Quine-The Roots of Reference (1990) Cs

the paul carus lectures
the roots of reference
w. v qu1ne
Edgar Piere Professor of Philosophy
Harvard University
197
open court
la salle, illinois
contents
Prefce
ix
I ntroduction
x1
I. PERCEI VI NG AND LEARNING
I . Reception and perception I
2. Cause
4
3. Disposition 8
4. Some questions and answers on dispositions 12
5. Si milarity 1 6
6. I nterfrence from within 20
7. Traces and salience 24
8. Pleasure
27
I I . BREAKI NG INTO LANGUAGE 33
9. Mentalism and language 33
I 0. Observation sentences 37
11. Ostensive learning 41
VIII Contents
1 2. Assent 45
1 3. Values 49
1 4. Masses and bodies 52
1 5. Individuation of bodies 55
1 6. Observational compounds 59
1 7. Predication and the categorical 63
1 8. Serendipity 67
1 9. Color and shape 70
20. Truth functions 75
2 1 . Analyticity 78
I I I . REFERRING TO OBJECTS 8 1
22. Narrowing the subject 8 1
23. General and singular 84
24. Relative clauses 89
25. Substitution and ' such that' 92
26. Quantifiers and variables 97
27. Quantifying over abstract objects 1 0 1
28. Set theory 1 05
29. Sour grapes 1 1 1
30. Identity and number 1 1 5
3 1 . Higher types 1 20
32. Psychogenesis summed up 1 23
33. Past and future 1 27
34. Ontological sophistication 1 3 1
35. Ontological economy 1 34
36. Relative empiricism 1 37
REFERENCES 1 43
I NDEX 1 47
p
reface
Relatively l ittle mystery enshrouds the ways in which we learn
to utter observation sentences, and to assent to them or dissent
from them when asked. Speaking of objects, however-abstract
objects, physical objects, or even sensory objects-is neither so
quickly achieved nor so readily accounted for. To speak of ob
jects beyond peradventure the child must master a considerable
apparatus of linguistic particles-' same' , ' another' , ' that' , ' it' ,
and the plural '-s' , and more-that are inaccessible at the level of
observation sentences. In Word and Object (p. 93) I wrote that
the contextual learning of these various particles goes on
simultaneously, we may suppose, so that they are gradually ad
justed to one another and a coherent pattern of usage is evolved
matching that of society. The child scrambles up an intellectual
chimney, supporting himself against each side by pressure against
the others.
Understandably unsatisfied by so brief and metaphorical an ac-
x
Preface
count of the matter, I have pondered it further; and the result is
the present essay.
I began it early in 1 970, on being invited to give the Paul Carus
Lectures at the end of the next year. By the summer of 1 97 1 my
ideas had taken the form of a hundred-odd pages of rough draft,
and these I expounded in the course of the Summer I nstitute of
Philosophy at Irvine. The lively critical response, especially on
the part of Gil bert Harman, Donald Davidson, Oswaldo
Chateaubriand, David Kaplan, Richmond Thomason, Edwin A.
Marti n, Jr. , and Stephen Stich, was of great value to me and
sparked substantial revisions. Some of my indebtedness is
speci fied at appropiate points in the book. From the revised
manuscript I abstracted the three Carus Lectures for oral presen
tation at a meeting of the American Phil osophical Association in
New York, December, 1 97 1 . Si nce then I have revised and ex
tended the little book.
I presented portions of an interim draft at Valencia last April
in a lecture entitled "Refexiones sobre el aprendizaj e del
languaj e. " Thi s appeared soon after i n Teorema. The
Hagerstrom Lectures, which I am to give at U ppsala next
month, will be based on the fi ni shed book.
I have been helped by comments on various stages of the
manuscri pt from Dagfinn Fllesdal, Richard Herrnstein, Robert
Harris, and especially Burton Dreben. The work was supported
by the National Science Foundation under Grant GS26 l 5.
W. V. Quine
Boston, December 1 972
introduction
By Professor Nelson Goodman
to First Carus Lecture by Professor Quine
The title of Profssor Quine's best-known philosophy book is
Word and Object. From the title of these lectures, I gather he is
going to di scuss an i mportant relation of words to objects-or
better-of words to other objects, some of which are not
words-or even better, of objects some of which are words to
objects some of which are not words.
I am sure that in every case the exact degree of opacity of
refrence wi ll be made entirely transparent, even though the
roots of refrence must be an even di rtier subject than refrence
itself, which as we all know is di rty enough.
In Profssor Quine's many encounters with refrence he has
always insisted on such sterling pri nci ples as: "Don't refr to
what isn' t"; "Don't suppose that merely by talking you are say
ing anything about anythi ng"; but on the other hand, "I f you do
Xll Introduction
say something about somethi ng, don't think you can escape the
consequences by saying you were only talki ng. "
I have no idea what the roots of refrence are, but I suppose
that whatever has powers, as refrence does, also has roots. But
here I am probably guilty, i n the language of Method of Logic, of
making, i n the full sweep of a fll swoop, what amounts to a full
swap or even a foul swipe.
Your attention need not be called to Quine's remarkable
productivity. In this era of control, it is positively indecent. He
gives birth to an i mportant new book more often than the Old
Farmer puts out a new Almanac. His articles are by now in
denumerable. And it is high time for a meta-anthology drawn
fom the anthologies of his own works al ready published.
But my pleasure in yielding the platform to Profssor Quine
arises not because he is a very distinguished philosopher but
because he is a very distinguishing one.
Nelson Goodman
p
art I

p
erce1V1n
g
and learnin
g
1. Reception and perception
Gi ven only the evidence of our senses, how do we arrive at our
theory of the world? Bodies are not given in our sensations, but
are only infrred from them. Should we follow Berkeley and
Hume in repudiating them?
What are given in sensation are smells, noises, fels, fashes,
patches of color, and the like: such were the conditions of the
problem for Berkeley and Hume. But in the present century the
Gestalt psychologists reacted against these conditions. Experi
ment suggests, and introspection as well , that what are sensed are
not primarily those sensory elements, but significantly structured
wholes. Confronted with seven spots equally spaced around a
center, the subject responds rather to the composite circular form
than to any component. Confronted with a solid, he di rectly
senses a body in depth. He goes through none of Berkeley' s in
frential construction of the depth di mension, for he i s unaware
2 The Roots of Reference
of the two-dimensional data of that construction. A painter has
to train hi mself to abstract those two-dimensional patches from
the living scene.
Are we to conclude then that the old epistemologists' problem
of
_
bridging a gap between sense data and bodies was a
pseudoproblem? No; the problem was real but wrongly viewed.
The old epistemologists may have thought that their atomistic at
titude toward sense data was grounded in introspection, but it
was not . It was grounded in their knowledge of the physical
world. Berkeley was bent on deri vi ng depth from two
dimensional data for no other reason than the physical fct that
the surfce of the eye is two-di mensional . But he and the other
old epistemologists would have resi sted this statement of the
matter, because they saw their problem as one of challenging or
substantiating our knowledge of the external world. Appeal to
physical sense organs in the statement of the problem would have
seemed circular. The building blocks had to be irreducibly men
tal, and present to consciousness. Given these ground rules, the
Gestaltists win hands down.
This far of circularity is a case of needless logical timidity,
even granted the project of substantiating our knowledge of
the external world. The crucial logical point i s that the
epistemol ogist is confronting a challenge to natural science that
arises from within natural science. The challenge runs as follows.
Science itself teaches that there i s no clairvoyance; that the only
i nformation that can reach our sensory surfces from external
objects must be li mited to two-dimensioal optical projections
and various impacts of air waves on the eardrums and some gas
eous reactions in the nasal passages and a fw kindred odds and
ends. How, the challenge proceeds, could one hope to fi nd out
about that external world from such meager traces? In short, if
our science were true, how could we know it? Clearly, i n confron
ting this challenge, the epistemologist may make free use of all
scientific theory. Hi s problem is that of fnding ways, in keeping
with natural science, whereby the human animal can have pro
jected this same science from the sensory i nformation that could
reach him according to this science.
Ancient skepticism, in its more primitive way, l ikewise
challenged science from withi n. The skeptics cited familiar i i-
Perceiving and Learning 3
lusions to show the fllibility of the senses; but this concept of il
lusion itself rested on natural science, since the qual ity of illusion
consisted simply i n deviation from external scientific reality.
It was science itself, then as in later times, that demonstrated
the li mitedness of the evidence for science. And it would have
befitted the epistemologist, then as now, to make free use of
science in his effort to determine how man could make the most
of those li mited sources.
Once he recognizes this privilege, the epistemologist can scorn
the Gestalt psychologist' s strictures against sensory atomi sm. He
can appeal to physical receptors of sensory sti mulation and say
that for him what is disti nctive about sense data is mere proximi
ty to these receptors, without regard to awareness. Better still, he
can drop the talk of sense data and talk rather of sensory
stimulation. Our li berated epistemologist ends up as an empirical
psychologist, scientifically investigating man's acquisition of
science.
A far cry, this, from old epistemology. Yet it is no gratuitous
change of subject matter, but an enlightened persistence rather in
the original epistemological problem. It is enlightened in
recognizing that the skeptical challenge spri ngs from science
itself, and that in coping with it we are free to use scientific
knowledge. The old epistemologi st filed to recognize the
strength of his position.
The epistemologist thus emerges as a defnder or protector.
He no longer dreams of a fi rst philosophy, frmer than science,
on which science can be based; he is out to defnd science from
withi n, against its self-doubts. His project becomes one of major
scientific and philosophical interest, moreover, even apart from
protective motives-even apart from any thought of a skeptical
challenge. For we can fully grant the truth of natural science and
still raise the questi on, within natural science, how it is that man
works up his command of that sci ence from the limited im
pi ngements that are available to his sensory surf aces. This is a
question of empi rical psy9hology, but it may be pursued at one or
more removes from the laboratory, one or another level of
speculativity. Its philosophical interest is evident. If we were to
get to the bottom of it, we ought to be able to see just to what ex
tent science is man' s free creation; to what extent, in Eddington's
phrase, it is a put-up job. And we ought to be able to see whatever
there is to see about the evidence relati on, the relation borne to
theory by the observations that support it.
We have undercut the Gestalt psychologist' s criticism of sen
sory atomi sm by dropping the awareness requirement and talk
i ng di rectly of physical input at the sense receptors. This,
however, is only half the story. Awareness and Gestalt still claim
an i mportant place. Sensory receptors operate at the level of
reception, and Gestalt operates at the level of percepti on. The old
antagonism was due to the epistemologist' s straining toward
reception while still requiring awareness, which belongs to
perception.
Reception i s fagrantly physical. But perception also, for all its
mentalistic overtones, is accessi ble to behavioral criteria. It
shows itself in the conditioning of responses. Thus suppose we
provide an ani mal with a screen to look at and a lever to press.
He fi nds that the pressed lever brings a pellet of food when the
screen shows a circular stripe, and that it brings a shock when the
screen shows merely four spots spaced in a semicircular arc. Now
we present hi m with those same four spots, arranged as before,
but supplemented with three more to suggest the complementary
semicircle. I f the animal presses the lever, he may be said to have
perceived the circular Gestalt rather than the component spots.
When conceived thus in behavi oral terms, the notion of
perception belongs to the psychol ogy of learning: to the theory of
conditioning, or of habit formation. Habits, inculcated by con
ditioning, are di spositi ons. The subject, having learned his lesson,
is thereafter di sposed to make the response in question whenever
activated by the stimulus in question. We shall do well, then,
before venturing further, to come to terms with the notion of a
di spositi on.
2. Caue
There is an evident affnity between the idioms of cause and
disposition. Solubility in water, for instance, i s the di sposition to
di ssolve i n water; and being in water causes the soluble body to
P
erceiving and Learning 5
diss
olve. I n their combination of utility and disreputability,
m
oreover, the two idioms are much alike. The trouble with
causation is, as Hume poi nted out, that there is no evident way of
distinguishing it fom mere invariable sucession. And why is this
troublesome? Because then, if we take any two classes of events
such
that each event i n the one class is foll owed by an event in the
other, we have to say that the events in the one class cause those
in the other . Thereupon any arbitrary event a can be said to have
caused any succeeding event b; for, we can just take the two
classes as the unit classes of a and b. We are caught in the fllacy
of post hoc ergo propter hoc.
There is the same quandary over dispositions. If there is no
distinguishing between a thing's disposition to act i n a certain
way i n certain ci rcumstances and the mere fct of its so acting in
those ci rcumstances, then whatever the thing may do can be laid
to a di spositi on, by defining the circumstances narrowly enough.
Stephen Stich has made this point stri kingly with refrence to
i nnate dispositi ons. Why not attribute a man's every act to an in
nate di sposition? True, i f the circumstances in which he now acts
in a certai n way are circumstances in which he once filed so to
act, we seem to have grounds for denying an innate disposition;
but the trouble is that any circumstances whatever can be said to
be of an unprecedented ki nd by defining the ki nd narrowly.
Of these two wayward idioms, the causal and the dispositional,
the causal is the si mpler and the more fundamental . It may have
had its prehi storic beginnings in man's sense of effort, as in
pushing. The i mparting of energy still seems to be the central
idea. The transfr of momentum from one billiard ball to another
i s persistently cited as a paradigm case of causality. Thus we
might seek a si mpleminded or root notion of causality in terms of
the fow of energy. Cause and effect are events such that all the
energy in the effect fl owed from the cause. Thi s ther
modynamical image requi res us to picture energy, like matter, as
traceable from point to point through time. Thus let us picture an
event si mply as any fragment of space-time, or the material and
energetic content thereof. Given an event e, then, imagine all its
energy traced backward through time. Any earlier event that in
t
ercepts all of these energetic world li nes quali fes as a cause of e.
According to this account each event e has countless causes,
distri buted over past time. The remoter the cause, usually, the
more di ffuse it will be in space, since it must intercept every
energetic world l i ne that is desti ned to get into e.
Let us now sort out the good and the bad fatures of thi s no
tion of cause. A possible objection is that it is too special, apply
ing only to physics. My answer is material istic. Causality is a
relation of events, and all events, mental and social ones includ
ed, are a matter ultimately of the action of physical forces upon
particles. My concern here is di fferent from Hume's; his was with
the epistemological basis for a causal relation, while mine is with
the ontological nature of the causal relation as an object of scien
tific theory. All will agree, materialists and others, that causal ef
fcacy within the material world, at any rate, is compounded of
microphysical forces, despite our incapacity to single out all
those components in every particular case.
Another possible objection is that my appeal to energy and
world l i nes is too sophisticated to be appropriate in explicating so
primordial a notion of common sense as the notion of cause. My
answer is that we may reasonably allow concepts, however
primordial, to evolve and sharpen with the progress of science.
After all, even the scope or subject matter of a science may not be
definable until the science has made great strides; so it was with
chemistry. Let the explication of causality not be hampered by
restraints on the use of sophisticated scientifc notions.
A third and opposite objection, and one that I can share, is
that my appeal to energetic world li nes is not sophisticated
enough. On what basis can an earlier and a later bit of work be
associated as two manifstations of one and the same continuing
bit of energy? The very di stinction between matter and energy
wavers in modern physics, and even the notion of the identity of
an elementary particle from moment to moment has fllen on
evil days, what with quantum jumps. Now I take this considera
tion to suggest simply that a notion of cause is out of place in
modern physics. Nor can this come as a surprise. Clearly the
term plays no technical role at austere levels of the subject. And
anyway, modern physicists are notorious for scouting primordial
concepts. They have taken away our absol ute time dimension,
our absolute si multaneity. They have taken away even our
relative velocity, or further relativized it. They have put l i mits on
divisibility, on velocity, on antiquity, and even on theoretical ex
actness of position and velocity. Taking away our causal ity is the
least of it.
For moderate velocities and respectable magnitudes, however,
the Newtonian mechanics retains its utility despite relativity and
quantum mechanics . For terrestrial concerns even the geocentric
fame of refrence retains its util ity despite Copernicus. Si milar
ly a notion of cause based on the not very sophisticated notion of
the fow of energy, or thermodynamics, could retai n utility.
Precisely this util ity, however, is called into question by a
fourth objection. The proposed explication covers only total
cause. It does assign multi ple causes, as we saw, but each of them
is total in this sense: each of them i ntercepts all the energetic l ines
that lead into the effect. These total causes di ffer from one
another only in date or duration or in the capricious inclusion of
superfuities. Yet contributory cause, rather than total cause, is
what we usually care about i n practice. The catalyst introduces
little energy into the chemical reaction that it sets off. The killer
introduces li ttle energy of his own when he triggers the ftal ex
plosion in his gun. The speaker i mparts little energy to the
hearer's eardrum, yet his words may spur the hearer to a frenzy
of energetic output .
We can accommodate contributory causes by defining them as
intercepting merely some of the energetic li nes that lead into the
effect . But one's interest in contri butory causes is conspicuously
i ndependent of the proport i on of energy cont ri buted .
Overwhelmi ng sources such as solar radiation will commonly be
passed over because taken for granted. When in practice we ask
the cause of something, or cite the cause, we are concerned only
with some contributory cause that is of especial interest to the
context. All the rest of a total cause may be uni nteresting because
already known; this one contri butory part is all that was needed
to complete our understanding of the case. Or the rest of the total
cause may be uninteresting because it would be i mmaterial to
some practical end, such as allocation of responsibility.
In practice, indeed, the words 'cause' and ' because' often con
note no contribution of energy at all. They are stretched to apply
to logical premi sses, to purposes, to dispositions. Something' s
merely being paper, or salt, is seen as a contri butory cause of its
taki ng fi re, or di ssolving in water. But all this strikes me as ex
tended usage, contribution of energy being the kernel idea. Of the
extended usage, a large part is best viewed under the head rather
of dispositions.
3. Disposition
The proposed thermodynamical conception of cause does
some justice to the notion of cause on its theoretical or
philosophical side, and i t makes sense to the extent that a notion
of energy fow makes sense. This is better than we can evidently
hope to achieve i n a theoretical account of the notion of di sposi
tion. For, even supposing the notion of cause to be in acceptable
theoretical shape, how would we define di sposition in terms of it?
The disposition is a property, in the object, by virtue of which the
circumstances c cause the object to do a. The ' by virtue' here is
what defies explication. An extensional conditional, a universally
quantified material conditional, does not bridge the gap. Thus
consider a gold piece that is destined never to be in water. The
fmiliar point i s that though we may vacuously and truly say in
an extensional sense that the gold piece di ssolves whenever in
water, we are not therefore to count it soluble. Now the further
point to note i s that even an appeal to cause does not mend
matters. For we can equally well say of the gold piece, vacuously
and truly, that whenever it is in water its being in water causes it
to dissolve. The perenni al fl sity of the antecedent of this univer
sally quantified material conditional has its triviali
_
zing effect
regardless of whether we talk of cause in the consequent. Cause i s
not the mi ssing li nk.
Where the strong connection i s wanted i s between the di sposi
tion (solubility i n water) and its realization (dissolving when in
water). The body di ssolves when in water by virtue of having the
dispositi on. The "virtual" connection is itself analogous to a
causal one; ' by vi rtue of ' is almost ' because of ' . But whereas we
were afforded some hint of an expl ication of cause in the fow of
Perceiving and Learning
9
energy from cause to effect, we can in general look to no such
fow fom the di sposition to its real izati on.
At an uncritical level the usual paraphrase of the di sposition
idi om is an intensional conditional. To say that a body i s soluble
in water is to say that it would di ssolve if it were i n water. This
strengthened conditional does its work at a curious remove.
Where the problematic link ' by virtue of ' was needed was for
linking ' solubility in water' with 'dissolves when in water'. But
the intensional conditional, the ' would i f were' , visibly li nks the
di ssolving rather with the being in water. This is not where a
strengthened li nk was needed; cause, we noted, was not the mi ss
ing li nk. Yet there is no denying that in its bumbling way this in
tensional conditional somehow conveys the force of the dis
positional idiom.
There are those who uncritically accept the di spositional idiom
as a clear matter of ordinary language. Say what a thing i s dis
posed to do i n what ci rcumstances, and the disposition holds no
further mystery for them. Solubility i n water i s the di sposition to
dissolve when i n water, and there is no plainer English than that.
Such is Ryle' s position in The Concept of Mind, where he under
takes to cl arify other more obscure and troublesome notions i n
dispositional terms and is content to leave them thus . Again there
are those who acquiesce in the general defi nition of the dis
positional idiom i n terms of the i ntensi onal conditional . This
group is not to be distinguished from the 'other, since the dis
positional idiom and the corresponding conditional are in
terchangeable i n ordinary language as a matter of course.
One who fi nds discomfort in intensional conditionals is at a
loss for a satisfctory defnition of the di spositional idiom. It was
on this account that Carnap, in his extensionalist period, 1
resorted to a theory of so-called "reduction forms, " as a means
of i ntroducing terms into a theory by partial explications short of
definiti on. These explications were partial in that they were i n
sufficient to render the terms eli mi nable, as true defi nitions
would do. They were "meaning postulates," i n Carnap's later
termi nology. Hi s reduction form or meaning postulate for
l. Testability and meaning."
1 0 The Roots of Reference
solubility in water stipulated that if a body is in water, then it is
soluble in water i f and only i f it is dissolving. The reduction form
for the di sposition to do a in circumstances c stipulated that if
something is in circumstances c then it has the disposition i f and
only if i t is doing a.
Thus Carnap, like Ryle, acquiesced i n an undefined notion of
disposition, but unlike Ryle he acquiesced grudgingly. The no
tion of disposition was short on meaning, in Carnap's eyes, to the
degree that the "reduction form" or "meaning postulate"
stopped short of definition.
I am with Carnap in not settling for defi nition of dispositions
by the intensional conditional. Unlike Carnap, however, I am not
concerned to establish the di sposition idiom as a technical idiom
of scientific theory at all, either by hook or by crook: either by
defini tion or by "meaning postulate. "
Each di sposition, in my view, is a physical state or mechanism.
A name for a specific disposition, e.g. solubility in water,
deserves its place in the vocabulary of scientific theory as a name
of a particular state or mechanism. In some cases, as i n the case
nowadays of solubil ity i n water, we understand the physical
details and are able to set them forth explicitly in terms of the
arrangement and interaction of small bodies. Such a formula
tion, once achieved, can thenceforward even take the place of the
old disposition term, or stand as its new definition.
Where the general di spositional idiom has its use is as follows.
By means of it we can refr to a hypothetical state or mechanism
that we do not yet understand, or to any of various such states or
mechanisms, while merely specifying one of its characteristic
effects, such as di ssolution upon immersion in water . There are
dispositions, such as i ntell igence, whose physical wo, rkings we
can scarcely conjecture; the dispositional characterization is all
we have to go on. Intelligence is the di sposition to learn quickly,
if I may oversimplify. By intelligence I still mean some attribute
of the body, despite our ignorance concerning it; some durable
physical state, perhaps a highly disjunctive one. A term for this
attri bute is entitled to a place in our theoretical vocabulary, even
if all we know about the attribute is that an animal that has it is
quickly conditioned. After all, we do not restrict our theoretical
Perceiving and Learning 1 1
vocabulary to things we understand completely. Ignorance i s
everywhere, and i s a matter of degree. .
I remarked that the fow of energy characteristic of cause i s
not characteristic of di sposition. Still it is not excluded. The dis
positions are hypothetical physical states or mechanisms, and the
mechanisms will pass energy.
Terms for specific di spositions have a legitimate place, then, in
the theoretical vocabulary. As theory progresses, some of these
terms can be paraphrased, li ke 'water soluble' , into terms of the
mechanics of small bodies . Others, like ' i ntelligent' , may stay on
as unel i minable components of a fw theoretical statements. The
general dispositional idiom, however, may best be viewed as ex
ternal to these growing theories in which the particular cases of
the dispositional idiom turn up. By 'the general dispositional
idiom' I mean the general technique of applying the suffix ' -ile'
or '-ble' to verb stems and of using the word ' disposition' and, for
that matter, the corresponding intensional conditional. This
general idiom is programmatic; it plays a regulative rather than a
constitutive role. It forms fmilies of terms on the basis not of
structural or causal affi nities among the physical states or
mechanisms that the terms refr to, but on the basis only of a
sameness of style on our own part in earmarking those states or
mechani sms. The suffix ' -ile' or '-ble' in ' soluble' , ' portable' ,
'visible' , 'ductile' , ' fragile' , 'combustible' , and 'comestible' con
note a sameness in the style of the cues or tests that we are off er
ing for recognizing or identifying these seven physical attributes.
"Put the thing into water and see if it di ssolves, " "heft it and see
if you can carry it," "fce it and see if you can see it," and so on.
The seven physical attributes that are more or less recognized
through these cues can be conveyed in furth\r detail in terms of
size, shape, density, and mi nute physicochemical structure, and
there is no significant physical principle that sets the seven apart
from others. The dispositional idiom is indifferent to the physical
subject matter and serves only to signal how we are getting at it.
So, if I were trying to devise an ideal language for a fini shed
theory of reality, or of any part of it, I would make no place in it
for the general di spositional idi om. In developing a theory, on
the other hand, the idiom is indispensable. Just as in writing an
essay one commonly sketches various ulterior paragraphs before
completing the font ones, so in developing a theory one sketches
in a fw key traits of what is meant ultimately to emerge as a
satisfctorily explanatory mechanism. Such is the role of the
general dispositional idi om. And si nce scientific theory is always
developing, the idiom is here to stay.
. Some questions and answers on dispositions
An i nfirmity of the dispositional idiom, or the intensional con
ditional, is its dependence on a vague proviso of caeteris paribus.
The usual disposition is not surefire. Will anything that is soluble
i n water unfili ngly di ssolve when in water, or must we make
allowance for low temperatures and high ionization and other
possible interfrences as yet unforeseen? One expedient that has
been suggested for accommodating this difficulty is a retreat to
probability: 1 anythi ng that i s soluble in water will probably dis
solve when in water. Now one is bound to agree to the truth of
this probability statement, but it raises questions when proposed
as a means of explication. What sort of probability is intended? I f
it i s subjective probability i n some sense, then this explication of
solubil ity seems wide of the mark; solubility should be a physical
property of the soluble body, even when only dimly understood.
Or i f objective probabi lity is intended, hence relative frequency,
then what is the refrence class? Evidently the class of all soluble
bodies; but that was what we wanted explained. Perhaps the fir
answer is that this probability statement is meant only as a par
tial explicati on of a strictly undefned di sposition term, on a par
with Carnap's reducti on form.
,
Carnap's own response to the caeteris paribus problem was
di fferent: not a resort to probabil ity, but a recognition that his
reduction form could be seen as at best an i nstructive idealiza
tion, because of its unqualified demand that anything soluble in
water dissolve when in water. I f the progress of science were to
l. E.g. see the quotation from Chomsky a page or two hence.
reveal the need of excepti ons for low temperatures or other ci r
cumstances, the reduction form might be fitted out with added
complexities to accommodate them. Now in thi s matter Car
nap' s view resembles the one I am urging: the view of a disposi
tion as a partially discerned physical property that will be more
fully identified, we hope, as science progresses . But a con
spicuous di fference in our views is that for him all such patching
and adjusting of reduction forms was something like redefinition,
giving rise to newly analytic sentences, sentences true by the
meanings of the words. I, on the other hand, am invoking no dis
tinction between analytic sentences and others .
A quandary of Stich's over innate dispositions was noted early
in 2. What can be said of i nnate dispositions now? When I posit
an innate di sposition I am assuming some specific though un
specified arrangement of cells or perhaps some combination of
such arrangements. It could be a nerve tract or a gl and. It could
consist of several structures, variously situated in the organism.
It could be one structure in one individual and some di fferent
structure to the same speci fed effect in another individual . Its in
nateness consists i n its being complete at birth.
Innate refexes, Holt has well argued, are no di fferent in ki nd
from postnatally conditioned refexes. To acquire a refex is to
acquire a neural path of lowered resistance; according to my
philosophy of dispositions, i ndeed, that path is the refex. Some
such paths are establ i shed by rei nforcement of random
movements of the infant, and others, according to Holt, by rein
forcement of random movements of the foetus. The i nnate dis
positions, then, are a mi xed bag: innate refexes are learned in
utero, while innate dispositions of deeper sorts are handed down
from generati on to generation through genetic coding i n the
chromosome. They are a mi xed lot of structures, speci fed
primarily by what they make the ani mal do in what ci r
cumstances, and grouped together by the accident of being com
plete at bi rth.
The attribution of a behavioral disposition, learned or un
learned, is a physi ological hypothesis, however fragmentary. It is
the assumption of some physi ological arrangement such that, if
we were ever to succeed in identifying and analyzing it, we should
arrive at a satisfactory understanding of the mechanism of the
animal behavior i n question.
I have descri bed the primary role, as I see it, of the dis
positional i diom. The idiom i s handy also, though not indispen
sable, in historical contexts . Thus take again the example of
solubil ity in water. Some scientist's discoveries have enabled us
to specify this physical attri bute i n microphysical terms, thus
bypassing the verb stem and dispositional suffx. But how is the
historian to give this scientist his due, once solubility itself is
redefned as nothing other than this microphysical equivalent?
He gi ves him his due, of course, by cleavi ng to the naively dis
positional sense of the disposition term. This, however, is a mi nor
practical poi nt, since the historian would fnd no di fficulty i n
describing the scientist's contribution without resorting to the
dispositional idiom. He has only to say that the scientist showed
that any substance endowed with the microphysical structure i n
question will dissolve when in water, and any substance not so en
dowed will fail to di ssolve when in water.
I once expressed my view of dispositions by saying that a
disposition term is a promissory note for an eventual descri pti on
in mechanical terms. Goodman noted (p. 45n) that these
mechanistic terms will in the last analysis probably be i mplicitly
dispositional i n turn, thus affording no escape from the circle.
This objecti on neatly bri ngs out a di fference between my attitude
toward the problem and the attitude of both Carnap and Good
man. They were seeing the problem as that of defning or
somehow explicating the dispositional idiom in a more strictly
empi rical idiom. The circularity apprehended by Goodman
would matter there. I, on the other hand, am content t9 rest with
a theoretical vocabul ary some of whose primitive physical
predicates were frst learned with help of the dispositional idi om.
Nor am I bent on fi nding a respectable pl ace for the general dis
positional idiom in a regimented theoretical language. I describe
the heuristic role of that idiom in the working up of a scientific
theory, and then I use it .
There are two curious criticisms that I would briefy take up
before closing this general discussion of dispositions, lest the mi s-
conceptions that underlie them be shared by any present readers .
Ziff observes, rightly enough, that every sentence that a man is
capable of using or understanding must correspond to a di sti nct
disposition on hi s part and therefore, on my view, to some dis
tinct mechanism or enduring condition in his body. Ziff fi nds it
i mplausible that there should be so many di sti nct mechanisms or
concurrent physical conditions in the body. Now the reason it is
not i mplausible is that we are not to imagine countless discrete
mechanisms side by side, nor physical states of discrete parts o
the body, one for every sentence. The several mechanisms cer
tainly share their parts, much as the sentences themselves shar
their vocabulary and share their grammatical constructions. One
wonders, indeed, what alternative Ziff has i n mi nd. If two men
now were physically exactly alike down to the smallest particle,
might there be a sentence that the one man is now capable of un
derstanding and the other man not? Maybe the trouble was a
confusion between disti nctness of mechanisms and di screteness
of mechanisms.
The other point was Chomsky' s, i n reference to my "definition
of ' language' as a 'complex of di spositions to verbal behavi or. ' "
Presumably, a complex of dispositions is a structure that can be
represented as a set of probabilities for utterances in certain
definable "circumstances" or "situations." But it must be
recognized that the notion "probability of a sentence" is an entire
ly useless one .... On empirical grounds, the probability of my\
producing some given sentence of English ... is indistinguishable
from the probability of my producing a given sentence of
Japanese. (P. 57.)
Let us not forget that dispositions have their conditi ons. The
probabil ity that a given lump of salt will dissolve at time t is as
may be, but the probability that it will dissolve i f i mmersed in
water is high. Chomsky's worry may have been a more specifc
di fficulty: that of setting conditions for the triggering of verbal1
dispositions. This is an important problem, and happily it has an
easy solution-a soluti on, i ndeed, that was promi nent in the
book that Chomsky was commenting on. It is the procedure of
query and assent, which I shall take up in 1 2.
5. Similarity
Having refected on the general notion of disposition, let us
return now to the notion of perception; for that was what brought
dispositions up. The animal had been trai ned to press the lever
when confronted with the circular stripe and to refrain from
pressing i t when confronted with the fur spots. Then, when he
was confronted with the seven spots, his pressing of the lever was
the criterion of his perceiving the circular Gestalt.
But we gain fexibility i f, instead of speaking thus fatly of
what is or is not perceived, we allow for di fferences of degree.
This can be done by speaki ng of perceptual similarity; thus the
configuration of seven spots proves to be perceptually more
si milar to the ci rcular stripe, for this ani mal, than to the con
fguration of four spots. Better still, we may take perceptual
similarity as relating moments or brief episodes of the subject's
life.
This shift from perceptions to perceptual simil arity bri ngs not
only fexibility but also a certain gain i n ontological clarity, by
dismissing the percepts or percepti ons. Ontologically the
episodes that are related by perceptual similarity may be un
derstood si mply as brief stages or temporal segments of the
perceiving subject's body. They are ti mes i n his lif. Thus they
are global episodes, including all i rrelevancies. But the percep
tual si milarity that relates them is no overall point-by-point
si milarity. It can be as partial as you please, focussing on where
the action i s.
Readers familiar with Carnap's A ufau wi l l be remi nded here
of his Elementarer/ebnisse and A ehnlichkeitserinnerung. The
parallel is no accident .
A theory of perceptual similarity, then, is the place for Gestalt
principles . Perceptual si milarity contrasts with receptual
si mil arity; this is mere physical si milarity of impact on the sen
sory surfces, regardless of behavior. Both of these si milarity
rel ations may be viewed to begin with as triadic: episode a is
more si milar to b than to c. Episodes are receptually si milar to
the degree that the total set of sensory receptors that are
triggered on the one occasion approxi mates the set triggered on
the other occasi on. Perceptual similarity, on the other hand, is a
bundle of second-order dispositions to behavior. Rather than try
to defne the notion at this stage, let us take it provisionally as a
theoretical notion about which some substantial things can be
said. Refecting on the example of the circular stripe and the
spots, we see how perceptual si milarity is mani fsted in behavior.
To explain how, we are apt to say that an episode a is proved to
be perceptually more si mi lar to b than to c when the subject has
been conditioned to respond in some fashion to b and not to so
respond to c, and then is found to respond in that fashion to a.
But let us remember that a, b, and c are individual dated concrete
occasions in the subject's life, whereas conditi oning is di rected
rather to repeatable types of occasions.
This discrepancy can be corrected by enlisting the aid of recep
tual si mil arity, so as to appeal not just to unique episodes a, b,
and c, but more generally to episodes that are receptually si milar
to these. Now receptual si milarity is, we know, a matter of
degree. Full receptual identity would never be realized, or, if
realized, recognized; for it would be a triggering of all and only
the same sensory receptors on the subject' s surface on both oc
casions, no surface barred. However, the mathematical idea of a
neighborhood can be put to use here.
The term ' neighborhood' is one that makes sense oniy in
special contexts, and the key word of those contexts is ' all' . When
we attribute some property to all points in the neighborhood of a
point p, we mean, in the vernacular, that every point "sufficiently
near top" has the property; or, to be quite explicit, we mean that
there i s a point q that is disti nct from p and is such that every
point that is nearer top, than q i s, has the property. Applying this
idea to receptual si mi larity, let us attribute a property to all
episodes in the receptual neighborhood of an episode a when
what we mean is that there i s an episode d that is not receptually
identical with a and is such that every episode that is receptually
more si milar to a, than d is, has the property in questi on.
Now we can correct our formulation of the behavioral condi
tion for perceptual si milarity, to read thus: a is shown to be
perceptually more similar to b than to c when the subject has
been conditioned to respond i n some fashion to all episodes i n the
receptual neighborhood of b, and to withhold that response from
all those in the receptual neighborhood of c, and is then found to
so respond to those in the neighborhood of a.
Perceptual similarity is a question of the subject's disposition
to submit to conditi oni ng in one way and another; hence of his
disposition to acquire or change his habits of response. These
habits are themselves dispositions to behavior, and thus it is that
perceptual similarity is a bundle of second-order dispositions to
behavior.
Perceptual si milarity is no doubt a very disconnected relati on.
That is, there would be many episodes for which i t would make
no evident or useful sense to say that this one was perceptually
more or less si mi lar to that one than to the other. For that
matter, we cannot consider ourselves to have defned perceptual
si milarity even for the best of cases. For such cases I have
propounded a behavioral condition that is a sufficient condition
but not a necessary one. We can console ourselves for this want
of definition by recalling our general refections on dispositi ons.
This behavioral condition for perceptual si milarity serves merely
to earmark a hypothetical mechanism in terms of one of its key
traits.
My characterization of perceptual similarity as tri adic-a is
more si mil ar to b than to c-can be depended upon to have
triggered in the reader's mi nd the more general tetradic idea: a i s
more si milar to b than c i s to d. Pursuing thi s idea, the reader will
have noticed that our way of experi mentally testing for percep
tual similarity comparisons of the triadic ki nd admits of no ob
vious extension to the tetradic. Nor do I see a need for this
tetradic relation, i n a theory o( learni ng.
There i s reason, however, for a polyadic extension of another
form: a is more similar to b,, ... ,bm than to c,, . . . , en.
Thus, to anticipate an example that will receive closer attention
i n 8, let us i magine a certain response rei nforced i n the presence
of a red ball and penalized in the presence of a yellow rose. A red
rose, then, will perhaps not elicit the response, given its favorable
color but unfvorable shape. But if the response was reinforced
also in the presence of a red shawl, the red rose will elicit it. So
we do not want to say that the episode of the red rose was prcep
tually more similar to that of the red ball than to that of the
yellow rose, but we do want to say that it was perceptually more
si milar jointly to the episodes of the red ball and the red shawl
than to that of the yellow rose.
One i s incli ned to distinguish respects of perceptual si milarity;
thus shape versus color. This complication is convenient i n prac
tice, but I think it is dispensable in theory, by spreading the
similarity polyadically as i n the above example.
Let us now turn away from the logical technicalities and con
template the i mportance of the relation of perceptual similarity.
If an individual learns at all, differences in degree of similarity
must be i mplicit in his learning pattern. Otherwise any response,
if rei nforced, would be conditioned equally and indiscri mi nately
to any and every future episode, all these being equally si milar.
Some implicit standard, however provisional, for ordering our
episodes as more or less si mi lar must therefore antedate all learn
ing, and be innate.
Perceptual similarity is always confined within an individua
.
l;
the episodes that it relates are episodes in his li f, and they are
more and less si milar for hi m. One cannot easily give meaning,
indeed, to a general objective similarity relation among things in
the worl d. Might we say that a thing is more si milar to one than
another if it shares more properties with the one than with the
other? But what counts as a property? Classes, certainly, show no
fvorites; a thing shares no more class-membership with any one
th
ng than with any other.

Yet the innate sense of perceptual similarity has, for all its sub
jectivity, a degree of objective validity. After all, man's inductive
expectations are reached by extrapolating along li nes of percep
tual si milarity: experiences that begin si milarly are expected to
turn out i n si mi lar ways . Our innate standards of perceptual
similarity show a gratifying tendency to run with the grain of
nature. This concurrence is accountable, surely, to natural selec
tion. Si nce good prediction has survival value, natural selection
will have fostered perceptual si milarity standards i n us and i n
other ani mals that tend accordi ngly. Natural selection wi ll have
fav
ored green and blue, as avenues of inductive generalization,
and never grue. '

These thoughts are not meant to justiy induction. For that
l . See Goodman, pp. 74-10.
20
The Roots of Reference
purpose the appeal to a law of natural selection would be un
warranted, since that law rests in turn on induction. In the matter
of justifying i nduction we are back with Hume, where we
doubtless belong. Asking for a justi fcation of i nduction is like
asking for a frst philosophy in support of science. What natural
selection contri butes, rather, in the foregoing argument, is a
reason why i nduction works, granted that it does. We have here,
indeed, an illustrative part of an answer to what we recognized
earl ier as the central question of enlightened epistemology: how,
i f our theory of the external world is true, could we ever hit upon
it?
6. Interference from within
We saw that some i mplicit standards of perceptual simi larity
must be i nnate. The standards change markedly, however, with
.
experience.
Pavlov ... mentions an experiment that can easily be interpreted
as the acquisition of a similarity sensation between three stimuli.
A dog was subjected to three types of pairing situations in which
the sound of a buzzer, the sound of a metronome, and a tactile
stimulus were paired with food .... Later, one of these stimuli was
inhibited and it was observed that the inhibition generalized to the
other two stimuli.1
Because of this instabil ity of one's standards of perceptual
si mil arity, a problem arises over what to count as perceptual
si milarity at all. To appreciate this problem, remember that a
subject's standards of perceptual si milarity are to be known only
from his behavior. Some response on his part was reinforced on
occasion b and penal ized on occasion c and then elicited on occa
sion a, and this is supposed to show that a was perceptually more
si mi lar to b for him than to c. Now the trouble with this criteri on
is that it does not screen out the possible effects of the i nternal
states, however transitory, that the subject was enjoying when
these i mpingements overtook hi m. What of his current purposes,
l . Stemmer, p. 211. His reference is to Pavlov, pp. 55f.
Per
ceiving and Learning
2 1
his passing memories, his interrupted train of thought? I speak
mentalistically, but I refer to factors of his physical state,
whatever their physical mechanism. These factors could en
courage or inhi bit the response in questi on, on occasion a or b or
c, overriding the force of current impingements . How to make
proper allowance for these internal interfrences is a delicate
problem, for we do not want to screen out the subject' s actual
contributi on to perception.
Episodes i n the subject's li f ought, one fels, to admit of three
sorts of si milarity. There is receptual si milarity, having to do
only with sensory i nput. There is behavioral si milarity, at the
other extreme, which relates episodes accordi ng to the output of
overt behavior at those times, regardless of causal factors. A
theoretical defnition of behavioral si milarity is readily i magined.
It might be sought in terms of the total set of fbres of stri ped
muscle that are contracted or released on one occasion and on
another, or a more functional approach might be devised.
Perceptual si milarity, fi nally, should be somehow intermediate
between receptual and behavioral similarity. It should be
refected i n the behavioral output of the episode rather than just
current i nput, but it should be refected i n only so much of the
behavioral output as is somehow distinctive to the current input.
However fully the subject's present inner l i f is determined by
the totality of past impingements, it is not determined by present
ones . This is why similar behavior is not in general to be expected
in receptually si mi lar episodes . But perceptual similarity is not to
be expected of receptually similar episodes either, because of
changes i n standards of perceptual similarity, changes i n second
order dispositions. So, if a response that was rei nforced on occa
sion b and penalized on occasion c is elici ted on occasion a, how
are we to judge whether a is indeed more si milar to b than to c by
the subject' s current perceptual standards, or whether some
perceptually irrelevant i nternal state has i ntruded? Our trouble is
that we are groping for a notion of perception just here, with no
other to check against.
The
notion of perceptual si milarity seemed straightforward
enough when we were considering the animal experiment. We
reward a response in one situation and penal ize it in another and
22
then check for it in a third. This idea was the basis for the ex
plication of perceptual simi larity that we undertook in 5 . As
usual where dispositions are concerned, the explication was par
tial: a purportedly sufficient condition for perceptual si milarity
in behavioral terms. Now we are fnding, as is usual where dis
positions are concerned, that our behavioral condition is not hard
and fast even as a sufficient condition. It bogs down in the
caeteris paribus effect (). This happens, we know, to the best of
dispositioqs: interferences intrude. But then it is our job to look
for ways of screening out such i nterfrences as best we can.
In practice this is scarcely a problem. We might expl ore a
human subject's si mil arity standards simply by aski ng hi m which
of various pairs of stimuli are more alike, thus trusting his inar
ticulate sense of what i nterfrences to screen out. Or, using more
cunning, we mi ght condition some refexes; and in this event we
would circumvent any li kely i nterfrences by dint of an i nar
ticulate sense of our own. Theory, however, must be articulate.
What are we sorting out?
At thi s poi nt some heuristi c value can be got from
evolutionary considerations. There must be, we saw, an i nnate
standard of perceptual si milarity. It underl ies our primitive in
ductions, and is accountable to natural selection by virtue of its
survival val ue. So we may be confident that what we are looking
for under the head of perceptual similarity must persist rather
stably, and mani fst itself in the subject' s behavior a good part of
the time, despite sporadic interferences from his ongoing i nternal
states . If it were not thus domi nant and persistent, it would not
have been so important for survival; it would not have hel ped our
ancestors so much in recognizing the wholesome and the toxic,
the predator and the prey. Moreover, such being the nature of
this hypothetical physiological state or mechanism, we may ex
pect it to change only slowly under the infuence of experience.
These refections suggest how one might in principle dis
tinguish between the perceptually relevant episodes, as we might
call them, and the epi sodes in which the behavior is due largely to
internal interference.
-
If an episode is perceptually relevant, then
most episodes that are not very remote in time from that one,
Per
ceiving and Learing 23
and are receptually si milar to i t, should be behaviorally si mi lar to
it . These perceptually relevant epi sodes are the ones to count in
gauging a subject's standards of perceptual si milarity. It is a
matter of detecting regular trends beneath the perturbati ons.
One is remi nded of Fourier analysis of wave patterns, though
matters are vaguer here.
The same considerations of i nnateness and natural selection
suggest also another and better index of what behavior to reckon
to perceptual si milarities. Namely, we can count on considerable
social uni formity i n perceptual similarity standards. We may ex
pect our i nnate similarity standards to be much alike, since they
are hereditary in the race; and even as these standards gradually
change with experience we may expect them to stay signifcantly
alike, what with our shared environment, shared culture, shared
language, and mutual i nfuence. So, if we fnd that one subject' s
episodes a and b tend to be perceptually more or less similar ac
cordi ng as another subject's episodes a
'
and b' are perceptually
more or less si milar, wherever a is receptually very simi lar to a
'
and b to b', we may be encouraged to believe that our plotting of
perceptual similarities for these two subjects is proceedi ng nicely.
Perception being such a private busi ness, I fi nd it ironical that
the best evidence of what to count as perceptual should be social
conformity. I shall not pause over the lesson, but there is surely
one there.
A certain lacuna must be acknowledged, however, in this
matter of a, b, a
'
, and b' . Receptual similarity was defned i n 5
in terms of how close the class of all the receptors that were ac
tivated in one episode came to matching the class of those ac
tivated in another episode. At that point we were thinking of the
episodes and receptors as all belonging to one subject. But now
we have appealed to receptual simil arity between episodes a and
a
'
of two subjects. The subjects share no receptors, so it is no
longer a question of matching the two classes of receptors on the
score of thei r sharing most of their members . It becomes a ques
t
ion rather of how nearly homologous, anatomically, most of the
members of one class are with those of the other. Vagueness
mounts, si nce the receptors of different subjects are fr from
homologous. 2 Nor is anything to be gained by trying rather to
match the distribution of the external forces i mpinging on the
two subjects; for we would have to require that the subjects be
oriented alike to the i mpi ngement pattern, and this revives the
homology questi on. In practice, of course, psychologists fi nd no
di ffculty i n such i ntersubjective equating of sti mulus situations;
they si mply see that there are no physical differences that are apt
to matter. We shall do well to take the same l ine, having just
noted i n passi ng that there is more to the equating of
sti mulations than meets the eye, or indeed perhaps rather less
than seems to do so.
7
. Traces and salience
Episodes leave traces. Memory is a case in point. I shall not
here discuss memory as such, for that concept covers too broad
and di ffuse a range of phenomena to be helpful i n an analysis of
the learning process. We of course cannot avoid positing various
hypothetical mechanisms, partially speci fyi ng them by their re
quired effects and leaving their physical mode of operati on un
determined. But it behooves us to keep these functi onal
specifications as si mple and speci fic as we can, in hopes of
hasteni ng the day when their physical mode of operation may be
understood. Such posits may in part fall within the broad range
of what one calls memory, but they will be more li mited.
A subject's perceptual si milarities are refected in his behavior:
i n the rei nforcement and exti nction of his responses: i n a word, i n
his learning. Perceptual similarity relates his present episode to a
past episode. If perceptual si milarity is to have its requi red effect
on hi s present behavior, then, or

indeed any effect on anythi ng,
the subject must harbor some physiological condition that was
brought about by that past episode. Otherwise that episode
2. Nerve nets differ markedly in structure from one member of a species to
another even at the level of insects. "I should never have expected that the
branching of the main nerves close to the great central ganglion of an insect
would have been variable in the same species; . . . yet quite recently Mr. Lub
bock has shown a degree of variability in these main nerves in Coccus, which
may almost be compared to the irregular branching of the stem of a tree. "
(Darwin, pp. 45f. )
Perc
eiving and Learning 25
would be lost to present perceptual comparison. Such traces,
whatever their physiol ogical nature, are essential to all learning.
The trace of an episode must preserve, i n some form, enough in
formation to show perceptual si milarity between that episode
and later ones. (Recall agai n Carnap's A ehnlichkeitserinnerung. )
Each episode, it will be recalled, is a brief time in the life of the
subject in his bodily enti rety. All impingement is i ncluded, spar
ing no bodily surface. The trace of course preserves no such full
i nformation, nor would much of it be useful for perceptual
similarity. For it is only receptual si milarity that accepts all the
activated receptors on an equal footi ng, unrestricted to what the
subject notices. Perceptual similarity hinges more on notici ng,
and so it is with traces.
Noticing is a matter of degree, and perceptual si milarity is sen
sitive to this variati on. Thus suppose a cat is visible at ti mes a, b,
and c; suppose that the broad visual setting of the cat is much the
same at times a and c but quite different at b; but suppose the cat
i s salient at times a and b, because of motion or spotlighting or
focal positi on, and not at c. Then the subject may fi nd a percep
tually more similar to b than to c, despite the sameness of
landscaPe at a and c. Perhaps a is receptually much more similar
to c than to b; still, salience has the power to swing perceptual
similarity the other way.
Psychologi sts ordi narily speak simply of the stimulus, where I
am speaking of what is salient i n the episode. One reason for my
speaki ng this way is that there can be multiple salience, and i n
varying degrees, within an episode. In classical terms one would
speak, in such a case, of simultane_ous sti muli of unequal
strengths. But the salience version suggests a fi eld of gradations
rather than just one or several clean-cut stimuli, and this I fnd
good. Further, the salience version encourages us to think of the
overall episode as basic, and to think of its operative components
or fatures as abstracted from these episodes by the psychologist
on the basis of coll ati ons of the subject's behavior.
Sal ience shows itself i n behavior through the behavioral
evidence for perceptual si milarity. How salience is shown by
perceptual similarity is evident from the above example of the
cat. Talki ng of sali ence does, however, invite a convenient
modicum of mentali stic idi om. The color of some object may be
said to be salient in one episode, and the shape of the same object
may be said to be salient in another episode. The color of the ob
ject is salient i n the one episode because of its brilliance and
saturation, and the shape is sal ient i n the other episode because
of boundary contrast or movement. It would be intolerable to
deprive ourselves of these quick and vivid ways of speaking. But
let us remember that this i s all meant to be, in the end, a matter
of physiological mechanisms, manifested in behavior.
We noted various conditions of sal ience: focal position, mo
tion, brightness, boundary contrast, gaudy color. Salience can
also be induced by the lingering traces of an earl ier episode. The
trace will tend to accentuate the si milarity of that past episode to
the present one by enhancing the sal ience of the fatures of the
present episode in which that similarity resides. Thus it is that
present sal ience is affected by past experience. The other con
ditions of salience just now l isted are, in contrast, innate. It is as
i f, along with our acqui red traces, we had a fund of innate traces
that were inducing the salience of the bright colors and shape
boundaries and the rest. It is as if there were racial mem
ory-and indeed there is, if we choose to speak thus of natural
selection.
The trace of a past episode fuctuates in vividness. Its vividness
will depend partly on how si milar that episode was to the present
-e and partly on the strength of the trace. By its strength I mean
its capacity to be enlivened by present similarities. That strength
will depend partly on recency; traces tend to wear out. But Freud
has warned against overrati ng this tendency.
Between the trace of a past episode and the present episode, we
see, the enlivening effect is reciprocal . Si milarities enliven the
trace; here is the familiar ma_tter of our being reminded of the
past episode by si milarities i n the present. And conversely, as just
now stated, the trace enhances the sal ience of the present episode
at its points of si milarity to the past one. We can account i n this
way for the power that the sound of the word 'dog' has to draw
our attention to a dog that we would otherwise have overlooked.
The account i s as follows: a trace survives of a past episode of im
pingement from which we learned the word; an episode, that, i n
which the creature was vividly seen and the word heard. The
Per
ceiving and Learning
27
pr
esent
epis
ode of i mpingements resembles that one in part,
nam
ely in the sound of the word. Consequently the trace of the
past
epi sode enhances the sal i ence of other poi nts of
resem
blance, and lo the dog.
We
tend to think of the enlivened trace of an ocular impinge
ment as if it were itself a visual experience si milar to what was
occasioned by that impi ngement, only fi nter: a visual image.
There is even a certai n aptness i n so regarding it, on account of
the effect of the trace uppn sal ience. If we think of the enlivened
trace as an i mage, then in the above exam pie we can think of the
enhanced salience of the dog as brought about by superposing the
trace upon the otherwise i nconspicuous real dog of the current
scene. Likewise the power of verbal suggestion to induce visual i
lusion falls plausibly i nto place. This mentalistic angle could have
heuristic value for the neurophysiologist, by suggesting that the
neurophysiology of the trace may resemble the neurophysiology
of the original sensati on. This shared mechanism may someday
be explained, just as the genes, posited at fi rst as hypothetical
bodies, were fnally explai ned by molecular biology. We do well
surely to avail ourselves of any such heuristic benefits of the men
talistic idi om, while keeping the dangers of an uncritical men
talism frmly in mi nd. I shall say more of this i n 9.
8. Pleasure
Foregoing sections have been lightly strewn with allusions i n
the
classical idiom to the rei nforcement and extinction of
resp
onses to sti mul i . Certain deviant ways of speaking of these
matters seem to off er advantages, but I am slow about edging
over to them for fear of hi ndering communication.
Since 5 the polyadic relation of perceptual similarity has
?
omi n
ated the scene, as a relation among episodes in the sub
ject' s l i f.
Each episode is brief, but it embraces everything that
the
subjec
t is doi ng and suffering at the time. Any focus on
sp
ecial
aspects of this broad cross section is the selective work of
the
relation of perceptual similarity. I n 7, then, I invoked
salience as a way of capturing or coagulating the product of this
selective work on the part of perceptual si milarity. Sal ience took
on some of the burden ordi narily carried by the notion of
stimulus. Now I shall propose a further shift of model or idiom,
which tempers the classical polarity between sti mulus and
response. But the classical idiom has vi rtues of succinctness, and
I shall still lapse into it from time to time.
Episodes are pleasant or unpleasant in varying degrees . It has
already been explai ned that the strength of a trace varies with
recency; also that the trace preserves information needed for
perceptual si milarity. I must now add that the strength of the
trace varies also with the pleasure or discomfort of the episode,
and that the trace preserves an index of pleasure or discomfort. '
Where the trace is that of a pleasant episode, the subject is im
pelled so to exert himself as to increase the si milarity of the
present episode to that pleasant past one; and this drive will vary
in strength with the vividness of the trace. Correspondi ngly,
where the trace is that of an unpleasant episode, the subject is im
pelled to decrease the similarity or hinder its increase.
The drive to increase or decrease the si milarity will therefore
vary with the degree of pleasantness or unpleasantness of the
earlier episode. This follows because the strength of the trace
varies in that fashion, and its vividness varies with its strength
(7), and the drive with the vividness.
The dri ve wi ll vary also with the degree of perceptual similar
ity that already obtains . This follows because the vividness of the
trace varies i n that fashion (7) and the drive with the vividness .
It is a convenient effect, for it sets the subject to trying to recap
ture pleasant episodes on which he already has a head start, or to
avert unpleasant ones that have al ready begun to recur.
Learning, thus viewed, is a matter of learning to warp the
trend of episodes, by intervention of one' s own muscles, i n such a
way as to simul ate a pleasant earlier episode. To learn is to learn
to have fun. Behavi orally, the shoe is on the other foot: an
episode counts as pleasant i f, through whatever unidentified
mechanism of nerves and hormones, it i mplants a drive to
reproduce it. The pleasure is measured by the strength of this
I . See Young, p. 62 1 , on mnemons.
drive. And all this applies also in reverse, to the avoiding of the
unpleasant.
Thus consider the learning of the word ' red' . Suppose the child
happens to utter the word in the course of the random babbl i ng
that is standard procedure in small children, and suppose a red
ball happens to be conspicuously present at the time. The parent
rewards the child, perhaps only by somehow mani fsting ap
proval . Thus in a certai n brief mi nute in the history of overall im
pingements on the child's sensory surfaces there were these
fatures among others: there were l ight rays i n the red frequen
cies, there were sound waves in the air and i n the child's head
bones caused by the child's own utterance of the word ' red' , there
were the i mpacts on the proprioceptors of the child' s tongue and
larynx occasioned by that utterance, and there were the impacts,
whatever they were, that made the episode pleasant. On a later
occasion a red shawl is conspicuously present. Its color makes
for a degree of perceptual similarity between the pleasant earlier
episode and the present, thus enl ivening the trace of that episode.
The child contorts his speech muscles so as to add what more he
can to the si milarity: he agai n says ' red' , and we may hope that
the similarity is yet further enhanced by a recurrence of the
reward.
Or take again the animal (I , 5). He had been through a
pleasant episode whose salient fatures included the circular
stripe, the pressing of the lever, and the emergence of food. Hi s
present episode i s perceptually si mi l ar to that one to the extent of
the circular stri pe, or, what is fai rly si milar for hi m, the seven
spots. He adds what more he can to the si milarity by again press
i ng the lever.
Similarity i s partial . As a history of variously similar episodes
accumulates, the traces compete and determine the net resultant
drive as . by the adding of vectors, or composition of forces. Thus
let us prolong the above example. The episode of the red shawl
proved pleasant, like that of the red ball. But there was also an
unfortunate episode of a yell ow rose. This episode started out
bearing little perceptual si milarity to the two previous ones, but
still the traces of those previous ones had thei r share of vividness
si mply because those episodes were so recent and pleasant. The
two traces j oi ntly proved suffcient to impel the child to recreate
what he could of those episodes; so again he contributed the
sound ' red' , inducing that much present si milarity to both of
those past epi sodes. However, the further fature that had made
those episodes pleasant was not forthcoming this time. On the
contrary, this episode proved unpleasant: a window slammed.
And now we move to yet another episode, i n which the child i s
confronted with a red rose. Its color enlivens, through perceptual
similarity, the traces of the pleasant epi sodes of the red ball and
the red shawl; and its shape enlivens the trace of the unpleasant
episode of the yellow rose. Will he say ' red'? To do so would in
crease the perceptual similarity of the present epi sode to that of
the yellow fower; insofar as he is i mpelled not to. But it would
increase the perceptual si milarity to the two pleasant episodes, so
the score is two to one in favor. He says it, and the story con
cludes on a happy note.
All this is of course easily enough descri bed in classical terms
of the reinforcement and inhibition of responses by conditioning.
But a certain advantage of the present approach i s suggested by
the next example. Imagine a pleasant episode one of whose
various salient fatures was the sound of the word ' preposterous'
spoken by a parent. Later, prompted by the recurrence of some
others of those conspicuous fatures, the child undertakes to
enhance the overall similarity of the present epi sode to that
pleasant one by sounding the word himself. Now the parent's
saying ' preposterous' in the fi rst episode bel ongs on the stimulus
side, and the child's saying it i n the second episode belongs on the
response side. But by speaking of perceptual similarity of
episodes as above, rather than in the polar terms of stimulus and
response, we exploit the auditory si milarity of the two utterances
in the child' s ear and abstract from their di fference of source.
There are many incidental differences anyway between any two
episodes, and i n the present case we are enabled to relegate the
difference of speaker to this category. Radical separation of the
sti mulus and response would have obstructed this account.
People and other primates are i nveterate imitators and im
itatrices. We are said to have an instinct for imitati on. I just now
explained i mitation in one instance, however, without special
assumption. For the child would be said in this i nstance to have
i mitated the parent, not only in saying ' preposterous' but in say-
ing
it in circumstances si milar to those in which the parent said it.
Much else that goes by the name of i mitation i s perhaps si milarly
covered.
The child had of course al ready acqui red the muscular tech
ni ques for producing desi red sounds. Whereas perceptual simi
larity and the pleasure principle are concerned with motivation,
this is the question of engineering. I shall not be concerned with it
except to recognize its existence. How does the small child know
which muscles to pull? It is the general question of acquiring
muscular skills. We must recognize a disposition to make ran
dom movements i n an experi mental spirit. Infnt babbling is an
example. And we must assume some neural mechanism, of the
nature of feedback, whereby practice makes perfct . 2
Clearly pleasure is not constant under receptual similarity, nor
under perceptual similarity. Episodes receptually or perceptually
very si milar to an episode a cannot be depended upon to be as
much fun as a. For let us not forget the internal states. They are
consequences largely of earlier episodes, and bel ong neither to
current reception nor to current perception. We have been at
pains to screen them out of any perceptual simil arity com
pari sons. But they are part still of the episode a, and can con
tribute to its pleasure. The episode may be pleasant because of
some idyllic memory that the subject was al ready entertaining at
the time, or because of some uncommonly eupeptic phase in his
digestive cycle. Later he will fex such muscles as are calcul ated
to induce a perceptually si milar episode, lnrewarding though
perceptual similarity may in this case be. 3 For that matter, he
may also manage to fex agai n those unidenti fed l ittle muscles,
perhaps laryngeal among other, that went into the entertaining of
the idyllic memory. I nsofr, he is enhancing not just perceptual
similarity but something more. (The eupeptic fature will have to
go by the board, however, since the smooth musulature i s
beyond his control . )
Summed up, here i s how action looks i n terms of perceptual
si mi
larity and the pleasure principle. The subject basks in present
i
mping
ements and puts his best foot forward. Traces of past
2. Se Holt, Chapters VI-VI I I, on the physiology of all this .
3. Herrnstein' s theory of superstition fts in here.

episodes tell hi m what to seek and what to avoid. Similarities
point his strategy, which is that of exploiting the head starts, im
proving the advantages. The i nductive method is i mplicit in that
strategy, for in effect that strategy consists in reproducing some
components of a past episode in the hope that other components
will accompany them, or in averting some components of a past
episode for fear that others may accompany them.
This hedonistic model is of the ki nd that Troland (p. 278f
approvi ngly calls "hedonism of the past" and associates with
Thorndi ke. Hedonistic theories of action have long had their dis
tinguished champions and their distinguished critics. One
counterargument is that pleasure, being mental, cannot move
muscles. Thi s point need not detain us, si nce pleasantness is for
me an attribute of bodily episodes, manifested in behavior. There
are plausible notions of its physiol ogical mechanism; thus Holt
(pp. 226, 232) pictures pleasantness and unpleasantness in terms
of harmony and confict of motor i mpul ses .
A more serious objection to hedonism is si mply that it is un
realistic as an account of responsible adult behavior. For instance
my self-i mposed activity, thinking and writing, is not fun. Very
well; responsible adult behavior is an obscure and complicated
matter. Still, when we are looking for the elements of the learn
ing process at its simplest, perceptual simil arity and the pleasure
principle afford a reasonable schema.
p
art I I
breakin
g
into lan
g
ua
g
e
9. Men ta/ism and language
The mental istic idiom is at the ti p of one's tongue. In treating
of perceptual simil arity I have stressed behavior, but the term
has strong overtones of introspection. In talki ng of the learning
process we commonly deal in mentalistic terms of i nduction and
expectation. The ani mal confronted with the seven spots may be
said to expect the lever to deliver a pellet and not a shock, and he
may be said to have reached this expectation by i nduction from
past epi sodes.
The talk of a pleasure principle was a conspicuous instance of
mentalistic idiom, however behavioral the i ntent. Si mi larly for
salience and traces. And we even found heuristic value in think
ing of activated traces as i mages (7).
Mental ism thus has its uses as a stimulant. Like other
sti mulants, it should be used with caution. Mental entities are un
objectionable if concei ved as hypothetical physical mechanisms
and posi ted with a view strictly to the systematizing of physical
phenomena. They should be posited i n the hope of their submit
ting someday to a full physical explanation in turn. Any
vaguenesses or complexities that mi ght obstruct that objective
should be mi ni mized. One must not mistake the familiarity of
mentalistic talk for clarity, and thus be tempted into a dream
world of introspecti on. Philosophers and Titchener-type psy
chologists have worried about whether one's triangle i mage is
equilateral or scalene or oscillatory among various angularities.
Some have wondered whether one's image of a speckled hen had
an odd or an even number of spots or perhaps neither, and, if
neither, how a number could be neither odd nor even. If we thi nk
of i mages aseptically as hypothetical neural states, these queer
problems do not arise. A triangle i mage or a speckled-hen i mage
is a neural state that requi res no commitment to angle ratios or
speck numbers.
Images, traditionally, were ideas. They were the least tenuous
of the ideas (apart from sensations themselves), and hence the
more insidious i n luring the unwary frther into that mi sty real m.
Berkeley and Hume were wary; they even drew the l i ne at
the abstract triangle idea, admitting i mages of the speci fic
triangular shapes only. At any rate it is the facile resort to ideas
i n general , not just i mages, that renders the mentalistic idiom dis
astrous . Questi ons of the mechanics of learning subside into idle
questions of the causal connections of ideas. An illusion of un
derstanding is created by pushing the problems back into a realm
that is too di m for their easy detection.
Our dissociation from the old epistemologists has brought
both freedom and responsibility. We gain access to the resources
of natural science ahd we accept the methodol ogical restraints of
natural science. In our account of how sci ence mi ght be acquired
we do not try to justify science by some prior and frmer
philosophy, but neither are we to maintain less than scientifc
standards. Evidence must regularly be sought i n external objects,
out where observers can joi ntly observe it. Speculation is
allowable i f recognized for what it is and conducted with a view
to the possible access of evidence at some future stage. C. S.
Peirce has well stated that "the only way of investigating a psy
chological question is by inference from
_
external facts" (5. 249).
Br
eaking into Lnguage 35
We want to know how men can have achieved the conjectures
and abstractions that go into scientifc theory. How can we pur
sue such an inquiry while talki ng of external things to the exclu
sion of ideas and concepts? There is a way: we can talk of
language. We can talk of concrete men and their concrete noises.
Ideas are as may be, but the words are out where we can see and
hear them. And scientifc theories, however speculative and
however abstract, are i n words . One and the same theory can be
expressed in di fferent words, so people say, but all can perhaps
agree that there are no theories apart from words. Or, if there
are, there is l ittle to be lost in passing over them.
In turning away from the ideas and looking to the words, we
are taking the nomi nalist strategy. Perversely, there persists an
old and stubborn tendency of the opposite kind: to appeal to the
ideas when theorizing about the words. Language, we are told,
serves to convey ideas. When we learn language we learn to
associate its words with the same ideas with which other speakers
associate them. Now how do we know that these ideas are the
same? And, so far as communication is concerned, who cares?
We have all learned to apply the word ' red' to blood, tomatoes,
ripe apples, and boi led lobsters. The associated idea, the
associated sensation, is as may be. Language bypasses the idea
and homes on the object. Than the idea there is l ittl e less useful to
the study of language.
Infant learning is a bright domain, and there behavioristic psy
chology blooms. The beginni ngs of language are learned osten
sively. The needed stimuli are right out there in front, and
mystery is at a mi ni mum. The old-time tal k of ideas, ideas
grasped and ideas conveyed, is given up without a wrench.
Subtleties and obscurities crowd in, however, when we press to
less primitive levels of language learning. The child learns to
recombine his growing vocabul ary into new sentences of hi s own
and to use them properly. Up to a point, this process in turn
stands fairly well to reason. The child learns some brief sentences
as wholes in the obvious way, by hearing them from adults in the
appropriate observable circumstances; and then he makes new
ones by analogical substitution, supplanting a component word
of an acqui red sentence by some other word of his acquired
vocabulary. Soon, though, his learning process becomes much
36
T
he Roots of Reference
harder to picture and to conjecture about. He gets to producing
sentences that bear no i mportant relation to concurrent observ
able circumstances at all . He utters sentences about the past and
the future-sentences whose only connection with present cir
cumstances is that they are tri ggered by someone's present
remark. In the fullness of time he gets to producing sentences
that bear no very clear relation even to observable circumstances
in the past or future; sentences, these, of myth or theory or con
jecture. The best of these sentences do enjoy some ki nd of con
nection still with observation, but the philosophers who devote
themselves to the logic and methodology of science have been
hard put to it to say in explicit detail just what the appropriate
connecti ons might be.
It is proverbial, or used to be, that man i n his study of nature
flls back on the old-time religion to fll in where his scientifc ex
planations leave off. It is at least equally true that man in his
study of language falls back on the old-time mentalistic seman
tics to fll in where his scientifc explanations leave off. Men
talism, supernaturalism, and other unwholesome cultures thrive
in dark places.
A healthy suspicion has been getting around that the idea idea
is not quite respectable, and for this reason it is tending to exert
its power less overtly than it did in the days of Kant, Hume, and
Locke. It is still ill concealed, certainly, under the name of
propositi on; for a propositi on, when not taken as a mere
sentence, is the idea that a sentence expresses. Happily there is an
increasing tendency to be guarded in one's tal k of propositions.
However, a philosopher who is chary of talking of ideas and
propositions is apt still to talk as blithely as a layman on the sub
ject of translation. He tends uncritically to accept the relation of
a sentence to its translation, as a relation that is i ntelligible out
right. He thinks it makes sense to ask, of just about any sentence
in any language, for an Engl ish translati on. When I try to picture
hi s uncritical acceptance of this relation, I can picture it only as
an unconscious old-fashioned acceptance of the idea idea: one
sentence is a translation of another if it expresses the same idea,
the same thought, the same meaning, the same propositi on.
Br
eaking into Language 37
What is wanted for a better understanding of the mechanics of
language and language learning is a continuing adherence to ex
ternals. Conjectures about i nternal mechanisms are laudable in
sofr as there i s hope of their being supported by neurol ogical
fndings. But the idea idea of our fathers will be only in the way.
10. Observation sentences
Midway i n the precedin section we concluded that to account
for man's mastery of scientifc theory we should see how he ac
quires theoretical language. Our reason was that we can study
words more responsibly than ideas. There is also a second
reason, having to do with the relation between a scientifc theory
and the observations that support it. For this relation has, besides
its epistemol ogical aspect, a semantical aspect . Besides being the
relation through which the sentences affrmed i n the theory gai n
their support, _it is the relati on through which they gain their
meaning. For we learn the language by relating its terms to the
observations that elicit them. Now this learning process is a
matter of fct, accessible to empi rical science. By exploring it,
science can in effect explore the evidential relation between
science itself and its supporting observations .
The reason for the basic role of observati ons, both in the sup
port of theory and in the learning of language, is their inter
subjective i mmedi acy. They are what witnesses will agree about,
on the spot. They are the common ground on which to meet when
there i s disagreement. Hence their basic role i n the support of
theory. And i n their i ntersubjective immediacy they are basic
also to language learning, because we learn the language from
other people in shared circumstances. Though we learn it largely
by learning to relate strings of words to strings of words,
somewhere there have to be nonverbal refrence points, nonver
bal ci rcumstances that can be i ntersubjectively appreciated and
associated with the appropriate utterance on the spot . Ostensive
learning i s fundamental, and requi res observabil ity. The child
and the parent must both see red when the child learns ' red' , and
one of them must see also that the other sees red at the time.
The two roles of observations, their role in the support of
theory and their role in the learning of language, are inseparable.
Observations are relevant as evidence for the support of theory
because of those very associations, between observable events
and theoretical vocabulary, whereby we learn the theoretical
vocabulary in the fi rst place. Hence, of course, the com
monplaces of the veri fication theory of meaning. The meaning of
a sentence lies i n the observations that would support or refute it.
To learn a language i s to learn the meaning of its sentences, and
hence to learn what observations to count as evidence for and
against them. The evidence relation and the semantical relation
of observation to theory are coextensive.
But the old champions of a verification theory of meaning
went wrong in speaking too bl ithely of the meani ng of individual
sentences. Most sentences do not admit separately of obser
vational evidence. Sentences interl ock. An observation may
refute some chunk of theory comprising a cluster of sentences,
and still leave us free to choose which of the component sentences
to continue to count as true and which to abandon. The evidence
rel ation is thus intricate and indirect. The same, of course, i s true
of the semantical relation. The semantical rel ation of observation
to the theoretical language is similarly intricate and indirect,
since we learn the language only partly by associating terms or
sentences directly with observation, and partly by linking them to
one another. The evidence relation, in all its intricacy, and the
semantical relation, in all its intricacy, are coextensive still .
Let us now come more nearly to grips with these matters.
What are observations? They are vi sual, auditory, tactual , olfac
tory. They are sensory, evidently, and thus subjective. Yet it was
crucial to the use of observations, both as evidence and as seman
tical reference points, that they be socially shared. Should we say
then that the observation is not the sensation after all, but the
shared environmental circumstances? No, for there i s no
presumption of i ntersubjective agreement about the environing
situation either; two men will assess it differently, partly because
of noticing di fferent features and partly because of entertaining
different theories.
Breaking into Language 39
There i s a way out of this diffi culty over the notion of observa
tion. It consists in talking neither of sensation nor of environing
situation, but of language: talki ng of language at the obser
vati onal end no less than at the theoretical end. I do not suggest
that observations themselves are something verbal , but I propose
that we drop the talk of observation and talk instead of observa
tion sentences, the sentences that are said to report observations:
sentences like 'This is red' , 'This. is a rabbit' . No matter that sen
sations are private, and no matter that men may take radically
different views of the environing situation; the observation
sentence serves nicely to pick out what witnesses can agree on.
Since I propose to dodge the problem of defi ni ng observation
by talki ng i nstead of observation sentences, I had better not
define observation sentences as sentences that report obser
vations. Nor do I need to. The requirement of intersubjective
agreement al ready affords us just the definition we need. A
sentence is observational insofr as its truth value, on any occa
sion, would be agreed to by just about any member of the speech
community witnessing the occasi on. This definition depends still
on the idea of membership in the speech community, but that
presents no problem; we can recognize membership in the speech
community by mere fuency of di alogue, something we can
witness even without knowing the language.
We might want to hedge our defini tion of observation sentence
a little, so as not to count as observation sentences those
platitudes on which all speakers will agree come what may. Such
sentences, which I call "stimulus-analytic" (2 1 ), can be
eli mi nated by requiring that an observation sentence be an occa
sion sentence, that i s, a sentence that does not command assent
or dissent once and for all, but only variably from occasion to oc
casi on. However, these are uninteresting legali sms. What is
worth noticing is that we have here a behavioral criterion of what
to count as an observation sentence. It does not appeal to sense
data or other epistemological preconceptions.
We noticed earlier, when tal ki ng vaguely still of observations
rather than observation sentences, that the ability of witnesses to
agree in their observations was crucial on two counts. It was
necessary for evidential purposes, as providing common ground
on which to meet when there is di sagreement about theory. And
it was necessary for semantical purposes, as enabling our elders
to assess the appropriateness of the occasions with which we were
associating our newly acqui red words and sentences. Such, then,
are the two crucial roles of observation sentences: evidential and
semantical . Observation sentences are sentences on which scien
tists can reach agreement when they are trying to reconcile their
theories, and they are sentences that can be socially checked
against their occasions of utterance when we are picking up a
language. Because of this semantical trait of observation
sentences it is they that are learned most readily, affording the
entering wedge in the acquisition of one's language. Observation
sentences are the gateway to language, as to science.
Typical observation sentences are ' Red' (or 'This is red' , 'I see
red' ), ' Rabbit' , ' It is raining' . Mostly they treat not of sensations
but of external things, since they admit of public veri fication.
Their di stinctive trait i s the sufficiency of present i mpi ngements.
To appreciate the sense i n which we may say that present im
pingements suffice for observation sentences, consider sentences
of the other sort: a remark about ancient Egypt or the nucleus of
the atom or the destiny of man. Now it must be granted that
these latter sentences also are triggered by present i mpi ngements:
perhaps by the sight of a page or a bas relief or a photographic
plate or someone's question. And it must be granted, conversely,
that one's readiness even to affi rm or assent to an observation
sentence-' It i s raining' , ' Thi s i s red' , 'That i s a rabbit'-is
dependent still on one's earlier training, one's rudi mentary
language learni ng. But we know the social criterion that dis
tinguishes the two sorts of sentences. If the remark about ancient
Egypt is put as a question to two fuent speakers in the presence
of impingements as nearly alike as you please, one speaker may
assent and the other may not. Si milarly for the remark about the
nucleus of the atom, or about human destiny. But you will get
like verdicts if in the presence of the same impingements you
query an observation sentence.
There has been a tendency in recent philosophy to question the
notion of observation, or of observation sentence. One complaint
i s much the same as the Gestalt psychol ogist's objection to sen-
sory atom ism ( I ); viz. , the ubiquity of unconscious i nfrence.
But observati on sentences in the present sense are not open to
that objection, for they are not about sense data.
A second complaint is that what count as observati ons for the
specialist often do not count as such for the layman. An answer
to this objection is that the notion of observation sentence i s
relative to a linguistic community. If a sentence would qual ify as
an observation sentence for the scientist and not for the layman,
it is couched i n a technical sublanguage i n which the layman is
not a fuent communicant . A better answer i s to insist on adher
ing to the broad li nguistic community for the philosophical
criterion, thus not counting the specialist' s recondite sentence as
an observation sentence in the strict sense. For the speci alist can
always be driven back to the less techni cal levels of evidence,
though in practice he cites only what is needed for reassuring hi s
trained colleagues.
A third complaint is that no datum is wholly saf from
repudiation, if it conficts with a theory that has overwhel ming
support from other quarters. Now we can accept this point as
true but not as an objecti on. Our definition of observation
sentence speaks only of concurrence of present witnesses, and
sets no bar to subsequent retractions.
The definition does raise a subtler problem, however-a
problem that was al ready noted in another connection in 6. The
defnition speaks of joint witnessing. In a more precise state
ment, it would speak of witnesses subject to receptually simi lar
impingements; and thus it would raise agai n the homol ogy ques
tion that we noted at the end of 6.
Still, the definition i s as sharp as the notions of witness and
linguistic community on which it depends. It i s good as
behavioral concepts go.
1 1 . Os tensive learning
Observation sentences are the gateway to language. We can
learn them fi rst because we have only to key them to current
episodes; there is no arriere pensee, no need of deduction or con-
jecture, no searching of memory. This is because other speakers,
whom we imitate and who encourage our own behavior, have
associated the sentences with those same ranges of concurrent
impingement. Not that each of us learns all his observation
sentences by di rect conditioning i n this way. Each of us happens
to learn many observation sentences in indi rect ways, by verbal
explanation or by verbal context or by analogical construction
from component words previously learned. Some of us master a
particular observation sentence in one way, some in another.
Still, an observation sentence is one that can be learned by di rect
conditi oning. It is within the scope of standard animal training.
Other sentences, not observational, can be partially learned by
that same di rect conditioning. The child can learn ' fi msy' , or ' I t
i s fi msy' , ostensively from conspicuous instances, and may
wrongly treat it as observational for a while, thus withholding it
from things not visibly fimsy. When fully mastered, ' fi msy' fils
the test of observational ity because of covert fimsiness. A con
trivance may be declared fi msy on sight by one observer and not
by another, according to prior experience with things of the same
make. ' Sick' i s another example: there are the patently sick and
there are those who are sick only to the trained eye.
The term ' red' , functioning as a one-word sentence, i s an
observation sentence that al ready figured conspicuously i n il
lustrations of the learning process in 5 and 8. Mentalistically
described, the child' s learning of this sentence consists in his
proceeding i nductively from observati on and experiment to a
general unspoken knowledge of the ci rcumstances in which to ex
pect adults to assent to ' red' . Descri bed in passive and less septic
terms, he i s being trained by successive rei nforcements and ex
tinctions to say ' red' on the right occasions and those only. He
exerts hi mself to enhance the overall similarity of present i m
pingements to pleasant past episodes that i ncluded the sound
' red' and to di minish similarity to unpleasant episodes that in
cluded that sound.
The child' s success in learning this and other observation
sentences depends on substantial agreement between hi s si milar
i ty standards and those of the adult . For he is antici pating the
adult's reacti ons by extrapolating from past samples according
Breaking into Language
43
to hi s own similarity standards. Clearly, therefore, the similarity
concerned here is perceptual similarity, that being fai rly uniform
over society. The child may make some wrong extrapolations
because of random i nterferences from his fux of inner states, but
the
se wrong tries will be extinguished as the conditioning goes
on.
I n 6 we considered how to i solate pure perceptual similarity,
screening out the i nterferences from withi n. Aid i s now visible
from the side of language: perceptual simil arity relates the
episodes that warrant assent to an observation sentence.
The learning of an observation sentence amounts to deter
mi ni ng, as we may say, its similarity basis. By this I mean the
disti nctive trait shared by the epi sodes appropriate to that obser
vation sentence; the shared trait in which their perceptual
similarity consists. In learning the sentence the chi ld may ap
proximate its similarity basis little by little. In learning ' red' he
has to learn that it is a question of sight, not some other sense.
He has to fi nd the proper di rection in the scene, and how much to
count: how big a patch. He has to learn what aspect of the patch
to count; he might think that what mattered in hi s fi rst red patch
was the shape and not the color. Also there is the question of
chromatic latitude: how orange can red get? Having been reward
ed for saying ' red' on one exposure, he can only conjecture what
similarity might warrant saying it again. We may expect hi m to
have to make a number of trials and eli mi nate a number of errors
before he i s shunted onto the right track for good.
Chromati c latitude is the worst of it. Si milarity by the child' s
inna
te lights ascends the spectrum in an unbroken chain, and red
dwindles only by a gradual di mi nution of rei nforcement on the
part of society. Some portions of the spectrum show steeper sub
jective gradations than others, perhaps, but there are no natural
segments, as witness the different segmentations of color
nomenclature i n different societies. '
The other matters that the child needed to learn about the
simil arity basis of the word ' red' could in principle be left to long
term inductive resoluti on, like the indeterminacy of how far up
l . Cf. Lenneberg and Roberts.
the spectrum to apply ' red' . Among the myriad fatures of
episodes of overall impingement, those fatures that are i rrele
vant to ' red' would in the long run cease to compete. Times when
the sound ' red' was rei nforced would show their common
fatures ever more clearly as their irrelevant fatures continued
to vary at random, until at last the child-mentally a child still, if
physically well along in years-would get to using the word ' red'
at just the right times. I n practice, of course, things move faster,
thanks to sal ience. No multiple inductive steps are needed to
eli mi nate from competition all fatures of the original scene out
side the relevant color patch itself, if that patch i s set off in any of
various effective ways: focally situated, brightly l ighted, garishly
colored, emphatically outli ned, moving against a background.
Or, agai n, the child would be spared the mi stake of extrapolating
on the basis of shape if on fi rst exposure the red happened to be
bright and its boundaries vague.
Salience thus expedites the learning of observation words ex
ceedi ngly. Here it is that pointing confrs its benefits. The scene
is selectively enl ivened by the conspicuous intrusion of a fi nger in
the foreground of a chosen object, or by the motion of a fi nger
outl i ni ng a chosen region.
Wittgenstein noted some perplexities of pointing. 2 How do we
know how much or what aspect of the ostended region is in
tended? How do we even recognize pointing as pointing? How do
we know that an ostensively defined term is not a term for the
pointing fi nger? How do you ostend ostensi on? Well, mi stakes do
occur. The indri, a lemur of Madagascar, owes this name to a
Malagasy expression meaning "There it goes. " The French
natural ist Pi erre Sonnerat supposed at the time that the ostend
i ng native was naming the animal . 3 Still, ostension i s brought
under control on the whole; and let us thi nk how. Pointing, we
saw, contri butes by heightening the salience of a portion of the
visual fi eld. Primitively thi s salience i s confrred on the pointing
fi nger and its i mmedi ate background and neighborhood i n
discrimi nately, through the fmiliar agency of movement and
2. Paragraphs 33-38, 454.
3. American Heritage Dictionary; reference courtesy of David K. Lewis.
Also Encyclopedia Britannica, 1911.
contrast. Even in thi s primitive effect there i s a gain: most of the
irrelevant stretches of the scene are eli mi nated from attention,
and much laborious eli mi nation by induction i s thus averted.
Some l i mited induction may remain to be done, some reinforce
ment of the verbal response in the absence of the pointing fi nger
and some extinction of it in the presence of the pointing
fnger, before the subject succeeds in eli minating the pointing
fi nger i n favor of the thing or fature for which the word is in
tended. In later cases we skip even this li mited bit of fumbli ng,
having learned the pointing custom.
12.
Assent
Pointing, we see, serves to expedite the learning of terms that
might otherwise be learned only through a long sifting of
similarities and di fferences of overall impingement patterns.
Now another device that greatly expedites such learning is as
sent . Certainly assent and di ssent must very soon supplant the
routine of saying ' red' i n a red presence and being rewarded. I f
the child were to persist i n volunteering the names of passing
colors and other observables, he would soon bore his purveyor of
rewards beyond the point of di mi ni shi ng returns. Increasingly the
child contents hi mself with answering 'yes' and ' no' when asked,
and with asking and taking ' yes' and ' no' as answers. I n the con
tinuing enterprise of ostensive learning, these two vocables are
the laconic refrai n.
I have made much of the learning of ' red' because i t was easy
to talk about, and it was for the same reason that I have dwelt on
that aspect of the learning of ' red' that consists in learning to
volunteer the word in the presence of red. If this emphasis has
given a di storted i mpression of the early phases of language
learning, the refections on assent to which we are now turning
will help to redress the balance.
To begin with, it would be wrong to suppose that learning
when to volunteer statements of f act or to assent to them is all or
most of what goes into language learning. Learning to react in
appropriate nonverbal ways to heard language i s equally i mpor-
tant. The child learns to react appropriately to many words
before being moved to volunteer them. Dogs learn to act ap
propriately on some words without learning to volunteer any.
Much of what is earliest and most urgent in language learning,
furthermore, is a matter of neither stati ng nor assenting nor
acting upon statements, but of i mportuning.
But statement learning is what i s relevant to our study, which
ai ms at understanding the acquisition of scientific theory. As for
concentrating on the child's statements rather than hi s nonverbal
responses to statements, this i s a clear matter of strategy; verbal
behavior can be more handily classified than the other, and we
can get whatever we need by probing from this side. And note by
the way that though the child learns most of language by hearing
the adults and emulating them, his very ability to i mitate is ac
counted for (if 8 is right) by the more basic mechanisms that
have been our concern.
We must by all means exami ne the learning of assent, for thi s
device i s indispensable to a child' s progress i n the art of state
ment. The child must learn how and when to assent to parents'
queries, as lately remarked, because of the parents' li mited
tolerance of childi sh initiative. Conversely, the child who has
reached the point of wanting to verify and i mprove hi s own usage
must learn to query statements for parents' assent; for the
utterances volunteered by the parents are too sporadic to meet
his needs. These are two reasons why assent is i ndi spensable, and
a fi rmer reason still can be added. For there are observati on
sentences whose proper occasions of use cannot be di scovered by
mere watchful waiting even with the best of luck; they can only
be checked by query and assent. I refer to observation sentences
with overlapping ranges. By just passively noting the epi sodes of
i mp
)
ngement i n which one's elders volunteered ' rabbit' or
' animal ' , one would have no way of maki ng sure that all the
things cited as rabbits would count also as ani mals, or whether
any of the things cited as animals might count also as rabbits. By
query and assent, on the other hand, it is the work of a moment.
It is worth noting, in passing, that the method of query and as
sent is indispensable li kewise, and for the same reasons, to the
field linguist who is breaki ng into an unknown language by in-
Br
vestigating native speakers. At first there i s little he can do but
watch and wait; but hi s progress will be slight until he has noted
and guessed enough to enable hi m to start volunteering native
words or observation sentences on his own and querying them for
assent. It is the indispensable means of tapping the reservoir of
li nguistic di spositi ons. A native sign of assent is apt to be
recognizable by this obvious partial criterion: a speaker will as
sent to a sentence, if queried, in circumstances sufficiently
similar to those in which he volunteers the sentence. (For the
meaning of ' suffi ciently' see 5 on neighborhoods. ) Fi rmer con
trol can be gained by finding also what might plausibly be taken
as a sign of di ssent; and the partial criterion for this is that a
speaker will di ssent in no circumstances sufficiently similar to
those in which he volunteers the sentence. Troubles will be com
pounded, of course, i f the sentences tried happen not to be obser
vation sentences and hence not to be li nked to concurrent cir
cumstances. But one does what one can.
Let us return then t o home ground and consider how our child
might get on to the trick of assent. The mechanism of perceptual
similarity and the pleasure principle proves to cover the case
quite wel l . One of the child' s rewarding episodes may be sup
posed to have included a conspicuous show of _red together with
the sound ' red' from his own mouth, followed by the sound 'yes'
from the parent. In a later episode there i s agai n the col or and
agai n the sound ' red' . Such i s the partial similarity of the later
episode to the earl ier. There are of course i ncidental differences,
and one of these just happens to be that the sound ' red' i ssued
from the parent this time, actually with i nterrogative intent.
Anyway the child i s moved as usual to heighten the resemblance,
so he supplies another element of the earlier episode, the sound
' es' . Rewarded again, he has learned to say 'yes' i n the presence
of the col or red and the sound ' red' . Unpleasant epi sodes will dis
courage him from saying 'yes' when he hears the sound ' red' in
the absence of the col or.
In this account we see again an advantage of thinking i n terms
of perceptual similarity and the pleasure principle rather than i n
the polar terms of sti mulus and response. It i s the same advan
tage that was seen in the treatment of i mitation (8). For note
48
that ' red' is spoken by the child and ' yes' by the parent in the fi rst
episode and vice versa in the second.
We have imagined how the child might learn to assent o the
specific observation sentence ' red' . He would learn similarly to
assent to other queried observation sentences, one by one, as
readily as to ' red' . Soon he would learn the general trick of
assenting when an observation sentence previously learned i s
queried for the first time. He would learn this by generalizing
from a few cases that he had learned separately in the way in
which we i magined him to learn to assent to ' red' . But now what
might the method or mechanism of such generalizati on be?
It is here, I thi nk, that we must appeal to the language
dependent learni ng of language: to learning that depends on
other locutions previously learned. What is learned here, in par
ticular, is an equival ence: assent to a sentence entails the same
rewards or penalties as a repetition of the sentence would entail .
This learning depends, like other learning, on an appreciation of
similarity; but this time it is a language-dependent si milarity.
The shared feature in which the similarity lies is perhaps an in
trospective sense of willingness to repeat the heard sentence; a
sense of freedom from inhibiti on. I n venturing to speculate thus
on i nward sense I rel ax my behaviorism, but not much; I speak
only of an i ncipient drive toward overt behavior.
The child' s learning of di ssent would not begin in quite the way
in which we i magined his learning of assent to have begun. That
began with hi s learning to assent to a specific sentence, ' red' . Hi s
learning of this depended on a rewarding episode in which he said
' red' and the parent said ' yes' . If we try a parallel account of dis
sent, the trouble is that an episode in which the child said ' red'
and the parent said ' no' would tend not to have been a rewardi ng
one which the child would try to recreate. If by chance the child
babbled ' no' when the parent said ' red' i n the absence of the
color, then indeed he would be rewarded and would have his
start; but this line i s too improbable. We can perhaps more easily
suppose that the child eventually achi eves a general second-order
learning of di ssent without having fi rst learned to di ssent to
various specific observation sentences independently. Mastery of
assent could be a hel pful prel i minary step to this general mastery
Br
of di ssent . Movi ng from the one to the other is a matter of com
ing to appreciate that dissent i s rewarded where assent is penal
ized and vice versa. If the shared fature underlying the general
mastery of assent was a sense of willingness to repeat the heard
sentence, a sense of freedom from inhibiti on, then the feature un
derlying the general mastery of di ssent is si mply the sense of in
hib
ition.
13. Values
Scienti fic theory stands proudly and notoriously aloof from
value judgments. Let us look briefy i nto this relati on, or want of
relati on, from the point of view of language learning. At fi rst let
us concentrate on 'good' as applied to things to eat.
Without doubt the term is learned at fi rst on a par with obser
vation terms such as ' red' . 1 The child learns perhaps to assent to
it in the presence of toffee. A l ittle subsequent trial and error
shunts him off the wrong similarity di mensions, such as
sticki ness, and settles him properly on pleasure as the feature
shared by good epi sodes. It i s a curious case of convergence of
factors. The two factors that make for learning, in general , are
perceptual simi larity and the pleasure principle; but in this case
pleasure does double duty, serving also as the similarity basi s.
The similarity basis and the rei nforcement coincide here. Under
such auspices the term 'good' must come through as a red-l etter
word indeed.
In 1 1 we noted ' fi msy' and ' sick' as terms that could be partly
learned by ostension but did not qual i fy as observation terms.
We noted that the child might even mi suse
-
such terms as obser
vational at fi rst, by di ssenting from thei r application to covert
cases i nstead of properly wi_thholding his verdict . We noted that
in
adult usage these terms fil of observational ity according to
our criterion; observers may di ffer in their verdicts to ' fimsy' or
' sick' on present evidence because of di fferent past experience.
l .
"Good and bad," Peirce wrote (5. 247), "are feel ings which frst arise as
p
redi
cates. "
50
Now the term 'good' turns out somewhat similarly: tasters give
different verdicts, di ffering as they do in what they li ke. The child
gets his good start on 'good' , as on ' fi msy' , from cases where
there is general agreement; we al/ li ke toffee. 2 Later, failures of
unanimity force hi m to have second thoughts on ' good(-tasti ng)'
as on ' fi msy' .
But these second thoughts take di fferent li nes. In the case of
' fi msy' and ' sick' there are, he fi nds, ways of resolving the dis
agreements by pressing an investigati on. These are terms of
scientific theory, and evidence for their appl icati on can be
marshalled that leads back to really observational terms. With
'good' , said of taste, the case is otherwise; the di sagreements are
a dead end. De gustibus non disputandum est.
The above refecti ons on goodness of favor clearly apply
equally to the aesthetic good generally. Wht i s perhaps less ob
vious, they apply i n considerable part to the moral good as well .
The child' s early acts of obedience are agreeably rewarded and
his early transgressi ons are spanked down. Thus we fi nd the same
convergence of f actors here as before: the simil arity basis of the
term 'good' , morally applied, i s the reward itself. Obedience
brings pleasure much as toffee does. The li kening of obediece to
toffee is indeed the very strategy of the parent' s trai ning
program.
I ntrospection would have incli ned me to expect that the 'good'
of pleasure and the 'good' of righteousness were the merest
homonyms, independently learned and calling perhaps for quite
heteronymous translations in various other languages. But not
so; thei r similarity bases are essentially one and the same, and
most stri kingly. The child can, however, easily learn a subordi
nate distinction between the aesthetic good and the moral good,
between the tasty and the virtuous. For, within the broad
simil arity that relates all reinforcements, there i s a sensible
difference between those sensory i npingements that taste or feel
good and those that herald what tastes or feel s good. The
difference grows, i n the fullness of time, through failures in the
heralding; virtue sometimes fails, I regret to say, to be rewarded.
2. W. S. Gi lbert.
Br
eaking into Language
5 1
At this point it is anybody's guess whether the child will readjust
his usage of the term 'good' , through filure of expected rein
forcements. Some children turn out one way and some, I regret
to say, another.
It i s one thing to learn the difference between right and wrong,
and another thing to suit the action to the word. The one i s a
matter of learning good Engl ish, learning to tal k; the other is a
matter of learning good manners, learning to behave. The
si milarity basis for learning the word 'good' i s reward, or i mpu
nity, and the basis of the moral training is the same, but still
the lessons learned are different. The child may go on properly
applying the word 'good' even after the training has- stopped and
the similarity basis has thus left off; for by then he has learned by
enumeration what acts are called good by the li ngui stic com
muni ty. Language confrs its benefits through conformity of
speech habits, and the individual stands to gain nothing from
private l i nguistic deviati on, however covert. But he may stand to
gain much from wicked behavior, once the sanctions are
dropped. It is remarkable how successful we often are in trai ning
the young to police themselves against their own selfish interests;
surely there is a native amenability at the bottom of it all, and it
has been favored by natural selecti on because of its survival value
for the race taken collectively. But this i s not language learning.
On the side of language learning we found that the 'good' of
aesthetics and the 'good' of ethics began together. We saw
further how the 'good' of aesthetics would subsequently di s
tinguish itself both from observation terms and from terms of
theoretical science. It di ffered from observation terms in that
people di sagree as to what tastes or looks good, and it differed
from scientific terms in that there is no di sputing about taste.
Now on these counts the moral 'good' turns out quite otherwise.
Normally and typically there is agreement i n the community as
to what to count as morally good; for morality, like language
itself, is
.
a community matter.
I thi nk that what sets morals apart from scienti fic theory i s a
s
ubs
tantive point of modern scientific theory itself: a scientific
doctrine as to the origins and basis of morality. Science sees the
moral law no longer as coeval with the cosmos, but as the work
of society. Therefore science addresses itself to the origins of the
moral law, among other things, but does not incorporate its con
tent. This divorce of science from moral values i s a sophi sticated
manifestation, refecting no significant quirk in language learn
ing.
1 4. Masses and bodies
' Red' is at once a term and a sentence. In past pages I treated
it under the head of sentences-observation sentences-along
with ' I t is raining' , because we were concerned with affirmation
and assent. When we speak of it rather as a term, we are foresee
ing its eventual use within longer sentences as a word naming or
descri bi ng objects: describing bodies or surf aces or naming a
color. Si nce my fi nal concern in this book is with objective
reference, I shall speak of ' red' and similar words mostly
hereafter as terms, observati on terms, rather than as sentences .
Much of what I said of ' red' in foregoing pages could be said
equally of proper names of bodies, e. g. ' Mama' , ' Fido' , ' Jumbo' .
For such names to quali fy as observation terms, under my defini
tion, we must indeed narrow the linguistic community to those
who know these individuals by name. Let us do so, since such
names belong with observation terms for purposes of the theory
of learning.
Much of what I said of ' red' could be said also of mass terms
such as 'sugar' and ' water' . The signi fcant thing about a mass
term is that it i s closed with respect to aggregati on. Two squares
do not together constitute a square, nor two apples an apple; but
when you add sugar to sugar the total is still sugar. On this score,
color words li ke ' red' behave li ke mass terms. From the point of
view of learning, moreover, these are all substantially alike, and
like ' Mama' and ' Fido' as well . They all are observation terms,
capable of being learned ostensively, and in each case to have
learned the term is to have learned when to assent to it or di ssent
from it as an occasion sentence. Mama and water are very un
li ke, but to learn either term is to learn by i nduction the ap
propriate similarity basi s. Mama, water, and red are recog
nizably recurring presences. Mama di ffers from water and red i n
Br
eak
ing into Language 53
being, for al l her sporadic comi ngs and goings, spatiotemporally
conti nuous; but this is a sophisticated matter of physical theory
with
li ttle bearing on the learning of observation terms.
Between Mama on the one hand and water or sugar or red on
the other there are indeed differences, significant even at the
level of ostensive learning. Sugar can be present in simultaneous
separate porti ons; so can water; and so can red. Mama, on the
other hand, when present at all, is visibly continuous, unless par
tially ecli psed by intervening bodies . And these casual ecli pses
offer li ttle threat to Mama's i ntegrity as a single Gestalt; for they
are feeting and they are independent of Mama' s movements,
responding rather with parallax to the movements of the
observer.
Another di fference between Mama and the masses is that the
masses are amorphous. The similarity basis of the mass term
' sugar' , or ' water' , or ' red' , has nothing to do with shape. On the
other hand the similarity that l i nks the various presentati ons of
Mama i s very much a matter of shape. Her visual shapes are
many in the course of her various orientations and contortions,
but they are joi ned by observed conti nuity of deformati on. We do
not keep Mama under observati on, but we do watch her often
enough as she changes visual shape i n our field. The sim
'
rity
that li nks the many presentati ons of Mama for us under the one
name is not just static visual similarity, but a si milarity that
depends also on continuity of deformation and di splacement. We
may next recognize Mama because our new glimpse i s visually
similar to some earlier view that observably evolved from
attested views by conti nuous deformation. Chains of such links
are what hold Mama together.
I suggested that ostensive gestures do thei r work by heighten
ing sal ience. A sweeping gesture of ostension can serve to
heighten Mama' s sal ience by fol l owi ng her through a
characteri stic brief episode of continuous deformati on. Dynamic
gestures serve nicely to differentiate the use of two terms by
di recting the sal ience differently through time. Ostension accom
panied by utterance of ' Mama' continues to follow Mama after
she casts off her red shawl; ostension accompanied by utterance
of 'red' takes the other turning, followi ng the shawl .
54
The si milarity basis of ' Mama' was rather a long story. Each
view of her is a continuous patch? Not quite; there was the little
matter of ecli pse and parallax. Each view of her is similarly
shaped? No indeed, but there is the continuity of deformati on.
Thi s rather tortuous sort of si milarity is the unifying princi ple
not only of Mama but of Fido and indeed of bodies generally.
For all its tortuousness, it is apparently a sort of similarity that
we are innately predisposed to appreciate. The well-known
Gestalt effect i s basic: the readi ness to recognize a simple and
uni fed figure, ignoring interruptions of outl i ne. A similar read
iness to recognize the persistence of an object in uniform motion,
despite temporal i nterrupti on, is reported in early i nfancy: the
baby will see an object pass behind a screen and show surprise
when it does not duly emerge at the other side. 1 It is no wonder
that bodies, bodily identity and bodily persistence, are the
mai nstay of ontol ogy. Bodies, for the common man, are basically
what there are; and even for the esoteric ontologist bodies are the
point of departure. Man is a body-minded animal, among body
mi nded animals. Man and other animals are body-mi nded by
natural selecti on; for body-mi ndedness has evident survival value
i n town and jungle.
I see little point, now or later, i n trying to make the notion of
body precise. Bodies are things like Mama and Fido and other
animals, also apples, cups, chairs. The considerations that
associate the presentati ons of Mama are what make for a body at
thi s stage of our speculati ons: there is her synchronic visi ble con
tinuity, i nterrupted only casually by partial ecli pses, and there is
her diachronic continuity of displacement and deformati on.
When the time comes for the precision of physical science, the
notion of body can give way to the more i nclusive, more recon
dite, and more preci se notion of physical object . Any arbitrary
congeries of particle-stages, however spatiotemporally gerry
mandered or disperse, can count as a physical object
. More of
thi s in 23, 34.
I
. See Bower.
Bre
aking into Language
55
1 5 . Individuation of bodies
Mama i s a body; red and water are not. This is no small
difference. Still, all three terms share a certain semantic simplic
ity when contrasted with terms like 'dog' . We have learned those
three terms insofar as we are able merely to tell whether Mama,
red, and water are present, or salient i n the scene. To learn 'dog'
we have to learn more than presence. We have to learn also the
indivi duative force of the term, the division of reference. We have
to learn what to count as one dog and what to count as another.
In one way, ' dog' i s like ' red' and ' water' and unlike ' Mama' .
Red and water can come in simultaneous scattered porti ons.
Si milarly we can be confronted by many dogs at once. Mama is
radically different; she comes at any one time as a single Gestalt,
even if subject fl eetingly to partial ecl i pses. But on thi s score
'dog' resembles also ' Mama' in a way: each dog comes at any one
time as a si ngle Gestal t. Also each dog resembles Mama 'i n what
counts for its uni fication over time; namely, continuity of dis
placement, continuity of visual distorti on, continuity of dis
colorati on. Each dog, like Mama, is a body.
We refected that these body-unifying considerations, though
com pl ex, are rooted in i nstinct. Often they are enough to assure
the right extrapolati on from a si ngle ostension. But not always.
We saw how Mama and her red shawl could part company, and
thereupon a continuing ostension could take either of two courses
according as ' Mama' or ' red' was the word to be explai ned.
Likewise in the case of the dogs, then, let us be prepared for
dynamic ostensions of some durati on.
Pleonasms are helpful in prolonging the ostension: we accom
pany the continuous gesture with not just the word 'dog' ,
however slowly spoken, but with the pleonastic elaborati on 'This
is a dog' . A more extravagant pleonasm may come into play for
purposes of further prol ongation: we may say ' This is the same
dog as this' .
The great difference between the ostensive learning of a name
O mass term like ' Mama' or ' red' or ' water' and the ostensive
learni ng of a general term like 'dog' is that the latter must go
with the grain. The sustained dynamic pointing that accompanies
the word ' dog' , or the words 'This is a dog' or a still longer
pleonasm, must not jump dogs; and it may be protracted and
repeated as necessary. The si milarity basis of such a term is
rather sophisticated: a second-order similarity, as it were, of
similarities.
We had to do in 1 2 with language-dependent language learn
ing: learning that depends on other locutions previously learned.
The present matter is different: a similarity of si milarities .
Al ready in learning the name ' Fido' the child depended on the
similarity of one presentation of Fido to another, and of one
phase of a sustained presentation of Fido to another. In learning
the general term ' dog' he has to appreciate a second-order
simi larity between the similarity basis of ' Fido' and the similarity
bases determi ning other enduring dogs. These various canine
simi larity complexes are more simi lar to one another, in the
successful child' s eye, than to the various further similarity com
plexes that determine the various rabbits, the various apples, the
various buckles; and thus it is that the child is to master each of
these general terms in the fullness of ti me, and to keep them
properly clear of one another.
Thanks to his insti nctive body-mindedness, he is an apt pupil
when the general terms are terms for bodies. He is able to ap
preciate not only that the second-order similarity of a dog to a
dog exceeds that of a dog to a rabbit, but also that the latter in
turn exceeds that of a dog to an apple or buckle. Thus it is that he
can learn not only the general terms 'dog' and ' rabbit' but also
the more general term ' ani mal' , which covers the dogs and the
rabbits but not the apples or buckles. And then there is the yet
sl ighter degree of second-order si milarity, residi ng in just those
very general body-uni fying considerations that preserve the iden
tity of each dog, each rabbit, each apple, each buckle, in short
each body. This would be a second-order similarity basis for the
child' s ostensive learning of the general term ' body' itself, or
' thing' , to take the li kel ier word.
Cutting across the dichotomy between mass terms and general
terms, there is another important dichotomy: that between ab
solute and rel ative terms. Among the relative general terms a
good observational example is ' smaller than' ; thus ' Fido i s
smaller than Jumbo' . Another example, not observational, i s
' mother of ' . But also, what i s seldom noted, there are relative
mass terms; thus 'darker than' . Whereas ' mother of ' or ' smaller
than' relates bounded wholes, 'darker than' can relate mere
points or undemarcated neighborhoods of ostension.
Like ' water' , ' sugar', and ' red' , the rel ative term 'darker than'
can be taught by simple punctual ostensions. It differs only in
taki ng the points two by two: ' Here is darker than here' . This is
why I call it a relative mass term; it stands to the absolute mass
terms si mply as two stands to one. On the other hand the osten
sive teaching of a relative general term, such as ' smaller than' ,
calls for a pai r of sweeping ostensions, adequate to suggesting the
bounds of the objects concerned: 'This is smaller than this' . More
restrained gestures will suffice only if the regions concerned are
already salient through other causes.
' Dark' passes for an absolute mass term, and ' small' for an ab
solute general term, but both of them make strict sense only as
relative: 'darker than' , ' smaller than' . On this score they are
typical of many terms. Much the same can be said even of ' red'
and other color terms; red is a matter of degree. In ' redder than'
we have a rel ative mass term, on a par with 'darker than' .
Some relative mass terms are intimately related to the in
dividuation of absolute general terms; thus ' same dog as' . We
point twice and say ' Here is the same dog as here' . I noted this
idi om on an earlier page as a mere temporizing device to prolong
an ostension; but we see now that it qualifies also as a relative
mass observation term in its own right. This reduction of an ab
solute term 'dog' to a relative term ' same dog as' is an idea that I
owe to Geach, 1 except that he does not distinguish between
rel ative mass and relative general terms.
The relative mass terms 'darker than' and ' redder than' are
transitive and asymmetri cal . The rel ative mass term ' same dog
as' , on the other hand, is transitive and symmetrical ; in a word,
e
q
uivalential. It is these equivalential ones, as Geach observes,
that provide individuation for absolute general terms.
l . "Ontological relativity and relative identity. "
There is a strict formal l i mitation to the sort of term that can
be individuated i n this way. The approach fils if the term i s to
denote things that overlap. For consider why the relative mass
term ' same dog as' suffices to individuate the dogs: each dog con
sists of just the points that are on the same dog as some one
point. This happy circumstance depends on the fct that no dogs
have points in common. A rel ative mass term ' same circle as'
would not suffice to individuate circles, in the sense say of visibly
outlined circular regions. Some of these circles may overlap, or
lie one withi n another; and then it will not be true that each circle
consists of all and only the points that are in the same circle as
some one point. This characterization would mi ss an embedded
circle altogether, and it would wrongly accord the title of circle to
an oblate region built up of two overlapping ci rcles .
The term 'circle' in the intended sense can indeed be taught by
ostension, but what is wanted is a series of sweeping ostensions,
each of which traces the outl ine and sweeps out the interior of
one of the circular regions. Pairwise ostension of points is not to
the purpose here, nor is a rel ative mass term 'same circle as' .
Absolute general terms of the type of ' dog' and ' apple' , which
can be taught by pairwise ostension of poi nts, constitute an im
portant subclass. They are probably the ones that are learned
fi rst . We gain an i mportant insight into those terms, and into the
i nception of the identity concept itself, by looking to the underly
ing relative mass term. But afterward we must recognize the ad
vent of further absol ute general terms whose individuation calls
for gestures more extravagant than pairwise pointing. Note also
that the schemati sm of an underlying relative mass term admits
of no evident extension that would implement the learning of
general terms of a rel ative ki nd, such as ' smaller than' . In that
example sweeping ostension was of the essence.
We have seemed to di scern the inception of the identity
predicate. It is a mere shared fragment of various relative ms
observation terms, e. g. ' same dog as' ; or, merer still, it is a tem
porizing vocable for prolonging an ostension. Such i s the humble
beginning of a predicate whose sense or util ity has been pondered
by Aristotle, Lei bniz, Locke, Hume, Frege, Wittgenstei n, and
59
Geach. For Locke and Hume identity made no sense except as
appl ied to earlier and later presentations of the same body. Our
present
.
refections on the primitive role of identity, in the in
dividuation of bodies, ft quite well with that view. But this role is
one that identity later outgrows.
Years earlier Geach argued that identity makes sense only
relative to one or another general term, as in ' same dog' . 2 We
have been seei ng that this i s decidedly true of the identity idiom
at its i nception. It remains true, moreover, as long as the siqes of
the identity sentence are demonstrative pronouns. There is no
sense in pointing twice and saying 'This is the same (one) as that' ;
we can only ask ' Same what?' One may have pointed at the same
dog and a different ear. When on the other hand we reach the
sophi sticated stage of making statements of identity with names
on either side, or descriptions, or variables, then evidently we can
affi rm identities without relativizing them. We can say outright
that a is identical with b; whether a is the same dog as b, or the
same ear as b, will depend on whether a is a dog or an ear. When
the identity idiom comes of age, Geach' s old relativism evidently
goes untenable. Yet there is a subtle sense in which identity is
relative stil l . I shall take it up in 30.
16. Observational compounds
Generations of lingui sts have rightly stressed that the distinc
tive
trait of language, not shared by the signal systems of lower
animals, is its productivity of combinations. We learn modes of
composition as well as words, and are thus prepared to produce
and to respond to complex expressions never heard before. One
such mode of composition i s what grammarians call attributve
composition, as in ' yel low paper' . We can imagine learning it as
fllows. The components 'yellow' and ' paper' have separately
been learned in the by now fmil iar way. Hearing these words
enlivens traces of episodes in which yellow was salient, and
episodes in which paper was salient; and these traces enhance
2. Reference and Generalty, pp. 39f.
60 The. Roots of Reference
the sal ience of any yellow or paper in the present scene. Sali ence
of yellow paper is thus enhanced doubly. All our mentor has to
do to perfect our training in the compound 'yellow paper' is dis
courage assent in those less stri ki ng cases where the yellow and
the paper are separate.
We noted (7) that there could be heuristic value in thi nking of
enlivened traces as mental i mages. Attributive composition af
fords another good case. The words ' yellow' and ' paper' induce
the two i mages, and heard together they induce them superim
posed.
Other attributive compounds may be acqui red individually in
parallel fashion: ' red wine' , ' red ball' , ' white rabbit' , ' sore
thumb' . But the child needs to learn something more general : the
art of making new attributive compounds on his own and prop
erly responding to new ones. In learning this, as in learning as
sent and dissent (1 2), he must be guided by a language
dependent brand of similarity. The occasions for assenting to at
tributive compounds are similar to one another in that they share
the following complex trait: always the two component terms
heighten the salience of some one part of the present scene. This
account not only depends on our present reasonable doctrine that
a learned word has power to enhance the sali ence of an ap
propriate part of a current episode; it also assumes that the
overlapping of two such verbal heightenings would itsel f be
noticed and used as a point of similarity.
Attributive compounds like ' yellow pager' , formed of observa
tion terms, are clearly observation terms in turn. We could learn
'yellow paper' as a whole by induction from ostensions in the
fmiliar way, as if it were a si ngle word. Attributive composition
i s just a device for the i nstant production of observati on terms .
Observation terms may, as remarked, be viewed as observa
tion sentences, which are a species of occasion sentences (cf. 1 0)'.
' Red' , ' Mama' , and the rest, seen as sentences, may also be
phrased pleonastically in the manner ' Here is red' , ' Red i s here' ,
' Here is Mama' . Si milarly for the compound observation term
' yell ow paper' : it does the work of the occasion sentence ' Here is
yellow paper' , ' This i s yellow paper' , indeed even 'This paper is
yellow' , 'The paper is yellow' . When rephrased i n such ways, the
6 1
little device of attri butive composition gives our language rather
a di scursive ai r. Hitherto there had been only the abrupt style of
a footman or a bird watcher, announcing or confirming arrivals.
There is something less ostentatiously ostensive in the phrasing
'The paper i s yellow' . Indeed no ostension is l i kely to be needed
for the occasion sentence ' yellow paper' , however phrased; for
the words themselves, al ready fami liar, heighten the salience of
the yellow paper sufficiently without further help from a pointing
fi nger . The sentence 'The paper is yellow' remai ns an observation
sentence and occasion sentence for all that, when used as here in
tended. But it has the explicit form of a predication too.
In the terminology of grammar, 'yellow' is said to occur at
ributively in 'yellow paper' and predicatively in 'The paper is
yell ow' . But it seems that this distinction, like the distinction
between observation term and observation sentence, has little
bearing on the fundamentals of language learning.
Attributive composition affords access to a rich vein of
predicati ons. Thus take ' Mama i s smiling' . ' Smil ing' is an obser
vation term that can be learned on a par with ' red' , ' water' ,
' Mama' . We learn to apply or assent to the term when something
is sal iently smiling. Then ' Mama is smil ing' , or ' smiling Mama' ,
comes through by attributive composition on a par with 'yellow
paper' , thanks to the i ntersecting of the pertinent saliences; the
smiling occurs on Mama.
Attributive composition i s one of many dyadic constructions
on terms. Another is ' i n' . Thus suppose we have learned 'the gar
den' , at fi rst not as a compound but simply as an individual name
for our own garden, learned on a par with ' red' and ' Mama' .
Then we learn to affrm or assent to ' Mama in the garden' , or
' Mama is i n the garden' , just when the respective regions that are
rendered salient by these two terms are combined in a certain
pattern: the one surrounded by, or embedded in, the other. We
generalize from various examples a i n tr and end up by
associating ' i n' with this manner of embedment. 1 The learning of
l . My convention of Greek letters and quasi quotation is explained in
Mathematical logic, 6. But perhaps it is already clear enough from the con
text.
a in /, is thus just like the learning of attributive composition,
af, , except that there it was a matter merely of a partial coin
cidence or overlapping of the two relevant masses, while now it
is a matter more specifcally of an embedding of the frst in the
second.
Here again, and I suppose in all learning of modes of composi
tion of expressions, we use a language-dependent kind of
simi larity. The occasions for the ' in' compound resemble one
another by virtue only of all having this complex trait: terms are
uttered whose associated portions of the scene are embedded the
one in the other.
One mode of composition deserves notice because of its
peculiar i mmedi acy. It turns one observation term into another
by adding the suffx '-l i ke' . Having learned 'dog-l i ke' , 'tree-l i ke' ,
and a fw other such cases as simpl e observation terms, we l earn
to form further such terms on our own; and the language
dependent simi larity this time is curiously simple. What the
propr occasions for assenting to r a-like1 have in common is a
percepti ble but insuffcient i mpulse to assent to a itsel f.
We are sti l l dealing in observation terms and occasion
sentences. We can still view them indifferently as terms or as
sentences, unless they occur embedded in longer terms or
sentences. There is no need to distinguish between the observa
tion term ' Mama in the garden' and the occasion sentence
' Mama is in the garden' , nor between 'yel low paper' and ' Yellow
paper is here' or 'The paper is yel low' , as l ong as the expression
in q'uestion is not part of a longer one. But we do best not to
equate the term 'yellow' or ' paper' to the sentence ' Yellow is
here' or ' Paper is here' when the term stands in the broader con
text 'yellow paper' . While sti l l equating the term 'yel low paper'
to ' Yel l ow paper is here' , we cannot naturally regard this
sentence as a compound of 'Yellow i s here' and ' Paper i s here' .
Certainly it is not a truth function of them. Yellow and paper
may both be present, in a sweepingly ostended region, and yel low
paper sti l l be absent.
63
17 . Predication and the categorical
Observation sentences are occasion sentences, as opposed to
)
standing sentences. Assent to an occasion sentence has to be
prompted anew, whenever the sentence is queried, by what is
currently observable. The sentence simply has no truth value
apart from the occasion. A standing sentence, once assented to,
remains as a standi ng commitment for a while at l east . Among
the standing sentences there are, at the extreme, the eternal
sentences . Their truth val ues are fxed for good, regardless of
speaker and occasion-though speakers may sti l l di sagree about
them, through error, or change their minds .
An eternal sentence may be general in i mport, or it may report
a speci fc local event. In the latter case it will gain its specifcity
through explicit use of names, dates, or addresses. The eternal
sentences most characteristic of scienti fc theory are of course
general .
Now what does language learning amount to in the case of
eternal sentences? In the case of occasion sentences i t amounts to
l earni ng what occasions warrant assent to the sentences, or dis
sent. But this opportunity for a relevant continuing check against
current episodes of sensory i mpingements lapses when we turn to
eternal sentences. We are disposed to assent to ' Dogs are
animals' or ' Water is liquid' whenever asked, and this disposition
leaves us no scope for semantic distinctions on the score of oc
casions. Eternal sentences shed the passing occasi ons. And sim
ply being disposed to assent to an eternal sentence once and for
all, or to dissent once and for all, surely cannot be regarded as
summing up our semantic understanding of such sentences.
Assent and dissent, assertion and denial, do not exhaust the
uses of sentences, not even of decl arative sentences. Even an oc
casion sentence affirmed on a present occasion will be affrmed
for an ulterior motive: perhaps for the expectation that it may
engender of some future event, through systematic connections
that are embodied i n some rude or refned body of theory. Eter
nal sentences often play useful parts without being asserted or
denied, bel ieved or disbel ieved, at any point. They may do their
work as intermedi ate l i nks i n a theoretical devel opment, e. g. as
64
case headi ngs: as alternative hypotheses in some dilemmatic
argument by cases. Eternal sentences owe their util ity to their
systematic participation in theory, rude or refned, and occasi on
sentences owe much of their utili ty to the same. But what makes
the occasion sentences less puzzling semantically than the eternal
sentences is that we do not need to trace their systematic connec
tions with theory in order to probe their meaning; they are at the
peri phery, where their meaning can be empi rically assessed by
checki ng them for assent and dissent, occasion by occasi on. It i s
through thi s periphery that science and language i mbibe all em
pi rical content or meaning. To trace out the meaning of an eter
nal sentence deep inside the theory, on the other hand, we have
nothi ng to go on but its multifrious connections within the
theory and ultimately, indi rectly, with the periphery. Each of
these strands being descri bable only by its interrelations with
others, there ceases to be any clear sense in asking the meaning of
a single such sentence at all . The sentence can be paraphrased in
terms of others of its kind, and perhaps a substantial fbric of
such sentences can be given some joi nt explanation in terms of
thei r combined net bearing on observations and occasion
sentences .
This predicament of the semantics of eternal sentences is one
that we have largely and long fai led to appreciate, because of an
uncritical retreat into mental ism. I f ways are to be found of cop
ing with it and developing a scientific semantics, the most
hopeful approach would seem to be through a retracing, real or
hypothetical , of the process by which we acqui re out command of
such language in the first place.
Consider, then, an exampl e: ' Snow is white' . At the end of 7
we noted that when a word has been learned and then is heard
again, it enlivens the trace of the learning episode. I ntui tively
speaki ng, the word 'snow' induces a snow i mage. This way of
putting the matter is good, I urged, in its suggestion that the
enl ivened trace can have effects somewhat l i ke those of ocular
sti mulation. Now such an assumption of l ikeness of effect stands
us in good stead i n expl ai ni ng the learning of ' Snow is white' . The
child has previously l earned the observation terms 'snow' and
'white' ; that is, he has learned on what occasions to assent to the
65
query ' Snow?' and to the query 'White?' . Now the parent queries
the eternal sentence ' Snow is white' . Its frst word induces the
snow i mage, and then the queried sequel 'white?' el icits the
child's assent as if he were seeing snow and not just having the
i mage.
The mechani sm that I am suggesting is, in the fmiliar ter
mi nology, a transfr of conditioning. The child has been con
ditioned to assent to the query 'White?' when snow is presented,
and then this response becomes transferred from the snow
sti mulus to the associated verbal sti mulus, the word 'snow' . My
account i s an account of such transfer in terms of traces and
sal ience.
I n learning to understand and use the observation sentences,
we depended very di rectly on truth-value considerations; for this
l earning consisted simply in learning the circumstances in which
to assent to or di ssent from the sentences. Comi ng to eternal
sentences, we noticed with some misgivings that this approach
was no longer suitable, because of the fxity of the truth val ues.
But we see now that the variability of truth value has withdrawn
merely to a higher level of abstraction. A predication may be
saddled with one truth value for all eternity; the predicational
mode of composition, however, takes on varying truth val ues,
yielding truth for some pai rs of terms and flsity for others. First
and last, in learning l anguage, we are learning how to distribute
truth values. I am with Davidson here; we are learning truth con
ditions. In learning an occasion sentence, we learn in what cir
cumstances to count it true and in what circumstances flse. In
learning the eternal predicati onal construction, we are learning
how to judge whether a given pair of terms produces a true
predication, true for good, or a flse one, flse for good.
The predication ' Snow is white' i s the simplest sort of standing
sentence, joi ning as it does two mass terms. The next to simplest
sort i s i l l ustrated perhaps by the predication ' Fido is a dog' . Here
the subject, a singular term, is on a par sti l l with the mass terms
so fr as our theory of learni ng is concerned; but the predicate
now is a general term, 'dog' , i nvolving individuation. Still, ' Fido
is a dog' could be learned in much the same way as ' Snow is
white' . The child has learned to assent to ' A dog?' in the presence
of Fido and his i l k, and then by transfr he assents to 'A dog?' on
hearing ' Fido' . He thus assents to ' Fido is a dog?' .
The case is l ess simple than that of 'Snow is white' . That case
hi nged on the chi ld's having learned to assent to ' White?' when
we point at snow. Now the child has also quite properly learned
to assent to 'A dog?' when we point merely at Fido's head; but
whereas any of the snow counts as white, the head is not to count
as a dog. To account for the l earning of ' Fido is a dog' we must
look back rather to the dynamic ostension, suitably extended,
that suffced to distinguish whole dogs from heads. That was the
basis of the child's learning of 'dog' , and of ' Fido' . The sensed or
surmised presence of Fido in all his wholeness, then, or of Fido's
ilk, is what suffces to prompt assent to the query ' A dog?' . Final
l y, by transfer as before, the chi ld assents to ' A dog?' on hearing
the word ' Fido' . The word has induced an i mage of the i ntegral
animal .
Relative general terms, l i ke absolute ones, adjoin singular
terms in predication to form standing sentences; thus ' Fido is
smal ler than Jumbo' . Having learned the three component terms
ostensively, the child could l earn this standing sentence in much
the same way as ' Fido is a dog' .
The next more complicated sort of predication is ill ustrated
perhaps by 'A dog is an animal '-really a universal categorical ,
Every a i s a {, . Here the subject as wel l as the predicate is a
general term. Still, much the same l earning pattern suggests itself
here as in the preceding cases . Having learned the term ' animal ' ,
the child is disposed to assent to the query ' An animal?' if he sur
mi ses the presence of dogs or other animal s. Then, by transfr, he
comes to assent to 'An animal?' on hearing the words 'A dog' .
He assents to ' A dog is an animal?'
The child not only l earns speci fic eternal sentences 'A dog is an
ani mal ' , ' Snow is white' , ' Fido is a dog' , but also, by generalizing
from them, he learns the modes of com position that they il
l ustrate; for he will make further such compounds on his own. He
comes to sense si mi larity among the ways of coming to ap
preciate that a dog is an ani mal , that snow is white, and that Fido
is a dog. What he senses, as common to each such pair of terms,
67
is perhaps a tendency in the sound of the fi rst term 10 di spose him
to assent to the second. Again I tread dangerous ground in
speculating on i nward sense. And agair I plead in extenuation
that it is only a question of awareness of inci pi ent drives toward
specifc overt behavi or. '
Such singular sentences as ' Fido is a dog' and ' Snow is white'
must be di stinguished, as eternal, from other singular sentences
such as ' Fido is wet' which are occasion sentences. The sentence
' Fido is wet' resembles the standi ng predication ' Fido is a dog'
grammatically, but psychol ogically it bel ongs rather with the at
tri butive compounds such as ' red ball' and ' white rabbit' and, for
that matter, ' Mama is in the garden' . We learn ' Fido' and ' wet'
by di rect exposure on a par with 'white' and ' rabbit' , and then by
what I described as a convergence of i mages we learn when to as
sent to ' wet Fido' , or ' Fido is wet' . It is remarkable what a
semantic gulf is bri dged by the simple form of predication ' Fido
is a dog' , ' Fido is wet' . The Spaniards, with thei r two copulas ' es'
and ' esta' , are more sensitive to it than we.
'Snow is white' contrasts sharply with 'The snow is white' , for
the latter is an observation sentence like 'The paper is yel low' , a
mere variant of the attributive com pound ' white snow' . We
should picture the child as somehow learning these predicative
variants of attri butive composition only well after 'Snow is white'
and other eternal predications; otherwise the proposed way of
learning the eternal ones would surely fil. 2
18 . Serendipity
We have welcomed the idea that the sound of a word can have
somewhat the same effect as the sight of its object. This has
enabled us to account for the l earning of standing predications
and of universal categoricals. It has enabled us to bridge the gulf
l . Piaget (Chaps. 4, 7) has investigated the child' s mastery of class inclusion
by setting him to sorting objects. There should be a signifcant relation between
this turning point and the mastery of the universal categorical idiom.
2. I am indebted here to Lawrence Powers.
between occasion sentences and standing sentences, putting us
well on our way from observati on sentence to scientific theory.
But there is a sordid side. What we have here is of the essence
of the notorious confusion of sign and object, or of use and men
tion. Evidently this confusion, with all its evi ls, is integral to the
very woof and (yea) warp of the fbric of science.
Language is rooted i n what a good scientific language eschews .
There is not just the confusion of use and mention; there is the
whole matter of indicator words. Our first words as I represented
them were observation terms or occasion sentences, and these are
of the essence of indicator language; the occasion sentence ' red'
says ' Here now red' . There is no such thing as a standing
sentence until we make our fltering way to ' A dog is an animal'
and ' Snow is white' and ' Fido is a dog' , and this we manage only
by dint of that hanky-panky over use and mention. Language is
conceived in si n and science is its redempti on.
Evolution by natural selection i s a history of the survival of
happy accidents. Man' s ascent to l anguage and to science i s i n
ki nd. Truth can issue from fal l acious proof. To judge the out
come by its fl lacious origins is to add the genetic fallacy to what
had gone before. Let us rather count our blessi ngs.
Proverbially, what di stinguishes language from its subhuman
antecedents is its productivity of new combi nations. But there is
another distinguishing fature that is nearly as fundamental, and
it is the standing sentence. The signal systems of ani mals are
l i mited to simple occasion sentences; and such also are the
human sentences on which a dog l earns to act.
Serving as it does as the medi um of science and hi story, the
standing sentence-indeed the eternal sentence-must be ac
counted useful . It confers one conspicuous benefit straightaway
in the domain of its origin, moreover, as an aid to ostension
itself. Universal categoricals and standing predications serve ad
mi rably in speeding up the ostensive learning of new terms . Our
teacher ostensively introduces Jumbo, gesturing and saying
' [ This is] Jumbo', and then, i nstead of persisting in dynamic
gestures to point the way to the intended diachronic ex
trapolations, he finishes the lesson in short order by saying ' Jum
bo is an animal ' . Thanks to our pri or acquisition of the term
' animal' with its buil t-in style of identifying individual animals
69
through time, and thanks to our general acquisition of the ' is' of
standing predication, we can manage all the extrapolation of
' Jumbo' ourselves on the basis of the initial ostension together
with the guiding admonition 'Jumbo is an animal' .
Sometimes this expedient saves us nothing. We might
successfully introduce Jumbo by a si ngle ostension anyway, say
ing just ' Jumbo'-thanks to his sal ience and his vivid ani mality.
But the expedient can help when a nicely salient presentation i s
hard to arrange.
The pair of sentences 'This is Jumbo' and ' Jumbo is an animal'
admits of a fmiliar apocopation: 'This animal is Jumbo' .
The device helps not only when the object to be named is short
on sali ence, but when it and other sal i ent objects compete. Thus,
to use an example I have used elsewhere, an ostension accom
panied by the words 'This is Nadejda' could be misinterpreted
as defining the color or the material of the woman's garment,
pending further ostensions; whereas the words 'This woman
is Nadejda' enable a single ostension to suffice.
The device is equally helpful i n i ntroducing new general terms.
When we point to the i nconspicuous form in the tree and say
'This is a marmoset' , we have just begun; more l essons may be
called for. But the added words 'A marmoset is an animal' would
expedite matters; or, apocopating, 'This animal is a marmoset' .
Some further ostension may, however, still be wanted to settle
the allowable range of variation from one marmoset to another;
a general term such as ' marmoset' di ffers from a singular term
such as ' Jumbo' in having this additional degree of freedom.
In these moves one exploits the individuative force of some
more general term, already learned, to obviate the separate
teaching of an i ndividuative principle for the subsumed term that
one is concerned to introduce. The term ' body' , because of its ex
treme general ity, can be especi ally useful in this connection. We
may point once and say 'This body is a buckle' , or 'This body is
an apple', or even 'This body is an animal ' . This single ostension
is l i kely to have settled the new term ' buckle' or 'apple' or
' animal' once and for all .
The single ostension will in some cases demand care, however,
and a certain elaborateness. If one were to point to Jumbo and
say 'This body is an animal' , one would risk the mi stake of call-
ing Jumbo's trunk an animal; for his trunk is indeed a body in its
own right, and one may in pointing out Jumbo have poi nted to
his trunk. But this mi sadventure can be avoided by a sweeping
gesture. What is wanted in gestural support of the words 'This
body is an animal' i s ostension that is sufficiently sweeping to
enhance the salience of the elephant as a whole and not just that
of his trunk.
I t might seem that people could have maximized this access of
effciency by learning 'body' as the frst general term and then
letting all the more special terms for bodies accrete parasitically:
'This body is Mama' , 'This body is Fido' , 'This body is an
ani mal ' , 'This body is a buckle' , and so on. However, there is at
least one good reason why this has not happened: it presupposes
a general mastery of the 'is' of standing predication. Such
mastery can be acqui red, it would seem, only by abstraction from
a stock of separately learned examples such as ' Mama is an
ani mal ' , ' Mama is a body' , ' An animal is a body' , ' An apple is a
body' , ' An elephant is an animal' . Thus, while we may exploit the
term ' body' or indeed ' ani mal ' to expedite the acquisition of
further terms such as ' marmoset' , ' Jumbo' , ' Nadejda' , it was
necessary frst to have acqui red a representative stock of such
terms the hard way.
19. Color and shape
The general terms that we have been studying up to now are
terms for bodies . Such terms are at the heart of individuati on,
given man's body-mindedness. Other general terms may have
devel oped by anal ogy. But others there are. One i s 'col or' . Let us
examine i t.
' Red' was learned l i ke ' Mama' . The two terms di ffered only i n
the course of thei r ostensions, as when Mama went one way and
her shawl another. Moreover, just as Mama is an animal, so red
is a col or. But as soon as we try to treat the learning of the term
'color' i n parallel to the l earning of the term ' animal' , curious
di fferences emerge. In l earning 'animal' we had to l earn two
things: (a) when an animal was being indicated at all, and (b)
7 1
when the same animal was being indicated. In learning col or
there i s no parallel to (a), since col or i s everywhere. What then of
(b ), what of sameness of color? There is an odd twist of usage
here. We speak of sameness of color to mean matching, and most
red surfces do not match. Sameness of color, ordinari ly so
called, i s not sameness of color in the sense in which red is a
col or.
The very notion of a color, i n the latter sense, i s unnatural .
Whether some arbitrary interval in the spectrum is a color, in
this sense, depends on the casual matter of thei r being a word for
it; and this matter of vocabulary varies from culture to culture.
The notion of a col or, in this sense, is less basic than the notion of
a color word.
The color words themselves, ' red' and ' blue' and the l i ke, are
good observation terms on a par with 'water' , ' Mama' , 'dog' , and
the rest . I am not downgrading them, but rather the general term
'color' as thus applied. For an analogy, consider the people
whose telephone numbers are prime. This is an unnatural notion,
but the people that fall under it are real people. Correspondingly
for red and blue.
Unl i ke the unnatural notion based on telephone numbers, the
unnatural notion of color or of col or word is i mportant and
useful . Like other general terms, 'color' in this sense is useful in
expediting ostension. Thus take the observation sentence 'This is
mauve' . It can be taught laboriously by brute ostension, repe1ated
suffciently to eliminate the misconception that the similarity
basis might involve shape or substance or texture. By adding the
eternal sentence ' Mauve is a color' , the teaching i s expedited.
The observation sentence and the eternal sentence apocopate
in the usual way, to 'This col or is mauve' . Mauve is a col or in the
sense in which red is color: not a precise tint, but a vague range.
Consequently the pupil may sti l l need several exposures to get a
feling for the al l owable latitude. Still, calling it a color is a big
hel p.
Another principal use of the word 'color' i s in describing
something by resembl ance to something fmiliar, as by saying
that it is l i ke an orange in color, or a pomegranate. This is not
quite a matter of matching, but neither is it subject to the ac-
cidents of color words. It is a matter of chromatic si milarity.
Perhaps something quali fes as pomegranate-colored if it is
chromatically bracketed by some pair of pomegranates; that is, i f
it is chromatically more si mi l ar to each of the pai r than they are
to each other. Chromatic si mi larity is fundamental to color tal k,
and matching is i ts l i mi t. Matching i s a question of "exact color"
or tint, which is independent of vocabulary.
The word 'color' serves poorly as an observation sentence,
'Color here' , 'This is a color', si nce, as remarked, there i s color
everywhere. But there are three contexts in which the word i s
needed and needs to be l earned: '-colored' (as in ' pomegranate
colored'), 'is a color' (as in ' Red is a color'), and 'This is the same
color [ = tint] as this' .
Take '-col ored' . First the child learns some examples as wholes
in the usual ostensive way: 'orange-colored' , 'coffee-colored' , on
a par with 'red' . Also he learns the component observation terms
in the usual ostensive way: 'orange' , 'coffee' . Finally he abstracts,
from such examples, the art of fbricating new color terms of the
form a-colored1 along similar l ines on his own. The similarity
relation that links such formations is, as at the end of 1 5, a
simil arity of simi larities. The occasions for assenting to a
colored1 are similar to the occasions for assenting to a, and
similar in a special way: chromatical ly. I assume, plausibly
enough, that chromatic simil arities bear a second-order similar
ity to one another. Chromatic similarities constitute, for the
child, a discriminable species of similarities .
What now of the eternal sentences of the form a is a col or 1 ?
Grammatically they are predications l i ke ' Fido i s a dog' and
'Snow is white' . These latter were handled in 1 7 by treating
each utterance of such a sentence as an occasion sentence in
which the sound of the frst word is the occasion for affirming or
assenting to the rest . Si mi l arly we may treat each utterance of a
is a col or 1 as an occasion sentence i n which the sound of a is the
occasion for affrming or assenting to 'is a color' . To master this
idi om, then, the child must learn to distinguish col or words from
others; just that. Now the bond of similarity among the color
words is their capacity to induce purely chromatic images. ' Red' ,
for instance, enlivens the traces of episodes in which the red
73
came in al l shapes; and these shapes cancel out. Having al ready
assumed that the child can distinguish chromatic si mi larity from
other si mi l arity, I do not hesitate to assume that he can dis
tinguish chromatic images from others.
The remaining context that was to be' learned is a straight
observation sentence: 'This is the same color [ = tint] as this' .
Each occasion when this observation sentence may appropriately
be assented to is an occasion of double ostension whose salient
regions are chromatically similar in the extreme. So the simil ar
ity basis of this observation sentence is the second-order similar
ity' that obtains among extreme chromatic similarities.
Along with the locutions a-colored\ a is a color\ and 'This
is the same tint as this', one might look also for a is a tinf . Ac
tually no use is made of this construction, outside of technical
circles; names are given not to exact tints but only in a haphazard
way to broader colors.
This is a point of contrast between our color vocabulary
and our shape vocabulary. Terms for exact shapes are abun
dant: 'square' , 'circl e', 'parabol a', ' equi lateral triangle' , 'Star of
David', ' regular enneagon' , 'sphere' , 'cube' , 'regular icosa
hedron' . This is because various specifc shapes have distinctive
properties of a conspicuous or useful kind, while specific tints do
not, except in technical pursuits such as spectroscopy. Terms for
broad ranges of shapes, analogous to the color words, are abun
dant too: 'oblong' , 'rectangle' , 'ell ipse' , 'triangle' , 'enneagon' ,
' parallelepiped' , 'cone', ' pyramid' .
Two useful contexts of the word 'shape' are exactly parallel i n
form and meaning to two of the contexts of the word 'color' that
we have exami ned. One of them is a-shaped, , e. g. 'egg-shaped' .
Something is perhaps egg-shaped when it i s morphically more
similar to each of some pair of eggs than they are to each
other. Morphic si mi l ari ties, like chromatic simi larities, may be
assumed to constitute for the child a di scriminable species of
simi l arities; and on that basis the child can master the idiom a
shaped1 i n just the way that we have imagi ned hi m to master the
idiom r a-col ored, .
The other useful context of 'shape' that parallels a context of
'col or' is the observation sentence 'This is the same shape as this' .
The ostension accompanying this sentence would consist in the
tracing of two outl ines-or simply in pointing at two fgures if
they are al ready salient . The si mi larity basis of this observation
sentence is the second-order si mi larity that obtains among ex
treme morphic similarities. Extreme morphic simil arity is what
geometers simply call similarity.
' The idiom a is a shape, is unlike a is a col or 1 because of an
accident of usage. Color words, we saw, normally cover broad
ranges of tints; and so it is with the a of a is a color, . Shape
words, on the other hand, exist both fr exact shapes and for
ranges of shapes, and normally the context a is a shape1 calls for
th name of an exact shape. It is natural to say 'Square is a
shape' ; less natural to say ' El l i pse is a shape' or 'Obl ong is a
shape' : The usage of 'is a
shape' thus harmonizes with that of

' same shape', whereas the usage of ' is a color' was at variance
with that of 'same col or' .
We noted the utility of ' Mauve is a_ col or' in expediting osten
sion. An idiom to similar effect is wanted in expediting ostension
in the case of shape words, even when, like 'oblong' , the word
does not name a unique shape. 'Oblong is a shape' will not do, we
saw. But something else wi l l : we can say, outright, ' 'Obl ong' is a
shape word' . We al ready saw reason to think of ' ' Red' is a color
word' as mo' re to the point than ' Red is a color' , and we can ap
prove this resort on the same score. As for the method of learn
ing, we have al ready speculated on how the child would learn a
is a color1 ; it was a matter of his learni ng to distinguish color
words from others. These speculations carry over in exact
parallel to the learning of a is a shape word, .
There remains, then, the idiom a is a shape, in its proper
sense, for exact shapes: ' Square is a shape' . The child l earns
'square' as an observation term, and its simi larity basis is of
course sameness of shap. We al so had hi m learning the observa
tion term 'same shape' itself; that is, the observation sentence
'This is the same shape as this' . Now in any episode of double os
tension where the chi l d i s disposed to assent to 'Thi s is the same
shap as this' he is also disposed to assent to 'square' doubly or
not at al l . This general disposition to dispositions would be, if he
could sense it, his ground for assenting to ' Square is a shape' . But
it is a good deal to ask. He could learn to assent to specifc cases
75
by parroting his teacher, but he must master the princi ple if he i s
to cope with new sentences a is a shape, on the strength of
knowing a. Probably this is beyond hi m, pnding mastery of
other apparatus. But he al ready has the benefts of accelerated
ostension that are afforded by the easier idiom r a is a shape
word1
20. Truth functions
We have seen nothing as yet of compound sentences, that is,
sentences with other sentences as parts. ' Yellow paper' , though
qual i fying as an occasion sentence, was seen (1 6) not to contain
'yellow' and ' paper' as sentences . The si mpl est construction that
produces sentences from sentences is negation. It presents a
strangely simple case of learning language in relation to prior
language. An occasion sentence is uttered by the pupil, or as a
query by the teacher, in the presence of relevant other im
pingements, and the teacher or pupil di ssents, saying 'no' .
Al ready we have here a passable negation, i f we just think of the
occasion sentence and the di ssent as joi ned up continuously. For
it i s precisely this combination that i s appropriate to occasi ons to
which the occasion sentence by itself is not . Thus we may think of
our negation sign as basically a postpositive 'no' .
We have been thinking here of negation as applied to occasion
sentences. It is i n this appl ication that the theory is so pat. We
want the negation of an occasion sentence to be conditioned to
just the occasions where the original sentence deserves a dissent
ing response; so the 'no' of dissent just fl ls the bil l, here, of a
postpositive negation sign. When we move out to standi ng
sentences, this talk of occasions loses its point. Still, assent and
dissent carry over to standing sentences, and with much the same
force that they had for occasion sentences. When the child
uttered an occasion sentence on the wrong occasi on, the parent's
'no' betokened negative conditioni ng; it was the harbinger or ac
compani ment of the slap, or the substitute for i t. The parent's
'no' served the same purpose when the child transgressed in
other, nonsentential ways; and its force is not lost on standing
sentences. On these quite as well as occasion sentences, then, the
' no' of di ssent al ready serves automatical ly as a postpositive
negation sign.
What of conjunction? A governing circumstance that goes fr
toward fxi ng its meaning is that a conjunction commands assent
when and only when each component does. The learning is
trivial; each component simply gets prompted on its own i n the
circumstances, and the circumstances are joint. The ' and' is
pleonastic. The mechanism here is just what we saw i n attributiv
e
composition or predication, 'yellow paper', 'The paper is yel low',
except that there is no l onger the censorship of cases where the
yel low and paper are di screte. Conjunction gives 'yel l ow and
paper' , ' Here are yel low and paper', where attributive composi
tion gave 'yel l ow paper', ' Here is yellow paper' . In a way the

'and' is not pleonastic after all, si nce it preserves this distinction.
But a more natural course would have been to use an additional
symbol to disti nguish attributive compositi on, si nce that was
where the additional consideration came i n. To affrm a conjunc
tion is simply to affrm its components, and could have been left
at that if attributive composition had not intruded and pre
empted the notation of simple juxtaposition.
If we were content al ways to affrm conjunctions or leave them
alone, then what has just been said would be the whole story. The
components get affrmed, frst one and then the other, and there
is no cause to cal l this history of affrmations a compound. It is
in dissent that the rub comes: i f we are to di ssent from a conjunc
tion, and not just from one or other or each of its components on
its own, then we need the compound to dissent from. The ci r
cumstances of dissent from a conjunction have to be mastered in
dependently of the excessively si mple rule of assent. Still, one of
the rules of dissent is simple enough: the conjunction commands
dissent whenever a component does. This is a uniformity which,
though language-dependent, would be quickly learned: pople,
whenever queried, are observed to di ssent from a conjunction in
all ci rcumstances where they dissent from either component.
Conjunction has its blind spot, however, where neither compo
nent commands assent nor di ssent . There is no di rect way of
mastering thi s quarter. In some such cases the conjunction com
mands di ssent and in others it commands nothing. This sector is
mastered only later in theory-laden ways. Where the components
are ' It is a mouse' and ' It is a chipmunk' , and neither is affrmed
nor denied, the conjunction will still be denied. But where the
components are ' It is a mouse' and ' It is in the kitchen' , and
neither is affrmed nor denied, the conjunction will perhaps be
left i n abeyance.
A contrast thus emerges between truth functions and
something more primitive, verdict functions. Verdict logic is
three-val ued, the three verdicts being assent, dissent, and absten
tion. A compound sentence is a verdict function of its com
ponents if a verdict to the compound is determi ned for each
assignment of verdicts to the components. Negation is at once a
verdict function and a truth function
.
The verdict to the com
pound is assent, abstention, or di ssent according as the verdict to
the component is dissent, abstention, or assent. Conj unction, on
the other hand, is a truth function that does not quite qual i fy as a
verdict function. Its verdict table is incomplete, as fol l ows.
q
assent abstain dissent
p
assent assent abstain dissent
abstain abstai n
?
dissent
dissent dissent dissent dissent
What now of alternation? The question how we learn alterna
tion is of l ittle moment, since we could construct it from negation
and conjunction in the wel l-known way. Let us then just pause to
observe that its behavior under assent and dissent is similar and
dual to that of conjunction. Alternation, like conj unction, has its
blind quarter where neither component commands assent nor
dissent. We might assent to the alternation of ' It is a mouse' and
' I t is a chipmunk' , or we might abstain. The incomplete verdict
table for al ternation is as follows.
q
assent abstain dissent
p
assent assent assent assent
abstain assent
?
abstain
dissent assent abstain dissent
Verdict functions approximating to conjunction and alterna
tion could be forged by speci fying abstention at the center of the
tables. These are more primitive than the genui ne truth- func
tional conjunction and alternation, in that they can be learned
by induction from observation of verdictive behavior. They
are independent of our parochial two-valued l ogic, and indepen
dent of other truth-value l ogics. Truth val ues represent a more
advanced, more theory-laden level of li nguistic development; and
it is in terms of theory, di fferent theories for di fferent subject
matters, that we eventually learn (if at all) what verdict to give to
the cases of conjunction and alternation that are indeterminate at
the center of the verdict tables. Two-val ued logic is a theoretical
devel opment that is l earned, like other theory, in indi rect ways
upon which we can only speculate. Some theorists, notably the
intuitionists, fvor another logic, and there is nothing i n the
observable ci rcumstances of our utterances that need persuade
them to assign meaning to our two-valued scheme.
2 1. Analyticity
Carnap maintained, and Frege befre hi m, that the laws of
logic held by vi rtue purely of language: by vi rtue of the meanings
of the logical words. In a word, they are analytic. I have
protested more than once that no empirical meaning has been
given to the notion of meaning, nor, consequently, to this
linguistic theory of l ogic. But now in terms of the learning
process can we perhaps fnd some sense for the doctrine? We
learn the truth functions, I just now suggested, by fnding connec
tions of dispositions; e. g. that people are disposed to assent to an
alternation when disposed to assent to a component. The law that
an alternation is i mplied by its components i s thus learned, we
might say, with the word ' or' itself; and similarly for the other
laws.
Some such l i nking of meaning and truth is of course
characteristic of language l earning generally, also apart from the
l ogical particles. We learn to understand and use and create
decl arative sentences only by learning conditions for the truth of
such sentences. This is evident enough in the earl iest language
learning, the learning of observation sentences, since such learn
ing is simply a matter of learning the circumstances in which
those sentences count as true. The case is not quite thus with eter
nal sentences, since the truth value of an eternal sentence does
not vary with circumstances. Still, the l earning of ' A dog is an
animal' as I represented it consisted in learning to assent to it,
and this hinged on the truth of the sentence. It hi nged anyway on
our having l earned to assent to 'dog' only in circumstances in
which we l earned also to assent to 'animal ' . If we learned to use
and understand 'A dog is an animal' in the way I described, then
we learned at the same time to assent to it, or account it true.
It would seem reasonable, invoking the controversial notion of
analyticity, to say that by this account the sentence 'A dog is an
animal' is analytic; for to learn even to understand it is to learn
that it is true. Where the rub comes, however, is in numbers: the
number of different universal categorical sentences and the
number of persons learning them. My hypothesis is that each of
us learns his frst fw universal categorical sentences in the
described way, but that di fferent persons will begin with di fferent
sentences. Afterward, by abstraction from such samples, each of
us masters the universal categorical construction as such, and is
able to form new universal categoricals on a do-it-yourself basis.
This general mastery of the universal categorical construction
bri ngs mastery of countless universal categorical sentences that
no one would call analytic nor even true. I f the
-
samples frst ac
quired qual i fy as analytic, still they gain thereby no distinctive
status with respect to the language or the community; for each of
us will have derived his universal categorical powers from
different frst samples. Language is social, and analyticity, being
truth that is grounded in language, should be social as wel l . Here
then we may at last have a line on a concept of analyticity: a
sentence is analytic if everybody learns that it is true by learning
its words. Analyticity, like observational ity, hinges on social
uni formity.
The formulation wants some refning. We should l i mit the peo
ple to those who learn the language as mother tongue. Also we
should allow for chains of proof; we would want a recondite
sentence to count still as analytic if obtainable by a chain of in
frences each of which individually is assured by the learning of
the words.
Perhaps this version of analyticity succeeds i n drawing a rough
line between sentences like 'No bachelor is married' or 'We are
our cousins' cousins', which are ordinarily said to be analytic,
and sentences that are not. At any rate it would seem that we all
learned ' bachelor' uniformly, by learning that our elders are dis
posed to assent to it in just the ci rcumstances where they will as
sent to ' unmarried man' .
In making analyticity hinge thus on a community-wide uni
formity in the learni ng of certain words, we reopen the question
of analyticity of l ogical truths; for what about disagreement over
l ogical truths, e. g. on the part of intuitionists? We should fnd
perhaps that some logical truths are analytic and some not. I
suggested in particular that we do learn that an alternation is im
plied by its components, with the very learning of the word ' or';
and this is al l very well , for it is a logical law that the intuitionists
do not contest. I suggest that the law of excluded middle, which
they do contest, is not simi larly bound up with the very learning
of 'or' and ' not' ; it l ies rather in the blind quarter of alternation.
Perhaps then the law of excluded middle, though true by our
lights, should be seen as synthetic.
In Word and Object I defned a stimulus-analytic sentence as
one to which every speaker is disposed to assent. The analytic
sentences i n the present sense are a subclass of those, and a
somewhat nearer approximation to the analytic sentences un
critically so called. Even so, we have here no such radical
cleavage between analytic and synthetic sentences as was called
for by Carnap and other epistemol ogists. In learning our
language each of us learns to count certain sentences, outright, as
true; there are sentences whose truth is learned in that way by
many of us, and there are sentences whose truth is learned in that
way by fw or none of us . The former sentences are more nearly
analytic than the latter. The analytic sentences are the ones
whose truth is learned in that way by all of us; and these extreme
cases do not di ffer notably from their neighbors, nor can we
always say which ones they are.
22. Narrowing the subject
p
art I l l
referrin
g
t objects
We have been speculating on the mechanisms of language
learni ng, with emphasis on cognitive language. Our general ob
jective was a better understanding of how scientific theory can
have been achieved. We have now reached the end of our
speculations on the primitive steps . Our speculations on the sub
sequent steps toward theoretical language will be l i mited to one
i mportant aspect: the refrential aspect, the acquisition of an ap
paratus for speaking of objects. Somehow we do learn to speak
effortlessly of objects, and not only of physical objects but of at
tri
butes, numbers, sets, all sorts of abstract objects. To what do
we owe al l this virtuosity of objective refrence, and wherein does
it consist?
When can a child be said to have learned to refr to the color
red? Suppose he has learned to respond, on demand, in disti nc
tive verbal ways according as red i s conspicuously present or not.
Can we then say that he has learned to refr to red? No, this i s
not enough for what I mean by refrence. We can credi t the child
at this point with being able to discriminate red, to recognize red.
We i n confrring these credits do refr to the child and to the
color; these refrences we will readily own. But to say that he
refers to the color would be to impute our ontology to hi m.
The gratuitousness of thus i mputing our ontology can be ap
preciated by considering i n pl ace of the child a foreign adult who
gives _ si mi l ar evidence of recognizing red. There is a verbal
response that he makes on demand in case red is present and not
otherwise. Must this response be construed as a name of the
color? Might it not be, instead, a general term by which he
denotes each red portion of surfce? or a general term by which
he denotes each whole visible red patch but no smaller parts of
such patches? or a general term by which he denotes each body
that shows a conspicuously red portion of surfce? or a general
term by which he denotes each whole episode or specious present
that faunts red conspicuously? Under these different choices the
object of refrence varies. Under the one choice it is a col or.
Under other choices i t is a patch, and a di fferent patch from oc
casion to occasion. Under sti l l other choices it is a body, or an
episode, and a different one from occasion to occasion.
Thus, while we may determi ne straightforwardly enough that a
given word of the foreigner's language serves to attest to the
presence of red, this leaves us a long way from settling what
thing or thi ngs, abstract or concrete, his word may refr to, i f in
deed to any. We would settle this rather by working up a manual
of English translation for a substantial portion of the foreigner's
language. Within that systematic structure, the word that we
were worrying about would be given an English translation com
pati ble with the role that the word pl ays in that broader setting.
Then, presumably, we could answer the question of refrence.
We could come out with di fferent answers by devel oping
different manuals of translati on, each consistent with the man's
verbal behavior. Such is the doctrine of indeterminacy of transla
tion that I have urged elsewhere; it will not be a theme of this
book. What I want to bring out now is merely that the refrence
of the forei gner's word is not settled by hi s use of it in attesting to
Referring to Objects
83
the presence of red, and that the refrence is settled only by
translating a good deal of the foreign l i nguistic apparatus.
The cases of our own small chi ld and that of the adult
foreigner are superfcially ali ke: each of these persons has
mastered a word attesting to the presence of red. But the two
cases di ffer in that the refrence of the foreigner' s word has yet to
be assessed, whereas the refrence of the chi ld' s word has yet to
be acquired. Assessment of the refrence of the foreigner's word
awaits only our systematic English manual of translation of his
elaborate language. Such a manual would enabl e us to fll back
upon our fmi l i ar English refrential apparatus and so decide
whether to regard his word as designating the col or red or as
denoting patches or as denoting bodies or whatever. Di fferent
manuals may lead to di fferent answers, but one will do. The
child, on the other hand, is too young to have acqui red any ap
paratus, English or otherwise, whereby to di stinguish among
these various possible refrences. We can credit him with the
knack of responding distinctively to red episodes, but there the
credit stops.
I shall speculate on the steps by which the child might progress
from that pri mitive stage until we are satisfed by his easy com
munication with us that he has mastered our apparatus of
refrence. Thus I shall be concerned only with our language, not
with translati on. I spoke of translation from the foreigner' s
language just in order to make one point clear: to show that
refrence involves more than the simple abil ity to acknowledge a
presence. I showed this by pointing out that a word adequate to
acknowledging red episodes could be drawn from any of various
refrential roles; and I needed the foreign setting so as to keep
the questi on of refrence open. Now that this point is made, I can
forget about foreign languages and translation.
It may be objected that adherence to the home language af
fords no escape from theoretical problems of translati on, since
we are in principle still translating from idiolect to idiolect even i f
our manual of translation happens to be the null manual of
homophonic translati on, the identity transformation. Wel l , this
l ast part is true: the indeterminacy of translation holds also at
home. But adherence to the home l anguage can nevertheless af-
ford escape from problems of translation. It all depends on what
we are trying to do. It depends on whether we are going to make
capital of relations of sameness and difference of meaning. In
translation we do, whereas here I shall not. I shall be speculating
on the steps by which I may plausi bly have got to the point of
bandying certai n portions of English as I do-portions that I call
the refrential apparatus and can roughly speci fy by enumerating
some English pronouns, copulas, plural endings. I shall be equal
ly i nterested i n the steps by which you all may plausibly have
arrived at much the same usage as mine; but the sameness that I
have in mind here is merely the sameness that is tested by
smoothness of di alogue, and not a sameness of hidden meanings.
The child l earns this apparatus by somehow getting a tentative
and fulty command of a couple of i ts component

devices,
through i mitation or analogy perhaps, and then correcting one
against the other, and both against the continuing barrage of
adult precept and example, and going on in this way until he has
a working system meeting social standards. This is a vague pic
ture of how it has to be. I want a less vague picture. This refren
tial part of language learning needs to be better understood
because it is so central to our conceptual scheme. Our under
standing of the psychogenesis of refrence could enhance our
understanding of refrence itsel f, and of ontol ogy: of what i t
means to posit something. It could enhance our understandi ng, i n
particular, of universals.
23. General and singular
What i s the refrential apparatus? I mentioned pronouns,
plural endings, copulas. There is the copula of identity and there
is the copula of predication that joins general terms to singular
terms. Also the contrast . between general and singular term is
itsel f part of the mechanism.
A general term is true of any number of objects, from none up.
A singular term designates a single object, when all goes wel l .
Predication j oins the two terms to form a sentence to the effect
that this designated object is one of the objects of which the
general term is true.
85
' Mama' and ' Fido' are singular terms, though our categorizing
them as such i s a sophisticated bit of retrospection that bears lit
tle relevance to what the learning child is up to. ' Ani mal ' , 'dog' ,
' apple', ' buckle' , and ' body' are general terms, retrospectively
speaking, and what makes them so is the built-in individuation.
I n these examples the objects are bodies. The general terms are
true of bodies and the singular terms ' Mama' and ' Fido'
designate bodies, one apiece. But those two singular terms were
learned as observation sentences i n the same way as other obser
vation sentences, such as ' red' and ' water' and ' I t is raining' , that
do not designate bodies. Recurrence of Mama or Fido was
recurrence of a recognizable ci rcumstance, like recurrence of red
or rain. Thus the learning of these singular terms had nothing
distinctive to do with objective reference. It i s rather the learning
of the frst general terms, as we now cal l them, that may be said
to bring the child a step nearer to our patterns of objective
refrence, because of the individuation.
Indi viduation is initially the one fature that distinguishes
general from singular: 'dog' from ' Fido' . Their di fference of role
in predication is not significant at fi rst, because the way of learn
ing the predication ' Fido i s a dog' or ' Fido is an animal' is not
signi fcantly di fferent from the way of learning ' A dog is an
animal' or ' Snow is white' .
' Snow' , ' water', ' white' , and ' red' can be learned in the simple
manner of ' Fido' and ' Mama' . These all start out on a par, with
no thought of designation and no premium on bodies. The early
i ndi viduative terms, on the other hand, are general terms for
bodies. Bodies are the charter members of our ontology, let the
subsequent elections and expulsions proceed as they may.
' Color' seems, superfcially, to be a general term on a par with
the ones for bodies. ' Red' was quite like ' Fido' in manner of
learni ng, and ' Red is a color' matches ' Fido is a dog' in form.
The sameness of form is no accident; the syntactical behavior of
the term 'color' fol l ows the analogy of general terms for bodies.
But we found ( 1 9) that the term ' color' was defective: i t did not
admit of the context 'This is a color' , whereas 'This is a dog' ,
'This is an apple', etc. were basic contexts for these general terms
for bodies. And we found even that ' Red is a col or' , ' Mauve is a
color' , etc. were flse fronts in a way; they were curiously
language-bound. ' ' Red' is a col or word' , ' ' Mauve' is a col or
word' seemed more to the point.
Language thrives on anal ogical formation. The surfce
resemblance of the grammar of the term 'color' to the grammar
of the general terms for bodi es is quite the way of language, and
it is not to be set down to some early belief that col ors, like
bodies, were objects. I think matters devel oped in the opposite
order. As one came increasingly to think in refrential ways, one
came to objecti fy colors along with bodies on the strength of the
superfci al grammatical parallels.
'Shape' i s another term that seems superfcially to behave like
the general terms for bodies. When we looked deeper we found it
intermediate in behavior between 'color' and the body terms. It
resists the context 'This i s a shape' ; on this score i t i s l i ke 'col or' .
On the other hand predications such as 'Square is a shape' have
none of the language-bound character of ' Red is a color'; they
have rather the objectivity of ' Fido i s a dog' and other predica
tions regardi ng bodies.
So much for the analogy between ' Square is a shape' and ' Fido
i s a dog' . Observe now a conficti ng analogy that casts ' square' i n
the role no longer of ' Fido' but of 'dog' . We can poi nt and say
'This is a square' , ' This i s the same square as this', every bit as
signi fcantly as ' This i s a dog' , 'This is the same dog as this' .
'Square' i ndividuates. We have to learn when we are poi nting
twice to the same square and when to two squares. We can say
' Hoboken i s square' , or ' Hoboken i s a square', quite on a par
with ' Fido is a dog' . In these ways 'square' is unlike ' Fido' , ' red' ,
'snow', and 'white', and l i ke 'dog', ' animal' , and 'tint' . 'Square' i s
a general term afer all, by our best criterion thus fr: individua
tion.
Logicians resolve this confict of analogies by declaring an am
biguity. ' Square' i s general in ' Thi s is a square' and in 'This i s the
same square as this', they declare, and singular i n ' Square i s a
shape' . They di spel the ambiguity when they wish by rendering
the singular term as 'squareness' . It i s an abstract si ngular term,
they say, in contrast to the concrete singular term ' Fido' . It is
abstract by virtue of its intimate connection with a general term,
'square' . Here i s that frst portentous step down the primrose
Referring to Objects 87
path of abstract ontology. Shapes are abstract objects, univer
sals; and squareness i s. one of them.
' Squar
' and other shape words afford thus an enteri ng wedge

for Platonism. Color words would not have done so. Formally
the di fference is that a union of patches of a given color preserves
the color, while a union of patches of a given shape does not
preserve the shape. We can speak of snow as white and of blood
as red without treating ' white' and ' red' as general terms. We can
treat ' white' and ' red' as on a par with 'snow' and ' blood' . Snow
is a scattered porti on of the world that i s part of a more extensive
scattered portion of the world, the white. Similarly for blood and
red. Si mi larly for Mama smiling and Mama, for that matter, ex
cept that the temporal scatter of Mama's comi ngs and goings
gets taken up i nto a spatiotemporal continuity as science ad
vances. Fido' s being a dog, on the other hand, does not come
down to his being part of that more extended part of the world
that i s made up of dogs; for this could be said also of each of his
ears. Si mi larly for something's being square.
' Dog' , though i ndividuative like ' square' , would not have
served Platonism as enteri ng wedge for another reason: 'dog' i s
less likely than 'square' to be drawn i nto the role of si ngular term
as subject of a predication. ' Dog is a species' is unlikely in early
discourse; 'Square is a shape' is more likely. ' Dogs are animals'
i s of course beside the point; the various dogs are various
animals. What 'Square i s a shape' says is that square or
squareness i s a si ngl e shape, as Fido is a single animal.
' Dogs are numerous' is i ndeed a fir analogue of 'Square i s a
shape' , for it cannot be read as qualifying each separate dog. It is
on a par rather with ' Dog is a speci es' , or ' Dogs are a species',
and it has the advantage of being likely even in early discourse. If
its idiomatic form had been ' Dog is numerous', like ' Square i s a
shape' , it would have been a good wedge for Platonism too.
Once the wedge is i n, analogy drives it further. Color words
come to be thought of as general terms, so that ' Snow is white'
and ' Blood is red' are assimilated to ' Fido i s an animal' instead
of being seen as mere subsumptions of one mass term under
another. And then the color word in subject position, ' Red is a
color', comes to count as an abstract si ngular term like ' square'
in ' Square is a shape' rather than a concrete one as in ' Fido is an
ani mal' .
Under ordi nary usage, we see, the contrast between concrete
general and abstract singular terms is indistinct. Identity,
another component of our refrential mechanism, likewise fogs
up under ordi nary usage. Often we speak seemi ngly of objects of
one or another queer sort for which we can supply no principle of
i ndividuati on. Take characters of fcti on. Do two drafts of a
novel have the same hero? How dissimilar can they be? For that
matter, how dissimilar can the drafts be and still be drafs of the
same novel? Or take gods. Was Baal the devi l? Were the Algon
quins, in their worship of the Great Spirit, worshipping God? All
these exam pies belong under the general head of identity between
i ndividuals in di fferent possible words. To turn to a soberer ex
ample, take attri butes. Coextensiveness of attributes does not
make them the same attri bute; but then what does? We are
fobbed off with the answer ' necessary coextensiveness', but this
only puts another name to the problem.
Putting our house in ontological order is not a matter of mak
ing an already i mplicit ontology explicit by sorti ng and dusting
up ordinary language. It i s a matter of devising and imposing.
Genetically what we have beforehand i s just a play of gram
matical analogies that mask differences i n learning patterns.
Centrally situated there is what we retrospectively classi fy as talk
of bodies. Here i s where the apparatus of objective refrence gets
its frst development. Bodies are the prime reality, the objects par
excellence. Ontology, when it comes, is a generalization of
soma to logy. Steps i n this di rection have al ready occurred in the
development of ordinary language, in the emergence of such
general terms as 'color' and ' shape' by grammatical analogy to
the general terms for bodies. In forgi ng this grammatical analogy
we make our frst fltering allusion to i ncorporeal things. Gram
mar i s thereby simpli fed, while ontology i s multiplied.
Science seeks organization and simplicity. I n this spirit the on
tologist may try to render our ontology less heterogeneous. He
fnds a fi rly natural way, perhaps, of generalizing some one of
his categories so as to cause it to subsume another of his
89
categories. Thus he generalizes body to physical object, taki ng
this term broadly enough to admit as a physical object any
material aggregate however scattered; and thereupon he is able
to treat a mass term such as ' water' or ' sugar' as a singular term
designating a single di ffuse object .
It is in deliberately ontological studies that the idea of objec
tive refrence gains full force and expl icitness. The idea is alien to
large parts of our ordinary language. Still it has its roots in or
dinary language. A distinction between concrete general and
abstract singular is sometimes visible in ordi nary usage, and
clean-cut standards of individuation are i mplicit i n ordi nary
usage for wide ranges of objects. It is in imposing this refrential
pattern all across the board that scientifc theory departs from
ordi nary language. We see the result : objective refrence is cen
tral to our scientifc picture of the world. This is why it seems
desirable to trace the roots of refrence i n language learning.
24. Relative clauses
.
The pronoun was one of the items that I mentioned as making
up the referential apparatus. It has a crucial place i n that ap
paratus, as we shall see, and it is the prototype of the variable of
logic and mathematics. I n ordinary language it does i ts i mpor
tant work as an adjunct of the relative clause. So I propose to
speculate on the learning of the relative clause.
First let us look to its functi on. We have a complex sentence
about some object, say Fido, and the object is mentioned deep i n
side the sentence, or mentioned more than once; say ' I bought
Fido from a man that fund hi m' . Then the relative clause
enables us to segregate the object from what the sentence says
about it. We get the relative clause, 'that I bought from a man
that found hi m' , from the sentence by substituting the relative
pronoun 'that' or 'which' for the name of the object (and then
moving this pronoun to the front). This relative clause serves as a
general term which, when predicated of the object (Fido), reaf
frms the original sentence.
Geach has well argued, to the contrary, that the relative clause
is not a general term. 1 He would equate the relative pronoun
'that', in ' I bought Fido fom a man that found him', to 'and he' ;
thus ' I bought Fido from a man and he [ had] found him. I n other
examples he equates 'that' (or 'which' or ' who') to ' and it' , or ' i f
he' , or ' i f it' , or 'si nce he' ; but always to a conjuncti on (in the
grammarian's sense of the word) pai red with a simple pronoun.
On this view, which Geach calls the Latin prose theory of relative
pronouns, it is wrong to treat 'that found him' or 'man that found
him' a a . term or as a self-contai ned grammatical entity at all.
For, to switch to a mediaeval example that he adduces, take 'man
that owns a donkey'-as i f to say 'donkey-owner' . The sentences:
Any man that owns a donkey beats it,
Some man that owns a donkey does not beat it
would reduce to nonsense:
Any donkey-owner beats it,
Some donkey-owner does not beat it .
On his analysis, which renders ' that' as ' i f he' or 'and he' and in
cludes changes in word order, the sentences remain coherent:
Any man, if he owns a donkey, beats it.
Some man owns a donkey and he does not beat it.
A more challengi ng example, adapte by Geach from Emmon
Bach, runs thus:
A boy that was only fooiing her ki ssed a gi rl
that really loved him.
We cannot treat ' boy that was only fooling her' as a term in its
own right, filing a reference for the dangling pronoun ' her' ; nor
can we seek that refrence i n 'girl that really loved him' , fi ling a
refrence for the dangling pronoun ' him' . This reci procal passing
of the buck is nicely resolved by Geach's analysis:
1 . Reference and Generalty, pp. 1 1 5- 1 22; also "Quine's syntactical in
sights. "
A boy ki ssed a girl and she really loved him
but he was only fooling her.
9 1
Geach' s analysis resolves out the relative clauses and accords
prior status to quanti fcation. For the above fur paraphrases
along his line amount to these:
( 1 )
(3x) (x i s a man and I bought Fido from x and x had found
Fido),
(2) (x) () (if x is a man and y is a donkey and x owns y then x
beats y),
(3) ( 3x) (3y) (x is a man and y i s a donkey and x owns y and not
(x beats y) ),
(4) (3x) ( 3y) (x was a boy and y was a girl and x kissed y and y
really loved x but x was only fooling y).
The clearest and neatest account of the grammar of a fully
acqui red language is one thing; the most plausible account of the
child' s steps of acquisition of that grammar, or of the historical
stages of its evolution, i s another. Harmony between the two ac
counts would be grati fying and reassuri ng. According to Halle,
Chomsky, and Bloomfeld, it should be sought and expected. 2 In
the present instance, however, I fvor a contrary position: a
dualistic one. I accept Geach' s Latin prose theory as a descrip
tion of the accomplished grammatical fct, and I fnd at the same
time that the steps of acquisition are most readily imagi ned by
assimilating the primitive relative clauses to general terms. Most
contexts i n which we use relative clauses, and all the simplest
ones, are contexts li ke the Fido example, which submit in
di fferently to the view of relative clauses as general terms and to
the Lati n prose theory. I picture the child as masteri ng this much
and then being led on by evident analogies to further contexts
where crucial pronouns stray i rrecoverably afeld, as in Geach's
three examples. I picture him as being thus lured by analogies
between ' that' or 'which', on the one hand, and 'and he' , ' i f it' ,
etc. on the other-the very analogies that afterward become the
central structure of the fnished grammar of relative terms in
Geach's frmulation.
2. See Halie's paper and his references.
The fllowing considerations i ncline me to this theory of the
learning process. We cannot easily picture the child as learning
the relative pronoun i nitially as a versatile surrogate for ' and he' ,
' i f it' , ' si nce he' , and the rest, when we consider the complex
dependence of these alternatives upon differences of context.
Furthermore, if the child did manage to begin thus, he would still
have to learn the quantificational idioms as of ( l )-(4), or their
vernacular equivalents, before learning relative clauses. The one
course of plausibly short steps of transfrmation and analogy
that I have managed to devise, leading ultimately to a command
of something equivalent to quantifcation, is a course that
achi eves the relative clause frst and then moves with its help to
quanti fcation.
Anyway I am not bent even upon a fctual account of the
learning of English, welcome though it would be. My concern
with the essential psychogenesis of refrence would be fulfilled in
fir measure with a plausible account of how one might proceed
from i nfncy step by step to a logically regi mented language of
science, even bypassing English. This terminal language could
very well have complex general terms as its nearest analogues of
relative clauses, without prejudi ce to a Latin prose theory of
proper English.
25. Substitution and 'such that'
Let us proceed, then, with our psychogenetic speculati ons. I
shall not quite desert English, but with a view to the woods as
over agai nst the trees I shall depart from the most idiomatic
English. Already in English there i s a variant of the relative
clause that obeys somewhat simpler rules: the ' such that' con
structi on. It di ffers none from the idiomatic relative clause in
respect of Geach' s strictures, but it is simpler in word order. In
this idiom the relative clause 'that I bought from a man that
found him' becomes ' such that I bought him from a man that
found hi m' . Here the word order of the original sentence about
Fido is preserved: 'I bought him from a man that found hi m' , ' I
bought Fido from a man that found him' .
93
Tangles of cross-refrence quickly arise, in the ' such that' con
structi on no less than in ordi nary relative clauses. This is il
lustrated al ready by our present example, as soon as we put the
part 'that found him' i nto the ' such that' idi om. We get ' such
that I bought him from a man such that he found him' : who
whom? Well, it is mai nly mathematicians that talk i n ' such
that' s, and they settle matters by turning their pronouns to bound
variables:
'
x such that I bought x from a man y such that y found
x
'
. This, the ' such that' construction with bound variables, is the
rectifed relative clause, rid of crotchets that could only com
plicate our speculations on the essential psychogenesis of
refrence. Let us hereafter picture relative clauses thus, as i f the
child were growing p di rectly into this idiom i nstead of the
traditional one.
The relative clause enables us to put any sentence about an
object a i nto the form of a predication,
'
a is P', where ' P' i s a
general term. It enables us to render ' I bought Fido from a man
that found him' as a predication, ' Fido is such that I bought him
from a man that found him' . Here the relative clause, or ' such
that' clause, fgures as a general term in adjectival form. To turn
it i nto substantival form, which will be more useful hereafter, we
merely apply an adaptor, the word 'thing' : thus ' Fido i s a thing
such that I bought it from a man that found it' . Dog lovers will
please excuse my sudden depersonalization of Fido; it will be
helpful to waive logically i nsigni fcant disti nctions. Anyway, the
' i t' is on its way out now in fvor of variables: ' Fido is a thing x
such that I bought x from a man that found x
'
.
To say that the relative clause enables us to put any sentence
about a into the form of a predication,
'
a is P' , is not to say that
we should want to. The uti lity of the relative clause lies rather in
contexts where the clause is not thus predicated, but where it
plays rather one of the other roles of a general term. Now one
such role i s that of the a of the universal categorical construction,
An a i s a {1
,
_
' Every a is a /1 This is not a predicati on. It
couples two general terms.
Thus take the sentence ' Everything that we salvaged fom the
wreck i s i n the shed' . If this example is to make sense as a univer
sal categorical, Every a is {' , we need a general term to play the
role of a; and the term fr the purpose is precisely the relative
clause 'thing that we salvaged from the wreck' . Regimented in
our ' such-that' idiom, it i s 'thing x such that we salvaged x from
the wreck' . The whole sentence now runs: ' Everything x such that
we salvaged x from the wreck is i n the shed' . Clearly the utility of
the universal categorical construction depends heavi ly on this use
of relative clauses.
How, then, may the child have learned the relative clause?
The obvious way is by an equivalence transformation. The
mechanism of learning an equivalence transformation seems
simple: the learner i s merely brought to see, by abundant ex
amples, the i nterchangeabi lity of certain constructi ons. What he
is brought to see for our present purposes is that w can in
terchange 'I see the moon' with 'The moon is a thing that I see',
or, i n our regimentation, 'The moon is a thing x such that I see x' .
We can i nterchange ' Fa' with 'a is a thing x such that Fx' . It is a
substitution transformation: substitution of ' a' fr 'x' in ' Fx'. The
child learns this transfrmation by fnding inductively that peo
ple wi ll, if asked, assent to ' Fa' in all and only the situations
where they will assent to ' a is a thing x such that Fx'. He learns i n
context, in this way, the relative clause 'thing that I see' and in
deed the general construction 'thing x such that Fx' . He learns it
i n just the predication context, this being the context i n which the
relative clause is explained away by the substitution transforma
tion.
The acquisition would have little value i f the child left i t at
that. Why say 'The moon i s a thing that I see' when you can say
'I see the moon'? The substitution transformation explai ns the
relative clause i n just the one positi on, the predi cative, where it i s
useless. This i s no coi ncidence; the clause is useless there because
the equivalence transfrmation can elimi nate it there. What gives
the relative clause its utility is something else: the child' s pursuit
of analogy. General terms and relative clauses take predicate
position; so relative clauses are analogous to. general terms; so
the child lets the relative clause into other positions, notably in
the categorical construction, where he is accustomed to using
general terms. More exactly, what he does i s to emulate his
elders i n this maneuver without excessive bewilderment, thanks
95
to the analogy and despite there being no equivalence transfor
mation to explai n the relative clauses away from these contexts.
The substitution transformation that starts the relative clause
on its way could be used more li berally if one did not care about
producing something analogous to a general trm. The transfr
mation carried ' Fa' into 'a is a thing x such that Fx' . The words
'a i s a thing x such that' (or 'a i s a thing that' ) serve, we see, as a
substitution operator; here is the essence of the relative clause.
Apply the operator to ' Fx' and you get ' Fa' . To stress this sub
stitutional aspect, let us temporarily condense the words 'a is a
thing x such that' to read simply 'a vice x' . This is the Latin vice,
which I pronounce [ vai : si] more ang/ico. The result of applying
this substitution operator 'a vice x' to an expression does not
designate the result of substitution; it i n effect is the result of the
substituti on. 'Fa' i s equated to 'a vice x Fx', or 'a is a thing x such
that Fx' . 'I see the moon' is equated to 'The moon vice x I see x' .
Now that the words 'thing' and 'such that' are suppressed from
view, we can easily di ssociate our ' a' and 'x' from the category of
singular terms; for this substitution operator makes sense for any
grammatical category. You could transform ' How do you do'
and say ' Do vice x how x you x' .
The relative clause was learned in predicate position but
became useful by wandering from predicate position, i n emula
tion of general terms; by wanderi ng into the categorical . In so
doing it lost its eliminability. There is no lure, similar to the
general term, to unsettle this more general substitution operator
'a vice x'. Even so, it can get itself into i nextricable positions.
Consider this combi nati on: ' (x x vice x) vice x x vice x' . Bewil
deri ng? Well, we have our instructions; let us carry out the indi
cated substitution and see what the result says. If you try it on
paper you come out with just what you started with: '(x x vice x)
vice x x vice x' . '
Thus fr, little harm. Failure of eliminability had to be fced
anyway when the relative clause wandered into the categorical.
But there i s worse ahead, as you wi ll guess: Russell's paradox.
I . The parentheses can be mechanized by writing this as '(x x vice x) vice (x x
vice x)', and
'
a vice x Fx' in general as
'
a vice (x Fx)'.
Just repeat my earlier example with two negation signs i nserted:
' (x not (x vice x) ) vice x not (x vice x)' . It prescribes a substitu
tion which, when carried out, produces the negation of the whole
frmula itself.
Our child could learn this general substitution operator,
'
a vice
x' , as easily as he learned the relative clause. For the words
'
a is a
thing x such that' are just a speci al case of ' a vice x'; they are the
case for singular terms. The equivalence transformation by
which the general case would be learned is just the same, and i n
fct the general case
'
a vice x' is easier than the speci al case in not
being foll owed up by analogical extension to other than sub
stitutional contexts.
There, then, but for the grace of God, goes our child, blithely
down the garden path and into the very jaws of Russell's
paradox. Or maybe he would be stopped short by a saving quality
of realistic level-headedness or want of i magination. Up to a
point he would take the ' vice' construction in his stride because
he can elimi nate i t by carrying out the substituti on, but maybe
his tolerance of it would lapse when the case was so frftched
that he could not see for the l i f of him how to eli mi nate it even
fom full substituti on- contexts. 2 Maybe this would happen
already in the affrmative case, which was i rresoluble and queer
even though not yet self-contradictory. 'x x vice x indeed! ' he
lisps i ndignantly, and it warms one's heart to hear him. Or i f,
headlong and unheeding, he extrapolates too fr, he is presently
confounded by the paradox and thus receives his overdue lesson
in critical thi nking.
The paradox is not quite the same as Russell' s, si nce this idiom
is one of sheer substituti on. There i s no appeal to classes, no clear
case of objective refrence at all, nor any intrusion of semantic
concepts. It is i nteresting that the paradox can be got at this
level.
In any event the paradox goes ungrammatical when we confne
the substitution operator to the ' such that' case, the case where
the variable takes the position of si ngular terms. Transcri bed for
this case, the previously paradoxical line would run thus:
2. This criterion of meaninglessness is reminiscent of Church on lambda con
version. See his page 1 7.
(thing x such that x is not a thing x such that) is a thing
x such that x is not a thing x such that.
One is prepared to find this ungrammatical.
The runaway substitution operator 'a vice x' has 'a i s a thing x
such that' as its special case where ' a' represents a singular term
and 'x' takes the positions of singular terms. This is the case that
accounts fr relative clauses, and it is innocent of paradox. It is
so innocent that it can be translated i nto elementary logic as
' (3x) (a x and' , or, equivalently, as '(x) (if a x then' ; fr
Fa = ( 3x) (a x and Fx) = (x) (if a x then Fx) .
But this translation does not depict the learning process. Quan
ti fcation i s a later acquisiti on.
26. Quantiers and variables
I n 1 7 we speculated on how the child might master the uni
versal affrmative categorical constructi on, r Every a is a /1
Now an application of the ' such that' construction, or relative
clause, may be seen in the derivation of the other categorical
forms. Where the general term f i n ' Every a i s a /1 i s the relative
clause thing x such that x i s not a i 1
'
the whole becomes
' Every a i s a thing x such that x is not a i
,
,
which may be abbrevi ated as ' No a is a i ' -the universal
negative categorical. The two particular categoricals, then,
' Some a is a i 1 and ' Some a is not a {' , are got by negating the
sentences ' No a i s a i
,
and ' Every a i s a /1
Quantification i s forthcomi ng too. This was already evident in
24 from the example ' Everything that we salvaged from the
wreck is i n the shed' . for it amounts to a universally quantifed
conditional. In general the universally quantified conditional '(x)
(if Fx then Gx)' i s forthcomi ng as a universal categorical ' Every
a is a /1 with relative clauses for a and {, thus:
Every thing x such that Fx i s a thing x such that Gx.
From the particular categorical ' Some a is a '1 we have m
similar fshion the existentially quanti fed conjunction '( 3x) (Fx
and Gx)' , thus:
Some thing x such that Fx i s a thing x such that Gx.
Also we i mmediately get straight quantifcatwn, universal and
existenti al, since ' (x)Fx' and ' (3x)Fx, can be explai ned as:
(x) (i f not Fx then Fx), (3x) (Fx and Fx).
These derivations are artificial, but thei r existence suffces to
dull one's interest in what the actual learning process may have
been.
There are two attitudes toward quanti fcation, and toward
variables, that must be carefully distingui shed; fr thei r
differences are subtle but fr-reaching. Viewed in one way, the
variable i s strictly a placeholder for the constants that can be
substituted for it. Such variables do not purport to refer to ob
jects as values. The constants that may be substituted fr them
need
.
not be names at all; they may belong to any grammatical
category. We saw in 25 that that category had to be fxed
somehow, on pain of paradox; but it does not have to be fxed as
the category of names, or of singular terms. When its variable i s
conceived thus substitutionally, a universal quantificati on counts
as true i f and only if the open sentence following the quantifer
comes out true under every substitution fr the variable; and an
exi stential quanti fcation counts as true i f and only i f the open
sentence comes out true under some substituti on.
Objectually construed, on the other hand, the variable refrs to
objects of some sort as its values; and these need not even be ob
jects each of which is separately speci fable by name or descrip
tion. This is how variables are understood when we give the quan
ti fers ' (x)' and ' (3x)' the classical readings 'everything x i s such
that' and ' something x i s such that' .
Substitutional quanti fcation di ffers from objectual not only i n
being available. to other grammatical categories besides that of
singular terms. It differs still in i ts truth conditions when applied
to singular terms. A universal quanti fcation in the objectual
sense can be flsified by some individually unspeci fable value of
its variable, while the same universal quantifcation in the sub
stitutional sense remains true; and an existential quanti fcation i n
the objectual sense can hold true by virtue of some unspeci fed
value, while the same exi stential quantification in the sub
stitutional sense fils for lack of a speci fable example. But these
divergences tend to be unobtrusive, hinging as they do on un
speci fable examples.
Ruth Marcus construes quanti fcation substitutionally, and so,
less explicitly, did Lesniewski : she for reasons havi ng to do with
modal logic, he for reasons of nominalism. Some writers,
careless of the distinction between use and mention of ex
pressions, are hard to sort out. Eclectic readers have read
Whi tehead and Russell's quantifcation as substitutional on the
strength of some frthright passages, but dogged reading of
Whitehead and Russell supports the objectual i nterpretati on.
The variable of the ' such that' construction, which i s i n effect
the relative pronoun, is a substitutional variable at its i ncepti on.
The words ' i s a thing x such that' are learned by an equivalence
transformation that i s explicitly substitutional in character. And
this variable, surely, is the variable at its most primitive. It is a
regimentation of the relative pronoun. Variables begin as sub
stitutional .
Once the relative clause or ' such that' construction has done its
i mportant work of siring quanti fcation, a vital change takes
place in the character of its pronoun or variable: it goes objec
tual. Si nce the categorical construction An a is a /1 is learned
through such examples as ' An apple is a fruit' , ' A rabbit is an
animal' , it would be inappropriate to read '(x) (i f Fx then Gx)' i n
the substitutional way as meaning merely that every substituted
name that verifies ' Fx' verifies ' Gx' . It is unnatural if not absurd
to imagine names, or singular descriptions either, for all apples
and rabbits.
I see thi s switch from substitutional 'x' to objectual 'x' as an
i rreducible leap in language learning. We already noticed this
leap in part when we pictured relative clauses as wandering i nto
categoricals where they could not be eli mi nated by the sub
stitutional equivalence transfrmation. The further point to
notice now about this leap is that along with forsaki ng
eli minabi lity it frsakes the substitutional status of the variables
themselves.
Once the substitutional variable gos objectual, it goes objec
tual with a vengeance. It becomes the distilled essence of on
tological discourse. When we talked in si mple categoricals, we
tal ked i n li mited ranges: all dogs are animals, all rabbits are
animals, some apples are red, without prejudice to what objects
there may or may not be apart from animals and apples. But the
x and y of quanti fcation are anything.
Quantification i s a welcome encapsulation of the refrential
apparatus. Once a theory is frmulated in quanti fcational style,
its objects of refrence can be said simply to be the values of its
quanti fed variables. This of course i s explicit in the intended
readings of the universal and existential quanti fers: 'everything x
is such that' , ' something x is such that' . And the convenience of
thi s encapsulation becomes evident when you try to say in some
other way what the objects of a theory are. If you say they are the
objects named by the singular terms, you omit objects that you
might want to include even though individually unspeci fiable:
various electrons and transcendental numbers, perhaps, if not
also some remote grains of sand and star dust. Also you run
against the question what terms to count as singular terms and
which of these to count as naming.
Taking other li nes, you may say that the objects of a theory are
what the general terms are true of; or, agai n, what the pronouns
can refer to. These versions do amount pretty much to saying
that the objects are the values of the quanti fed variables; but
quanti fcation i s conveniently graphic and explici t.
Quanti fcation, i n the frm in which we have come to know
and love it, i s less than a hundred years old. Still it i s in principle
a combi nation and exci sion of preexisting idiom. It can be
paraphrased i nto old and ordinary English. By considering what
steps could lead the small child or primitive man to quantifica
tion, rather than to the less tidy referenti al apparatus of actual
English, we arrive at a psychogenetic reconstruction i n skeletal
outline. We approximate to the essentials of the real psy
chogenesis of reference while avoiding inessential complications.
Referring to Objects 1 01
Seei ng the refrential apparatus as epitomized i n quanti fca
tion, we see it as consisti ng essentially of two sorts of device:
there are the quantitative particles ' every' and ' some' , as applied
to general terms in the categorical constructions, and there are
the variables or pronouns as used in abstracting new general
terms in the frm of relative clauses. The relative clause and the
categorical thus stand frth as the roots of refrence. The objec
tual variable is an outgrowth of these two roots, not of one alone;
for the variable of the relative clause begins as substitutional.
27 . Quantiying over abstract objects
I have been hoping to deepen our understanding of refrence,
and of object, by an i maginative reconstruction of how people
and peoples might achieve refrence to objects . I have been try
i ng to devise a series of plausibly easy stages, plausibly short
leaps, that might bring the emulous individual or the evolving
society to the point of handling something tantamount to quan
tificati on and thus refrring i n the fullest sense to objects of some
sort.
Such a study commands added interest where i t touches
abstract objects, since we tend to be mysti fed regarding thei r
nature and doubtful regardi ng their credentials. The frst
abstract objects to gain recognition are perhaps properties, or at
tributes. One already has general terms, concrete general terms,
to start one down the garden path. One has then only to treat
such a term as a singular term; the attri bute is what this singular
term names . And we already noticed some general terms that slip
over i nto the role of singular term remarkably easily: we say
' Hoboken i s square' and we say ' Square i s a shape' . Such terms
beat a path for the others.
Anything we can say about an object is treated as assigning an
attribute to i t. This highly creative doctrine of attributes is the
i nevitable outcome of two fctors. One fctor i s the shi ft from
concrete general to abstract singular; i t i s thus that we project at
tri butes from all our general terms. The other fctor is the
relative clause; fr it assures a general term encapsulating
anything we can say about an object. We thus end up with at
tri butes correspondi ng to everythi ng we can say about anythi ng.
I should like now to develop my hypothetical series of stages of
language learning more explicitly, to the point where these
abstract objects enter the ontological scene unequivocally as
values of quanti fed variables. In order to set aside some ex
traneous issues, however, I shall treat not of attri butes but of
classes. The only di fference i s that classes are taken always to be
identical i f they have the same members, whereas attributes are
not always taken to be identical when they hold of exactly the
same objects. A trouble with attri butes is that we are never told,
or anyway not in clear enough terms, what the further conditions
of their identity might be. Talk of attributes does ft ordinary
language more closely than talk of classes, and I think I know
why. I suspect, as usual, a li ngering tendency to confuse use and
menti on. One fels the attributes to be di ferent when one fels
dissimilar attitudes toward the corresponding relative clauses.
Still classes do accompli sh all the scientific work that the at
tri butes would accomplish, si nce the one trait that sets attributes
apart from classes i s imprecision. And fnally, what matters for
present purposes, a class has all the abstractness of an attribute.
By coping with classes we shall be coping with the full problem of
abstract objects. So let us speculate on the possible linguistic
origins of set theory.
A curiously myopic view of this matter has been mani fsti ng
itself of late. ' There is a hindsightful reaction, after two
generations, to the paradoxes of set theory. The new view is that
even before the paradoxes it was not usual to suppose there was a
set, or class, fr every membership condition. The view is
defnded by citing Cantor as having already entertai ned
certain restraints on the existence of classes before Burali- Forti
publi shed the frst of the paradoxes. Fraenkel has undercut this
argument by clai mi ng that Cantor had already sensed paradox. 2
What i s myopi c about the view, in any event, i s that it looks back
only to the first systematic use in mathematics of the word ' set'
I . E.g. in D. A. Martin's review. See also my reply.
2. See a biography by Fraenkel in Cantor, p. 470.
or ' Menge' , as i f this were uncaused. For surely it i s traditional to
talk as i f everything we say about an object assigned an attri bute.
It i s evident nowadays, further, that this attitude toward at
tri butes is i nvolved in paradoxes just like those of set theory. And
it should be evident that classes, or sets, are wanted simply as the
extensional distillates of attributes. It is i mplausi ble that Cantor
or anyone else would narrow this universe of classes for other
than sophisticated reasons, either nominalistic scruples or far of
paradox.
In trying to i magine a psychologically fasible genesis of set
theory I shall start with general terms, includi ng the relative
clauses, and use substitutional quanti fcati on. The genesis will
not be a matter of el i mi native definiti ons. It will proceed by
i rreducible leaps, but plausibly short ones.
The namelessness of apples and rabbits was what showed us
that our vari ables had gone objectual. But we might continue to
use substituti onal variables for other purposes. I n fct it is
precisely in the use of substitutional variables for general terms
that I can i magine an origin of an ontology of attributes, or
classes. I shall now develop this idea.
Quanti fcation over bodies owes its origin in part, I suggested,
to the previ ous learni ng of the categorical construction r Every a
is a {1 This line was forced on me by the objectual character of
the quanti fcation. I f a variable can be held to substitutional
status, on the other hand, our pupil can easily learn to quantify it
without ever thi nki ng of the categorical construction. He
observes that universal quanti fcation, of the substitutional ki nd,
command assent if and only i f each substitution i nstance com
mands assent. He observes further that the quanti fcati on com
mands di ssent if a substitution i nstance does. Once he has per
suaded hi mself of these regularities i n the behavior of the
speakers of the language, he has gai ned a pretty good command
of universal substitutional quanti fcation. There remains just one
limited blind quarter that he must master later in theory-laden
ways: i f none of the instances of the quantification command dis
sent, but some command abstention, then he will not know in
general whether the quantifcation will command dissent or
abstenti on. The situation is like that of conjunction (20).
Existential substitutional quantification is parallel . It com
mands di ssent if and only i f each substitution i nstance commands
di ssent, and it commands assent i f some substitution instance
commands assent. It has its blind quarter where none of the in
stances command assent but some command abstention; here the
quanti fcation may command either assent or abstention. The
situation i s like that of alternation.
So we see, well enough, how substitutional quanti fcation can
be learned. Different styles of substitutional quanti fcati on can
be learned i n this way, corresponding to different syntactical
categories of instantial expressions-di fferent categories of ex
pressions substitutable for the quanti fed variables. The syntac
tical category that i nterests me here is that of general terms, for
it is by substitutional quanti fcation with respect to general terms
that we can simulate set theory. In so fr as we thi nk of this sub
stituti onal quanti fcation as simulating objectual quantifcation,
we are thinking of the general terms as simulating abstract
singular terms: si mulating names of attributes, or of classes.
Summarizing then, I shall sketch our pupil' s past and present
progress down this garden path. He learns his frst variables, we
saw (24), by learning relative clauses; for, in my caricature, the
relative clause has the form 'thing x such that Fx' . These fi rst
variables are substitutional, fr he learns the relative clause i n
predicative position as a substitution idiom. Independently of
this he learns also the categorical copula, wi thout variables, as
joi ning general terms: r Every a is a
{
1

Next, we saw, he com
bines these two idioms, on the strength of an analogy between
relative clauses and general terms. The relative clause is
analogous to a general term in that it is learned i n predi cative
position; this is the very position that gave the relative clause its
intelligibility as a substitution idiom in the frst place. So, pur
suing this analogy, our pupil slips relative clauses i nto the
categorical s. Thenceforward the relative pronoun, or ' such that'
variable, fgures as an objectual variable rather than a sub
stituti onal one; for the animals and other bodies are mostly
nameless. Our pupil is engaging in quantifcation now, fr
categoricals with relative clauses give quantifcation; and it is
objectual quantifcation over bodies. All this happened by the
end of 26.
Next he learns another sort of quantification, whose variables
take the positions of general terms. Formally, structurally, it
resembles the objectual quanti fcation that he already learned.
But he learns it di ferently. Objectual quanti fcation came of a
fusi on of two idioms, the categorical copula and the sub
stitutional ' such that' ; and the fusion turned the substitutional
variable objectual. On the other hand the new quantification can
be learned whole, without any such fusion, because it is sub
stitutional and is to stay substitutional until further notice. It i s
learned, we saw, by learning how the conditions of assent and dis
sent relate these quanti fcations to thei r substitution instances.
Of course thi s learni ng is faci l itated by the outward
resemblance of this quanti fcation to the objectual quantification
that was already learned. This resemblance has also a more
profound effect: it induces a resemblance between general terms
and singular terms, since general terms are the substituends fr
these substitutional variables and si ngular terms are substituends
for the objectual variables. The resemblance makes the general
terms feel like names of something or other, and our pupil is not
going to put too fne a poi nt on it. Such is the unconscious
hypostasis of attributes, or classes. Our pupil has progressed well
down the pri mrose or garden path.
28. Set theory
I n recounting this supposititious psychogenesis in terms of
quanti fcation, I am adhering to my caricature. My conjecture is
that our actual learning of the ordi nary idiomatic apparatus runs
parallel to this caricature. But I gain visibility by bypassing the
sinuosities of ordinary language.
The substituends fr the substitutional variables are the
general terms. Among these are the relative clauses, or ' such
that' clauses. As substitutional quantification si mulates objectual
quantification, then, so the ' such that' clause simulates a class
name. The clause 'thing x such that Fx' , strictly a concrete
general term, takes on the guise of the abstract singular term '{ x:
Fxl ' , 'the class of all things x such that Fx' . The ' such that' con
struction takes on the jUise of class abstracti on.
1 06
Let us promote the si mulation by writing the ' such that'
clauses as class abstracts, ' { x: Fx } ' . The old substitution transfor
mation that explai ned the ' such that' construction now reappears
as the set-theoretic law of concretion. As a substitution transfor
mation it equated ' Fy' with 'y is a thing x such that Fx'. As law of
concretion it equates ' Fy' with 'y
E { x: Fx } '. The old copula of
predication, ' i s a', has become 'is a member of ' , or epsilon.
Classically, quanti fcation over classes i s objectual, class
abstracts are singular terms, and epsilon i s a two-place predicate
or relative general term. I am now suggesting an avenue to this
classical estate from humbler beginnings. The 'such that' con
structi on is a humble enough accomplishment, and o is sub
stitutional quantification governing general terms.
The set-theoretic law of comprehension:
( 1 ) ( 3Z) (x) (x E Z = Fx)
is frthcoming. It assures a class for every membership
conditi on. For we have, to begin with, the tautology ' (x) (Fx =
Fx)' . Transforming its left side by the substitution transforma
tion, we get:
(x) (x is a thing y such that Fy = Fx),
or, i n our new notati on,
(2) (x)
(x
E { y: Fy } = Fx) .
But this i s a substituti on i nstance of the substitutional existential
quanti fcati on ( 1 ). So ( 1 ) fllows.
( 1 ) combines substitutional and objectual quantifcation. The
combination is not new. It has been made by Wilfid Sellars and
more recently by Charles Parsons and in my Philosophy of Logic
(pp. 93f.
I n the above proof of ( 1 ) I assumed that ( 1 ) i s closed, or devoid
of free variables. I assumed that there are no free object variables
hidden along with 'x' i n the clause of ( 1) that is represented as
' Fx' . For suppose there were a frther free variable
'
w
'
. Let us
rewrite ' Fx' accordingly as ' Gwx' . The desi red law ( 1 ) then runs
thus:
1 07
(3)
(3 Z) (x) (x E Z = Gwx)
and its basis (2) thus:
(4) (x) (x E { y: Gwy} = Gwx).
But open sentences, sentences with free variables, are not true or
flse; rather they are satisfed, or not, by values of thei r free
variables. I f a substitutional quanti fcation is open, as (3) is, then
what it calls for is not a truth condition but a satisfction condi
tion. The satisfction condition for an open substitutional ex
i stential quanti fcation, such as (3), is just this: a value of the fre'
variable (
'
w' here) satisfes the quantifcation i f and only if it
satisfes some i nstance thereof, obtained by putting some class
abstract fr the 'Z'. (Parsons, p. 235. ) Now every value of '
w'
does indeed satisfy (3), because every value of ' w' stisfes the in
stance (4) of (3).
What are true are thus not (3) and (4) themselves, but thei r
universal closures:
(5) (w) (3 Z) (x) (x E Z = Gwx)"
(6) ( w) (x) (x E { y: Gwy J = Gwx) .
In this way the law
( 1 ) of comprehension i s still forthcomi ng in
its full generality, with and without hidden fee variables. The
truth condition for substitutional quantifcation is what yields ( 1 )
when there are no hidden variables, and the satisfction condi
tion for substitutional quantifcation i s what yields
( 1 ) when there
are hidden variables.
The unrestricted law of comprehension is a fmous source of
paradox. But there i s no such worry here, because of the two
styles of variables; the class vari able 'Z' cannot supplant the in
dividual variable 'x'. We have two types, in Russell' s sense.
Let us clearly understand where our pupil now stands. He
seems to be talking of classes. He even says he is, though we may
question the meaning of his saying so. At any rate his law of
comprehension i s unrestricted, so fr as classes of frst type are
concerned-classes of individuals. Yet all that i s really afot is
substitutional quanti fcation with concrete general terms as sub-
stituends. Can he get the beneft of all these classes without the
onus of really assuming them?
He cannot. ' There are elementary truths of set theory that fil
under the substitutional i nterpretation. The si mplest example I
know i s what we may call the law of unit subclasses: Any class
that has members has some unit subclasses. This i s unacceptable
when substitutionally interpreted. For, thus i nterpreted, what it
says is that whenever we can write a class abstract or relative
clause that is true of a lot of i ndividuals, we can write another
that i s true of exactly one of those individuals.
It is unacceptable for the same reason that substitutional
quanti fcation over physical objects was unacceptable (26). It
says that whenever we can somehow demarcate a multiplicity of
physical objects we can also speci fy a unique sample; and this
assumption is about as unwelcome as assuming a distinctive
designation fr every physical object. If we were prepared to
make such assumptions, we could as well accept substituti onal
quanti fcation across the board.
It is interesting to observe where an apparent proof of the law
of unit subclasses from (5) bogs down. From (5) we make fur
successive steps of deducti on, as follows:
(7) (w) ( 3 Z) (x) (x E Z = w x),
(8)
(
Y
( w) ( w E Y

(3 Z) ( w E Y
(x) (x E Z =
w
x) ) ),
(9)
( Y
( ( 3 w) (w E
Y ( 3 w) (3 Z) (w E Y

(x) (x E Z =
w
x) ) ),
(
1 0)
( Y
( ( 3 w) ( w E
Y
( 3Z) ( 3 w) (w E Y . (x) (x E Z = w
x) ) ).
This last line is a precise formulation of the law of unit sub
classes. Where then have we gone wrong?
The fllacy occurred in the last step, where I switched ' (3 w)
( 3Z)' to '( 3 Z) (3 w)' . 2 Commutation of consecutive existential
quanti fers (and of universal ones) is emi nently allowable in or
dinary logic, but not when one quantifer is objectual and the
other substitutional.
I . Discussion with Gilbert Harman and Oswaldo Chateaubriand started me
on this track.
2. I am indebted here to William Talbott.
1 09
For, pi cture ' Y' as some class abstract or relative clause that is
true of various individuals but true of none that can be uniquely
singled out. Each such i ndi vidual
w
will satisfy:
w E Y (x) (x E { y: w y} = w x)
and hence will satisfy:
( 3Z) (w E Y
(x) (x E Z = w x) ).
Thus
(3 w) (3 Z) ( w E Y . (x) (x E Z = w
x) ).
Yet not
(
11
)
(3 Z) (3 w)
(w E Y . (x) (x E Z =
w
x) ),
since this would require there to be a closed class abstract, devoid
of free variables, that singles out some such individual w unique
ly.
Commutation of consecutive universal quanti fers fils along
with that of existential ones. For, since '(3 w) (3
Z)' amounts to
'
-( w) (Z)
-
' and '(3 Z) (3 w)' amounts to
(Z (w) -
' , com
mutativity of the universal quanti fers would imply commutativ
ity of the existential ones.
There i s a still worse anomaly. Each member u of Y satisfes
the open sentence:
{ y: u y} is a unit subclass of Y . u u.
Hence, by our satisfction condition fr substituti onal quan
tification, each such u satisfes also the open sentence:
( 3Z) (Z
is a unit subclass of Y . u u).
Yet '
(3 Z) (Z is a unit subclass of Y )' without the 'u u' i s
simply ( 11 ), and thus fl se. It i s intolerable that such deletion of a
vacuous clause 'u u' should reduce a satisfable sentence to a
flse one.
In classical set theory, which is wholly objectual, the law of
unit subclasses of course goes without saying; nor is there any
trouble about switching consecutive existential quantifers, or
1 1 0
universal ones. But the combination of objectual quanti fcation
fr i ndividuals and substitutional quanti fcation fr classes i s like
oil and water. Substitutional quanti fcation i s inadequate fr
classes unless we are prepared to make an assumption tan
tamount to substitutional quanti fcation over individuals as wel l .
And surely for indi viduals, physical objects, the case fr objec
tual quantification was conclusive. Quanti fcation over physical
objects was objectual because of its categorical root, i n sentences
like ' Rabbits are animals' that treat of individually nameless ob
jects. Of course each rabbit and even each grain of sand can in
principle be systematically specifed and accorded a descriptive
name, e. g. with help of spatiotemporal coordi nates. But such an
artifice is wildly i rrelevant to genetic considerations, and has its
pl ace only at the level of a conscious reworking of explicit scien
tifc theory.
I do not see this clash, or crash, as refting my genetic
speculations. I see it as confict in the actual genetic frces. We
saw another such debacle already in 25, when we found that
something l i ke Russel l' s paradox could be generated simply by
overdoing the substitutional idiom that gives us relative clauses.
It is a historical accident that that paradox was not noticed
befre the rise of set theory, and it is a historical accident that the
present clash was not noticed more explicitly than it seems to
have been.
Once it is noticed, steps need to be taken in the way of revision:
revision not of the psychogenetic theory, but of the set theory or
other science to which those psychogenetic processes have led.
Any scientifc theory has its psychogenesis, and occasionally a
theory does prove to demand revision. The present case i s
remarkable only i n lying so deep.
The revi sion that I would expect is a revi sion of status of the
quantifcation over classes: a shi ft fom substitutional to objec
tual. Already, indeed, we have seen such a shift from the original
ly substituti onal variable of relative clauses to the objectual
variable of quantification over physical objects. When we
reconstrue class quanti fcation as objectual, the hypostasis of
classes is complete. Class abstracts attain the full stature of
abstract singular terms.
Referring to Objects 1 1 1
Schematism has carried us fr from the human situation. Let
us restore something of the connecti on. Mostly people do not ex
plicitly quantify over classes at all . Occasionally they speak of
properties or attributes i n ways that answer nearly enough, for
our schematic purposes, to quanti fcation over classes. Also they
will say things that involve no explicit mention of classes or at
tri butes but that come nevertheless to call fr quantificati on over
classes when we paraphrase them into economically schematic
language. The Geach-Kaplan example, ' Some critics admi re
only one another' , is a good case of this. 3 It is only the
mathematicians and ki ndred spirits that go beyond these
sporadic i mplicit uses and press the class apparatus on principle.
It will be to such spi rits, almost excl usively, that the holdi ng or
filing of the law of unit subclasses or the law of commutation of
quantifers or related laws will make any detectable di fference
even i mplicitly. Even these mathematicians, moreover, will most
ly be heedless of an option between substitutional and objectual
set theory, and heedless of a swi tch from one to the other. We,
however, can tell that they have switched, however uncon
sci ously. For we may be sure that nominalism is right psy
chogenetically: classes or attributes are first concei ved through
substitutional variables for general terms. And then, when we
catch the mathematician conforming to the law of unit subclasses
and the like, we know that he has gone objectual .
So the di stinction is absurdly remote fom the typical language
learner. But it concerns us because we are interested in the theory
of language learning for the l ight that it may throw on the nature
of scientific theory.
29. Sour grapes
Our loss, in gi vi ng up the substitutional view of class
quantifi cati on, is less acute than might at fi rst be supposed. We
must recognize that even substitutional quantification, for all its
reassuring air of easy access, is not altogether a fee ride. It owes
3. See my Methods of Logic, third edition, p. 238 of later printings. It was
stated incorrectly in early printings of the third edition.
1 1 2 The Roots of Reference
its easy i ntelligibi lity to its transparent truth conditions: a univer
sal quantification is true if and only i f each substitution i nstance
is true, and an exi stential quantification is true i f and only i f
some substitution i nstance is true. But those truth conditions, for
all their transparency, are not elimi native definitions. Adoption
of substitutional quantification over classes was thus already an
i rreducible assumption of some sort, and who i s to say that it is
not an indi rect assumption of some sort of new objects such as
classes? There is a problem here of foreign exchange: what bor
rowings in terms of substitutional quantificati on are equivalent
to what explicit borrowings of objects of objectual quan
tification?
By way of frther consolation it should be noted that those
truth conditions of substitutional quantification were not really
so transparent anyway, if the class abstracts that are used as sub
stituends happen to contain class quantifiers in turn. For, thi nk
what might happen. A substitutional quantification over classes
has its truth condi tions, we tend to think, in the truth conditions
of those simpler sentences that are substitution i nstances of that
quantificati on. But the fct i s that those substitution i nstances
need not be any si mpler. The class abstract that we substitute for
the quantified variable, in getting one of the substitution in
stances, can itself contain a quantification more formidable than
the one we are i nstantiating.
The i mportant distinction emerges between predicative and
impredicative class abstracts. The impredicative ones contain
bound class variables. (There is more to the distinction when
classes of classes are at hand; but this will do for now. ) As long as
we adhere to predicative class abstracts, the circularity that we
just now observed does not occur. Classical set theory, however,
demands the i mpredicative line. Adopting it, we would have had
to recogni ze that those substituti onal truth conditions afford only
a partial semantical account of the quanti fiers. For a full truth
defi ni tion we would have to revert to Tarski ' s style, as if the class
quantifiers were objectual . This refection may help further to
reconcile us to our objectual attitude toward class quantificati on.
But it is still not to deny that class quantification was sub
stituti onal i n origin. I am persuaded that abstract objects owe
1 1 3
their acceptance to what is essentially substituti onal quantifica
tion, cast in natural language.
Substitutional quantification has already been widely regarded
as i nadequate for classical set theory, but for a wrong reason.
People have reasoned fom indenumerability, as follows.
Substituti onal quantification allows no nameless classes as
values of the variables. Since the available expressions are
denumerable, then, substitutional quantification allows only
denumerably many classes. Any of the classical versions of set
theory, on the other hand, assumes i ndenumerably many classes.
The class N of the natural numbers itself has indnumerably
many subclasses.
Because of our genetic approach, we have been picturing
classes only of physical objects. Still, numbers must somehow
make their entry in due course. For the moment let us antici pate
them, so as to meet the above argument on its own terms.
Actually there i s no clear contradiction between substituti onal
quantification and indenumerabi lity . 1 No function f enumerates
all classes of natural numbers; this Cantor shows by citing the
class { n: not (n E f (n) ) } as one that is mi ssed by the enumeration
f Does substitutional quantification require the contrary-that
some fnction f enumerate all classes of natural numbers? At
fi rst it seems so: it seems we could produce /by lexicographically
numbering all class abstracts. However, the function that
numbers the abstracts is not quite the required /; it is a di fferent
functi on, g. I ts values are abstracts, whereas the f that would
contradict Cantor's theorem would have classes as values. After
all, the substitutional character of our quantifiers and variables
does not mean that the classes are the abstracts; the substitutes
fr the variables are not names of abstracts but the abstracts
themselves, the purported or si mulated names of classes. The
fnction / that would confict with Cantor's theorem is rather the
function such that f (n) i s the class named by the nth abstract
g(n). But there is no prospect of speci fying such a function in the
notation of the system; for the naming rel ation is notoriously un
specifiable, on pain of the fmiliar semantic paradox of Grelling
I . I am indebted here to a remark of Saul Kripke' s.
or Richard. The apprehended contradi ction over Cantor's theo
rem is just that same semantic paradox.
The reasoning that I have just rebutted was needlessly devious.
With a little permuting and shortcutting it comes down to the
observation that there is a class that is named by no abstract;
namely the class
( l ) { x: x i s an abstract and not a member of the class it
names } .
This way of putting the matter cuts through the talk of numbers
and i ndenumerability and treats di rectly of the expressions and
classes of them. The paradoxical character becomes explicit,
since
( l ) itself is visibly an abstract. And the paradox is hum
drum, hingi ng visibly on the naming relati on.
I agree that a satisfctory substitutional foundation for
classical set theory is not forthcoming. But I have been at pains
just now to separate good fom bad reasons.
There i s also a bad argument to the opposite effect, purporting
to show that substitutional quantification is adequate to classical
set theory and indeed to any theory. This argument appeals to
the Skolem-Lowenheim theorem. Accordi ng to that theorem,
any consi stent theory has a model in the natural numbers. And
obvi ously the natural numbers are amenable to substitutional
quantification, there being a numeral for each.
To see the Skolem-Lowenheim theorem in its proper perspec
tive, let us bypass the talk of models and turn to fundamentals.
Consider a set of truths regardi ng an indenumerable domai n.
The theorem tells us that we can reinterpret those truths as a set
of truths regarding a denumerable domai n. Reinterpret? I f we
are all owed to reinterpret every sign capriciously, we can make
any strings of signs say anything we like. No; the idea of the
theorem is that we are to disturb only the terms and the range of
the variables, while preserving the meanings of the truth-function
signs and quantifiers. In short, we keep elementary logic and
change the rest. But then what the Skolem-Lowenheim theorem
tells us is merely that quantificati onal and truth-functional struc
tures are by themsel ves i nsensitive to the di stinction between
1 1 5
denumerable and indenumerable. The theorem tells us merely
that the di stinction is not that elementary, and that it can be ex
pressed only with help of one or another term-epsilon, for in
stance, or, for that matter, 'denumerable' . Quantificational and
truth-fnctional structures are by themselves i nsensitive, of
course, to most di stinctions; they are sensitive to the di sti nction
between fi nite and i nfi nite, as i t happens, and i nsensitive to di s
tinctions between infinites.
When phrased in the ol d way i n terms of models, the theorem
engendered a fel ing that indenumerability is somehow a matter
of point of view or of perverse interpretation. Thi s feli ng tends
to subside when we recognize that what is involved is only the
degree of elementariness of the concept. Why should the theorem
suggest that only natural numbers are needed as values of our
variabl es?
I should l i ke also to ask what the notion of substitutional
quantification amounts to even as appl ied to the natural
numbers. I do so a fw pages hence.
30. Identity and number
We i magi ned i n 1 5 a primitive inception of identity as a mere
temporizing l ocuti on, helping to sustain a prolonged ostension.
This accounted for the identity sign only as restricted by some
general term, such as 'apple', and fanked by demonstratives:
'This i s the same apple as this' .
Now that we are picturing the child as learning various
linguistic constructions by language-dependent simi larities, we
readily picture hi m as learning unrestricted identity al ong the
following l i nes. He comes by degrees to appreciate that whoever
assents to sentences a 1 and a is a i
will assent also to r / is

a i In the end he becomes di sposed to assent to a 1' when
he can in general see his way to assenting to r B i s a i
on the heels
of a i s a i
fr arbitrary i. He i s not using quantification over

classes here; the regularity of behavior that he comes to ap
preciate and to emulate of course does not get put into words at
al l . He may, however, be presumed to have mastered the relative
clause. Thi s is valuable for the vast range of general terms that it
affords hi m in the role of i above. Thanks to relative clauses, a
i s a i
can in effect be any sentence containing a.

Even here, where identity stands free of any restrictive term, a
relativistic account of identity retains a certain force. For the
semantic standard of identity just now set forth remains relative
to the words available in forming the terms i. When some ad
di tional general terms accrue to the language, a sentence r a {1
that counted as true by the above standard could be turned false.
The fi neness of i ndi viduation of our universe of di scourse, or of
values of our variables, varies with the richness of our supply of
general terms. Indi viduation is in the eyes of the beholder, and
varies with the strength of his lexical spectacles. '
When class quantificati on i s at hand, 'x y' becomes defin
able in fmiliar fshion as meaning that y is a member of
every class of which x i s a member. This is of course not the route
of learni ng. Still, for the while that class quantification retains its
pri meval substitutional character, this definition j ust sums up the
semantic characterization of identity that went before: the
characterizati on by exhaustion of relative clauses.
Armed with the identity predicate, let us now turn to the
natural numbers. These are definable in set theory i n various
fmiliar fshions, but not withi n the low level of set theory that
we have thus fr considered: the theory of classes of indi viduals.
Anyway, those set-theoretic definitions of number are notorious
ly irrelevant to psychogenesi s. What perhaps does refect the
learning of number pretty well, when due allowance is made for
the arti ficiality of the formalism, is numerical existential quan
ti fcation as defi ned with help of identity. We have:
( 3x) Fx = not (3x)Fx
0
( 3x) Fx = (3x) (Fx and ( 3y) (Fy and not
(
x) ) ),
I
0
( 3x)Fx = ( 3x) (Fx a
n (3y) (Fy and not
(
x) ),
2 I
I . See my Set Theory and Its Logic, p. 1 5. For bringing the point to bear in
the present connection I am indebted to Geach, "Ontological relativity and
relative identity. "
1 1 7
and so on. There is no class quantification here at all; just
objectual quantification over indi viduals.
What now of quantification over numbers? Substitutional
quantification bogged down fr classes, but it seems perfctly
suited to the natural numbers, si nce each has a numeral to
designate it. As before, we do not define these quantifiers, but we
know how they can be learned. '(n)Gn' holds if and only if ' Gn'
holds under all substitutions of numerals, and ' (3 n)Gn' holds i f
and only i f ' Gn' holds under some substitutions.
' Gn' here stands for any sentential context of ' n'. But what
contexts are there? Thus fr, ' n ' can take only the subscript posi
tion beneath the existential quantifier. We can say:
(3 n) ( ( 3x)Fx and (3x)Gx)
R
R
to mean e. g. that there are as many fllows as girls. But what of
arithmetic, with all its sums, products, and equations? In pri nci
ple a learning routine can be l aid out, however absurd from a
human poi nt of view. People will assent to '(3x)Fx' in j ust the cir
cumstances in which they will assent to ' (3)Fx' . Thi s generality
N
warrants assent to the identity 'n + 0 n' , according to the iden-
tity standard noted a few paragraphs back; for the singular terms
a and f of that paragraph can be 'n + O' and
.
' n' , and the
general term i can be 'thi ng m such that ( 3x)Fx' .
D
By a si milar fntasy we may picture the learning of 'm + (n +
l ) (m + n) + l ' . This, with 'n + 0 n' , gi ves the classical
recursion for addition. The recursions for multi plication and ex
ponentiation can be handled si milarly, and also the equations '2
1 + l ' , '3 2 + l ', etc. that defi ne the numerals. Elementary
number theory is thus well begun.
The drab reality is di fferent, and less indicative. Children are
taught prefbricated algorithms intensively in elaborate in
stituti ons. Otherwise they would not learn arithmetic. The
routine of observati on and emulation of adult usage that we have
just now been i magi ni ng would bog down for want of example.
But I thi nk the above account shows the steps that would be re
quired in learning arithmetic under ordi nary conditions of
language learning, if it could be done. It would be done i f there
were a more general tendency among adults to chat about sums
and products withi n earshot of the young. Not that this is
desirable.
We talk of numbers as of things, to the extent of quantifying
over them. We discourse elaborately of them, but mostly in isola
tion fom the rest of the world; they enter into operations and
equations with one another. When they do relate to the world
beyond arithmetic, they do so at fi rst in just one way: in saying
how many things are thus and so, (3x)Fx. Numerical expressions
resi st other contexts. If we ask whether 3 is a class, and whether 5
is a member of it, we draw a blank or worse. This is why the set
theoretic interpretati ons of number enjoy such feedom and
variety: there are no preconcepti ons apart fom special contexts.
And it i s just those special contexts, the algorithmic ones and
' (3x)Fx', that are accounted fr in the above caricature of
N
number learni ng. I thi nk the caricature captures essential traits
of what really goes on.
This numerical setting i s opportune for some frther refec
tions on substitutional quanti fcation, its comforts and its vex
ati ons. A comfort of it is the unproblematical character of the
learning of it. The natural numbers seem peculiarly amenable to
it, thanks to their numeral s.
A substitutional number theory can even be combined with
objectual quantification over i ndi viduals without confict. The
troubles over classes in 28 hinged on our relative specification
relative to a free objectual variable-of an otherwise un
speci fable class. Those troubles do not extend to natural
numbers, because each of these is absolutely specifable by a
numeral .
But now consider: where is the comfort? I f the natural
numbers are not to stop short, i nfnitely many numerals must be
available to substitute for the numerical variables. What are
these numerals? We cannot say that each is a physical object, an
i nscription, for then the supply stops short. Cn we say they are
shapes? Shapes in what sense? Not in the sense of classes of the
i nscriptions that are the physical realizations of the shapes, for
then we are in trouble again over short supply; unreal ized shapes
would reduce to the null class and thus be indi sti nguishable from
one another. And not i n the sense of analytical geometry, where a
shape becomes a class of classes of pai rs of real numbers, for
Referring to Objects 1 1 9
there is no comfort in expl aining numbers in terms of numerals
that are explai ned ultimately in terms of real numbers.
The dilemma can be stated more si mply and generally. A sub
stituti onal explicati on of arithmetical quanti fcation brings no
ontological economy to elementary number theory; for either the
numbers must run short or the numerals are infnitely numerous.
If the explanatory talk of i nfnitely many numeral s is itsel f to be
understood in turn in terms of substitutional quantifcati on, we
confront a problem at least as grave as the one about numbers;
and if on the other hand our talk of infnitely many numerals is to
be understood in terms of objectual quantifcation, we might set
tle uncritically for objectual quanti fcation over numbers in the
frst place.
The truth conditions of substitutional quantifcation over
numbers can i ndeed be made clear by tal ki ng only of numerals
and their substitution, but the numerals, i f they are to serve this
purpose, must be as abstrat as the numbers themselves. Ex
pressions generally, if they are to be infnite i n supply, might as
well be identi fed with their Godel numbers; no other view of
them offers any visible reduction in degree of abstractness.
Between a substitutional and an objectual account of quan
ti fcation over the natural numbers, then, there is a material
di fference only insofr as the substitutionalist is prepared to
forswear the arithmetical law that every number has a successor.
He would be in the positi on contemplated by Goodman and me
in 1 947. Some number would be the last, though our sub
stituti onalist would not know which; it would depend on the fcts
of existence of inscriptions, past and future. His would be an
arithmetic with an unknown fnite bound-an arithmetic of the
sort that Essenin- Vol pin has called the theory of feasible
numbers.
Language is learned i n a succession of leaps, and the un
conscious transition from substitutional to objectual quanti fca
ti on over numbers is one more such leap. It is completed, to all
intents and purposes, when the learner fnds himself bel ieving
that every natural number has a successor. This crucial step is
abetted by ignorance: nobody knows how high the actual
numerical inscripti ons are destined ever to reach, whereas
everyone knows how to top any given one. And fnally, having
1 20
gone unconsciously objectual over numbers, he will not boggle at
objectual quantifcation over classes. It was somewhat thus, we
may imagine, that abstract objects made the scene.
3 1 . Higher types
We have thought about classes of individuals, in some sense.
Cl asses of physical objects, perhaps; or perhaps also, by now, of
numbers . What then of classes of classes? Resuming our psy
chogenetic speculations, we may or may not picture our aging
pupil as having caught up with the troubles over the laws of unit
subclasses and commutati on of quanti fers and thus been per
suaded to switch to an objectual doctrine of frst-type classes. In
either case we may picture the classes of second type as emerging
in the manner of their frst-type precursors: through sub
stitutional quanti fcation with relative clauses as substituends.
Early i n life our schematic pupil learned our schematic relative
clause, the ' such that' construction, by learning to equate the
predicati on
'
y is a thing x such that Fx' to ' Fy'. Now simi larly he
will learn relative clauses of the next type by learning to equate
the predicati on ' Y is a class X such that FX' to ' FY'. Relative
clauses are general terms, but the relative clause that he has now
learned is an abstract general term: ' class X such that FX. Next
he would learn second-type substitutional quantifcation, using a
new style of variable whose substituends are these abstract
general terms. But also he would repeat his old trick of pronoun
cing relative clauses as class abstracts. Thus these abstract
general terms become class abstracts of second type, hence
doubly abstract singular terms. He would take to saying, not ' Y
i s a class X such that FX' , but ' Y E { X: FX} ' . His new sub
stituti onal quantifcati on is now nominally a quantifcation over
second-type classes. Finally he might fnish the job by reconstru
ing this quanti fcati on as objectual, because of the law of unit
subclasses or similar troubles.
The caution that was sounded at the end of 28, by way of in
ducing some sense of contact with reality, is of course doubly
needed now. Natural cases of impl icit quanti fcation over higher
type classes are hard to fnd in ordinary discourse and even
1 2 1
in nonmathematical science. Il l ustrations comparable to the
Geach-Kaplan example (28) are not forthcoming for higher
types . We happy few who broach that domain at all learn it from
systematic treatises. My purpose in sti l l pressing on with the con
struction of a fctitious learning process is to understand better
why set theory came about: how it represents the firly natural if
not inevitable continuation of what is already at work at lower
l anguage level s.
The i maginary devel opment just now described can be
repeated for further types, step by step. The types can be taken
either as mutually exclusive or as cumul ative; it is merely a ques
tion of taking the ranges of substituend expressions excl usively or
cumulatively. I f they are taken cumulatively, there is no diff
culty in the way even of transfnite types. Class variables of type
w admit as substituends the class abstracts of all fnite types.
I have represented set theory as rooted in substitutional quan
ti fcation. Further I have represented it as rooted i n a theory of
types, by imagining analogical extensions from type to type. But
I have represented it also as outrunning its substitutional inter
pretation. Equally, and more promptly, it would l ose any type
theoretic structure. The analogical extensions were matters
surely of unconscious analogy, bringing wave after wave of new
classes or attributes; and these would naturally j ust accrete to the
growing uni verse without preservation of strata. Scruples over
mixing the levels would be as unlikely as scruples over im
predicative classes. Thus I do not see Russel l ' s theory of types as
dormant common sense awakened. Still I do see it as somewhat
aki n to that . I see Russell as making cleavages that were not
there before, but that nevertheless bear a signi fcant relation to
the psychogenesis of classes or attributes. The relation is roughly
this: the nth type compri ses just the things that would be
forthcoming i f our unconscious receptivity to analogical exten
sion were unnaturally restricted to n iterati ons. But this way of
phrasing the relation i s rough in two respects, one minor and one
maj or. The mi nor point i s that the phrasing fts rather the
cumulative types than the excl usive ones, and it is the exclusive
types that were hi storically Russel l ' s. The major point is that this
way of phrasing the relation neglects i mpredicative classes . The
impredicative classes would seem to have no better pedigree,
genetically speaking, than might be claimed for a class that
violated the theory of types. So let us not cease to see the theory
of types for what it is: partly a formalizati on of natural origi ns,
but primarily an arti fce for bl ocking the paradoxes .
There are of course alternative artifces-Zermel o' s and the
rest. These seem to be more remote from likely psychogenetic
patterns, for there is no evident way of getting them started in
terms of substitutional quantifcation. In the case of Zermel o's
system, ' or that of von Neumann and Bernays,2 or my "New
foundations, "3 what makes substitutional quantifcation hard to
apply is that in those systems there is no deciding in general
whether a class abstract succeeds in naming a class or not;
whereas in the theory of types every class abstract that is gram
matical at all is taken as naming a class . In the system of my
Mathematica/ Logic,4 and again in the system that has been
vari ously refrred to as von Neumann-Quine and Bernays-Quine
and Kelly-Morse, 5 the obstacle is diferent. In these systems
every abstract does name a class, but there is no evident way of
accommodating the disti nction between sets and ulti mate classes.
I f our notion of the real m of classes or attributes was
devel oped frst along l ines somewhat congenial to the theory of
types, as we have been imagining, then further steps to the other
set theories are easily pictured. Fi rst, within the theory of types,
there is the unconscious shift in the interpretation of quanti fca
tion over classes: a shif to the objectual . Once the theory of types
is seen objectually, the other set theories become intelligible as
deliberately fshioned alternatives.
Zermel o's system and others are someti mes rendered as pure
set theory: set theory without individual s. In that setting there is
no evident impediment to substitutional quanti fcation, except
that the truth condition is impoverished as always by im-
I . See my Set Theory and Its Logic, 38.
2. Ibid., 3.
3. Ibid., 0.
4. Ibid., 2.
5. Ibid., .
1 23
predicative classes. However, set theory plays its part in our
overall theory of the world only by goi ng i mpure.
32. Psychogenesis summed up
We have reviewed the genesis and devel opment of refrence.
Our review may be taken as applying to either the indi vidual or
the race, since it is imaginary in any event. An early phase of
reference, perhaps the earl iest worthy of the name, was the un
iversal categorical, as in 'A dog is an animal ' . Thus fr no
variable and so no substitution. The next phase was the relative
clause, or ' such that' construction. Here was the inception of the
variable, and a substitutional variable thi s emphatically was; for
the relative clause was learned by means of a substitution
transformation. But the contexts that could be resol ved by this
transformation were of course just the contexts where the relative
clause was l east needed, since el iminabil ity implies dispensabil
ity. The relative clause turned ineliminable and i ndeed i ndispen
sable when it slipped into the position of one of the general terms
i n the universal categorical . Here, moreover, its variable turned
objectual. Objectual quantifcation was thus at hand and objec
tive reference was in full sway. There was refrence to physical
objects as val ues of quantifed variables.
Quanti fers and variables are alien to the vernacular, and so,
for that matter, is 'such that' . Yet I fnd the foregoing account
plausible i n its essentials, especially as a fctual account of learn
ing by the child. Surely his learning of the rel ative clause begins
with his getting the knack, however unconsciously, of a substitu
tion transformation, since this transformation is the one evident
link between a relative clause and the feestanding sentence that
is its prototype. 'I bought Fido from a man who found him' and
'that I bought from a man who found him' go into each other by
substituting 'that' for ' Fido' and vice versa. My use of the 'such
that' form for the relative clause merely simplifes matters and
brings the substitution relati on into sharper relief by obviating
the idiomatic readjustment of word order in the relative clause.
The chi ld might have learned the 'such that' construction more
quickly than the classical relative clause, i f it had been in the air.
Perhaps he learned interrogatives earlier than rel ative cl auses:
' who?' and ' what?' earlier than the relative pronoun 'that' . Thi s is
just a question of the detail of his mastery of the substitution
transformati on for relative pronouns, since the same substituti on
transformation underlies the interrogative pronoun.
The variable, like ' such that' , is a departure fom the ver
nacular on my part to simplify the rules. Nested clauses generate
ambiguities of cross-reference otherwise, we saw, that are resolu
ble only by opportunistic paraphrases i f at all . Variables might
wel l , like ' such that' , have been easier for the child to learn than
the vernacular styl e of relative clauses and pronouns. In invoking
them I schematize my conjectures about language learning, but
what I am conjecturing is that the child learns the vernacular i n
the same way, i n between mastering the inessential compli cations
that are bypassed in the schematism.
When I i mpute a shi f fom substitutional variables to objec
tual, I am speculating at two removes; fr the child does not real
ly know about variables, let alone any distincti on between sub
stitutional and objectual . Still I am schematizing what I think is
the real devel opment . The child learns the relative pronoun
through the substitution transformation; it i s i n this sense that I
say hi s variables are substitutional at frst. But when he gets to
sayi ng sentences of the form ' Everything that we salvaged is in
the shed' , ' Everything I fnd is mine' , he is not counting on in
dividual designators; substitution is no longer the point. It is i n
this i mplicit way that he may be said to have switched from sub
stituti onal to objectual variables, though really no variables are
in sight. In seeing thi s as a switch I am assuming that the univer
sal conditional ' Everything that . . . is . . . ' was learned by slip
ping relative clauses into the categorical; but this seems rea
sonable.
The quantifer i s another departure on my part from the ver
nacular. With
_
its variable it conveniently encapsulates several
vernacular l ocutions, such as ' Everything that . . . is . . . ', 'There
are . . . ' , ' Something is . . . ' . Quanti fcation is translatable into
these idioms and vice versa, with the aid of truth functions. When
I speak conveniently of the child as learning quanti fcation,
therefore, the lie is distinctly a white one.
Talk of classes is another such heuristic departure on my part.
The vernacular usually has 'attribute' or 'property' . I n cleaving
to classes I avoid the dark side of attributes, while continuing to
examine the l earning of abstract terms and the hypostasis of
abstract objects. I thi nk the resulting account is true to that
learning process, excludi ng the dark side, the side that concerns
the individuation of attributes; and this excl usion is after all
pretty vacuous, since the i ndividuation of attributes remains
forever pretty much unlearned.
So my account of the l earni ng of frst-type set theory is meant
as a schematized account of the child' s actual learning of the or
dinary language of abstract terms. This learning is eased, I think,
by two blurs: the blur between the concrete general and the
abstract singular use of a word, and the blur between sub
stituti onal and objectual variables. Variabl es, I urged, are at frst
substitutional-which is to say, more li terally, that the relative
pronoun is learned by a substitution transformati on. The child
may learn them as easily when general terms are the substituends
as when singular terms are the substituends. The blur between
general and singular terms is partly a help here, no doubt, and
partly a consequence. Thus fr, then, we have brought the child
abreast of the relative pronoun of abstract refrence; that is, in
effect, the variable of ' such that' with abstract val ues. It is
abstract refrence or abstract val ues only by courtesy, still, for
the variable is only substitutional . The next step in the un
conscious simulation of abstract refrence is quantifcation of
this variable, or the vernacular equival ent-partly hidden,
perhaps, in such contexts as the Geach-Kaplan example. This
could be learned by easy analogy with the previously acquired
quanti fcation over individuals (or the vernacular equivalent). To
the child of course it is no conscious analogy, but just more of the
same. Here we have a blur between the substitutional variable
and the objectual , since the quanti fcation over bodies had one of
its
roots in the universal categorical and was thus i mplicitly ob

jectual . The new abstract quantifcation has no such root in the
categorical, and may thus stay substitutional for a time. As a
reward for this, moreover, it admits of an easily learned truth
condition in lieu of a root in the categorical . This substitutional
truth condition brings frmness to the idiom, analogically learned
at frst, of quantifcation over virtual classes (or the vernacular
equivalent).
I n a subtle way thi s substitutional quantifcation, with general
terms as substituends, depends on the prior objectual quanti fca
tion over individuals also quite apart from the matter of analogy.
For it needs the relative clauses, or 'such that' clauses, as its sub
stituends; and ' such that' is a variable-bi nding operator whose
variable has al ready been l ured over into objectual status by the
demands of categorical contexts.
Numbers are learned i n counting and perhaps other contexts,
and then the child learns substitutional quanti fcation over them
(in effect) on the analogy of what has gone before. The sub
stituti onal truth condition does its good work again here.
Al ready there have been short ontological leaps into the
abstract. These class quanti fers and number quantifers, for all
thei r di sarmi ngly substitutional character, are not eliminably
defned. Still there need be no thought, at frst, of anythi ng so
monstrous as abstract obj ects-numbers, classes. And observe
now with what cunning these monsters insinuate themselves. The
numerical quanti fer may as well be seen as objectual, I
suggested, i f we allow each number a successor. The class quan
ti fer went objectual when we admitted commutation of quan
tifers and therewith the law of unit

subclasses.
Again no awareness of a di stinction between substitutional and
objectual quanti fcation is to be imputed to the child, nor to he
layman. With some laymen the question of accepting the
successor law never arises. To them, then, this distinction
between two kinds of numerical quanti fcation does not apply.
Other laymen may commit themselves i n response to question
i ng. If they do, thei r acceptance of the successor law may be ex
pected.
I n the case of classes our criterion of objectual quantifcation
was acceptance of the law of unit subclasses and the commuta
tion of successive existential quantifers and of successive univer
sal ones . To test a lay subject by these criteria would require
cl ever questi oning. Failing that, we must put our subject down as
one to whom the distinction between substitutional and objectual
quantifcati on over classes does not even i mplicitly apply.
1 27
The probl em of extorting an implicit commitment to i m
predicative classes i s much the same. We can expect the subject
to use relative clauses that are the vernacular equivalent of im
predicative abstracts, or to admit of being led- by interrogation
into using them. But it is hard to construct nontechnical oc
casions for hi s using such clauses as substituends for variables.
When we press on to higher types, we have l ittle to do with
language-learning processes in child or laymen. I al ready
stressed this point in 3 1 .
33. Pat and future
The preceding section summarizes the child' s acquisition of
the apparatus of refrence as I imagine it. Reference i s my main
concern in this book, but we might glance at other aspects of
cognitive discourse. What of sentences about past and future?
I f we are to deal with the child' s passage to sentences about the
past, we must consider what to do about memory. To remember
someone' s name is, typically, , to be di sposed to say it when
queried in hi s presence. Memory in thi s sense is substantially a
matter of observation sentences agai n, and no past tense. ' Red'
and ' Mama' and ' Addison Sims of Seattle' are all l earned alike.
Forgetti ng someone' s name i s forgetting a bit of language.
Thus what we ordi narily speak of as memory i s al ready
covered in some small part at the level of ordi nary observation
sentences where no past tense i s involved. What now of the past
tense? Alongside the observation sentence ' Red' or ' I see red' we
may picture the report of past observati on: the si mple, undated ' I
have seen red' . However, there is little point in speculating on
how the child might acquire thi s particular use of the past tense,
because it is so useless; it goes without saying. I f the chi ld is com
petent to use the word ' red' , it stands to reason that he has seen
red; and otherwi se he is in no position to say so, however flsely.
An observational report i n the si mple past is idle i nsofr as there
is a presumption that its observation term was acqui red by di rect
conditi oning. Russell declared singular existence statements to
be meaningless when their subjects were genuine proper names;
and the reason here is the same.
So let us turn rather to reports of past observation where the
term is one that may be expected to have been acqui red by defi ni
tion rather than by di rect conditioning. Or, si nce we can supplant
any defined term by its defini ens, what the case comes to is the
report of past observation where the term i s complex; say 'I have
seen a black rabbit' . Now mastery of the sentence consists, as
always, in learning the ci rcumstances of assent . What, then, are
the cues that the child might learn for assenting to the query
' Seen a black rabbit?'? He has learned the terms ' black' and ' rab
bit' by di rect exposure, and he has learned attributive composi
tion, so he will recognize any occasion for assenting to the query
' Black rabbit?' if the occasion ari ses. I n the idiom of i mages we
might say that those words conjure up the right i mage, even fil
i ng the real thing. And now I far we may have to assume yet a
little more in the way of built-in fculties; namely, a di scri mi na
tion on the child's part between two kinds of i mages, images of
fncy and images of memory. Thi s sounds l i ke Hume, who
appealed lamely to vi vacity as the di stinguishing trait. Memory
is muted sensation and fncy is muted memory. But we should
like to dementalize the formulati on. An i mage i s a neural event
i nducing a state of readi ness for an appropriate i mpi ngement
pattern. Such a pattern may not come, but the child is aware
anyway, we now assume, of that monitory neural event itsel f;
that is, he must be able to respond di sti nctively to it, and di sti nc
tively even to two varieties of i t. One variety rests solely on the
assembling of learned elements, 'black' and 'rabbit' . The other
variety has been rei nforced by enactment in some actual i m
pingement pattern. The one rests solely on description while the
other has been rei nforced by acquaintance, to speak in Russel l ' s
terms. Now it is when the words 'black rabbit' i nduce this l atter,
more vigorous sense of readiness that the child may rightly assent
to the query ' Seen a black rabbit?' . He is said to remember seeing
one.
There are two conditions for learning when to assent to a
sentence: the cues must be prceptible i n themselves, and they
must get assigned to the sentence. These present speculations
about memory i mages have to do with the first fctor: the child's
cues for ' Seen a black rabbit' . The other question, how the child
could learn from his elders to connect such sentences with such
cues, is more easily answered. The parent rewards the child for
assenting to queries regarding past observations that the parent
saw the chi ld make. Also he penal izes di ssent in such cases. Also
the child fnds the parent reporting past observations that the
child saw the parent make.
Dated reports of past observations, next, can be smoothly ac
counted for i f we i magine for convenience of schematism that a
calendar _clock is in view. ' Bl ack rabbit on February 9 at 1 0: 1 5' i s
a pl ai n observation sentence that commands assent under any
i mpi ngement pattern that happens to include among its fatures
a glimpse of a black rabbit and a suitable gl i mpse of the hands of
the calendar clock. If we throw i n the past-tense verb, 'I saw', we
have a sentence of the same type as the preceding example ' I have
seen a black rabbit'; it di ffers only in that its observation term i s
more complex, applying no l onger simply to animals but to
ani mals and clocks i n combinati on. Addi ng one more equally
minor complication, we can accommodate observation reports
that are both dated and located. We have merely to imagi ne that
a signpost is visible saying where we are.
At l ast we have arri ved at protocol sentences, as they were
called in the Vienna Circle. They are the repository of scientific
data. Thanks to their dates, they are eternal sentences, like those
of scientific theory generally-or they become so when we drop
the personal idi osyncrasy of the 'I saw' and report simply ' Black
rabbit on February 9 at 1 0: 1 5 i n Sever Quadrangle' . The drop
pi ng of the 'I saw' is a primitive step of i nfrence from a report of
observation to an eternal sentence integral to the theory. These
eternalized reports comprise the archives of scientific evidence,
or would do so i n a bookkeeper' s dream world.
What of the future tense? As the past tense hi nged on memory,
so the future must hi nge on expectation. Expectation, at its most
primitive, is supposed to hi nge on induction from past ex
perience. If episode a was foll owed by epi sode b, then after an
episode a
'
si mi l ar to a the subject expects one b
'
similar to b. Now
thi s much is al ready provided by our earlier theory of i mages.
Accordi ng to it, an episode l eaves a trace from which a partially
similar episode can raise an i mage. Just let the one episode be the
temporally inclusive one, a followed by b. Then a subsequent
episode a' similar to a is partially si milar also to the tandem
pattern a followed by b, and so raises a tandem i mage a
'
followed
by b: It is perhaps this, primitively, that constitutes the expecta
tion of b' on the strength of a:
This account is suited nicely to the bl i nd expectation that
future successions will resemble past successions. It gives only
what we al ready saw (end of 7) in the enhancement of salience
by the sound of a word. The word 'dog' was sounded in view of a
dog, and afterward it serves to induce the dog image and enhance
the salience of any further dog. Does this mean that it makes us
expect a dog? I thi nk it does, pri mitively; and perhaps our fi rst
use of the future tense has no more than this as its tentative cue.
But our elders penalize us for poor predictions and reward us for
good ones, and so our use of the future tense is eventually
brought under control . The details of thi s mechanism remain
rather baffi ng, and not without interest; for predi ction, after all,
i s the payoff of scientific theory itself.
The steps of language learning that I have conjectured lead up
to that portion of language that is used in natural hi story-i n the
ki nd of science that i s sometimes called empiri cal as opposed to
theoretical . Measurement might be worked in; for we have con
sidered the acquisition of the natural numbers, and set theory
shows the way from them to the real numbers. But what of
theoretical posits, hypothetical forces, hypothetical particles?
I hold open some hope for that forbiddi ng domai n by thi nki ng
of the ki netic theory of gases as a paradigm case. Boyle' s law, to
begin with, makes sense at the level of natural hi story; for it can
be explai ned as relating expansible tanks and thermometers. The
subsequent positing of molecules, then, can be seen as essentially
just one more extrapolation along the time-honored lines of
similarity. But i t i s a matter, to begin with, not of similarity of
the molecules to anything; it is the expansible tank that i s similar
to a confi ned swarm, say, of bees. Aferward, by further analogy,
the molecules are posi ted as analogues of the component bees.
Even a perfcted psychology of science would not aspire to
keep causal track of the mi nds at the advancing front of natural
science. This would be no great l oss; for the psychologi cal theory
is concerned rather with the basic phenomenon of scientifc
knowledge than with its latest variati ons. And anyway the minds
at the advancing front are themselves aware of what they are do
mg.
1 3 1
34. Ontological sophistication
As summarized in 32, the chi ld' s acquisition of the apparatus
of refrence stopped short of any deliberate ontologizing on hi s
part . But the boundary i s not sharp. The learner progresses by
analogy and even by crude simplicity considerations, largely un
awares. The scientist or phi losopher who in a scientific spirit un
dertakes to clarify, organize, or simplify hi s ontology is doing
more of the same, but doing it better and 'in full awareness.
So it is with the set theorist, venturing ever upward and out
ward on hi s lofty ontological limb. The frther he ventures, the
less it matters even to natural science, let alone common sense.
But besides that steep limb there are other ontologi cal branches
to consider.
One stri ki ng scientific i mprovement on common sense is the
notion of a scattered four-di mensional physical object with tem
poral as well as spatial parts. Naively considered, Mama smiling
and Jumbo on the rampage are nothi ng of themselves; there are
just Mama and Jumbo, who smile or romp fom time to time. So
body-mi nded are we, we do not thi nk of the Evening Star and the
Morning Star as phases or aspects of Venus; they si mply are
Venus, and Venus may be refrred to by the one term or the other
dependi ng on the time of day. Si mi larly for Carnap's example of
Rumber, or Titi see; ' the lake was one and the same, but was ap
propriately refrred to by the one term or the other dependi ng on
the weather. Si milarly for Dr. Jekyll and Mr. Hyde; they would
probably be reckoned not as complementary temporal parts of a
whole and nameless man, but as an identical man with two
names appropriate to two of his modes or moods. Bodi es, in a
fi rst unsophisti cated ontology, have no parts but bodies. The
more general and more sophisticated notion of physical object
enables us to talk more systematically of the variously changing
bodies, by gi vi ng the stages of the bodi es an ontological status on
a par with the bodi es themselves. We can identify a fit of ague
with the concurrent temporal segment of the victi m' s body. We
can identify a battle, for that matter, with the physical object that
is the sum of the appropriate temporal segments of all the com-
I . Foundations of Logic and Mathematics.
batants. Also, as remarked, we can assimilate mass terms to
singular terms, each a name of a di ffuse physical object .
Less easily, colors can be accommodated too. Scarlet can be
identified with the scattered totality of those surfces whose fi ne
structure is of such ki nd, or of any of perhaps several ki nds, as
to radi ate light in the appropri ate spectral band or to refect such
light selectively when illuminated in a complete mixture of fre
quencies. And how thick are we to take this "surfce," as a
physical obj ect? Just thick enough to i nclude this optically
operative fi ne structure. To attack this account as a bad defini
tion of ' scarlet' would be off the point; it is an account rather of
how to ft that color into the expanded ontology.
Shapes are more stubborn than colors. Each shape, indeed,
such as square, or circle, is an abstract object . However, putting
these aside for the present, what about the particular squares and
.ircles? Even these are more stubborn than col ors. Can we iden
tify a particular square with a physical object that is a sum of
four suitable edges? It i s not clear how thick to take the edges:
there i s no causal condition to resort to as there was in the case of
colored surfces. Shall we identi fy the square rather with the
whole enclosed patch of surfce? Four objections crowd in. First,
the question how thi ck to take the surfce, as a physical object, is
i n the same di ffculty as the previous questi on of edges. Second,
there i s bound to be some irregularity at the edges, some devia
tion from ideal squareness, when we get down to elementary par
ticles; and thus there is indetermi nacy as to which peripheral par
ticles to include. Thi rd, the resort to the enclosed patch would be
of no avail when we turn fom squares to semici rcles or other
open figures. Fourth, there remains the problem of identi fying a
square as the same square through time.
Moreover this approach, even i f successful, would accom
modate only thi s and that particular square and not the shape
square itself, on a par with the color scarlet . We accommodate
scarlet as a scattered mass, and so we may, since unions of
scarlet regions are scarlet . But a trouble with squares is that their
unions are not in general square; nor can we even tell what
squares a given union of squares is meant to be a union of, since
we have by now passed the stage of counting as square only such
manifolds as are visibly marked out.
One classical solution is a double ontology: matter and space.
Spatial manifolds are aggregates of points as physical objects are
aggregates of elementary particles; and squares are spatial
manifolds. Physical objects are in space. A particular cross sec
tion of some physical object will almost exactly occupy a par
ticular square, and indeed it will almost exactly occupy each of
infnitely many almost coi ncident squares. There i s no longer a
problem of diachronic identi fcation; a square, a particular
aggregate of points, retains its identity through all time. Or,
when we integrate time with space as a fourth dimension, we take
squares as the appropriate manifolds in four-space. Points give
way then to point-instants. Purely spatial squares, i. e. squares
normal to the time axis, are on this approach instantaneous
rather than diachronically identi fed. The everlasting square of
the preceding account is still with us, but it is represented now as
a three-dimensional square parallelepiped of i nfnite length
oriented parallel to the time axis in space-time, or to somebody's
time axis.
If the previous objections are now overcome, two new ones
take their place. One of them has to do with the inelegance of a
tandem ontology: matter and space. The other, more serious, has
to do with the gratuitousness of a doctrine of absolute position.
Without absolute position, spatial or spatiotemporal, an on
tology of purely spatial or spatiotemporal mani folds seems in
coherent.
These two objections dri ve us back to a renewed effort to con
strue the manifolds i n attachment somehow to physical objects.
This attachment may be contrived by means of numbers and
measurement; and such is the fmi liar enlightened approach to
the matter. A point i s identifed with a triple of real numbers, or,
for spatiotemporal purposes, a quadruple. Squares and other
manifolds are identifed with the appropriate classes of such
triples or quadruples, accordi ng to analytical geometry. Thus fr
there is no talk of physical space nor of physical objects. Fi nally
the connection with physical objects is made by applying pure
number through physical measurement. When we say e. g. that
four villages are so related to one another as to form the vertices
of a square, we are talking of the arithmetical relation of the dis
tance measurements of these villages. We are saying that the
pairs of measurements fom any arbitrary rectangular coor
dinates, one pair for each village, will meet the arithmetical con
ditions of the vertices of a square according to analytical
geometry. Ulti mately we are merely saying something about the
relation of the distances of the four villages from one another;
namely, that four of these six distances are equal to one another
and the other two are equal to each other. The excursi on through
coordinate systems is just a device for handling such relations
systematically.
We have got rid of the ontology of manifolds, but only to fnd
ourselves dealing with a good deal more than physical objects.
Here are the numbers and their pairs and triples and quadruples
and the ch:sses of such. We have got rid of the ontology of
manifolds only to take on an ontology of abstract objects. What
with these and the physical objects, we have a tandem ontology
as before.
It must be said in extenuation that we would have found
ourselves needing this ontology of abstract objects anyway for
many purposes. We would have broached it even while retaining
the ontology of manifolds, as soon as we began talking of shapes:
of square, circle, etc. For the mani flds were just individual
squares and circles and the like; the shapes would be classes of
them, and thus objects of a higher abstractness.
35 . Ontological economy
We have been seeing ontology as expansi onist. Set theory in its
higher reaches adds exorbitantly to the population of the uni
verse, and it does so purely on ontological principle: by ex
trapolating single-mindedly and inexorably ad infnit um some
guidelines that are hinted at their hither ends by the more modest
and fltering ontology of common sense and natural science.
Even at the level of physical objects, moreover, ontological self
consciousness has tended rather to add entities than to weed
them out, i f we may judge from our present survey. Surely our
sophisticated universe of physical objects is denser than the naive
universe of bodies. Entities are being multiplied.
Occam' s stricture on such multiplication, however, is but part
of a broader counsel of economy or simplicity. The generalizing
1 35
of bodies to physical objects brought an important gain in
simplicity of organization, by regimenting our mass terms and
Heraclitus' s talk of time and the river under a clean-cut pattern
of general and singular terms and objective reference. Again the
acceptance of numbers and low-type classes had, among other
simplifcatory effects, that of clarifying our talk of spatial
manifolds and organizing its connecti ons with our talk of bodies.
And even the excesses of higher set theory come of a drive for
simplicity, afer all: a distaste for di scontinuities of principle. But
in the case of higher set theory one yearns for other useful and
simple principles without all that fecundity.
Ontology is not always expansionist. It has its contracti onist
side, though we have not had occasion in these pages to observe
it . By becoming clear on the nature and requirements of
reference we fnd that some seeming entities are better bypassed
by paraphrase, in the interest of not only ontological economy
but simplicity of theory generally. An example is impure
number, or units of measure. Ordinarily we talk of these as of
objects, by names and general terms; but they are better analyzed
in an idiom that relates pure numbers di rectly to physical objects
in various ways . Again there is the whole realm of intensions, at
tributes, propositional meanings, unactualized possibles. As we
become clearer and more explicit in ontological matters we come
to appreciate the urgency of individuati on principles, which are
weak or wanting in the case of intensions. Also we become aware
of some subtleties having to do with refrential opacity, which I
shall not pause over here. The result can be that we fnd he inten
sional entities to be less help than hindrance, less simplifcatory
than complicatory, and out they go.
The vernacular use of the refrential apparatus is indeed
careless and prodigal of objects, i f we read it in a literal-minded
ontological way. 'There is one thing about him that I don't like';
'He and Elizabeth have so many interests in common' . How
many thi ngs are there about hi m altogether, liked and disliked?
How many i nterests has he altogether, how many has Elizabeth,
and how many are in co1mon? We use those idioms without
countenancing such questions, and the questions may be blamed
as j ustly on an emergent literal-mindedness of ontology as on an
abortively ontol ogical vernacular. But when ontol ogy steps
forward to take these matters systematically i n hand, the effect is
apt to be rather contractionist than otherwise.
Talk of ontological economy or extravagance makes sense
only within a prescribed framework of allowable grammatical
constructions. It would be meani ngless to dismiss vast categories
of objects by invoki ng novel l i nguistic forms and constructions
and declaring these to be i nnocent of refrential intent. It would
be meaningless for want of a standard of refrential i ntent. We
give content to the ontological issue when we regiment the
language of science strictly within the famework of the l ogic of
truth functions and objectual quanti fcation. In so doing we l i mit
the resources of other than ontological kind, and are thus enabled
firly to assess and compare any costs or savi ngs i n ontology
from theory to theory. I remarked early in 29, for i nstance, that
substitutional quanti fcati on, fr from being ontologically i nno
cent, is si mply ontologically inscrutable except through some
stated translation i nto this objectual idiom.
Such regimentation accounts for the ontological expansion
lately noticed. Some of the burden came to be reckoned to on
tology that had previously been accommodated in incommen
surate ways through miscel laneous constructions .
Mostly i n this book I have speculated on causes, not
justi fcations. I have asked how our ontol ogical notions are
possible, not why they are right. Even in the case of bodi es, those
prototypical objects of refrence, I offered no hope of justifica
tion. I entertained no thought of translating talk of bodies into
talk of sense impressions, as Russell and Carnap dreamed of do
ing a half century ago. I asked how, given our stimulations, we
might have devel oped our corporeal style of tal k. And I asked
how we developed our abstract, set-theoretic style of tal k. One
could ask, i n the same spirit, how we developed our religious
tal k, and our talk of witchcraft, and our talk of analyticity and
l ogical modalities. If we managed to reconstruct these causal
chains of language learning, we would fnd that every here and
there the learner had made a little leap on the strength of analogy
or conjecture or confusion; but then the same seemed to be true
of our learning to talk of bodies. I n short, I speculated on causes
1 37
and not on val ues . Sheep are caused and goats are caused, and
they are caused in similar ways.
In the past two sections I have been concerned no longer with
the subj ect's learning of the refrential apparatus, but with his
deliberate ontologizi ng. But even here I have been concerned
more with the nature and meaning of what he is doing than with
what he or we ought to do. How then should we settle our on
tol ogy?
36. Relative empiricism
That last questi on is little less than the general question of
scienti fc method: the question how best to develop an incl usive
scientifc theory. We want to maximize prediction; that is, we
want a theory that will anticipate as many observations as possi
ble, getting none of them wrong. We develop the theory by
progressive observation and correction. When we have to modify
the theory to accommodate a wayward observation sentence, we
have various possible corrections from which to choose; and here
the guiding considerations are simplicity and conservatism. We
prefr the correction that makes for a simpler theory, by our sub
jective standards of simplicity, unless the other alternative is
more conservative, that is, a less drastic departure from the old
theory. But a big simpli fcation can warrant a firly drastic
departure. We arbitrate between these two interests, simplicity
and conservati on. They are related dialectically, to borrow a
term from my students.
In some measure, conservatism is imposed by our poverty of
imagination rather than freely chosen. But it can al so be a
deliberate matter of prudence, a matter of adhering to the tried
and true where we can instead of betting on wild hypotheses. As
we go on modifying theory to accommodate observati on, the
consideration of simplicity of theory may indeed so fr outweigh
conservatism that we give up our old belief in witchcraft; perhaps
also religion; perhaps modal l ogic; but there are limits .
These two forces, the force for simpli fcation and the force for
conservatism, are already at work in a primitive way in our
learning of language, if my causal speculations have been near
right. I have pictured the process as a series of short leaps, each
made on the strength of si mi l arities or analogies . In this pursuit
of si mi larities or analogies we see the force for si mpl i fcation,
and i n the shortness of the steps we see the force for conservation.
Learni ng language, learning gradually to quantify over bodies
and eventually over abstract objects, is one phase of a continuing
process that goes on to embrace also the learning and even the
further devel oping of high scienti fc theory. We are worki ng up
our science from i nfncy onward. Each of the leaps of language
l earning that I have pictured is a private little scienti fc revolu
tion, another step i n the development of a system of the world. I f
the leap is one that conduces to simplicity i n the child's evolving
conceptual scheme, then normatively speaking it is good scien
tifc method on his part, however unconscious. If it is a short
leap, then again it is good, on the score of conservatism.
I have called short l eaps conservative. It i s more illuminating
to cal l them empiricistic. They are governed by this maxim of
relative empiricism: Don't venture frther from sensory evidence
than you need to. We abandoned radical empiricism when we
abandoned the old hope of translating corporeal talk i nto sensory
tal k; but the relative variety still recommends itself. We
recognize that between the gl obal l y l earned observati on
sentences and the recognizably articulate talk of bodies there are
irreducible leaps, but we can still be glad to mini mize them, and
to mini mize such further l eaps as may be required for further
reaches of ontology. The maxim has evident practical value, i n
mini mizing our l iabil ity to backtracking when need arises for a
change of theory.
The maxim would have us try to preserve the substitutional i n
terpretati on of quanti fcation over abstract objects, i f I have been
right i n supposing that this was genetically the prior interpreta
ti on. Considerations of overall simpl icity of theory could out
weigh this consideration and sustain the objectual i nterpretation,
but at any rate there should be a deliberate weighing of con
siderations .
The substitutional interpretation of quanti fcation over
abstract objects appeals to the nominal ist temper. This appeal is
1 39
a mani festati on of relative emp1 nc1sm, for the urge to
nominalism is itsel f just that . But the wishful nominali st must
beware, we saw, of overesti mating the ontol ogical i nnocence of
substituti onal variables. In general the val ues of the objectual
variables of a theory may firly be said to exhaust the ontol ogy
only i f the theory is couched in terms sol ely of predicates, truth
functions, and objectual quanti fcati on. I f there are i rreducible
additional devices-modal operators, say, or substituti onal
quanti fcati on-then there is no assessing its ontology except
relative to some stated translati on.
Very wel l , the nominal ist may reply, let us grant that sub
stitutional quanti fcati on does not make a cl ean ontol ogical
sweep; still something is gained. Substitutional quanti fcation
over numbers, for instance, gets explained in terms of ex
pressions and substitution rather than in terms of abstract ob
jects and refrence. But here again we must di sappoint him; for
we refected in 30 that the expressions needed as substituends
are entities as abstract as the numbers.
Our thwarted nominalist may still cleave t o substitutional
quanti fcation where he can, as a way of tempering ontological
excesses . Nomi nalism aside, he might hope by substitutional
quanti fcation to reduce the extravagant ontology of real
numbers or of set theory to that of elementary number theory by
framing truth conditions for substituti onal quantifcation in
terms of Godel numbers. But a reductive program of this sort
should be seen as a project l ess of nominalism than of
Pythagoreani sm. It is a matter not of prizing the concrete and
abhorring the abstract, but of tolerating natural numbers and
abhorring most of the transcendentals. It is as i f to say, with
Kronecker, "God made the natural numbers; the rest are human
handiwork. "
However, this more modest objective is obstructed too.
Substitutional quantifcation over classes was found i n 28 to
confict with objectual quantifcation over physical objects-or
to confict, anyway, with whatever good reasons there might be
for adhering to objectual quantifcation over physical objects.
This threat to the substitutional interpretation of cl ass quan
tifcation could be met i f we could see our way to interpreting our
quanti fcation over physical objects substitutionally too. Our
reason against this was the namelessness of most rabbits, all
grains of sand, all electrons. But are these really nameless? Every
physical object is speci fable with help of spati otemporal coor
di nates, and so can be named by a singular description. This
desperate resort was too frftched to be i nteresting as l ong as we
were speculating on the psychological origins, but does it bear
consideration now that we are ontol ogizing on our own?
I thi nk not, sti l l . Consider the motivation. We want to inter
pret our quanti fcation over physical objects substitutionally i n
order to remove the obstacle to substitutional quanti fcation over
classes. And why do we want substitutional quantification over
classes? The motive was quasi-nominal istic, and ultimately a
matter of relative empiricism. But if relative empiricism speaks
for substitutional quantifcati on over classes, it speaks also for
objectual quantifcation over physical objects-these being the
versions that are cl osest to the respective genetic origins, if my
genetic speculations have been right . And anyway the device of
numerical coordinates seems a poor way of buttressing sub
stituti onal quantification over physical objects, when we recall
the quandaries of substitutional quanti fcati on over numbers .
Charles Parsons has proposed for quanti fcati on over classes a
semisubstitutional compromise between the substituti onal and
the objectual account. For the truth of an existential quanti fca
tion in this sense it is no l onger required that there be a true sub
stituti on instance; there need only be a substitution instance that
contains free objectual variables and is satisfied by some values
of them. Correspondingly a universal quanti fcation, even i f
devoid of free variables, no l onger claims merely the truth of all
substitution instances devoid of free variables; it requires further
that all substitution instances contai ning free objectual variables
be satisfed by all values of them. This version of class quanti fca
tion restores the law of unit subclasses and the usual permutabili
ty of quanti fers, and it resolves the related anomaly of the
vacuous clause (28). Its truth condition lacks the charm of the
transparent truth condition of strictly substitutional quantifica
tion, and of course it suffers equally in the presence of im-
predicative abstracts. Still it does retain something of the desired
nomi nalistic aura that i s wholly lacki ng in the objectual version,
and I see no reason to doubt that it meets the needs of set theory.
references
T. G. R. Bower, "The object i n the world of the i nfnt," Scien
ti fc American 225, no. 4 (Oct. , 1 97 1 ); pp. 30-38.
Georg Cantor, Gesammelte Abhandlungen mathematischen und
philosophischen Inhalts, E. Zermelo, ed. , Berl i n, 1 932.
Rudolf Carnap, Der logische A ujau der Welt, Berl i n, 1 928.
Translati on, The Logical Structure of the World, University
of
-
California Press, 1 96 7.
~ , "Testability and meaning," Phil osophy of Science 3
( 1 936), pp. 4 1 9-47 1 ; 4 ( 1 937), pp. 1 -40.
~ g Foundations of Logic and Mathematics, University of
Chicago Press, 1 939.
Noam Chomsky, "Quine' s empirical assumptions," i n Davidson
and Hi ntikka, pp. 53-68.
1 44
Al onzo Church, Mathematical Logic, mimeographed, Princeton
University Mathematics Department, 1 936.
Charles Darwin, The Origin of Species, 1 859; fcsi mile, Harvard
University Press, 1 96.
Donald Davidson, "Truth and meaning," Synthese 1 7 ( 1 967), pp.
304-323 .
_and Jaakko Hi nti kka, eds. , Words and Objections:
Essays on the Work of W. V. Quine, Dordrecht : Reidel,
1 969.
A. S. Esenine-Votine, "Le programme ultra-i ntuitionniste des
fondements des mathematiques, " Jnfnitistic Met hods
(Proceedi ngs of Warsaw symposi um, 1 959), Oxford:
Pergamon, 1 96 1 , pp. 20 1 -223.
Peter T. Geach, Reference and Generality, Ithaca: Cornell, 1 962,
1 968.
_, "Quine's syntactical i nsights," in Davidson and Hi n
ti kka, pp. 1 46- 1 57.
_, "Ontological relativity and relative identity," i n Mi lton
Munitz, ed . , Ontology, New York Universi ty Press,
forthcomi ng.
Nelson Goodman, Fact, Fiction and Forecast, 2nd ed. , In
dianapolis: Bobbs- Merri l l , 1 965.
__and W. V. Qui ne, "Steps toward a const ructive
nomi nalism, " Journal of Symbolic Logic 1 2 ( 1 94 7), pp. 1 05-
1 22.
Morris Halle, "Phonology i n generative grammar, " Word 1 8
( 1 962), pp. 54-72.
Richard J. Herrnstei n, "Superstition, " in W. K. Honig, ed. ,
Operant Behavior, New York: Century-Crofts, 1 966.
Edwin B. Holt, Animal Drive and the Learning Process, New
York: Holt, 1 93 1 .
eferences 1 45
E. H. Lenneberg and J . M. Roberts, "The language of ex
perience," Internati onal Journal of American Linguistics,
supplement, 1 956.
Ruth B. Marcus, "Modalities and i ntensi onal languages,"
Synthese 1 3 ( 1 96 1 ), pp. 303-322.
D. A. Marti n, review of Quine's Set Theory and Its Logic, Jour
nal of Philosophy 67 ( 1 970), pp. 1 1 1 - 1 1 4.
Charles Parsons, "A plea for substituti onal quanti fcation,"
Journal of Phil osophy 68 ( 1 97 1 ), pp. 23 1 -237.
I. P. Pavlov, Conditioned Refexes, Oxford University Press,
1 927.
C. S. Peirce, Collected Papers, Cambridge: Harvard, 1 93 1 - 1 958 .
Jean Piaget and Barbel Inhelder, La Genese des Structures
Logi
q
ues Elementaires, Geneva: Delachaux, 1 959.
W. V. Quine, Methods of Logic, 3rd ed. , New York: Holt, 1 972.
_, Word and Object, Cambridge: M. I. T. , 1 960.
_, Set Theory and Its Logic, Cambridge: Harvard, 1 963,
1 969.
__ , Philosophy of Logic, Englewood: Prentice-Hal l , 1 970.
__, " Reply to D. A. Marti n, " Journal of Phil osophy 67
( 1 970), pp. 247f.
Gi lbert Ryle, The Concept of Mind, London: Hutchinson, 1 949.
Wilfrid Sellars, "Abstract entities," Review of Metaphysics 1 6
( 1 963), pp. 627-67 1 .
-, "Cl asses as abstract entities and the Russell paradox, "
ibid. 1 7 ( 1 96), pp. 67-90.
Nathan Stemmer, "Some aspects of language acquisition," i n Y.
Bar- Hi l l el, ed. , Pragmatics of Natural Languages, Dor
drecht: Reidel, 1 97 1 .
Leonard Troland, Fundamentals of Human Motivation, New
York: Van Nostrand, 1 928.
Ludwig Wittgenstei n, Philosophical Investigations, Oxford:
Bl ackwel l , 1 953.
J . Z. Young, An Introduction to the Study of Man, Oxford:
Clarendon, 1 97 1 .
Paul Ziff, "A response to ' stimulus meaning' , " Philosophical
Review 79 ( 1 970), pp. 63-74.
Abstention, 77
Abst ract objects, 8 1 , 87, 1 02, 1 26; quan
ti fying over 10 I f, I 25f, 1 38
Abstract terms, 86, 87f, 1 0 I f, 1 05f, 1 20,
1 25
Abstract ion of classes, 1 05f, 1 20
Act ion, 32
Aesthetics, 49- 5 1
Aggregat ion, 52
Alternation, 77f, 80
Ambiguity, 86
Analogy: i n sentence structure, 42, 86-88;
in terms, 85-88, 1 05; in relative clauses,
9 l f, 94f, 1 0; i n quanti fcation, 1 05, 1 25
Analyticity I 2f, 78-80
Apocopation, 69, 7 1
Assent, 45-49, 63f, 77
Att ributes, I OI , 1 0f, 1 1 1 , 1 25; versus
classes, 1 9, 1 01 - 1 03, 1 25; identity, 88,
1 01 - 1 02, 1 35
Attributive, 59f, 76
Awareness, I f, 3f
i ndex
Babbling, 29, 3 1
Behavior, 4, 20f, 25f
Behavioral si milarity, 2 1 - 23
BERKELEY, George, I f, 34
BERNAYS, Paul, 1 22
BLOOMFI ELD, Leonard, 91
Bodies, I f, 52, 88, I 36f; uni fcation of, 52-
55, 85; and physical objects, 54, I 34f; in
dividuation of, 56, 69f, 85
BOWER, T. G. R. , 54n
Boyle's law, 1 30
BURAU- FORTI , Cesare, 1 02
Caeteris paribu. 1 2, 22
CANTOR, Georg, 1 02f, I I 3f
CARNAP, Rudolf, 1 6, 78, 80, 1 3 1 , 1 36;
on dispositions, 9f, I 2f, 1 4
Categoricals, 97f, I Of, 1 0f; universal, 66,
67f, 79f, 93-96
Categories, grammatical , 98f, 1 03f
Cause, 4-8
CHATEAUBRI AND, Oswaldo, 1 08n
1 48
CHOMSKY, Noam, 1 2n, 1 5, 91
CHURCH, Alonzo, 96n
Ci rcularity, 2, 1 4
Classes: versus attributes, 1 9, 1 01 - 1 03,
1 25; simulated, I 0f, 1 08, I 39f; abst rac
tion of, 1 05f, 1 20; hypostatized, 1 1 0-
1 1 2, I 34f, 1 40; of classes, 1 20- 1 22, I 26f;
ultimate, 1 22
Closed, 1 06f, 1 09
Color, 43f, 70-73; ontology of, 85-87, 1 32
Commutation of quant ifers, 1 08- 1 1 0, 1 20,
1 26
Compounds, 59f, 75
Compreh
e
nsion, 1 06- 1 08
Concrete terms, 87f, 1 0 I f, 1 05f, 1 25
Concretion, 1 06
Conditional, 8- 1 0
Conditioning, 3f, 1 6f, 27f, 75f; of refexes,
1 3, 20, 22; and similarity, 1 7-20, 22;
t ransfer of, 20, 65; of sentences, 30, 41 f
Confsion, 68, 1 02, 1 25
Conjuncti on: logical, 76- 78; grammatical,
90-92
Conservatism, 1 37f
Coordinates, 1 1 0, 1 33f, I 39f
DARWI N, Charles, 24n. See also Evolution
DAVI DSON, Donald, 65
Demonstratives, 55f, 59
Denumerability, 1 1 3- 1 1 5
Depth perception, I f
Derel ativizat ion, 56-58
Dialectic, 1 3 7
Disposition, 4f, 7- 1 0, 1 2; innate, 4f, 1 2- 1 4;
as physical, 1 0- 1 5; as heuristic, 1 0- 1 5; as
statistical, 1 2, 1 5; fr historian, I 3f;
second-order, 1 7, 2 1
Dissent, 45, 48f, 63f, 75f, 77
Drive, 28f
EDDI NGTON, A. S. 3
Empi ricism, 1 37-40
Energy, 5- 1 1
Episodes, I 6f, 2 1 , 24-28
Epistemology, 1 -4, 20, 34, 3 7
Epsilon, 1-6
Equivalential term, 57
ESSENI N-VOLPI N, A. S. , 1 1 9
Index
Eternal sentences, 63f, 68, 1 29; learning of,
64-67, 79
Ethics, 49-52
Events, 5f, 1 3 1
Evidence, 2-4. 37-4 1
Evolution, I 9f, 22f, 5 1 , 54, 68
Expectation, I 29f
Fluency, 39, 83f
Fourier analysis, 23
FRAENKEL, A. A. , 1 02
FREGE, Gottlob, 58, 78
FREU D, Sigmund, 26
Future, I 29f
GEACH, Peter T.: on ident ity, 57, 59, I 1 6n;
on relative clauses, 89-93; and Kaplan,
I l l , 1 2 1 , 1 25
General t erms, 55- 5 7, I OOf; versus
singular, 84-87, 96f, l Ol f, 1 05f; and sub
stitutional variables, 1 03; abstract, 1 20
Gestalt, 1 -4, 1 6; body as, 53-55
GI LBERT, W. S. , 50n
GOdel numbers, 1 1 9, 1 39
Good, 49-5 1
GODMAN, Nelson, 1 4, I 9n, 1 1 9
Greek letters, 61 n
GRELLI NG, Kurt, 1 1 3
Habit, 4, 1 8
HALLE, Morris, 9 1
HARMAN, Gilbert, 1 08n
Hedonism, 32
HERACLI TUS, 1 35
HERRNSTEI N, Richard J . , 3 1 n
HoL T, Edwin B. , 1 3, 3 1 n, 32
Homology, 23f, 4 1
HUME, David, I , 20, 58f, 1 28; on cause, 5f;
on ideas, 34, 36
Ideas, 34-37
Identity, 57-59, 88, 1 1 5- 1 7
I l lusion, 2f, 26f
I mages, 26f, 33f, 59f, 64f; chromatic, 72f;
as memory, 1 27- 1 29; as expectation, 1 29
I mitation, 3 1
I mpredicativity, 58, 1 1 2, 1 2 1 f, 1 26, I 40f
I n, 61 f
Index
I ndenumerability, 1 1 3- 1 5
I ndicator words, 68
I ndividuation, 55, 57-59, 68- 70, 84-86;
want of, 87f, l Ol f, I 24f,. 1 35
I nduction, I 9f, 32f, 42-44, I 29f
I nhibition, 48f
I nnateness, see I nstinct
I nner states, 20-23, 3 1 , 43
I nstinct, 4f, I 3f, 26; of similarity, I 8f, 22f;
for perceiving bodies, 54f
I nterference, I nternal states, 20-23, 3 1 , 43
I nterrogatives, 1 24
I ntuitionism, 78-80
I nward sense, 48, 66f
KANT, I mmanuel, 36
KAPLAN, Davis, 1 1 1 , 1 2 1 , 1 25
Kelly-Morse system, 1 22
KRI PKE, Saul, I I 3n
KuNo, Susumu, 9
Language, essence of, 59f, 68, 79f, 86
Language-dependent learning, 48, 56, 60,
62, 1 1 5
Lat in prose theory, 90-92
Learning, 4, 1 9, 24f, 28-3 1 ; of language,
35-37; language-dependent, 48, 56, 60,
62, 1 1 5
LENNEBERG, Erk, 43n
LESNI EWSK I , Stanislaw, 99
LEWIS, David K. , 44n
Li ke, 62
LOKE, John, 36, 58f
Manifolds, 1 32- 1 34
MARCUS, Ruth B. 99
MARTI N, D. A. , 1 02n
Mass terms, 52f; likened to singular, 52f,
85, 87, 89, 1 3 1 f; relative, 56-59
Meaning, 38, 6, 78f, 84, 1 35; postulates,
9f
Measurement, 1 30, 1 33- 1 35
Membershi p. 1 05f
Memory, 20f, 24f, 3 1 f, 1 27- 1 29; racial, 26
Mental ism: uses of, 2 1 , 25-27. 33f, 48, 66f;
evils of, 34-37, 6
Mnemon, 28n
Modal logic, 99, 1 36, 1 39
Models, 1 I 4f
Morals, 50- 52
Muscular skill, 3 1
Namelessness, 99, 1 03, 1 1 0, 1 40
Names, 55, 99, 1 05, I I 3f, I 23f
Natural hi story, 1 30
1 49
Natural select ion, I 9f, 22f, 5 1 , 54, 68
Negation, 75f
Neighborhoods, I 7f
Nomi nal ism, 99, 1 03, I 38f
Numbers, 1 1 3, I 1 4f; quantifying over,
I 1 4f, 1 1 7-20, 1 26, 1 39; learning of, 1 1 6-
1 8, 1 26; real, I 1 8f, 1 39; in measurement,
1 30, 1 33-35
Observation, 37-39, 40f; terms, 52f, 6-62,
85
Observat ion sentences, 39-43, 79; terms as,
52f, 60-62, 85; and scientifc theory, 6,
1 29, 1 36, 1 38f
OCCAM, Wi lliam of, 1 34
Occasion sentences, 39f, 63f, 67, 68; as
terms, 60-62
Ontology, 54, 81 f, 84, 85-87; deliberate,
88f, 1 3 1 , 1 34-37; by quant ifcation, 98-
1 0 I , 1 1 1 f; tandem, 1 33; under relative
empi ricism, I 38f
Ostension, 44f, 52f; sustained, 53, 55f, 57,
5 8f, 66, 1 1 5 ; pai red, 5 5 - 5 8 , 86
accelerated, 60f, 68, 7 1 , 74f
Paradox: of subst itution, 95-97, 1 1 0; in set
theory, 95f, 1 02f, 1 06, 1 22; semantic,
1 I 3f
PARSONS, Charles, 1 06f, 1 40
Particles, 1 30
Past, 1 27- 1 29
PAVLOV, I van P. , 20
PEI RCE, Charles Sanders, 34, 49n
Percept ion, 4, 1 6, 23
Perceptual relevance, 22f
Perceptual si mil arity, 1 6- 1 8, 24f, 27f, 31 f;
polyadic, I 8f, 29; innate, 1 9, 22f; mask
ed, 20f, 3 1 ; changing, 20-22; uniformity
of, 22f, 42f. See also Si milarity
Physical objects, 54, 89, 1 3 1 f, I 34f; quan
tifying over, 99, 1 08- 1 1 0, I 39f
1 50
PI AGET, Jean, 67n
Platonism, 87
Pleasure, 27-29, 31 f, 4 7f, 50f
Pleonasm, 55f, 76
Pointing, see Ostension
POWERS, Lawrence, 67n
Predication, 6 1 , 6 7, 68, 84, 1 05f; learning
of, 64-67, 70, 85
Predicative, 6 1 . See also I mpredicat ivity
Predict ion, I 29f, 1 37
Probability, 1 2
Productivity, 59f, 66, 68, 74f, 79
Pronouns, 1 0; demonst rative, 55f, 58f;
relative, 89-92, 99, 1 24; interrogative,
1 24
Properties, see Attri butes
Propositions, 36
Protocol, 1 29
Pythagoreanism, 1 39
Quantifcation 91 , 92, 97, 1 24; origin i n
cat ego r i cal , 97f, I 04 , 1 2 3 ; and
namelessness, 98f, 1 03, 1 08- 1 1 0, 1 4; i n
ontology, 98f, I 0, 1 1 1 f; over classes,
1 05- 1 1 1 , I I 2f; commuted, 1 08f, 1 20,
1 26. See also Substitutional quantifca
ti on
Quantum mechanics, 6f
Quasi quotat ion, 61 n
Reception, Receptors, 3f, 23f
Receptual similarity, I 6f, 22-24, 4 1
Reduction forms, 9f, I 2f
Reference, 81 -84, 89, 92; apparatus of, 84f,
88, 89, IOf, 1 35; schematized, I Of;
learning of, summarized, 1 23- 1 26
Refexes, I 3f, 20, 22
Relative clauses, 89-93, 1 20; util ity of, 93-
95; how learned, 93-95, 97, I 0, I 23f
Relative terms, 56-59, 66
Respects, 1 9
RICHARD, Jules, 1 1 4
ROBERTS, J. M. , 43n
RUSS ELL, Bertrand, 99, 1 07' 1 2 1 , 1 27. 1 36;
his paradox, 95f, 1 1 0
RYLE, Gi l bert 9
Salience, 25-28; in learning language, 26f,
43-45, 53, 59f, 69
Index
Satisfaction, 1 06- 1 1 0
Scientifc theory, 63f, 1 29, 1 30; data of, 2-
4, 37-39, 40f; acquisition of, 2f, 34f, 46,
67f, 81 f, 1 30; and values, 49, 5 1 f; goals
and norms of, 88, 1 30, 1 3 1 , I 37f
Second order, 55f, 72-74
SELLARS, Wi l frid, 1 06
Sensation, 1 -3
Sentences: standing, 63, 65-68; singular,
67; compound, 75; open, closed, 1 06f.
See also Eternal sentences, Observation
sentences, Occasion sentences
Set theories, 1 02f, I 22f, 1 25, 1 3 1 , 1 40f. See
also Classes
Shapes, 73-75, 86; abstractness of, 86-88,
1 32, 1 34; variously construed, I I 8f, 1 32-
1 34
Similarity: recept ual, I 6f, 2 1 , 22f, 3 1 ; ob
jective, I 8f; behavioral, 2 1 , 22f; basis,
43; language-dependent, 48, 56, 60, 62;
second-order, 55f, 72-74; chromatic, 72;
morph i c, 74. See also Pe rcept ual
similarity
Simpl icity, 1 34f, 1 37, 1 40
Singular sentences, 67
Singular terms, 84f, 96f, 1 0; mass terms
l i kened to, 52f, 85, 87, 89, 1 3 1 f; abstract,
86-88, 10 I f, 1 05f, 1 20, 1 25
Skept icism, 2f
Skolem- Lowenheim theorem, 1 1 4
Society, 22-24, 39, 4 1 , 79f
Solubility, 8- 1 0, 1 1 , 1 2f
SONNERAT, Pierre, 4
Space, 1 32- 1 35
Spanish, 67
Speech community, 39, 83f
Stages, 1 3 1
Standing sentences, 63, 65-68
STEMMER, Nathan, 20
STICH, Stephen, 5, 1 3
Sti mulation, Stimulus, 25f, 27f, 30f, 47f;
common to two subjects, 23f, 41
Substitution: in relative clause, 89, 94-97,
I 23f; for variables, 94-97, 99, 1 05f
Substitutional quanti fcation: truth con
ditions of, 98f, 1 03f, 1 08- 1 1 0; over
physical objects, 99, 1 08, 1 1 0, I 39f; for
general terms, 1 03- 1 05, I 25f; over
classes, 1 05- 1 08, 1 1 0; in confict with
Index
object ual , I 08- I I O, 1 26, I 39f; and on
t ol ogy, 1 1 1 , J 35f, I 38f; and i m
predicativity, I I f; and Lowenhei m, 1 1 4;
over numbers, 1 1 8- 1 20, 1 26
Successor, 1 1 9, 1 26
Such that, 92-94, I 23f; and subst itution,
94-97; and quanti fcation, 97-99, 1 0f;
as class abstraction, I 05f
TALBOT, Wi lliam, I 08n
Taste, 49f
Terms: observation, 52f, 60-62; relative,
56-59; abstract, 86-88, I OI f, I 05f, 1 20,
1 25; concrete, 88, I OI f, I 06, 1 25. See
also General t erms, Mas s t erms,
Si ngular terms
Theory, see Scientifc theory
THORNDI KE, E. L. , 32
Ti nt, 72, 73
TI TCHENER, Edward B. , 34
Traces, 24f, 29; enli vened, 26-28, 60, 64f
Transfer, 65f
Transformation, 92, 93-95, 99, I 06, 1 23f
Translation, 36, 82-84
TROLAND, Leonard, 32
Truth conditions, 65f, 78f, 98, I 07f, 1 1 2
Truth functions, 75-79
Types, I I O, 1 20- 1 22, I 26f, 1 40
Uni formity, social, 22f, 39, 79f, 83f
Unit subclasses, I 08- l I l , I 20f, 1 26
Universals, 81 f, 84, 86
Use and mention, 68, 99, I Ol f
Value, 49; of a variable, 98f
1 5 1
Variabtes: as pronouns, 89f, 92f, I Of, 1 23,
1 24; i n ont ol ogy, 89f, I OOf; sub
stitutional or objectual, 98- 1 0, I 03,
1 0, 1 24f; for general terms, I 03, I 05,
I 25f. See also Quant i fication and
Substitutional quantifcation
Verdicts, 76-79
Vice, 95f
Vienna Ci rcle, 1 29
Vi rtue, 49- 52; by of, Sf
VON NEUMANN, John, 1 22
WHI TEHEAD, Alfred North, 99
WI TTGENSTEI N, Ludwig, 4, 58
YOUNG, J. Z., 28n
ZERMELO, Enst, 1 22
ZI FF, Paul, 1 5

Quine-The Roots of Reference (1990) Cs

Uploaded by

Copyright:

Available Formats

Quine-The Roots of Reference (1990) Cs

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quine-The Roots of Reference (1990) Cs

Uploaded by

Copyright:

Available Formats

the paul carus lectures

the roots of reference

ng than with any other.

3. Herrnstein' s theory of superstition fts in here.

shape' thus harmonizes with that of

' and other shape words afford thus an enteri ng wedge

will assent also to r / is

fr arbitrary i. He i s not using quantification over

can in effect be any sentence containing a.

roots in the universal categorical and was thus i mplicitly ob

You might also like