance: The premises should be
not only mutually compatible in the formal sense but also semantically
congenial. This principle of argumentation limits the range of applicabil-
ity of the so called rule of augmentation of premises, according to which
if p entails q then q follows also from p conjoined with an arbitrary
premise r. This rule holds only if the additional premise is either idle or
both formally and semantically coherent with the original premises. Once
the initial assumptions have been chosen we may use a deduction tech-
nique that complies automatically with the requisite of semantical
closure and has the additional advantage that it does not require guessing
the conclusions beforehand, so that it may be used by a nonmathemati-
cian - e.g. a computer. (See, e.g., Hilbert and Ackermann, 1950, p. 24.)
This mechanical procedure for extracting reference preserving conclusions
is as follows. (i) Conjoin all the assumptions; (ii) expand the conjunction
in conjunctive normal form, i.e. as a conjunction of binary disjunctions;
(iii) detach every conjunct and every conjunction of such. The detached
formulas constitute the desired set, i.e. the maximal set of consequences
compatible with reference conservation. This subset of consequences is
semantically closed and finite, hence a smallish subset of the infinite set
of all the consequences of the given premises. We cannot have both all
the infinitely many consequences of a finite set of premises, and semantical
closure.
Our Theorem 2 of reference conservation allows us to determine the
reference class of any factual theory cast in an axiomatic format and
consequently it enables us to compare theories as to their domains. In
fact the theorem backs up the following conventions.
fJlF(T)nfJlF(T')# 0.
DEFINITION 2.32 Let T and T' be two axiomatic factual theories. Then T
is said to be referentially bulkier than T iff
fJlF(T') 2 fJlF(T) .
Example 1 Particle mechanics and the theory of electromagnetic fields
in a vacuum are incommensurable: they are not about the same things.
Example 2 Electrodynamics and mechanics are referentially commen-
surable: they have common referents - bodies. Example 3 Relativistic
mechanics is referentially bulkier than nonrelativistic mechanics because,
unlike the latter, it concerns both bodies and electromagnetic fields.
Hence they are referentially commensurable. Consequently they are
comparable - pace Kuhn and Feyerabend.
Remark 1 If two theories are referentially incomparable then they
are not comparable at all except in strictly formal respects. In particular
they cannot be compared as to explanatory and predictive powers: they
do not explain and predict any shared facts. Remark 2 The concept of
referential commensurability will be supplemented with that of method-
ological commensurability in Sec. 5.1. Remark 3 Our concept of refer-
ential commensurability is at variance with Kuhn's (1962). For Kuhn
two theories are incomparable iff they fit different theory paradigms and
thus are conceptually different even though they may concern the same
things. Behaviorist and cognitivist learning theories would be incommen-
surable in this sense. But, since they concern the same kinds of animal, to
us they are referentially commensurable and thus comparable. If they
were not they could not be regarded as rival. Remark 4 Nor does our
criterion of theory commensurability agree with Feyerabend's, which is
the sharing of statements, hence of concepts. It is of course true that
relativistic mechanics "does not, and cannot, share a single statement
with its predecessor" (Feyerabend, 1970, p. 82). But it is also true that
both refer to bodies (cf. Example 3 above). Moreover the basic concepts
of classical mechanics can be retrieved from the corresponding relativistic
concepts: for example the classical mass function is a certain restriction
of the relativistic mass function (see Bunge, 1970b). Therefore the two
theories are commensurable, hence comparable - which is why they are
being compared all the time. If this were not the case then there would be
no more ground for preferring one of the theories over the other than for
68 CHAPTER 2
spurious referent of it, no matter how many authorities may claim the
opposite. For example, if a theory of electrons fails to contain law state-
ments about instruments sensitive to electrons, then the theory does not
refer to such instruments, much less to those in charge of the latter. It
will not do to claim on philosophical grounds that some reference to
instruments or to observers must be "implied" or "presupposed" by the
theory, for otherwise the latter would be neither meaningful nor testable.
The preceding criterion suggests adopting the following
be concerned with certain types of scientific concept not with the empiri-
cal operation scientists call 'measurement'. So much so that they do not
involve any laws of either the measured object or the measuring instru-
ment, whence they are of no help in designing and interpreting any
measurement proper or even in estimating random errors of measure-
ment (see Bunge, 1973c). Example 3 It is usually held that the quantum
theory of measurement (or some variant of it) accounts for actual mea-
surements and moreover concerns the system-apparatus-observer com-
pound. However, this theory fails to contain variables representing any
properties of the observer. A fortiori the latter does not occur among the
referents of the law statements of the theory - to the point that the author
of the standard quantum theory of "measurement" acknowledged that
the observer "remains outside the calculation" in his theory (von Neu-
mann, 1932, pp. 224, 234). And although variables alleged to represent
properties of an instrument do playa role in this theory, they fail to
concern real instruments. As a consequence none of the formulas of this
theory predicts any experimental results. A genuine measurement theory
concerns a specific set-up, hence it contains specific law statements
accounting for the peculiar structure and composition of the apparatus
and its coupling with the measured thing. Such a specificity shows up in
the constitutive equations or, equivalently, in the occurrence of non-
universal parameters and constants, such as the refractive index, the
electric resistivity, or the magnetic permeability. The quantum theory of
"measurement" contains no such constitutive equations, hence it can
refer to no genuine measurement set up. In sum, the quantum theory of
"measurement" is ghostly: both the observer and the apparatus are spu-
rious referents of it. (For details see Bunge, 1967b, Ch. 5; 1973b, Ch. 4.)
(to observers) not autonomous reality. The facts that all basic law
statements are required to be frame free (or covariant), and that many
magnitudes (such as the spacetime interval, the electric charge, and the
entropy) are invariant, i.e. the same in (relative to) all frames, were not
taken as evidence against this operationist interpretation. They were just
ignored.
Two decades later quantum mechanics suffered a similar fate: its
formulas were read in the light of the then dominant philosophy of
physics. According to the usual or Copenhagen interpretation of quantum
mechanics, every formula of the theory is about some microsystem under
the action of an experimental set up controlled in an arbitrary fashion
by an observer. Because of this alleged control and of the supposed
docility of microsystems, which are imagined to behave as commanded
by the observer, the apparatus-observer complex is usually called "the
observer" - even if the whole experiment is automated. And dynamical
variables are christened "observables". In this way no objective physical
properties would seem to remain: they all become, just as with the
operationist interpretation of relativity, observer dependent. For ex-
ample, an eigenvalue is construed as a possible measurement result, an
eigenstate as the corresponding observed state, a superposition of
eigenstates as symbolizing our own uncertainty concerning the state of
the system for not being under observation, and so forth. Consequently
phenomenalism is pronounced victorious and realism dead. From here
on the transition to strait subjectivism is easy: "the basic principles of
physics, embodied in quantum-mechanical theory, deal with connec-
tions between observations, that is, contents of consciousness" (Wigner,
1970).
Do these non-realist interpretations of contemporary physics have
any ground other than authority? Could semantics help us to make a
rational choice between phenomenalism and realism? It can indeed,
namely by supplying the criterion and the definition of factual reference
given in Sec. 4.3. To reveal the genuine factual referents of a theory pick
its basic predicates, analyze them and disclose the way they function in
the central law statements of the theory - i.e. before the latter is applied
or tested. Of all the alleged independent variables only those will qualify
as genuine which are characterized by the theory and occur in the
latter's law statements: all others make no difference and thus are
74 CHAPTER 2
ghostly, ie. they concern spurious referents. Let us see how this criterion
works in weeding out ghostly variables in a couple of important cases.
Example 1 Consider the famous Lorentz "contraction" formula
L(b, f)=L(b, b) [1-u 2(b, f)/c 2(w)]1/2 (SR)
where b denotes a body (in fact any physical system), f a reference
frame, and w an electromagnetic wave in a vacuum. The formula gives
the length of b in (relative to) J, in terms of the length L(b, b) of b relative
to b itself, and of the ratio u(b, f)/c(w) of two velocities: the velocity of
the body with respect to the frame, and the (absolute) speed of light
c(w) in void. The phenomenalist construes f as an observer and inter-
prets L(b, f) as the apparent length measured by f But, according to
our criterion, this interpretation is inadmissible because the theory gives
a strictly physical characterization of J, i.e. one that contains no psy-
chological and sociological concepts. In sum, the referents of SR are b,
J, and w, all three physical objects. Hence realism is upheld in relation
to special relativity.
Example 2 Look at Heisenberg's indeterminacy formula:
(QM)
If a correct deduction from first principles is performed and analyzed,
it is seen that 1/1((1, (1') represents the state of a microsystem (1 (e.g. an
iron atom) in an environment r:I (e.g. a magnetic field) at a given instant
of time. It is further realized that Ll"'(a,a,)P((1) represents the width ofthe
momentum distribution of the microsystem (1 when the system-envi-
ronment complex is in the state represented by 1/1((1, (1'). Similarly
Ll"'(a,a')Q((1) stands for the width of the position distribution of the
microsystem. In short the object variables or referents of the formula
QM are (1 and (1'. The phenomenalist claims that (1' is the observer-cum-
apparatus complex. However, (a) the formula holds even in the absence
of an environment, i.e. when r:I happens to be the null individual; (b)
the formula holds whether or not r:I includes a measurement device -
for example it holds for the atoms in a star; and (c) the theory does not
specify the properties of either observers or measurement devices: it is
a strictly physical theory and a thoroughly general one, married to no
particular laboratory, let alone ·to a particular mind. Conclusion: the
phenomenalist or operationist interpretation of quantum mechanics is
REFERENCE 75
5. RELEVANCE
[e. g. biology to
psychology)
Factual relevance
Fact o--------------~_o Fact
[e. g. finance to politics)
fact may be said to be relevant to another fact if and only if the former
makes some difference to the latter. And a construct may be regarded
as being pragmatically relevant to an action iff the former is part of a
view or theory that is instrumental in bringing forth or preventing the
given action. We shall be concerned only with the first four types of
relevance. They will be characterized in the following way.
DEFINITION 2.40 Let T and T' be two factual theories. Then T and T' are
said to be methodologically commensurable iff there are empirical facts
that are evidentially relevant to both T and T.
DEFINITION 2.40 Two factual theories are commensurable iff they are
referentially as well as evidentially commensurable.
A final convention: Two theories may be regarded as being semantically
rival, or competing, iff, being commensurable, they have different senses,
REFERENCE 79
i.e. if they do not 'say' the same. (Examples: rival theories of learning, or
of social mobility.) If two theories had the same sense then they could
differ only in the organization of the material: in this case they would be
just different formulations or presentations of one and the same body of
knowledge. But the concept of sense lies far ahead.
Before applying the preceding ideas on relevance we note that they are
relevant to the intuitive notion of a "category mistake" (Ryle, 1949). In
the light of the preceding considerations it is clear that there are no
absolute category mistakes: that whether or not a given association of
predicates is mistaken depends on the theory that is adopted. It is
equally clear that any such mistakes are epistemic not linguistic mishaps:
they do not consist in violations of linguistic rules but in departures from
accepted bodies of substantive knowledge. Thus while in the context of
behaviorist psychology (adopted by Ryle) the coupling of mentalistic
concepts to either behavioral or neurological ones would constitute an
unforgivable category faux pas, in either a more backward or more
advanced psychological context such predicate combinations might be
correct. The science of the day, not either grammar or literary criticism,
is competent to judge whether any given predicate association is right or
wrong. In short, category mistakes are scientific mistakes.
6. CONCLUSION
REPRESENTATION
1. CONCEPTUAL REPRESENTATION
represent anything even when true. Hence not everything that refers
represents.
On the other hand every positive statement does constitute a partial
representation of its referents. In particular a positive factual statement
represents a fact or rather some facet of it. Thus rb grows faster than c-'
refers to band c and represents (truly or falsely) the fact that b grows
faster than c. The negate of the same proposition has the same reference
class, namely {b, c}, but does not represent the "negative" fact that b fails
to grow faster than c: it is just the denial of the former statement.
The distinction between reference and representation is no idle philo-
sophical technicality: it is relevant to our understanding of topical sci-
entific controversies. For example, biologists are still arguing about the
authentic referents of the synthetic (neodarwinian) theory of evolution.
So far they have produced no conclusive argument for either of the
theses in dispute: that the theory is concerned with individual organisms,
or with populations, or with species. However, a semantic glance at the
typical formulas of the mathematical theory of evolution shows the fol-
lowing. First, the theory refers to populations or aggregates of coexisting
and interacting members of a given kind (species). Hence it uses all three
concepts: those of individual, species, and population. Second the theory
represents both individual and collective traits - among other things the
occasional changes in kind (speciation and extinction) occurring in a
population. The failure to realize that there is no incompatibility among
the three concepts, because they perform different and complementary
roles, may be blamed not only on the obsolete dichotomy Platonism-
nominalism, but also on the backwardness of semantics, which has never
helped science find its way.
The difference between reference and representation becomes partic-
ularly clear in advanced theories such as those in physics. For example,
here a probability function will refer to some system or some state(s) of
it, while the values of that function may be taken to represent certain dis-
positions of the system - much as the mass function M refers to bodies
while a particular value M(c) of M will represent the mass of the body c.
In quantum mechanics, every dynamical property of a system, such as
its linear momentum, is represented by some operator in a Hilbert space.
That is, the operator represents a property of its referent. In statistical
mechanics the partition function of a multicomponent system refers to
REPRESENT A TION 87
the latter but fails to represent any single property of it. It accomplishes
much more: it generates the conceptual representatives of all the thermo-
dynamic properties of the physical system. And in electromagnetic theory
the value E(4), x, t) of the vector valued function E at the field 4> at the
place x and the instant t, represents the strength of the electric component
of the referent 4> at x and t. There are infinitely many constructs in the
theory, such as the powers and derivatives of E, all of them with the same
referent but which represent no trait of it.
To sum up. While in factual contexts the reference relation pairs a
construct off to a thing as a whole, or to a collection of things, the rep-
resentation relation matches a construct with some aspect or property
of the thing or collection of things. Just as we read '~c1' as 'c refers to 1',
so we may abbreviate 'c represents l' to 'c:::;' f'. If c happens to be a quan-
titative construct we may also read 'c:::;' l' as 'c represents the strength of
1', where f is a property not a whole thing or a whole fact. For example,
in mathematical neurobiology the element amn of a certain matrix is as-
sumed to represent the strength of the action (excitatory or inhibitory)
of neuron m on neuron n. Table 3.1 exhibits some typical examples of
conceptual representation that will guide our subsequent investigation.
2.1. A Characterization
The relation :::;, of conceptual representation pairs some constructs off
to some objects, whether conceptual or factual. We shall restrict our
study to the case when the represented object is factual, i.e. when it is a
thing or an aggregate of things, a property of either or a change in one
or more properties of a system - i.e. an event. The main types of rep-
resentative and their respective representees are shown in Table 3.2.
A few comments are in order. Firstly, we have included no individual
constants in our list. And this because they can denote but not represent.
Thus the man Socrates is denoted by his name but is represented only
by certain propositions about him as well as by some definite descrip-
tions of him. And an unspecified individual thing (an arbitrary element
of a set of things) is likewise denoted by an individual variable but not
represented by it. Only a collection of statements can represent an un-
specified individual thing. Individual things, whether specific or unspec-
88 CHAPTER 3
TABLE 3.1
Constructs: representing and nonrepresenting
-; 3
.~
'" Open set of R" Open set of an
.... E.
II>
n-manifold
g
II>
~
....
II>
0 Region ofa Body or force field
~ (.)
0 3-manifold
Z
Q(b) Charge of b Bodyb
] !!J P(s) Probability of s System in state s
"EII> Il>0- 6n dimensional Dynamical states of a System with n components
.... (.)
~
~ 0
II> (.)
manifold system with n components
~ Partition function Multicomponent system
!!J
The lake froze. A freezing of the lake The lake
~
II>
The lake felt cold. Joint property of lake Lake and subject
~
=
II>
There are no
and subject
People
.....
CIl
green people.
TABLE 3.2
What represents what
DEFINITION 3.2 Let T be a theory about entities of kind K and call S(x)
the collection of possible states (i.e. the state space) ofthingxEK. Further,
call S = U K S (x) the union of the state spaces of all the members of K
X E
getting any head in the trial. I~ is easily seen that E has a Boolean
structure. Now form the set T of statements describing such possible
events. This set, too, will be a Boolean algebra. Hence there is a bijection
,;, that maps E on to T in such a way that, if e 1 and e2 are in E, then
';'(e 1)=11 ET, ';'(e2)=12ET, ';'(el ne2)= ';'(el) & ';'(e2), ';'(el ue 2 )=
';'(e 1 ) v ';'(e2)' and finally ,;, (e) = I ';'(e). The weakness of this rea-
soning is, of course, that E is not a set of actual facts but of possible
facts: actual events are "positive" and "definite" (simple or composite
but never alternative). Hence there is no bijection mapping actual
chance events on to a theory. What is true is, that probability theory is
part of the formal background of any stochastic theory furnishing a
(never fully accurate) representation of some factual domain.
We postpone a fuller discussion of ,;, to Ch. 6, Sec. 3. Here we note
the relations between ,;, and other semantic relations, as summarized in
Figure 3.1.
Symbol
Construct
Thing Aspect(s) of a
as a thing or fact or
whole whole thing or fact
Fig. 3.1. Relations between designation (D), reference (91), denotation (,1), representation
( ,;, ), and proxying (II).
represents speed. But we may also say that (in the same context) , V' stands
for, or proxies, speed.
We close this section with Table 3.3, that exhibits the many items of
one of the simplest and most characteristic of specific scientific theories.
It shows clearly that the representation assumptions are an integral part
of the theory: without them the latter reduces to a mathematical
formalism.
Coordinate
patch in R 3
s
Fig. 3.2. The coordinate patch constitutes a construct representing another construct,
namely a region of a manifold. And each of them constitutes a sui generis representation
of a region of physical space.
every region of the number space may be mapped onto some other region
of the same space by means of a coordinate transformation. Since there
are infinitely many possible coordinate transformations, there are infi-
nitely many possible representations of a given region of the manifold
M, hence just as many numerical representations of the original region
94 CHAPTER 3
it will be the one occurring in the truest and most numerous law state-
ments concerning the given system.
DEFINITION 3.6 Let T and T' be two theories with the same factual refer-
ents. CalllP and IP' their respective predicate bases. Then T and T' are
said to be semantically equivalent (or to constitute equivalent representa-
tions of their referents) if and only if there exists a set of transformation
formulas for IP and IP' that effects the conversion of T into T' and vice
versa.
Example 1 Lagrangian and hamiltonian dynamics are equivalent rep-
resentations of systems in general even though their formalisms are dif-
ferent. Indeed, there is a bridge or transformation formula between the
two theories, namely H = pi] - L, that leaves the content invariant. Ex-
ample 2 On the other hand the geocentric and the heliocentric "systems
of the world" are not semantically equivalent if only because the former
has no equations of motion (but only equations for the trajectories).
Only the planet trajectories, when written in geocentric or in heliocentric
coordinates, are equivalent representations. Since such trajectories is all
one can observe, an empiricist must conclude to the overall equivalence
of the two representations. However, they are really different in every
other respect: "factual" and "empirical" are not identical concepts. For
example the Copernicus-Kepler-Newton representation of the solar sys-
tem refers not only to the bodies in the system but also to the gravitational
field that keeps them together, which the Ptolemaic representation did
not. (Cf. Bunge, 1961c.)
We close this section by laying down some tenets that are germane to
the preceding considerations although they belong in the pragmatics of
science rather than in its semantics. Whichever their proper location
here they come.
PI For any factual item (thing, thing property, event) it is possible to
build at least one construct that represents it.
P2 Given any representing construct it is possible to form at least one
other construct that is semantically equivalent to the former.
P3 Given any representing construct it is possible to build a seman-
tically stronger construct.
REPRESENT A TION 99
3. MODELING
resentations of the same system. For example, find out whether these
two network diagrams are equivalent. (They are: see Seshu and Reed,
1961, pp. 1-2.)
2 ~----~ r----~ 3
'------~ 2
",
~----MN~------~ 3
we have called the laws of nature are the laws of our methods of repre-
senting it. The laws themselves do not show anything about the world"
(Watson,1950, p. 52). True or false? Neither: just confused. The goal of
natural science is to represent nature - an objective it achieves when
finding its laws. The investigation of the patterns of representation, on
the other hand, belongs to psychology, epistemology and methodology.
An analysis of both laws statements and the concept of representation
would have avoided the confusion.
To conclude: all science aims at producing conceptual representations
of its referents. (See however MacKay (1969) for the claim that the
making of representations is the concern of information theory.) In the
process of building such representations scientists meet challenging
methodological problems that cannot be appraised if hypothesizing,
modeling and theorizing are conceived as data summarizing. Whereas
some such problems are peculiar to the given field of research others are
undisguised epistemological problems. Philosophy can help in under-
standing them or at least in realizing that they involve philosophical
problems.
conform to the theory. Eventually it was found that 7l:-mesons did satisfy
the theory reasonably well. Accordingly the original semantic assump-
tion was changed.
The semanticist cannot decide whether to change the formalism or the
semantics of a scientific theory in the face of adverse empirical evidence.
All he can do is to insist that the semantic formulas be formulated ex-
plicitly and clearly in order to better keep them under control. He can
also alert to the philosophical concomitants of any given semantic - but
this point deserves another section.
/0
strued in at least three different ways:
mental conditions, and (c) whether the experimental results are in fact
critically dependent upon the observer and his equipment. And if the
theory contains semantic assumptions of the subjectivist type, we should
check (a) whether the theory does issue predictions about the theoret-
ician's (or the observer's) mental states or behavior and (b) whether the
available empirical data do bear on the cognitive subject himself rather
than on objects external to him - e.g. other people. In conclusion, the
semantic assumptions of a scientific theory are not conventions and they
are not dicta beyond controversy: they are testable hypotheses. (Only,
they cannot be put to the test separately from the formulas they endow
a factual content with.) That they are hypotheses, and often controversial
ones, can be seen from the debates on the interpretation of the mathe-
matical formalisms of the quantum theories, to which we now turn.
TABLE 5
Two sets of rival semantic assumptions (operationist and realist) for the eigenvalue equation
and the eigenfunction expansion. Aop: operator other than hamiltonian
Symbol Mathematical Operationist Realist
status semantic assumption semantic assumption
designing pressure gauges. And the latter are often employed in testing
quantum mechanics but the very notion of pressure is absent from the
latter. Another example: electromagnetic fields affect the growth of plants
and their shape. Hence one might think of using plants as low sensitivity
instruments for measuring some traits of the electromagnetic field. But
it would be preposterous to claim that the electromagnetic theory, let
alone quantum electrodynamics, represent plants. In general, any state-
ments concerning the nature or the strength of the empirical evidence
relevant to a scientific theory are built with the help of at least one other
theory (Bunge, 1967a, 1973b). However, this is another matter - one for
methodology not semantics.
5. CONCLUSION
INTENSION
i.e. concepts with the same instances, may have different contents. For
example, "mammal" and "hairy" are isomorphic and coextensive, yet
they have ostensibly different contents. Likewise "avian" and "feathery",
"the successor of 1" and "the smallest prime number", and so on and so
forth. In short, a scientific construct is not characterized by its form and
extension alone.
Granted then that, outside logic, there is something, variously called
'content', 'sense', 'intension', or 'meaning', that is not to be written off.
Granted also that reference and extension are of little if any help in de-
termining contents, as coreferentials and even coextensives, such as
"human" and "cruel", can have different senses. Granted, in sum, that
contents or senses are sui generis objects, distinct from forms, reference
classes, and extensions, and thus far more difficult to pin down than
either of these. The question is: What do we do with those elusive ob-
jects? Two attitudes are possible: retreat or attack. The former is argued
for in this way: "Senses have always been obscure. They have defied the
best philosophical minds. Therefore they are hopelessly obscure. Conse-
quently we had better give up any attempt to clarify them". The outcome
of this policy is to allow the beast to roam in the wilderness of obscur-
antist philosophy: after all, "man is a sense-making animal" (Quine, 1966,
p. 175), and nothing human is reputedly alien to philosophers. We refuse
to adopt the defeatist stand We propose to launch an attack with a view
to taming the wild beast. Our motto will be "Divide and conquer".
In fact our point of departure is recognizing that there is not a single
concept of sense but rather three different concepts. We shall show later
on that they are related, but we begin by distinguishing them as so many
dimensions of sense. Indeed we shall propose and explain the thesis that
a construct may have the following kinds of sense:
(i) the totality of its conceptual determiners, or purport:
(ii) the set of constructs it subsumes or embraces - its intension;
(iii) the totality of its implicates, or import.
In each case the sense of a construct is a set of constructs. (Note the
difference with respect to both reference and extension: a scientific con-
struct will refer to, and extend over, sets of factual items.) In each case
the content of a construct is logically related to the construct itself. (Note
again the difference with respect to both reference and extension.) But
in the case of purport we look upwards, at the impliers; in the case of
INTENSION 117
'intensional'. What PM meant by this term was not one of the semantic
concepts of sense but a certain class of pragmatic concepts, often called
'propositional attitudes', "such as what somebody believes or affirms,
or the emotions aroused by some fact". Typical "intensional" terms
would be 'doubts', 'is puzzled by', 'believes', 'knows', and 'asserts', which
designate relations between a person and a proposition. Correspondingly
a statement of the form r x believes that y'" is often called an intensional
context. And any attempt to clarify and systematize such pragmatic
statements is often said to belong to intensional logic.
To find out what such misnomers designate let us examine a typical
example of an "intensional context". Consider the statements
(1)
and
q = rBaby believes that p"'. (2)
Whereas p is allegedly a perfectly "extensional" construct, q would be
an "intensional" one, for its truth value is not a function of the truth
value of the subordinate statement p alone. If it were, the truth value
of q would remain invariant upon replacing p by any of its equivalents
- which is not the case. Indeed, p is equivalent to
r=p&(sv-,s) (3)
which, substituted for p in (2), yields
t = rBaby believes that milk is good and (s v -, s).'" (4)
which, unlike q, is false.
All this has nothing to do with intensions although it might be claimed
to concern intentions. In fact, since tautologies are void of content, p
and r above have the same intension. But they fail to have the same
effect on all persons: whereas to a logician the utterance of r conveys
just as much information as p does, to a layman r may sound more
informative than p and to a baby r is likely to sound as gibberish. (More
on information in Sec. 3.2.) All of this is pragmatics not semantics.
The actual situation is simply this. The statement q in (2) depends
not just on the statement p but also upon a person, namely Baby, and a
circumstance or time t. In brief, q=B(Baby, p, t) where 'B' stands for
122 CHAPTER 4
2. A CALCULUS OF INTENSIONS
2.1. Desiderata
Our aim is to build a calculus of intensions enabling us (a) to exactify the
notion of intension and (b) to compute the intension of a complex con-
struct, such as a conjunction, out of the intensions of its components.
And we wish our calculus to formalize and articulate the following in-
tuitive desiderata:
DI All and only predicates and statements shall have an intension.
D2 The intension of a construct shall be a set.
D3 The intension of a construct in a context C=(S, P, D) shall be
comprised between 0 and C, i.e. if PeC, then 0s;;;J(P)s;;;C.
D4 If P and Q are both either predicates or statements, then
(a) J(P & Q)2J(P), J(Q)
(b) J(PvQ)s;;;J(P), J(Q)
D5 If J(P)s;;;J(Q), then
recall the rule in Ch. I, Sec. 1.3. This does not entail throwing poetry
away but just noting that some poetic expressions, though pragmatically
significant, are semantically nonsignificant.
J(P)
Fig.4.1. The intension of an equivalence equals the (symmetric) difference between the
intensions of the members of the equivalence (Corollary 3).
THEOREM 4.3 For any two predicates (or statements) P and Q, if P~Q
is defined then
J(P &(P~Q))=J(P)u(J(P)nJ(Q))=
= (J(P) u:7(P}Jn (J(P)uJ(Q)) by distributivity.
Consequently 'P', 'P & P' and 'P v P' are not different concepts but
different signs representing the same concept.
Finally a couple of results concerning the relations in intension between
constructs that hold logical relations.
?o ?0
~'Jf(O)
nJIQI
1'n
L lp~~ P"o
J )0
.I(P )
J(P) uJtO)
~ 6u
Fig. 4.2. The lattice of predicates and the lattice of intensions: anti-isomorphic.
130 CHAPTER 4
Definition I and the preceding theorems and their corollaries show that
there is a duality, or anti-isomorphism, between constructs and their in-
tensions, in the sense that f carries meets into joins and conversely.
Look at Figure 4.2.
b(P, Q)=J(P<=>Q).
Proof By Corollary 3 and Definition 8.
Although equivalences obliterate extensional differences, they bring
out differences in intension. This was to be expected in view of the anti-
isomorphism between constructs and their intensions noted at the end
of Sec. 2.3. Indeed, if the logical difference between two constructs is
exhibited by their inequivalence, their semantical difference must be
given by their equivalence. In other words, the function b defined by
Definition 8 does the opposite of its counterpart, the function
d: U x U ~ U that takes pairs of constructs into their inequivalences, i.e.
such that
d(P, Q)=(P & iQ)v(P(iP & Q)=P<l>Q for Pand Q in U.
Furthermore any mapping f: U" ~ U preserving all the values of the log-
ical distance d may be regarded as a tautological transformation. On the
other hand any mapping g: U" ~ U that preserves all the values of the
seman tical distance b may be interpreted as a transformation (whether
logically valid or not) that does nothing to bring the constructs either
closer together or farther apart. We shall.not make a detailed investiga-
tion of the algebraic structure given to U by b in analogy to the one
induced by d on U. (For the latter see Blumenthal and Menger 1970.)
For our purposes it will be enough to prove
l32 CHAPTER 4
THEOREM 4.10 Let 0 be a ring of intensions. Then the structure (0, (),
where (): 02 -+ 0 is the difference in intension, is a pseudoquasimetric
space, i.e. the function () satisfies all of the following requisites:
(i) {)(P, Q)~0;
(ii) {)(P, Q)={)(Q, P);
(iii) {)(P, Q) A{)(Q, R)={)(P, R);
(iv) {)(P, P)=0
for any P, Q, R in the substrate U of o.
Proof All four properties but the third are easily checked by recalling
Definition 8. The "triangle equality"_ (iii) follows from the associativity
ofthe Boolean sum and the equations: AAA=0 and AA0=A.
The interest of this theorem is twofold First, it reinforces our decision
to regard "J(P) AJ(Q)" as the difference or distance in intension be-
tween P and Q. We may now picture intensions as points on a line: see
INTENSION 133
x x
J(P) +-l)(P, Q)-+J(Q)
Fig. 4.3. The intension space is one-dimensional
topology - rather than natural language - is the natural tool for refining
concepts of vicinity and togetherness.
In closing we note that the semantic concept of family resemblance is
not reducible to pragmatic (e.g. psychological or linguistic) concepts.
Thus the semantic distance between two constructs is determined by the
theory in which they occur not by the way they are conceived or mentioned
by a subject under certain circumstances. In particular, semantic distance
is unrelated to both psychological association and linguistic correlation.
Whether or not two constructs are semantically close in a given context,
they may be strongly correlated for some subjects and uncorrelated for
others, particularly if they have never thought ofthem before. This being
so, the attempts to account for semantic sense in psychological or in
linguistic terms are bound to fail.
3.2. Information
Intension is often equated with information content. Thus it is frequently
asserted that, whereas synthetic propositions convey information, ana-
lytic formulas do not. No doubt there is a relation between intension and
information: the greater the former the richer the latter. However, the
two are not identical. To begin with a construct has a content whereas a
signal, such as a written sentence, conveys (carries, transmits) information
to someone. In other words, whereas intension is defined on the set of
constructs, information is defined on the set of pairs signal-subject, where
a subject is a viewer or hearer competent to decode the signal. Informa-
tion, to be received, requires a receiver, i.e. a system equipped with
suitable receiving and decoding devices. The sentence '( - 1)2 = 1', which
expresses or conveys the statement that the square of minus one equals
unity, transmits no information to a baby and it may convey the wrong
information to a mathematician under the influence of LSD. In short,
whereas the concept of intension is semantical, that of information is
pragmatical: the former is subject free, the latter is tied to a subject.
Upon shoving aside the all important information receiver and the no
less important information channel, it is possible to build an impersonal
or semantic concept of information. In fact, upon removing the concept
of a subject from the foregoing considerations and rendering them more
explicit, we are left with the following principles.
INF 2 If Sand S' are sets of signals representing the sets of propositions
P and P' respectively, then
(i) the information conveyed by S is larger than or equal to the one
136 CHAPTER 4
3.3. Testability
Peirce, Frege, the Vienna Circle and the first Wittgenstein held the
identity of meaning and verifiability. Carnap transformed this thesis into
the equation of meaning with test procedures: "the meaning of a sentence
is in a certain sense identical with the way we determine its truth and
falsehood; and a sentence has meaning only if such a determination is
possible" (Carnap, 1936). This thesis was subjected to such devastating
criticisms (see, e.g., Williams, 1937; Russel~ 1948; and Hempel, 1965)
that it has now been almost abandoned by philosophers. (Quine, 1971,
is about the only faithful left.) However, it survives among scientists and
even, in a watered-down form, in Popper's thesis that "the degree of
testability of a statement increases with its content" (Popper, 1963a,
1963b). Prima facie this thesis is plausible enough: tautologies have no
content and are insensitive to empirical tests, whereas factual statements
have a content and are presumably susceptible to such tests. Moreover,
INTENSION 139
4. CONCLUDING REMARKS
Our theory of intensions boils down to the calculus in Sec. 2. This cal-
culus enables one to clarify a number of obscure notions, such as those
of intensional inclusion and intensional independence. We shall see in
Ch. 9, Sec. 1.6, that our theory also allows us to state and even prove the
reciprocal relation between intensions and extensions. Moreover, it en-
ables us to compute the sense of a whole as a function of the senses of
its parts - provided we either know the latter or do not care too much
for them, for remaining satisfied with finding intensional relations. This
shortcoming will be remedied in part in the next chapter, where we shall
learn to find the full sense of a theoretical construct. It will turn out that
the intension of a construct is included in its full sense.
Our theory construes intensions as basic or irreducible semantic ob-
jects. It is couched in set theoretic terms but it does not reduce intensions
to either extensions or reference classes. Nor does our theory make use
of modal concepts. In particular, our Definition 4 is at variance with the
following construals of the notion of intensional inclusion (Lambert and
van Fraassen, 1970):
and
P is intensionally included in Q iff r All (possible) individuals
which are Pare Q'.
We have kept our semantics free from modalities and independent of
modal logics for several reasons. Firstly because we do not need them.
Secondly because it is far from clear how modal prefixes should be in-
terpreted If all 'necessarily' is intended to mean is logical necessity, then
the concepts of entailment do this job and much better: they elucidate
a notion of relative (not absolute) necessity - the necessity of a conclusion
relative to its premises and to the accepted rules of inference. And if what
is meant is ontic (or physical) necessity, then it has no place in semantics
- nor, for that matter, does modal logic supply an adequate elucidation
of that concept. In sum, I find no use for modal logic in semantics. On
the other hand semantics should be used to try and solve some of the
riddles of modal logic, such as whether or not the statement rrhat trick
may work' means the same as rrhat trick may not work'.
So far we have looked, into one of the dimensions of sense. We now
turn to the remaining dimensions - purport and import.
CHAPTER 5
Our analysis of sense in the previous chapter has been "local" or "hor-
izontal": it was limited to the given construct with little regard for its
logical relatives. The limitations of such an approach are obvious. For
one thing it cannot do justice to propositions of the form r A means B',
where 'means' stands for "entails" or for "is entailed by". For example,
that x is loved "means" (follows from) somebody loving x; it also "means"
(entails) that x is lovable. In the present chapter we shall supplement the
"horizontal" or "local" approach to sense with a "vertical" or "global"
analysis. We shall in fact ask ourselves the questions: What is the an-
cestry (or set of implicants) of a construct?, and What is the progeny (or
set of implicates) of a construct? In other words, we will elucidate the
notions we have called purport and import.
We shall formalize the ideas (a) that the purport of a construct in a
given context is the collection of constructs upon which it depends, or
which determine it (logically), and (b) that the import of a construct in
a given context is the collection of constructs that hang from it, or that
are determined by it (logically). The concepts of purport and import are
thus mutually dual. And both are context-dependent: the purport and
the import of a construct depend on the body of knowledge in which it
occurs. This relativization of sense to context curtails a certain freedom
- the freedom to make arbitrary meaning shifts. The advantage of such
a relativization is clear: it is well-nigh impossible to determine the exact
sense of a stray predicate - whence the endless semantic disputes in the
young growing fields as well as in the old undisciplined areas. Only
systemic predicates and statements, i.e. constructs belonging to definite
deductive systems, have definite purports and imports. (When trans-
planted to a different theory, if not rejected, a construct may well acquire
a new sense - i.e. it may become a somewhat new construct.)
Finally the full sense of a construct in a given context may be taken
to equal the union of its purport and its import in that context. Conse-
quently the intension or "inner sense" of the construct will be included
GIST AND CONTENT 143
TABLE 5.1
Conceptual systems
Set of constructs
Context Referential homogeneity
Closed context Boolean algebra Referential homogeneity
Theory Filter Referential homogeneity
Consistent theory UI trafil ter Referential homogeneity
We get a structured context if we keep D and [Ill fixed and allow all and
only logical operations in the sets [Ill and S. Such a context will be closed
both formally (syntactically) and as regards reference (semantically). The
former because logical processing will produce nothing outside S, the
latter because no referents foreign to D will be allowed to intervene in
the course of that processing. (Such a double closure need not stifle
research: we are always at liberty to leap to a different context.) More
explicitly, we lay down
DEFINITION 5.1 The structure C=<S, [Ill, D) is called a closed context iff
(i) C is a context and (ii) S is closed under negation, conjunction, disjunc-
tion, and generalization (both existential and universal).
From an algebraic point of view the statements in a closed context
constitute a complemented lattice. Indeed every member s of S has its
opposite number is in S and, for any two elements sand t of S, both
144 CHAPTER 5
the join s v t and the meet s /\ t (i.e. the conjunction s & t) are in S. More-
over v distributes over /\ and conversely, so that the lattice is distri-
butive in addition to being complemented. And because of the latter
property S contains a null element 0 as well as a universal element o.
Every extensionally void statement equals 0 and every extensionally uni-
versal statement equals O. In brief, we have
THEOREM 5.2 The totality of predicates defined on the same domain and
belonging to a closed context form a complemented distributive lattice
with zero element and unit element - i.e. a Boolean algebra.
By virtue of the algebraic similarity between the set S of statements
and the subset P of predicates in a closed context, we may call the two
by the generic name closed set of constructs. In this way Theorems 1 and
2 can be lumped into the·
and therefore a closed context, but since they are not defined on a
common domain they do not belong to a closed predicate context.
It is now easy to check that every closed set of constructs from which
the universal construct is missing, is a maximal ideal. For reference:
For every element x in a lattice L, the set {YEL Iy~x} is called the
principal ideal generated by x in L. Symbol: (x)L-
The adaptation of these algebraic concepts to our needs yields the
following
(X)C={YEC lyl-x}.
&'«-tC(X)=(X)C={YEC lyl-x}.
¢
(x) u (y) or (x) n (y). A simple counterexample is this:
Vy
x/\y
cepts and its basic assumptions (axioms). Since such constructs have no
logical ancestry other than themselves, they are their own purports and
moreover their own gists. More precisely, we have the following
sets is constituted by the basic factual formulas of the theory, such as the
hypotheses concerning the constitution, structure, and mode of change of
the systems concerned. The other set of basic elements is formed by the
assumptions concerning the mathematical nature of the concept occur-
ring in the previous statements. In other words, the axiom base A of a
scientific theory can he partitioned into the following sets:
M = The mathematical postulates (e.g. function differentiability)
F= The basic factual formulas (e.g. law statements)
S= The semantic formulas (designation and denotation rules and rep-
resentation assumptions).
M F 5
Fig. 5.1. The foundations of a scientific theory: the background B, the mathematical
assumptions M, the factual hypotheses F, and the semantic formulas S.
DEFINITION 5.8 Let The a theory with axiom base A=MuFuS, where
M is the set of mathematical axioms, F the specific factual assumptions,
and S the semantic formulas of T. Then
(i) the specific mathematical gist of T equals M u F;
(ii) the specific factual gist of T equals M u Fu S, i.e. the specific extra-
logical gist of T - in short the whole of A.
This is no concession to semantic holism. We are not claiming that the
GIST AND CONTENT 153
by a single axiom;
A =ro is a binary operation on the set S.'
Let us now add the condition that be associative. The ensuing struc-
0
DEFINITION 5.9 Let T and T' be two nonlogical theories with axiom
bases A and A' respectively. Then the difference in specific extralogical
gist between T and T' equals the symmetric difference between A and A' :
c5(T, T')=AAA'=A-A'uA' -A.
It sometimes happens that two scientific theories share a mathematical
formalism but interpret it differently: that is, they have the same M's
and F's but different S's. At times this semantic difference is due to a
difference in the referents, at other times the referents are the same but
the predicates are assumed to represent different traits of their referents.
An example of the first kind of semantic unlikeness is afforded by the
theories of contagion and of the spread of rumours. An example of the
second kind of semantic difference is the variety of rival interpretations
of quantum mechanics. In either case the difference lies in the semantic
formulas of the theory. More exactly we have
DEFINITION 5.10 Let T and T' be two theories with the same mathematical
gist but different sets Sand S' respectively of semantic formulas. Then
the difference in specific factual gist between T and T' equals the sym-
metric difference between Sand S':
c5(T, T')= SAS'.
From a semantic point of view it does not matter whether T and T'
are coexistent or successive. Nor is it indispensable, in order ot use the
concept of change in gist, to axiomatize the theories of interest. Axiom-
atization is needed only to determine the exact amount of change in
basic purport or gist.
The problem of changes in both full sense and reference, i.e. the ques-
tion of meaning change, will be taken up in Ch. 7, Sec. 3.3. We must now
tum to the dual of purport, namely import, which is the other component
of sense. (Since constructs deriving from the same source have the same
purport, they must differ in import if they are to differ at all.)
spring. Since the concept of a filter will allow us to elucidate the notion
of logical progeny, we may as well take a look at filters. We can do it
quickly because a filter is the dual of an ideal - a character we met in
Sec. 1.2.
In Sec. 1.1 we learned that every closed set of constructs is a Boolean
algebra B and, a fortiori, a lattice. When tracing down the followers of
an element (predicate or statement) in such an algebra we wish to avoid
the surprise of ending up in the null element, for this spells disaster just
as the unit element spells triviality. To get rid of 0 we take a certain
subset of B known as a proper filter. But first the general notion.
A filter in a lattice L is that subset F of L which contains all the fol-
lowers of any given element of L as well as the meets of any two ele-
ments of F. More precisely,
If L is a lattice, then F is a filter in L iff F is a nonempty subset of L
satisfying the following conditions:
Fl For all xeF and yeL, if x~y then yeF.
F2 For all x, yeF, x l\yeF.
If the null element is missing from F the filter is called proper. Since our
construct lattices are Boolean, our filters are more than proper: they are
ultrafilters. In other words, a proper filter F in a Boolean algebra B is
called an ultrafilter just in case, for each xe B, either x or its complement
:i is in F but not both.
Our closed sets of constructs are ultrafilters provided we remove the
null construct from them. This will suffice to guarantee that they are con-
sistent. For, if x is in F, then :i will be automatically excluded from F,
as a consequence of which x I\:i will be out as well. In short, we have
the following
)x(c={yeC I xf-y}.
3.2. Import
We are now in a position to elucidate the idea that the import of a
construct is the collection of constructs it generates or "contains":
Jmfic(x)=)x(c={yeC xf-y}. I
Example 1 The import of a tautology is nil. Example 2 Take identity
to be defined by Leibniz' law. Then the import of "=" includes the
reflexivity, symmetry and transitivity of" = ". (This would have puzzled
Austin, who regarded "same" as being devoid of positive meaning. But
this is excusable: neither Wittgenstein nor his followers had a semantic
theory of meaning.) Example 3 The import of the law of the vibrating
string includes the superposition principle fThe superposition of any
two oscillations is a third oscillation."')
If C happens to be a consistent closed set of constructs, i.e. one in
which no contradictions occur, then the import of a construct x in C
will be the principal ultrafilter generated by x in C. Because filters are
the duals of ideals, and ultrafllters the duals of maximal ideals, we see
that import and purport are the dual of one another, hence mutually
complementary rather than rivals. (More on this in Sec. 4.)
It would be pleasant, but it seems chimerical, to obtain general rela-
tions among the principal filters generated by arbitrary constructs in a
GIST AND CONTENT 157
.fmlle {A \, A2 }
/-A------------------ A--',
I"~ ... -2 -,
I
,
/-
, I 'v\
,,< \ I 1\
I \ \ I \
I \ I \
I \ I \
I \ / \
I \ I \
, \ I \
/ \ / \
/"
I \ ' \
Y m P.e ( A \ ) / \ / \ .f mi'e (A 2)
\ / \
I
Y
I
\ \
/ ,/ \ \
\ " ~ - ....}~ ;
'-- --------- --- -- ---':,. ....._----- - - ------- --- _./
158 CHAPTER 5
(ii) the factual import of the axiom base, or factual content of the
theory T, equals the offspring of B & M & F & S with exclusion of L, i.e.
C(jF(A & B}=.Ymft(B & M & F & S}-L.
That is, the factual content of a scientific theory coincides with the
factual import of its axiom base, which is in turn identical to its extra-
logical import. In other words, while scientific theories have detachable,
hence portable mathematical formalisms, their factual content is not
detachable from the formalism. (A fortiori they have no purely empirical
or observational content.) The extralogical import of the axiom base of
a theory in factual science is both mathematical and factual.
Note that the notion of content, though dependent upon the metalogi-
cal concept of deducibility, is independent of the semantic concept of
truth of fact. Hence a theory with a rich content may well prove to be
false whereas a more modest theory may turn out to be approximately
true. Consequently, pace Popper (1966), the degree of truth of a theory
cannot be measured by its content. If it were so measurable then no
actual empirical tests for truth would be called for: a semantic analysis
of a theory would determine its truth value.
We close by noting a possible application of the foregoing ideas to
model theory. Intuitively the richer an abstract theory ("formalized lan-
guage") the fewer models it is bound to have. Thus there are less ex-
amples of a semigroup with unit element than of a semigroup. This
suggests introducing a comparative concept of theory versatility, or its
dual, rigidity, namely thus:
160 CHAPTER 5
DEFINITION 5.16 Let T and T' be two abstract theories with comparable
mathematical contents ~(T) and ~(T') respectively. Then Tis more ver-
satile (or less rigid) than T' iff ~(T)!;;;~(T').
3.4. Empirical and Factual Content
The various philosophies of science may be regarded as so many views
on the content of scientific theories - i.e. as so many semantics of science.
Here goes, in telegraphic fashion, a sample of the most interesting among
those views:
(1) Conventionalism. Scientific theories are handy tools for the pro-
cessing of experience (or the systematization of appearances). They have
neither a factual nor an empjrical content: ~F(T)=~E(T) =0.
(2) Positivism. Scientific theories are economic summaries arid artic-
ulations of data They have an empirical (observational, operational)
content only: ~(T)=~E(T).
(3) Mellowed positivism (standard view). Scientific theories are system-
atizations of experience propped up by heuristic devices, i.e. the so-called
theoretical terms, which have no representational function. Part of the
content (our "import") of a statement s in a scientific theory is observa-
tional (empirical, phenomenal) provided s is linked to observational
items via correspondence rules: ~E(S)!;;;~(s).
(4) Empirical realism. Theoretical constructs may refer to unobserv-
abIes even though their meaning is given by the connections established
within the theory between the constructs and observational statements
built with the observational vocabulary of the theory. Every scientific
statement s has both an empirical content and a factual one (its "surplus
meaning"): ~(S)=~E(S)U~F(S).
Of these four views the closest to our own is the last, according to which
sc\entific hypotheses have "a surplus meaning over against their eviden-
tial basis" (Feigl, 1950). (For elaborations see Rozeboom (1962,1970). For
criticisms see Hempel (1950) and Nagel (1950).) Although timid realism
is preferable to the previous forms of empiricism, it is not completely
satisfactory: keeping the myth of the observational vocabulary as be-
longing to every theory and as the source of meaning of the theoretical
vocabulary does violence to scientific theories, all the concepts of which
are theoretical. Furthermore it has the fatal consequence that it becomes
impossible to ascertain what exactly the theory is about: reference be-
GIST AND CONTENT 161
etc.). The way this doctrine handles the concept of empirical or observa-
tional content is roughly as follows.
The standard view assumes, first of all, that every scientific theory
contains not only theoretical terms, such as 'temperature', but also ob-
servational or phenomenal ones such as 'hot'. Second, only the theo-
retical terms are assumed to be subject to the theoretical statements fa
of the theory. They are interpreted, albeit only in part, by reference to
the observational terms: this interpretation is performed by the so called
"correspondence rules" C of the theory. Third, the content of a statement
is, roughly, what we have called its import. And the observational (or
empirical) content of a statement s is defined as the class of all non-
logically true observational statements implied by s - i.e. the set of all
those consequences ors that contain observational terms but no theo-
retical ones. In particular, the observational content of s relative to a
theory composed of theoretical statements fa and correspondence rules
C (both sets taken conjunctively), is the observational content of s & fa
& c. (For a brief and lucid expose see Carnap (1963b). For details cf.
Suppe (1971). For the alleged definability of theoretical terms as functions
of empirical ones: Carnap (1961, 1966).)
The root trouble with this view is, of course, that it rests on the un-
questioned assumption that every scientific theory does contain obser-
vational or phenomenal terms such as 'hot', 'blue', and 'rough'. In point
of fact theories contain only theoretical concepts and these are not in-
terpreted with reference to qualia but to supposedly real things (e.g.
genes) and their properties (e.g. the mutation frequency of a given gene).
The semantic assumptions in a well constructed theory are not concept-
concept affairs in which one of the two partners concerns an experiential
item: they are concept-fact correspondences (Ch. 3, Sec. 4.1). Now, if
there are no purely observational (hence unanalyzed) concepts in a sci-
entific theory, then there is no observational (or phenomenal or oper-
ational or empirical) content left. And since (unlike Feigl) Carnap, Hempel
and Braithwaite make no room for factual reference, no content at all is
left.
This unintended disaster could have been averted if any real scientific
theories had been analyzed in the light of the standard semantics, and if
the quest for" the Holy Grail of the empirical "basis" of science had been
given up in time. If theories are regarded instead as partial and symbolic
GIST AND CONTENT 163
suggests that we stop chasing the ghost of the alleged observational con-
tent of theories as well as its attendant ghosts - Ramsey sentences and
Craig's reduction.
Now that operationism lies buried we can afford to write a nice epitaph
on it. The deceased, though mistaken in the means it employed, was
sincere in pursuing a worthy goal- namely the uprooting of inscrutables
and the testability of hypotheses. This explains the large following it had
and still has in the scientific community. Such a noble cause can now be
defended in the following way: Although theoretical statements may lack
an empirical import, they should acquire one when conjoined with suit-
able empirical statements relevant to the former. How this strategy can
be implemented may be exemplified as follows.
Consider the statement schema
DEFINITION 5.17 Let T and T' be two scientific theories with backgrounds
Band B', mathematical assumptions M and M', factual assumptions F
and F', and semantic assumptions Sand S' respectively. Then the dif-
ference in factual content between T and T' equals the symmetric dif-
ference of their factual contents:
b(T, T')=,fmjt(B & M & F & S) A,fmjt(B' & M' & F & S')-L.
Since T and T' are infinite sets, their difference may prove infinite as
well. On the other hand the gist differences introduced in Sec. 2.4 (Defi-
nitions 9 and 10) look more manageable because they concern only the
axiom bases. However, this greater simplicity of the difference in gist
does not render it more suitable as a measure of the difference in sense.
In fact, since a theory can usually be axiomatized in alternative ways,
the gists of two different axiomatizations of a given body of knowledge
are bound to differ even though the total content of the theory is un-
altered by the different axiomatic arrangement. Obviously we need both
measures. And none of them is needed for practical purposes, such as
rewarding the scientist who manages to produce the richest of two rival
theories. The assumptions, definitions and theorems in our semantics
serve only the modest purpose of elucidating some interesting meta-
scientific ideas. (We shall come back to meaning change in Ch. 7.)
4. FULL SENSE
11 Jmft(x)=)X(
Fig. 5.2. A fragment of a conceptual system. The two components of the sense of a
construct x in the system: the purport g.ott~(x) and the import
it can be inferred that the value of its time derivative j( at t=O represents
the initial velocity (or rate of change of concentration or population
density). For this reason only the factual primitives of a theory need be
assigned a factual sense by semantic formulas: the derived (defined or
proved) constructs get their meaning from the source, i.e. the axiom
base. In other words, in a scientific theory sense flows downwards, from
assumption to consequence - not upwards, as the positivist doctrine
has it.
Since a tautology follows from any proposition and entails all of the
ot~er tautologies in a given logic, the full sense of a tautological con-
struct in a context equals the totality of propositions in it plus the under-
lying logic. Likewise all contradictions are interdeducible, so that the
sense of any of them includes all; and since a contradiction entails any-
thing, its sense is the totality of propositions in its context. That is, we
have proved
L~Y'c(P)~SuL.
To obtain the extralogical (in particular factual) sense of a proposition
we must then subtract L from its full sense:
DEFINITION 5.20 Of two constructs with comparable senses, the richer (or
more complex) is the one including the other:
If c and c' are in C, then cis richer than c' = df 9'C(c) 2 9'c (c').
The richer of two concepts is also the more specific and complex, i.e.
the less generic and less simple. The last convention serves then as a defi-
nition of semantic complexity - or its dual, semantic simplicity. But of
course even if two constructs are not comparable in sense they may share
a grain of sense. This notion of sense overlapping is exactified by
9'eo~,,{CiEC 11~i~n}=dJ n
i= 1
S(ci )
Example Each of the different explicata ej, for 1 ~ i ~ n, of a given
coarse or intuitive concept c, is supposed to recapture a part of the sense
of the latter:
hand the greater the purport of a construct, Le. the larger its depen-
dence upon other constructs, the smaller its importance.
Because in a closed context C there is no source of sense other than C
itself, intension has no separate existence in it: it is part of the full sense.
More explicitly, we conjecture
5. CONCLUSION
Publications, Inc.
Carnap, R. (1961). On the use of Hilbert's e-operator in scientific theories. In Bar-Hillel
et al. Eds. (1961), 156--164.
Carnap, R. (l963a). Intellectual autobiography. In Schilpp Ed. (1963), pp. 3-84.
Carnap, R. (1963b). Replies and systematic expositions. In Schilpp Ed. (1963), pp. 859-1013.
Carnap, R. (1966). Philosophical Foundations of Physics. M. Gardner, Ed. New York:
Basic Books.
Castonguay, C. (1971). Meaning and Existence in Mathematics. Ph. D. thesis, McGill
University.
Castonguay, C. (1972). Meaning and Existence in Mathematics. Wien-New York: Springer-
Verlag. •
Chomsky, N. (1963). Formal properties of grammars. In Luce et al. 1963, Vol. II, pp.
323-418.
Chomsky, N. and G. A. Miller (1963). Introduction to the formal analysis of natural
languages. In Luce et al. 1963, Vol. II, pp. 269-321.
Church, A. (1951). A formulation of the logic of sense and denotation. In P. Henle, H. M.
Kallen and S. K. Langer, Eds., Structure, Method and Meaning: Essays in Honor of
Henry M. Sheffer, pp. 3-24. New York: The liberal Arts Press.
Church, A. (1956). Propositions and sentences. In The Problem of Universals: A Symposium,
pp. 3-11. Notre Dame, Ind.: University of Notre Dame Press.
Church, A. (1973/74). Outline of a revised formulation of the logic of sense and denotation.
Nous 7: 24-33, 8: 135-156.
Chwistek, L. (1949). The Limits of Science. London: Routledge and Kegan Paul.
Cohen, L. J. (1962). The Diversity of Meaning. London: Methuen.
Cole, M. and I. Maltzman, Eds. (1969). A Handbook of Contemporary Soviet Psychology.
New York and London: Basic Books, Inc.
Dummett, M. (1973). Frege. Philosophy of Language. London: Duckworth & Co. Ltd.
Feigi, H. (1950). Existential hypotheses. Phil. Sci. 17: 35--62.
Feigi, H. (1958). The 'mental' and the 'physical'. In H. Feigi, M. Scriven, and G. Maxwell,
Eds., Minnesota Studies in the Philosophy of Science, II: 370-497.
Peyerabend, P. K. (1970). Against method: Outline of an anarchistic theory of knowledge.
In Radner and Winokur Eds. (1970) pp. 17-130.
Fraassen, B. C. van (1971). Formal Semantics and Logic. New York: Macmillan.
Frege, G. (1879). Begriffsschrift. Repr. in Begriffsschrift und andere Aufsiitze. I. Angelelli,
Ed. Hildesheim: Georg Olms 1964.
Frege, G. (1891). Funktion und Begriff. In Angelelli (1967).
Frege, G. (1892). Uber Sinn und Bedeutung. In Angelelli (1967).
Frege, G. (1894). Rezension von: E. G. Husserl, Philosophie der Arithmetik.1. In Angelelli
(1967).
Frege, G. (1895). Le nombre entier. In Angelelli (1967).
Frege, G. (1912). Anmerkungen zu P. E. B. Jourdain, The Development of the Theories of
Mathematical Logic, etc. In Angelelli (1967).
Frege, G. (1969). Nachgelassene Schriften. H. Hermes, F. Kambartel and F. Kaulbach,
Eds. Hamburg: F. Meiner.
Freudenthal, H. (1971). More about Foundations of Physics. Foundations of Physics 1:
315-323.
Geach, P. T. (1962). Reference and Generality. Ithaca, N.Y.: Cornell University Press.
Geach, P. T. and M. Black, Eds. (1952). Translations from the Philosophical Writings of
Gottlob Frege. Oxford: Basil Blackwell.
176 BIBLIOGRAPHY
Rosen, R. (1958). The representation of biological systems from the standpoint of the
theory of categories. Bull. Math. Biophys. 20: 317-341.
Rozeboom, W. (1962). The factual content of theoretical concepts. In H. Feigl and G.
Rosenbloom, P. (1950). The Elements of Mathematical Logic. New York: Dover.
Maxwell, Eds., Minnesota Studies in the Philosophy of Science III: 273-357. Minneapolis,
Min.: University of Minnesota Press.
Rozeboom, W. (1970). The crisis in philosophical semantics. In Radner and Winokur, Eds.
pp. 196-219.
Russell, B. (1918). The philosophy of logical atomism. The Monist 28: 495-527; 29: 32-63,
190-222,345-380.
Russell, B. (1940). An Inquiry Into Meaning and Truth. London: George Allen & Unwin.
Russell, B. (1948). Human Knowledge: Its Scope and Limits. London: George Allen &
Unwin.
Russell, B. (1959). My Philosophical Development. London: George Allen & Unwin.
Ryle, G. (1949). The Concept of Mind. London: Hutchinson.
Schaff, A. (1962). Introduction to Semantics. Oxford: Pergamon Press.
Schilpp, P. A., Ed. (1963). The Philosophy of Rudolf Carnap. La Salle, Ill.: Open Court;
London: Cambridge University Press.
Scott, D. (1970). Advice on modal logic. In K. Lambert, Ed., Philosophical Problems in
Logic. Dordrecht: D. Reidel Publ. Co.
Searle, 1. R. (1969). Speech Acts: An Essay in the Philosophy of Language. Cambridge:
University Press.
Seshu, S. and M. B. Reed (1961). Linear Graphs and Electrical Networks. Reading, Mass.:
Addison-Wesley.
Shannon, C. and W. Weaver (1949). The Mathematical Theory of Communications. Urbana,
Ill. : University of Illinois Press.
Sharvy, R. (1972). Three types of referential opacity. Phil. Sci. 39: 152-161.
Shwayder, D. S. (1961). Modes of Referring and the Problem of Universals. Berkeley and
Los Angeles: University of California Press.
Suppe, F. (1971). On partial interpretation. J. Phil. 68: 57-76.
Suppes, P. (1960). Axiomatic Set Theory. Princeton, N.J.: D. Van Nostrand.
Suppes, P. (1967). Set-theoretical Structures in Science. Stanford University: Institute for
Mathematical Studies in the Social Sciences.
Suppes, P. (1969). Studies in the Methodology and Foundations of Science. Dordrecht: D.
Reidel.
Suszko, R. (1967). An essay in the formal theory of extension and intension. Studia Logica
20: 7-28.
Tarski, A. (1956). Logic, Semantics, Metamathematics. Oxford: Clarendon Press.
Tarski, A., A. Mostowski, and R. M. Robinson (1953). Undecidable Theories. Amsterdam:
North-Holland Publ. Co.
Vaihinger, H. (1920). Die Philosophie des Als Ob, 4th ed. Leipzig: F. Meiner.
Waddington, C. H. (1967). Discussion of Eden's paper. In P. S. Moorhead and M. M.
Kaplan, Eds., Mathematical Challenges to the Neo-Darwinian Interpretation of Evolution.
Philadelphia: The Wistar Institute.
Watson, W. H. (1950). On Understanding Physics. New York: Harper & Brothers.
Weingartner, P. (1972). Die Fraglichkeit der Extensionalitiitsthese und die Probleme einer
intensionalen Logik. In R. Haller, Ed., Jenseits von Sein und Nichtsein. Graz: Akade-
mische Druck-u. Verlagsanstalt.
Weingartner, P. (1973). A predicate calculus for intensional logic. J. Phil. Logic 2: 58-141.
BIBLIOGRAPHY 179
TABLE OF CONTENTS
PREFACE
6 Interpretation
7 Meaning
8 Truth
9 Offshoots
10 Neighbors
INDEX OF NAMES
INDEX OF SUBJECTS
BIBLIOGRAPHY